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

Comparing cvsroot/libeio/eio.c (file contents):
Revision 1.52 by root, Fri Jan 8 09:29:56 2010 UTC vs.
Revision 1.95 by root, Tue Jul 19 04:56:43 2011 UTC

1/* 1/*
2 * libeio implementation 2 * libeio implementation
3 * 3 *
4 * Copyright (c) 2007,2008,2009,2010 Marc Alexander Lehmann <libeio@schmorp.de> 4 * Copyright (c) 2007,2008,2009,2010,2011 Marc Alexander Lehmann <libeio@schmorp.de>
5 * All rights reserved. 5 * All rights reserved.
6 * 6 *
7 * Redistribution and use in source and binary forms, with or without modifica- 7 * Redistribution and use in source and binary forms, with or without modifica-
8 * tion, are permitted provided that the following conditions are met: 8 * tion, are permitted provided that the following conditions are met:
9 * 9 *
… …
35 * and other provisions required by the GPL. If you do not delete the 35 * and other provisions required by the GPL. If you do not delete the
36 * provisions above, a recipient may use your version of this file under 36 * provisions above, a recipient may use your version of this file under
37 * either the BSD or the GPL. 37 * either the BSD or the GPL.
38 */ 38 */
39 39
40#ifndef _WIN32
41# include "config.h"
42#endif
43
40#include "eio.h" 44#include "eio.h"
45#include "ecb.h"
41 46
42#ifdef EIO_STACKSIZE 47#ifdef EIO_STACKSIZE
43# define XTHREAD_STACKSIZE EIO_STACKSIZE 48# define XTHREAD_STACKSIZE EIO_STACKSIZE
44#endif 49#endif
45#include "xthread.h" 50#include "xthread.h"
… …
49#include <stdlib.h> 54#include <stdlib.h>
50#include <string.h> 55#include <string.h>
51#include <errno.h> 56#include <errno.h>
52#include <sys/types.h> 57#include <sys/types.h>
53#include <sys/stat.h> 58#include <sys/stat.h>
54#include <sys/statvfs.h>
55#include <limits.h> 59#include <limits.h>
56#include <fcntl.h> 60#include <fcntl.h>
57#include <assert.h> 61#include <assert.h>
58 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
76#if !defined(ENOTSOCK) && defined(WSAENOTSOCK)
77# define ENOTSOCK WSAENOTSOCK
78#endif
79
80static void eio_destroy (eio_req *req);
81
59#ifndef EIO_FINISH 82#ifndef EIO_FINISH
60# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0 83# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
61#endif 84#endif
62 85
63#ifndef EIO_DESTROY 86#ifndef EIO_DESTROY
… …
66 89
67#ifndef EIO_FEED 90#ifndef EIO_FEED
68# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0) 91# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)
69#endif 92#endif
70 93
94#ifndef EIO_FD_TO_WIN32_HANDLE
95# define EIO_FD_TO_WIN32_HANDLE(fd) _get_osfhandle (fd)
96#endif
97#ifndef EIO_WIN32_HANDLE_TO_FD
98# define EIO_WIN32_HANDLE_TO_FD(handle) _open_osfhandle (handle, 0)
99#endif
100
101#define EIO_ERRNO(errval,retval) ((errno = errval), retval)
102
103#define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1)
104
71#ifdef _WIN32 105#ifdef _WIN32
72 106
73 /*doh*/ 107 #undef PAGESIZE
108 #define PAGESIZE 4096 /* GetSystemInfo? */
109
110 #ifdef EIO_STRUCT_STATI64
111 #define stat(path,buf) _stati64 (path,buf)
112 #define fstat(fd,buf) _fstati64 (fd,buf)
113 #endif
114 #define lstat(path,buf) stat (path,buf)
115 #define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
116 #define mkdir(path,mode) _mkdir (path)
117 #define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
118
119 #define chmod(path,mode) _chmod (path, mode)
120 #define dup(fd) _dup (fd)
121 #define dup2(fd1,fd2) _dup2 (fd1, fd2)
122
123 #define fchmod(fd,mode) EIO_ENOSYS ()
124 #define chown(path,uid,gid) EIO_ENOSYS ()
125 #define fchown(fd,uid,gid) EIO_ENOSYS ()
126 #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
127 #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
128 #define mknod(path,mode,dev) EIO_ENOSYS ()
129 #define sync() EIO_ENOSYS ()
130 #define readlink(path,buf,s) EIO_ENOSYS ()
131 #define statvfs(path,buf) EIO_ENOSYS ()
132 #define fstatvfs(fd,buf) EIO_ENOSYS ()
133
134 /* we could even stat and see if it exists */
135 static int
136 symlink (const char *old, const char *neu)
137 {
138 #if WINVER >= 0x0600
139 if (CreateSymbolicLink (neu, old, 1))
140 return 0;
141
142 if (CreateSymbolicLink (neu, old, 0))
143 return 0;
144 #endif
145
146 return EIO_ERRNO (ENOENT, -1);
147 }
148
149 /* POSIX API only */
150 #define CreateHardLink(neu,old,flags) 0
151 #define CreateSymbolicLink(neu,old,flags) 0
152
153 struct statvfs
154 {
155 int dummy;
156 };
157
158 #define DT_DIR EIO_DT_DIR
159 #define DT_REG EIO_DT_REG
160 #define D_NAME(entp) entp.cFileName
161 #define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
162
74#else 163#else
75 164
76# include "config.h"
77# include <sys/time.h> 165 #include <sys/time.h>
78# include <sys/select.h> 166 #include <sys/select.h>
79# include <sys/mman.h> 167 #include <sys/statvfs.h>
80# include <unistd.h> 168 #include <unistd.h>
81# include <utime.h> 169 #include <utime.h>
82# include <signal.h> 170 #include <signal.h>
83# include <dirent.h> 171 #include <dirent.h>
84 172
173 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
174 #include <sys/mman.h>
175 #endif
176
177 #define D_NAME(entp) entp->d_name
178
85/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */ 179 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
86# if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__ 180 #if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
87# define _DIRENT_HAVE_D_TYPE /* sigh */ 181 #define _DIRENT_HAVE_D_TYPE /* sigh */
88# define D_INO(de) (de)->d_fileno 182 #define D_INO(de) (de)->d_fileno
89# define D_NAMLEN(de) (de)->d_namlen 183 #define D_NAMLEN(de) (de)->d_namlen
90# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600 184 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
91# define D_INO(de) (de)->d_ino 185 #define D_INO(de) (de)->d_ino
92# endif 186 #endif
93 187
94#ifdef _D_EXACT_NAMLEN 188 #ifdef _D_EXACT_NAMLEN
95# undef D_NAMLEN 189 #undef D_NAMLEN
96# define D_NAMLEN(de) _D_EXACT_NAMLEN (de) 190 #define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
97#endif 191 #endif
98 192
99# ifdef _DIRENT_HAVE_D_TYPE 193 #ifdef _DIRENT_HAVE_D_TYPE
100# define D_TYPE(de) (de)->d_type 194 #define D_TYPE(de) (de)->d_type
101# endif 195 #endif
102 196
103# ifndef EIO_STRUCT_DIRENT 197 #ifndef EIO_STRUCT_DIRENT
104# define EIO_STRUCT_DIRENT struct dirent 198 #define EIO_STRUCT_DIRENT struct dirent
105# endif 199 #endif
106 200
107#endif 201#endif
108 202
109#if HAVE_SENDFILE 203#if HAVE_SENDFILE
110# if __linux 204# if __linux
… …
126#endif 220#endif
127#ifndef D_INO 221#ifndef D_INO
128# define D_INO(de) 0 222# define D_INO(de) 0
129#endif 223#endif
130#ifndef D_NAMLEN 224#ifndef D_NAMLEN
131# define D_NAMLEN(de) strlen ((de)->d_name) 225# define D_NAMLEN(entp) strlen (D_NAME (entp))
132#endif 226#endif
133
134/* number of seconds after which an idle threads exit */
135#define IDLE_TIMEOUT 10
136 227
137/* used for struct dirent, AIX doesn't provide it */ 228/* used for struct dirent, AIX doesn't provide it */
138#ifndef NAME_MAX 229#ifndef NAME_MAX
139# define NAME_MAX 4096 230# define NAME_MAX 4096
140#endif 231#endif
141 232
233/* used for readlink etc. */
234#ifndef PATH_MAX
235# define PATH_MAX 4096
236#endif
237
142/* buffer size for various temporary buffers */ 238/* buffer size for various temporary buffers */
143#define EIO_BUFSIZE 65536 239#define EIO_BUFSIZE 65536
144 240
145#define dBUF \ 241#define dBUF \
146 char *eio_buf; \
147 ETP_WORKER_LOCK (self); \
148 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 242 char *eio_buf = malloc (EIO_BUFSIZE); \
149 ETP_WORKER_UNLOCK (self); \
150 errno = ENOMEM; \ 243 errno = ENOMEM; \
151 if (!eio_buf) \ 244 if (!eio_buf) \
152 return -1; 245 return -1
246
247#define FUBd \
248 free (eio_buf)
153 249
154#define EIO_TICKS ((1000000 + 1023) >> 10) 250#define EIO_TICKS ((1000000 + 1023) >> 10)
155
156/*****************************************************************************/
157
158#if __GNUC__ >= 3
159# define expect(expr,value) __builtin_expect ((expr),(value))
160#else
161# define expect(expr,value) (expr)
162#endif
163
164#define expect_false(expr) expect ((expr) != 0, 0)
165#define expect_true(expr) expect ((expr) != 0, 1)
166
167/*****************************************************************************/
168 251
169#define ETP_PRI_MIN EIO_PRI_MIN 252#define ETP_PRI_MIN EIO_PRI_MIN
170#define ETP_PRI_MAX EIO_PRI_MAX 253#define ETP_PRI_MAX EIO_PRI_MAX
171 254
172struct etp_worker; 255struct etp_worker;
… …
176static int eio_finish (eio_req *req); 259static int eio_finish (eio_req *req);
177#define ETP_FINISH(req) eio_finish (req) 260#define ETP_FINISH(req) eio_finish (req)
178static void eio_execute (struct etp_worker *self, eio_req *req); 261static void eio_execute (struct etp_worker *self, eio_req *req);
179#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) 262#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
180 263
181#define ETP_WORKER_CLEAR(req) \
182 if (wrk->dbuf) \
183 { \
184 free (wrk->dbuf); \
185 wrk->dbuf = 0; \
186 } \
187 \
188 if (wrk->dirp) \
189 { \
190 closedir (wrk->dirp); \
191 wrk->dirp = 0; \
192 }
193
194#define ETP_WORKER_COMMON \
195 void *dbuf; \
196 DIR *dirp;
197
198/*****************************************************************************/ 264/*****************************************************************************/
199 265
200#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) 266#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
201 267
202/* calculate time difference in ~1/EIO_TICKS of a second */ 268/* calculate time difference in ~1/EIO_TICKS of a second */
269ecb_inline int
203static int tvdiff (struct timeval *tv1, struct timeval *tv2) 270tvdiff (struct timeval *tv1, struct timeval *tv2)
204{ 271{
205 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 272 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
206 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 273 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
207} 274}
208 275
… …
215static unsigned int max_poll_reqs; /* reslock */ 282static unsigned int max_poll_reqs; /* reslock */
216 283
217static volatile unsigned int nreqs; /* reqlock */ 284static volatile unsigned int nreqs; /* reqlock */
218static volatile unsigned int nready; /* reqlock */ 285static volatile unsigned int nready; /* reqlock */
219static volatile unsigned int npending; /* reqlock */ 286static volatile unsigned int npending; /* reqlock */
220static volatile unsigned int max_idle = 4; 287static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
288static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
221 289
222static mutex_t wrklock = X_MUTEX_INIT; 290static xmutex_t wrklock;
223static mutex_t reslock = X_MUTEX_INIT; 291static xmutex_t reslock;
224static mutex_t reqlock = X_MUTEX_INIT; 292static xmutex_t reqlock;
225static cond_t reqwait = X_COND_INIT; 293static xcond_t reqwait;
226 294
227#if !HAVE_PREADWRITE 295#if !HAVE_PREADWRITE
228/* 296/*
229 * make our pread/pwrite emulation safe against themselves, but not against 297 * make our pread/pwrite emulation safe against themselves, but not against
230 * normal read/write by using a mutex. slows down execution a lot, 298 * normal read/write by using a mutex. slows down execution a lot,
231 * but that's your problem, not mine. 299 * but that's your problem, not mine.
232 */ 300 */
233static mutex_t preadwritelock = X_MUTEX_INIT; 301static xmutex_t preadwritelock;
234#endif 302#endif
235 303
236typedef struct etp_worker 304typedef struct etp_worker
237{ 305{
238 /* locked by wrklock */ 306 /* locked by wrklock */
239 struct etp_worker *prev, *next; 307 struct etp_worker *prev, *next;
240 308
241 thread_t tid; 309 xthread_t tid;
242 310
243 /* locked by reslock, reqlock or wrklock */ 311 /* locked by reslock, reqlock or wrklock */
244 ETP_REQ *req; /* currently processed request */ 312 ETP_REQ *req; /* currently processed request */
245 313
314#ifdef ETP_WORKER_COMMON
246 ETP_WORKER_COMMON 315 ETP_WORKER_COMMON
316#endif
247} etp_worker; 317} etp_worker;
248 318
249static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */ 319static etp_worker wrk_first; /* NOT etp */
250 320
251#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock) 321#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
252#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock) 322#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
253 323
254/* worker threads management */ 324/* worker threads management */
255 325
326static void ecb_cold
256static void etp_worker_clear (etp_worker *wrk) 327etp_worker_clear (etp_worker *wrk)
257{ 328{
258 ETP_WORKER_CLEAR (wrk);
259} 329}
260 330
331static void ecb_cold
261static void etp_worker_free (etp_worker *wrk) 332etp_worker_free (etp_worker *wrk)
262{ 333{
263 wrk->next->prev = wrk->prev; 334 wrk->next->prev = wrk->prev;
264 wrk->prev->next = wrk->next; 335 wrk->prev->next = wrk->next;
265 336
266 free (wrk); 337 free (wrk);
267} 338}
268 339
269static unsigned int etp_nreqs (void) 340static unsigned int
341etp_nreqs (void)
270{ 342{
271 int retval; 343 int retval;
272 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 344 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
273 retval = nreqs; 345 retval = nreqs;
274 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 346 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
275 return retval; 347 return retval;
276} 348}
277 349
278static unsigned int etp_nready (void) 350static unsigned int
351etp_nready (void)
279{ 352{
280 unsigned int retval; 353 unsigned int retval;
281 354
282 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 355 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
283 retval = nready; 356 retval = nready;
284 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 357 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
285 358
286 return retval; 359 return retval;
287} 360}
288 361
289static unsigned int etp_npending (void) 362static unsigned int
363etp_npending (void)
290{ 364{
291 unsigned int retval; 365 unsigned int retval;
292 366
293 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 367 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
294 retval = npending; 368 retval = npending;
295 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 369 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
296 370
297 return retval; 371 return retval;
298} 372}
299 373
300static unsigned int etp_nthreads (void) 374static unsigned int
375etp_nthreads (void)
301{ 376{
302 unsigned int retval; 377 unsigned int retval;
303 378
304 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 379 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
305 retval = started; 380 retval = started;
… …
319} etp_reqq; 394} etp_reqq;
320 395
321static etp_reqq req_queue; 396static etp_reqq req_queue;
322static etp_reqq res_queue; 397static etp_reqq res_queue;
323 398
399static void ecb_noinline ecb_cold
400reqq_init (etp_reqq *q)
401{
402 int pri;
403
404 for (pri = 0; pri < ETP_NUM_PRI; ++pri)
405 q->qs[pri] = q->qe[pri] = 0;
406
407 q->size = 0;
408}
409
410static int ecb_noinline
324static int reqq_push (etp_reqq *q, ETP_REQ *req) 411reqq_push (etp_reqq *q, ETP_REQ *req)
325{ 412{
326 int pri = req->pri; 413 int pri = req->pri;
327 req->next = 0; 414 req->next = 0;
328 415
329 if (q->qe[pri]) 416 if (q->qe[pri])
… …
335 q->qe[pri] = q->qs[pri] = req; 422 q->qe[pri] = q->qs[pri] = req;
336 423
337 return q->size++; 424 return q->size++;
338} 425}
339 426
427static ETP_REQ * ecb_noinline
340static ETP_REQ *reqq_shift (etp_reqq *q) 428reqq_shift (etp_reqq *q)
341{ 429{
342 int pri; 430 int pri;
343 431
344 if (!q->size) 432 if (!q->size)
345 return 0; 433 return 0;
… …
360 } 448 }
361 449
362 abort (); 450 abort ();
363} 451}
364 452
365static void etp_atfork_prepare (void) 453static int ecb_cold
454etp_init (void (*want_poll)(void), void (*done_poll)(void))
366{ 455{
367 X_LOCK (wrklock); 456 X_MUTEX_CREATE (wrklock);
368 X_LOCK (reqlock); 457 X_MUTEX_CREATE (reslock);
369 X_LOCK (reslock); 458 X_MUTEX_CREATE (reqlock);
370#if !HAVE_PREADWRITE 459 X_COND_CREATE (reqwait);
371 X_LOCK (preadwritelock);
372#endif
373}
374 460
375static void etp_atfork_parent (void) 461 reqq_init (&req_queue);
376{ 462 reqq_init (&res_queue);
377#if !HAVE_PREADWRITE
378 X_UNLOCK (preadwritelock);
379#endif
380 X_UNLOCK (reslock);
381 X_UNLOCK (reqlock);
382 X_UNLOCK (wrklock);
383}
384 463
385static void etp_atfork_child (void) 464 wrk_first.next =
386{ 465 wrk_first.prev = &wrk_first;
387 ETP_REQ *prv;
388
389 while ((prv = reqq_shift (&req_queue)))
390 ETP_DESTROY (prv);
391
392 while ((prv = reqq_shift (&res_queue)))
393 ETP_DESTROY (prv);
394
395 while (wrk_first.next != &wrk_first)
396 {
397 etp_worker *wrk = wrk_first.next;
398
399 if (wrk->req)
400 ETP_DESTROY (wrk->req);
401
402 etp_worker_clear (wrk);
403 etp_worker_free (wrk);
404 }
405 466
406 started = 0; 467 started = 0;
407 idle = 0; 468 idle = 0;
408 nreqs = 0; 469 nreqs = 0;
409 nready = 0; 470 nready = 0;
410 npending = 0; 471 npending = 0;
411 472
412 etp_atfork_parent ();
413}
414
415static void
416etp_once_init (void)
417{
418 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
419}
420
421static int
422etp_init (void (*want_poll)(void), void (*done_poll)(void))
423{
424 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
425
426 pthread_once (&doinit, etp_once_init);
427
428 want_poll_cb = want_poll; 473 want_poll_cb = want_poll;
429 done_poll_cb = done_poll; 474 done_poll_cb = done_poll;
430 475
431 return 0; 476 return 0;
432} 477}
433 478
434X_THREAD_PROC (etp_proc); 479X_THREAD_PROC (etp_proc);
435 480
481static void ecb_cold
436static void etp_start_thread (void) 482etp_start_thread (void)
437{ 483{
438 etp_worker *wrk = calloc (1, sizeof (etp_worker)); 484 etp_worker *wrk = calloc (1, sizeof (etp_worker));
439 485
440 /*TODO*/ 486 /*TODO*/
441 assert (("unable to allocate worker thread data", wrk)); 487 assert (("unable to allocate worker thread data", wrk));
… …
454 free (wrk); 500 free (wrk);
455 501
456 X_UNLOCK (wrklock); 502 X_UNLOCK (wrklock);
457} 503}
458 504
505static void
459static void etp_maybe_start_thread (void) 506etp_maybe_start_thread (void)
460{ 507{
461 if (expect_true (etp_nthreads () >= wanted)) 508 if (ecb_expect_true (etp_nthreads () >= wanted))
462 return; 509 return;
463 510
464 /* todo: maybe use idle here, but might be less exact */ 511 /* todo: maybe use idle here, but might be less exact */
465 if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ())) 512 if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
466 return; 513 return;
467 514
468 etp_start_thread (); 515 etp_start_thread ();
469} 516}
470 517
518static void ecb_cold
471static void etp_end_thread (void) 519etp_end_thread (void)
472{ 520{
473 eio_req *req = calloc (1, sizeof (eio_req)); 521 eio_req *req = calloc (1, sizeof (eio_req));
474 522
475 req->type = -1; 523 req->type = -1;
476 req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 524 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
… …
483 X_LOCK (wrklock); 531 X_LOCK (wrklock);
484 --started; 532 --started;
485 X_UNLOCK (wrklock); 533 X_UNLOCK (wrklock);
486} 534}
487 535
488static int etp_poll (void) 536static int
537etp_poll (void)
489{ 538{
490 unsigned int maxreqs; 539 unsigned int maxreqs;
491 unsigned int maxtime; 540 unsigned int maxtime;
492 struct timeval tv_start, tv_now; 541 struct timeval tv_start, tv_now;
493 542
… …
523 572
524 X_LOCK (reqlock); 573 X_LOCK (reqlock);
525 --nreqs; 574 --nreqs;
526 X_UNLOCK (reqlock); 575 X_UNLOCK (reqlock);
527 576
528 if (expect_false (req->type == EIO_GROUP && req->size)) 577 if (ecb_expect_false (req->type == EIO_GROUP && req->size))
529 { 578 {
530 req->int1 = 1; /* mark request as delayed */ 579 req->int1 = 1; /* mark request as delayed */
531 continue; 580 continue;
532 } 581 }
533 else 582 else
534 { 583 {
535 int res = ETP_FINISH (req); 584 int res = ETP_FINISH (req);
536 if (expect_false (res)) 585 if (ecb_expect_false (res))
537 return res; 586 return res;
538 } 587 }
539 588
540 if (expect_false (maxreqs && !--maxreqs)) 589 if (ecb_expect_false (maxreqs && !--maxreqs))
541 break; 590 break;
542 591
543 if (maxtime) 592 if (maxtime)
544 { 593 {
545 gettimeofday (&tv_now, 0); 594 gettimeofday (&tv_now, 0);
… …
551 600
552 errno = EAGAIN; 601 errno = EAGAIN;
553 return -1; 602 return -1;
554} 603}
555 604
605static void
556static void etp_cancel (ETP_REQ *req) 606etp_cancel (ETP_REQ *req)
557{ 607{
558 X_LOCK (wrklock); 608 req->cancelled = 1;
559 req->flags |= EIO_FLAG_CANCELLED;
560 X_UNLOCK (wrklock);
561 609
562 eio_grp_cancel (req); 610 eio_grp_cancel (req);
563} 611}
564 612
613static void
565static void etp_submit (ETP_REQ *req) 614etp_submit (ETP_REQ *req)
566{ 615{
567 req->pri -= ETP_PRI_MIN; 616 req->pri -= ETP_PRI_MIN;
568 617
569 if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN; 618 if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
570 if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 619 if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
571 620
572 if (expect_false (req->type == EIO_GROUP)) 621 if (ecb_expect_false (req->type == EIO_GROUP))
573 { 622 {
574 /* I hope this is worth it :/ */ 623 /* I hope this is worth it :/ */
575 X_LOCK (reqlock); 624 X_LOCK (reqlock);
576 ++nreqs; 625 ++nreqs;
577 X_UNLOCK (reqlock); 626 X_UNLOCK (reqlock);
… …
596 645
597 etp_maybe_start_thread (); 646 etp_maybe_start_thread ();
598 } 647 }
599} 648}
600 649
650static void ecb_cold
601static void etp_set_max_poll_time (double nseconds) 651etp_set_max_poll_time (double nseconds)
602{ 652{
603 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 653 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
604 max_poll_time = nseconds * EIO_TICKS; 654 max_poll_time = nseconds * EIO_TICKS;
605 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 655 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
606} 656}
607 657
658static void ecb_cold
608static void etp_set_max_poll_reqs (unsigned int maxreqs) 659etp_set_max_poll_reqs (unsigned int maxreqs)
609{ 660{
610 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 661 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
611 max_poll_reqs = maxreqs; 662 max_poll_reqs = maxreqs;
612 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 663 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
613} 664}
614 665
666static void ecb_cold
615static void etp_set_max_idle (unsigned int nthreads) 667etp_set_max_idle (unsigned int nthreads)
616{ 668{
617 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 669 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
618 max_idle = nthreads <= 0 ? 1 : nthreads; 670 max_idle = nthreads;
619 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 671 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
620} 672}
621 673
674static void ecb_cold
675etp_set_idle_timeout (unsigned int seconds)
676{
677 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
678 idle_timeout = seconds;
679 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
680}
681
682static void ecb_cold
622static void etp_set_min_parallel (unsigned int nthreads) 683etp_set_min_parallel (unsigned int nthreads)
623{ 684{
624 if (wanted < nthreads) 685 if (wanted < nthreads)
625 wanted = nthreads; 686 wanted = nthreads;
626} 687}
627 688
689static void ecb_cold
628static void etp_set_max_parallel (unsigned int nthreads) 690etp_set_max_parallel (unsigned int nthreads)
629{ 691{
630 if (wanted > nthreads) 692 if (wanted > nthreads)
631 wanted = nthreads; 693 wanted = nthreads;
632 694
633 while (started > wanted) 695 while (started > wanted)
634 etp_end_thread (); 696 etp_end_thread ();
635} 697}
636 698
637/*****************************************************************************/ 699/*****************************************************************************/
638 700
701static void
639static void grp_try_feed (eio_req *grp) 702grp_try_feed (eio_req *grp)
640{ 703{
641 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 704 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
642 { 705 {
643 grp->flags &= ~EIO_FLAG_GROUPADD; 706 grp->flags &= ~EIO_FLAG_GROUPADD;
644 707
… …
651 break; 714 break;
652 } 715 }
653 } 716 }
654} 717}
655 718
719static int
656static int grp_dec (eio_req *grp) 720grp_dec (eio_req *grp)
657{ 721{
658 --grp->size; 722 --grp->size;
659 723
660 /* call feeder, if applicable */ 724 /* call feeder, if applicable */
661 grp_try_feed (grp); 725 grp_try_feed (grp);
… …
665 return eio_finish (grp); 729 return eio_finish (grp);
666 else 730 else
667 return 0; 731 return 0;
668} 732}
669 733
734static void
670void eio_destroy (eio_req *req) 735eio_destroy (eio_req *req)
671{ 736{
672 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); 737 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
673 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); 738 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
674 739
675 EIO_DESTROY (req); 740 EIO_DESTROY (req);
676} 741}
677 742
743static int
678static int eio_finish (eio_req *req) 744eio_finish (eio_req *req)
679{ 745{
680 int res = EIO_FINISH (req); 746 int res = EIO_FINISH (req);
681 747
682 if (req->grp) 748 if (req->grp)
683 { 749 {
… …
691 if (grp->grp_first == req) 757 if (grp->grp_first == req)
692 grp->grp_first = req->grp_next; 758 grp->grp_first = req->grp_next;
693 759
694 res2 = grp_dec (grp); 760 res2 = grp_dec (grp);
695 761
696 if (!res && res2) 762 if (!res)
697 res = res2; 763 res = res2;
698 } 764 }
699 765
700 eio_destroy (req); 766 eio_destroy (req);
701 767
702 return res; 768 return res;
703} 769}
704 770
771void
705void eio_grp_cancel (eio_req *grp) 772eio_grp_cancel (eio_req *grp)
706{ 773{
707 for (grp = grp->grp_first; grp; grp = grp->grp_next) 774 for (grp = grp->grp_first; grp; grp = grp->grp_next)
708 eio_cancel (grp); 775 eio_cancel (grp);
709} 776}
710 777
778void
711void eio_cancel (eio_req *req) 779eio_cancel (eio_req *req)
712{ 780{
713 etp_cancel (req); 781 etp_cancel (req);
714} 782}
715 783
784void
716void eio_submit (eio_req *req) 785eio_submit (eio_req *req)
717{ 786{
718 etp_submit (req); 787 etp_submit (req);
719} 788}
720 789
721unsigned int eio_nreqs (void) 790unsigned int
791eio_nreqs (void)
722{ 792{
723 return etp_nreqs (); 793 return etp_nreqs ();
724} 794}
725 795
726unsigned int eio_nready (void) 796unsigned int
797eio_nready (void)
727{ 798{
728 return etp_nready (); 799 return etp_nready ();
729} 800}
730 801
731unsigned int eio_npending (void) 802unsigned int
803eio_npending (void)
732{ 804{
733 return etp_npending (); 805 return etp_npending ();
734} 806}
735 807
736unsigned int eio_nthreads (void) 808unsigned int ecb_cold
809eio_nthreads (void)
737{ 810{
738 return etp_nthreads (); 811 return etp_nthreads ();
739} 812}
740 813
814void ecb_cold
741void eio_set_max_poll_time (double nseconds) 815eio_set_max_poll_time (double nseconds)
742{ 816{
743 etp_set_max_poll_time (nseconds); 817 etp_set_max_poll_time (nseconds);
744} 818}
745 819
820void ecb_cold
746void eio_set_max_poll_reqs (unsigned int maxreqs) 821eio_set_max_poll_reqs (unsigned int maxreqs)
747{ 822{
748 etp_set_max_poll_reqs (maxreqs); 823 etp_set_max_poll_reqs (maxreqs);
749} 824}
750 825
826void ecb_cold
751void eio_set_max_idle (unsigned int nthreads) 827eio_set_max_idle (unsigned int nthreads)
752{ 828{
753 etp_set_max_idle (nthreads); 829 etp_set_max_idle (nthreads);
754} 830}
755 831
832void ecb_cold
833eio_set_idle_timeout (unsigned int seconds)
834{
835 etp_set_idle_timeout (seconds);
836}
837
838void ecb_cold
756void eio_set_min_parallel (unsigned int nthreads) 839eio_set_min_parallel (unsigned int nthreads)
757{ 840{
758 etp_set_min_parallel (nthreads); 841 etp_set_min_parallel (nthreads);
759} 842}
760 843
844void ecb_cold
761void eio_set_max_parallel (unsigned int nthreads) 845eio_set_max_parallel (unsigned int nthreads)
762{ 846{
763 etp_set_max_parallel (nthreads); 847 etp_set_max_parallel (nthreads);
764} 848}
765 849
766int eio_poll (void) 850int eio_poll (void)
… …
775# undef pread 859# undef pread
776# undef pwrite 860# undef pwrite
777# define pread eio__pread 861# define pread eio__pread
778# define pwrite eio__pwrite 862# define pwrite eio__pwrite
779 863
780static ssize_t 864static eio_ssize_t
781eio__pread (int fd, void *buf, size_t count, off_t offset) 865eio__pread (int fd, void *buf, size_t count, off_t offset)
782{ 866{
783 ssize_t res; 867 eio_ssize_t res;
784 off_t ooffset; 868 off_t ooffset;
785 869
786 X_LOCK (preadwritelock); 870 X_LOCK (preadwritelock);
787 ooffset = lseek (fd, 0, SEEK_CUR); 871 ooffset = lseek (fd, 0, SEEK_CUR);
788 lseek (fd, offset, SEEK_SET); 872 lseek (fd, offset, SEEK_SET);
… …
791 X_UNLOCK (preadwritelock); 875 X_UNLOCK (preadwritelock);
792 876
793 return res; 877 return res;
794} 878}
795 879
796static ssize_t 880static eio_ssize_t
797eio__pwrite (int fd, void *buf, size_t count, off_t offset) 881eio__pwrite (int fd, void *buf, size_t count, off_t offset)
798{ 882{
799 ssize_t res; 883 eio_ssize_t res;
800 off_t ooffset; 884 off_t ooffset;
801 885
802 X_LOCK (preadwritelock); 886 X_LOCK (preadwritelock);
803 ooffset = lseek (fd, 0, SEEK_CUR); 887 ooffset = lseek (fd, 0, SEEK_CUR);
804 lseek (fd, offset, SEEK_SET); 888 lseek (fd, offset, SEEK_SET);
… …
808 892
809 return res; 893 return res;
810} 894}
811#endif 895#endif
812 896
813#ifndef HAVE_FUTIMES 897#ifndef HAVE_UTIMES
814 898
815# undef utimes 899# undef utimes
816# undef futimes
817# define utimes(path,times) eio__utimes (path, times) 900# define utimes(path,times) eio__utimes (path, times)
818# define futimes(fd,times) eio__futimes (fd, times)
819 901
820static int 902static int
821eio__utimes (const char *filename, const struct timeval times[2]) 903eio__utimes (const char *filename, const struct timeval times[2])
822{ 904{
823 if (times) 905 if (times)
… …
831 } 913 }
832 else 914 else
833 return utime (filename, 0); 915 return utime (filename, 0);
834} 916}
835 917
918#endif
919
920#ifndef HAVE_FUTIMES
921
922# undef futimes
923# define futimes(fd,times) eio__futimes (fd, times)
924
925static int
836static int eio__futimes (int fd, const struct timeval tv[2]) 926eio__futimes (int fd, const struct timeval tv[2])
837{ 927{
838 errno = ENOSYS; 928 errno = ENOSYS;
839 return -1; 929 return -1;
840} 930}
841 931
… …
845# undef fdatasync 935# undef fdatasync
846# define fdatasync(fd) fsync (fd) 936# define fdatasync(fd) fsync (fd)
847#endif 937#endif
848 938
849/* sync_file_range always needs emulation */ 939/* sync_file_range always needs emulation */
850int 940static int
851eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags) 941eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
852{ 942{
853#if HAVE_SYNC_FILE_RANGE 943#if HAVE_SYNC_FILE_RANGE
854 int res; 944 int res;
855 945
… …
872 /* even though we could play tricks with the flags, it's better to always 962 /* even though we could play tricks with the flags, it's better to always
873 * call fdatasync, as that matches the expectation of its users best */ 963 * call fdatasync, as that matches the expectation of its users best */
874 return fdatasync (fd); 964 return fdatasync (fd);
875} 965}
876 966
967static int
968eio__fallocate (int fd, int mode, off_t offset, size_t len)
969{
970#if HAVE_FALLOCATE
971 return fallocate (fd, mode, offset, len);
972#else
973 errno = ENOSYS;
974 return -1;
975#endif
976}
977
877#if !HAVE_READAHEAD 978#if !HAVE_READAHEAD
878# undef readahead 979# undef readahead
879# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 980# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
880 981
881static ssize_t 982static eio_ssize_t
882eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) 983eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
883{ 984{
884 size_t todo = count; 985 size_t todo = count;
885 dBUF; 986 dBUF;
886 987
… …
891 pread (fd, eio_buf, len, offset); 992 pread (fd, eio_buf, len, offset);
892 offset += len; 993 offset += len;
893 todo -= len; 994 todo -= len;
894 } 995 }
895 996
997 FUBd;
998
896 errno = 0; 999 errno = 0;
897 return count; 1000 return count;
898} 1001}
899 1002
900#endif 1003#endif
901 1004
902/* sendfile always needs emulation */ 1005/* sendfile always needs emulation */
903static ssize_t 1006static eio_ssize_t
904eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) 1007eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
905{ 1008{
1009 eio_ssize_t written = 0;
906 ssize_t res; 1010 eio_ssize_t res;
907 1011
908 if (!count) 1012 if (!count)
909 return 0; 1013 return 0;
910 1014
1015 for (;;)
1016 {
1017#ifdef __APPLE__
1018# undef HAVE_SENDFILE /* broken, as everything on os x */
1019#endif
911#if HAVE_SENDFILE 1020#if HAVE_SENDFILE
912# if __linux 1021# if __linux
1022 off_t soffset = offset;
913 res = sendfile (ofd, ifd, &offset, count); 1023 res = sendfile (ofd, ifd, &soffset, count);
914 1024
915# elif __FreeBSD__ 1025# elif __FreeBSD__
916 /* 1026 /*
917 * Of course, the freebsd sendfile is a dire hack with no thoughts 1027 * Of course, the freebsd sendfile is a dire hack with no thoughts
918 * wasted on making it similar to other I/O functions. 1028 * wasted on making it similar to other I/O functions.
919 */ 1029 */
920 {
921 off_t sbytes; 1030 off_t sbytes;
922 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 1031 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
923 1032
1033 #if 0 /* according to the manpage, this is correct, but broken behaviour */
924 /* freebsd' sendfile will return 0 on success */ 1034 /* freebsd' sendfile will return 0 on success */
925 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */ 1035 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
926 /* not on e.g. EIO or EPIPE - sounds broken */ 1036 /* not on e.g. EIO or EPIPE - sounds broken */
927 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0) 1037 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
928 res = sbytes; 1038 res = sbytes;
929 } 1039 #endif
1040
1041 /* according to source inspection, this is correct, and useful behaviour */
1042 if (sbytes)
1043 res = sbytes;
930 1044
931# elif defined (__APPLE__) 1045# elif defined (__APPLE__)
932
933 {
934 off_t sbytes = count; 1046 off_t sbytes = count;
935 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0); 1047 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
936 1048
937 /* according to the manpage, sbytes is always valid */ 1049 /* according to the manpage, sbytes is always valid */
938 if (sbytes) 1050 if (sbytes)
939 res = sbytes; 1051 res = sbytes;
940 }
941 1052
942# elif __hpux 1053# elif __hpux
943 res = sendfile (ofd, ifd, offset, count, 0, 0); 1054 res = sendfile (ofd, ifd, offset, count, 0, 0);
944 1055
945# elif __solaris 1056# elif __solaris
946 {
947 struct sendfilevec vec; 1057 struct sendfilevec vec;
948 size_t sbytes; 1058 size_t sbytes;
949 1059
950 vec.sfv_fd = ifd; 1060 vec.sfv_fd = ifd;
951 vec.sfv_flag = 0; 1061 vec.sfv_flag = 0;
952 vec.sfv_off = offset; 1062 vec.sfv_off = offset;
953 vec.sfv_len = count; 1063 vec.sfv_len = count;
954 1064
955 res = sendfilev (ofd, &vec, 1, &sbytes); 1065 res = sendfilev (ofd, &vec, 1, &sbytes);
956 1066
957 if (res < 0 && sbytes) 1067 if (res < 0 && sbytes)
958 res = sbytes; 1068 res = sbytes;
959 }
960 1069
961# endif 1070# endif
962 1071
963#elif defined (_WIN32) 1072#elif defined (_WIN32) && 0
964
965 /* does not work, just for documentation of what would need to be done */ 1073 /* does not work, just for documentation of what would need to be done */
966 { 1074 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
1075 /* libeio guarantees that the file offset does not change, and windows */
1076 /* has no way to get an independent handle to the same file description */
967 HANDLE h = TO_SOCKET (ifd); 1077 HANDLE h = TO_SOCKET (ifd);
968 SetFilePointer (h, offset, 0, FILE_BEGIN); 1078 SetFilePointer (h, offset, 0, FILE_BEGIN);
969 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0); 1079 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
970 }
971 1080
972#else 1081#else
973 res = -1; 1082 res = -1;
974 errno = ENOSYS; 1083 errno = ENOSYS;
975#endif 1084#endif
976 1085
1086 /* we assume sendfile can copy at least 128mb in one go */
1087 if (res <= 128 * 1024 * 1024)
1088 {
1089 if (res > 0)
1090 written += res;
1091
1092 if (written)
1093 return written;
1094
1095 break;
1096 }
1097 else
1098 {
1099 /* if we requested more, then probably the kernel was lazy */
1100 written += res;
1101 offset += res;
1102 count -= res;
1103
1104 if (!count)
1105 return written;
1106 }
1107 }
1108
977 if (res < 0 1109 if (res < 0
978 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1110 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
979 /* BSDs */ 1111 /* BSDs */
980#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */ 1112#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
981 || errno == ENOTSUP 1113 || errno == ENOTSUP
982#endif 1114#endif
1115#ifdef EOPNOTSUPP /* windows */
983 || errno == EOPNOTSUPP /* BSDs */ 1116 || errno == EOPNOTSUPP /* BSDs */
1117#endif
984#if __solaris 1118#if __solaris
985 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1119 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
986#endif 1120#endif
987 ) 1121 )
988 ) 1122 )
… …
992 1126
993 res = 0; 1127 res = 0;
994 1128
995 while (count) 1129 while (count)
996 { 1130 {
997 ssize_t cnt; 1131 eio_ssize_t cnt;
998 1132
999 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 1133 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
1000 1134
1001 if (cnt <= 0) 1135 if (cnt <= 0)
1002 { 1136 {
… …
1014 1148
1015 offset += cnt; 1149 offset += cnt;
1016 res += cnt; 1150 res += cnt;
1017 count -= cnt; 1151 count -= cnt;
1018 } 1152 }
1153
1154 FUBd;
1019 } 1155 }
1020 1156
1021 return res; 1157 return res;
1158}
1159
1160#ifdef PAGESIZE
1161# define eio_pagesize() PAGESIZE
1162#else
1163static intptr_t
1164eio_pagesize (void)
1165{
1166 static intptr_t page;
1167
1168 if (!page)
1169 page = sysconf (_SC_PAGESIZE);
1170
1171 return page;
1172}
1173#endif
1174
1175static void
1176eio_page_align (void **addr, size_t *length)
1177{
1178 intptr_t mask = eio_pagesize () - 1;
1179
1180 /* round down addr */
1181 intptr_t adj = mask & (intptr_t)*addr;
1182
1183 *addr = (void *)((intptr_t)*addr - adj);
1184 *length += adj;
1185
1186 /* round up length */
1187 *length = (*length + mask) & ~mask;
1188}
1189
1190#if !_POSIX_MEMLOCK
1191# define eio__mlockall(a) EIO_ENOSYS ()
1192#else
1193
1194static int
1195eio__mlockall (int flags)
1196{
1197 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1198 extern int mallopt (int, int);
1199 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1200 #endif
1201
1202 if (EIO_MCL_CURRENT != MCL_CURRENT
1203 || EIO_MCL_FUTURE != MCL_FUTURE)
1204 {
1205 flags = 0
1206 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1207 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1208 }
1209
1210 return mlockall (flags);
1211}
1212#endif
1213
1214#if !_POSIX_MEMLOCK_RANGE
1215# define eio__mlock(a,b) EIO_ENOSYS ()
1216#else
1217
1218static int
1219eio__mlock (void *addr, size_t length)
1220{
1221 eio_page_align (&addr, &length);
1222
1223 return mlock (addr, length);
1224}
1225
1226#endif
1227
1228#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1229# define eio__msync(a,b,c) EIO_ENOSYS ()
1230#else
1231
1232static int
1233eio__msync (void *mem, size_t len, int flags)
1234{
1235 eio_page_align (&mem, &len);
1236
1237 if (EIO_MS_ASYNC != MS_SYNC
1238 || EIO_MS_INVALIDATE != MS_INVALIDATE
1239 || EIO_MS_SYNC != MS_SYNC)
1240 {
1241 flags = 0
1242 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1243 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1244 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1245 }
1246
1247 return msync (mem, len, flags);
1248}
1249
1250#endif
1251
1252static int
1253eio__mtouch (eio_req *req)
1254{
1255 void *mem = req->ptr2;
1256 size_t len = req->size;
1257 int flags = req->int1;
1258
1259 eio_page_align (&mem, &len);
1260
1261 {
1262 intptr_t addr = (intptr_t)mem;
1263 intptr_t end = addr + len;
1264 intptr_t page = eio_pagesize ();
1265
1266 if (addr < end)
1267 if (flags & EIO_MT_MODIFY) /* modify */
1268 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1269 else
1270 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1271 }
1272
1273 return 0;
1274}
1275
1276/*****************************************************************************/
1277/* requests implemented outside eio_execute, because they are so large */
1278
1279static void
1280eio__realpath (eio_req *req, etp_worker *self)
1281{
1282 char *rel = req->ptr1;
1283 char *res;
1284 char *tmp1, *tmp2;
1285#if SYMLOOP_MAX > 32
1286 int symlinks = SYMLOOP_MAX;
1287#else
1288 int symlinks = 32;
1289#endif
1290
1291 req->result = -1;
1292
1293 errno = EINVAL;
1294 if (!rel)
1295 return;
1296
1297 errno = ENOENT;
1298 if (!*rel)
1299 return;
1300
1301 if (!req->ptr2)
1302 {
1303 X_LOCK (wrklock);
1304 req->flags |= EIO_FLAG_PTR2_FREE;
1305 X_UNLOCK (wrklock);
1306 req->ptr2 = malloc (PATH_MAX * 3);
1307
1308 errno = ENOMEM;
1309 if (!req->ptr2)
1310 return;
1311 }
1312
1313 res = req->ptr2;
1314 tmp1 = res + PATH_MAX;
1315 tmp2 = tmp1 + PATH_MAX;
1316
1317#if 0 /* disabled, the musl way to do things is just too racy */
1318#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1319 /* on linux we may be able to ask the kernel */
1320 {
1321 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1322
1323 if (fd >= 0)
1324 {
1325 sprintf (tmp1, "/proc/self/fd/%d", fd);
1326 req->result = readlink (tmp1, res, PATH_MAX);
1327 close (fd);
1328
1329 /* here we should probably stat the open file and the disk file, to make sure they still match */
1330
1331 if (req->result > 0)
1332 goto done;
1333 }
1334 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1335 return;
1336 }
1337#endif
1338#endif
1339
1340 if (*rel != '/')
1341 {
1342 if (!getcwd (res, PATH_MAX))
1343 return;
1344
1345 if (res [1]) /* only use if not / */
1346 res += strlen (res);
1347 }
1348
1349 while (*rel)
1350 {
1351 eio_ssize_t len, linklen;
1352 char *beg = rel;
1353
1354 while (*rel && *rel != '/')
1355 ++rel;
1356
1357 len = rel - beg;
1358
1359 if (!len) /* skip slashes */
1360 {
1361 ++rel;
1362 continue;
1363 }
1364
1365 if (beg [0] == '.')
1366 {
1367 if (len == 1)
1368 continue; /* . - nop */
1369
1370 if (beg [1] == '.' && len == 2)
1371 {
1372 /* .. - back up one component, if possible */
1373
1374 while (res != req->ptr2)
1375 if (*--res == '/')
1376 break;
1377
1378 continue;
1379 }
1380 }
1381
1382 errno = ENAMETOOLONG;
1383 if (res + 1 + len + 1 >= tmp1)
1384 return;
1385
1386 /* copy one component */
1387 *res = '/';
1388 memcpy (res + 1, beg, len);
1389
1390 /* zero-terminate, for readlink */
1391 res [len + 1] = 0;
1392
1393 /* now check if it's a symlink */
1394 linklen = readlink (req->ptr2, tmp1, PATH_MAX);
1395
1396 if (linklen < 0)
1397 {
1398 if (errno != EINVAL)
1399 return;
1400
1401 /* it's a normal directory. hopefully */
1402 res += len + 1;
1403 }
1404 else
1405 {
1406 /* yay, it was a symlink - build new path in tmp2 */
1407 int rellen = strlen (rel);
1408
1409 errno = ENAMETOOLONG;
1410 if (linklen + 1 + rellen >= PATH_MAX)
1411 return;
1412
1413 errno = ELOOP;
1414 if (!--symlinks)
1415 return;
1416
1417 if (*tmp1 == '/')
1418 res = req->ptr2; /* symlink resolves to an absolute path */
1419
1420 /* we need to be careful, as rel might point into tmp2 already */
1421 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1422 tmp2 [linklen] = '/';
1423 memcpy (tmp2, tmp1, linklen);
1424
1425 rel = tmp2;
1426 }
1427 }
1428
1429 /* special case for the lone root path */
1430 if (res == req->ptr2)
1431 *res++ = '/';
1432
1433 req->result = res - (char *)req->ptr2;
1434
1435done:
1436 req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
1022} 1437}
1023 1438
1024static signed char 1439static signed char
1025eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) 1440eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1026{ 1441{
1027 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */ 1442 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1028 : a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0; 1443 : a->inode < b->inode ? -1
1444 : a->inode > b->inode ? 1
1445 : 0;
1029} 1446}
1030 1447
1031#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0 1448#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1032 1449
1033#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */ 1450#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1034#define EIO_SORT_FAST 60 /* when to only use insertion sort */ 1451#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1035 1452
1036static void 1453static void
1037eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1454eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1038{ 1455{
1039 unsigned char bits [9 + sizeof (ino_t) * 8]; 1456 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1040 unsigned char *bit = bits; 1457 unsigned char *bit = bits;
1041 1458
1042 assert (CHAR_BIT == 8); 1459 assert (CHAR_BIT == 8);
1043 assert (sizeof (eio_dirent) * 8 < 256); 1460 assert (sizeof (eio_dirent) * 8 < 256);
1044 assert (offsetof (eio_dirent, inode)); /* we use 0 as sentinel */ 1461 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1045 assert (offsetof (eio_dirent, score)); /* we use 0 as sentinel */ 1462 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1046 1463
1047 if (size <= EIO_SORT_FAST) 1464 if (size <= EIO_SORT_FAST)
1048 return; 1465 return;
1049 1466
1050 /* first prepare an array of bits to test in our radix sort */ 1467 /* first prepare an array of bits to test in our radix sort */
1051 /* try to take endianness into account, as well as differences in ino_t sizes */ 1468 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1052 /* inode_bits must contain all inodes ORed together */ 1469 /* inode_bits must contain all inodes ORed together */
1053 /* which is used to skip bits that are 0 everywhere, which is very common */ 1470 /* which is used to skip bits that are 0 everywhere, which is very common */
1054 { 1471 {
1055 ino_t endianness; 1472 eio_ino_t endianness;
1056 int i, j; 1473 int i, j;
1057 1474
1058 /* we store the byte offset of byte n into byte n of "endianness" */ 1475 /* we store the byte offset of byte n into byte n of "endianness" */
1059 for (i = 0; i < sizeof (ino_t); ++i) 1476 for (i = 0; i < sizeof (eio_ino_t); ++i)
1060 ((unsigned char *)&endianness)[i] = i; 1477 ((unsigned char *)&endianness)[i] = i;
1061 1478
1062 *bit++ = 0; 1479 *bit++ = 0;
1063 1480
1064 for (i = 0; i < sizeof (ino_t); ++i) 1481 for (i = 0; i < sizeof (eio_ino_t); ++i)
1065 { 1482 {
1066 /* shifting off the byte offsets out of "endianness" */ 1483 /* shifting off the byte offsets out of "endianness" */
1067 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8; 1484 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1068 endianness >>= 8; 1485 endianness >>= 8;
1069 1486
1070 for (j = 0; j < 8; ++j) 1487 for (j = 0; j < 8; ++j)
1071 if (inode_bits & (((ino_t)1) << (i * 8 + j))) 1488 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1072 *bit++ = offs + j; 1489 *bit++ = offs + j;
1073 } 1490 }
1074 1491
1075 for (j = 0; j < 8; ++j) 1492 for (j = 0; j < 8; ++j)
1076 if (score_bits & (1 << j)) 1493 if (score_bits & (1 << j))
1077 *bit++ = offsetof (eio_dirent, score) * 8 + j; 1494 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1078 } 1495 }
1079 1496
1080 /* now actually do the sorting (a variant of MSD radix sort) */ 1497 /* now actually do the sorting (a variant of MSD radix sort) */
1081 { 1498 {
1082 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base; 1499 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1083 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end; 1500 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1084 unsigned char *bit_stk [9 + sizeof (ino_t) * 8]; 1501 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1085 int stk_idx = 0; 1502 int stk_idx = 0;
1086 1503
1087 base_stk [stk_idx] = dents; 1504 base_stk [stk_idx] = dents;
1088 end_stk [stk_idx] = dents + size; 1505 end_stk [stk_idx] = dents + size;
1089 bit_stk [stk_idx] = bit - 1; 1506 bit_stk [stk_idx] = bit - 1;
… …
1168 } 1585 }
1169 } 1586 }
1170} 1587}
1171 1588
1172static void 1589static void
1173eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1590eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1174{ 1591{
1175 if (size <= 1) 1592 if (size <= 1)
1176 return; /* our insertion sort relies on size > 0 */ 1593 return; /* our insertion sort relies on size > 0 */
1177 1594
1178 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */ 1595 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
… …
1186 1603
1187/* read a full directory */ 1604/* read a full directory */
1188static void 1605static void
1189eio__scandir (eio_req *req, etp_worker *self) 1606eio__scandir (eio_req *req, etp_worker *self)
1190{ 1607{
1191 DIR *dirp;
1192 EIO_STRUCT_DIRENT *entp;
1193 char *name, *names; 1608 char *name, *names;
1194 int namesalloc = 4096; 1609 int namesalloc = 4096 - sizeof (void *) * 4;
1195 int namesoffs = 0; 1610 int namesoffs = 0;
1196 int flags = req->int1; 1611 int flags = req->int1;
1197 eio_dirent *dents = 0; 1612 eio_dirent *dents = 0;
1198 int dentalloc = 128; 1613 int dentalloc = 128;
1199 int dentoffs = 0; 1614 int dentoffs = 0;
1200 ino_t inode_bits = 0; 1615 eio_ino_t inode_bits = 0;
1616#ifdef _WIN32
1617 HANDLE dirp;
1618 WIN32_FIND_DATA entp;
1619#else
1620 DIR *dirp;
1621 EIO_STRUCT_DIRENT *entp;
1622#endif
1201 1623
1202 req->result = -1; 1624 req->result = -1;
1203 1625
1204 if (!(flags & EIO_READDIR_DENTS)) 1626 if (!(flags & EIO_READDIR_DENTS))
1205 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); 1627 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1206 1628
1207 X_LOCK (wrklock); 1629#ifdef _WIN32
1208 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 1630 {
1631 int len = strlen ((const char *)req->ptr1);
1632 char *path = malloc (MAX_PATH);
1633 const char *fmt;
1634
1635 if (!len)
1636 fmt = "./*";
1637 else if (((const char *)req->ptr1)[len - 1] == '/' || ((const char *)req->ptr1)[len - 1] == '\\')
1638 fmt = "%s*";
1639 else
1640 fmt = "%s/*";
1641
1642 _snprintf (path, MAX_PATH, fmt, (const char *)req->ptr1);
1643 dirp = FindFirstFile (path, &entp);
1644 free (path);
1645
1646 if (dirp == INVALID_HANDLE_VALUE)
1647 {
1648 dirp = 0;
1649
1650 switch (GetLastError ())
1651 {
1652 case ERROR_FILE_NOT_FOUND:
1653 req->result = 0;
1654 break;
1655
1656 case ERROR_INVALID_NAME:
1657 case ERROR_PATH_NOT_FOUND:
1658 case ERROR_NO_MORE_FILES:
1659 errno = ENOENT;
1660 break;
1661
1662 case ERROR_NOT_ENOUGH_MEMORY:
1663 errno = ENOMEM;
1664 break;
1665
1666 default:
1667 errno = EINVAL;
1668 break;
1669 }
1670 }
1671 }
1672#else
1209 self->dirp = dirp = opendir (req->ptr1); 1673 dirp = opendir (req->ptr1);
1674#endif
1675
1676 if (req->flags & EIO_FLAG_PTR1_FREE)
1677 free (req->ptr1);
1678
1210 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1679 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1211 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; 1680 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1212 req->ptr2 = names = malloc (namesalloc); 1681 req->ptr2 = names = malloc (namesalloc);
1213 X_UNLOCK (wrklock);
1214 1682
1215 if (dirp && names && (!flags || dents)) 1683 if (dirp && names && (!flags || dents))
1216 for (;;) 1684 for (;;)
1217 { 1685 {
1686 int done;
1687
1688#ifdef _WIN32
1689 done = !dirp;
1690#else
1218 errno = 0; 1691 errno = 0;
1219 entp = readdir (dirp); 1692 entp = readdir (dirp);
1693 done = !entp;
1694#endif
1220 1695
1221 if (!entp) 1696 if (done)
1222 { 1697 {
1698#ifndef _WIN32
1699 int old_errno = errno;
1700 closedir (dirp);
1701 errno = old_errno;
1702
1223 if (errno) 1703 if (errno)
1224 break; 1704 break;
1705#endif
1225 1706
1226 /* sort etc. */ 1707 /* sort etc. */
1227 req->int1 = flags; 1708 req->int1 = flags;
1228 req->result = dentoffs; 1709 req->result = dentoffs;
1229 1710
1230 if (flags & EIO_READDIR_STAT_ORDER) 1711 if (flags & EIO_READDIR_STAT_ORDER)
1231 eio_dent_sort (dents, dentoffs, 0, inode_bits); /* sort by inode exclusively */ 1712 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1232 else if (flags & EIO_READDIR_DIRS_FIRST) 1713 else if (flags & EIO_READDIR_DIRS_FIRST)
1233 if (flags & EIO_READDIR_FOUND_UNKNOWN) 1714 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1234 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */ 1715 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1235 else 1716 else
1236 { 1717 {
… …
1238 eio_dirent *oth = dents + dentoffs; 1719 eio_dirent *oth = dents + dentoffs;
1239 eio_dirent *dir = dents; 1720 eio_dirent *dir = dents;
1240 1721
1241 /* now partition dirs to the front, and non-dirs to the back */ 1722 /* now partition dirs to the front, and non-dirs to the back */
1242 /* by walking from both sides and swapping if necessary */ 1723 /* by walking from both sides and swapping if necessary */
1243 /* also clear score, so it doesn't influence sorting */
1244 while (oth > dir) 1724 while (oth > dir)
1245 { 1725 {
1246 if (dir->type == EIO_DT_DIR) 1726 if (dir->type == EIO_DT_DIR)
1247 ++dir; 1727 ++dir;
1248 else if ((--oth)->type == EIO_DT_DIR) 1728 else if ((--oth)->type == EIO_DT_DIR)
… …
1251 1731
1252 ++dir; 1732 ++dir;
1253 } 1733 }
1254 } 1734 }
1255 1735
1256 /* now sort the dirs only */ 1736 /* now sort the dirs only (dirs all have the same score) */
1257 eio_dent_sort (dents, dir - dents, 0, inode_bits); 1737 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1258 } 1738 }
1259 1739
1260 break; 1740 break;
1261 } 1741 }
1262 1742
1263 /* now add the entry to our list(s) */ 1743 /* now add the entry to our list(s) */
1264 name = entp->d_name; 1744 name = D_NAME (entp);
1265 1745
1266 /* skip . and .. entries */ 1746 /* skip . and .. entries */
1267 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1747 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1268 { 1748 {
1269 int len = D_NAMLEN (entp) + 1; 1749 int len = D_NAMLEN (entp) + 1;
1270 1750
1271 while (expect_false (namesoffs + len > namesalloc)) 1751 while (ecb_expect_false (namesoffs + len > namesalloc))
1272 { 1752 {
1273 namesalloc *= 2; 1753 namesalloc *= 2;
1274 X_LOCK (wrklock);
1275 req->ptr2 = names = realloc (names, namesalloc); 1754 req->ptr2 = names = realloc (names, namesalloc);
1276 X_UNLOCK (wrklock);
1277 1755
1278 if (!names) 1756 if (!names)
1279 break; 1757 break;
1280 } 1758 }
1281 1759
… …
1283 1761
1284 if (dents) 1762 if (dents)
1285 { 1763 {
1286 struct eio_dirent *ent; 1764 struct eio_dirent *ent;
1287 1765
1288 if (expect_false (dentoffs == dentalloc)) 1766 if (ecb_expect_false (dentoffs == dentalloc))
1289 { 1767 {
1290 dentalloc *= 2; 1768 dentalloc *= 2;
1291 X_LOCK (wrklock);
1292 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent)); 1769 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1293 X_UNLOCK (wrklock);
1294 1770
1295 if (!dents) 1771 if (!dents)
1296 break; 1772 break;
1297 } 1773 }
1298 1774
… …
1378 if (EIO_CANCELLED (req)) 1854 if (EIO_CANCELLED (req))
1379 { 1855 {
1380 errno = ECANCELED; 1856 errno = ECANCELED;
1381 break; 1857 break;
1382 } 1858 }
1859
1860#ifdef _WIN32
1861 if (!FindNextFile (dirp, &entp))
1862 {
1863 FindClose (dirp);
1864 dirp = 0;
1865 }
1866#endif
1383 } 1867 }
1384}
1385
1386#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1387# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
1388#else
1389
1390int
1391eio__msync (void *mem, size_t len, int flags)
1392{
1393 if (EIO_MS_ASYNC != MS_SYNC
1394 || EIO_MS_INVALIDATE != MS_INVALIDATE
1395 || EIO_MS_SYNC != MS_SYNC)
1396 {
1397 flags = 0
1398 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1399 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1400 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1401 }
1402
1403 return msync (mem, len, flags);
1404}
1405
1406#endif
1407
1408int
1409eio__mtouch (void *mem, size_t len, int flags)
1410{
1411 intptr_t addr = (intptr_t)mem;
1412 intptr_t end = addr + len;
1413#ifdef PAGESIZE
1414 const intptr_t page = PAGESIZE;
1415#else
1416 static intptr_t page;
1417
1418 if (!page)
1419 page = sysconf (_SC_PAGESIZE);
1420#endif
1421
1422 /* round down to start of page, although this is probably useless */
1423 addr &= ~(page - 1); /* assume page size is always a power of two */
1424
1425 if (addr < end)
1426 if (flags & EIO_MT_MODIFY) /* modify */
1427 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
1428 else
1429 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
1430
1431 return 0;
1432} 1868}
1433 1869
1434/*****************************************************************************/ 1870/*****************************************************************************/
1435 1871
1436#define ALLOC(len) \ 1872#define ALLOC(len) \
… …
1452{ 1888{
1453 ETP_REQ *req; 1889 ETP_REQ *req;
1454 struct timespec ts; 1890 struct timespec ts;
1455 etp_worker *self = (etp_worker *)thr_arg; 1891 etp_worker *self = (etp_worker *)thr_arg;
1456 1892
1457 /* try to distribute timeouts somewhat randomly */ 1893 /* try to distribute timeouts somewhat evenly */
1458 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 1894 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1459 1895
1460 for (;;) 1896 for (;;)
1461 { 1897 {
1898 ts.tv_sec = 0;
1899
1462 X_LOCK (reqlock); 1900 X_LOCK (reqlock);
1463 1901
1464 for (;;) 1902 for (;;)
1465 { 1903 {
1466 self->req = req = reqq_shift (&req_queue); 1904 self->req = req = reqq_shift (&req_queue);
1467 1905
1468 if (req) 1906 if (req)
1469 break; 1907 break;
1470 1908
1909 if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
1910 {
1911 X_UNLOCK (reqlock);
1912 X_LOCK (wrklock);
1913 --started;
1914 X_UNLOCK (wrklock);
1915 goto quit;
1916 }
1917
1471 ++idle; 1918 ++idle;
1472 1919
1473 ts.tv_sec = time (0) + IDLE_TIMEOUT; 1920 if (idle <= max_idle)
1474 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 1921 /* we are allowed to idle, so do so without any timeout */
1922 X_COND_WAIT (reqwait, reqlock);
1923 else
1475 { 1924 {
1476 if (idle > max_idle) 1925 /* initialise timeout once */
1477 { 1926 if (!ts.tv_sec)
1478 --idle; 1927 ts.tv_sec = time (0) + idle_timeout;
1479 X_UNLOCK (reqlock);
1480 X_LOCK (wrklock);
1481 --started;
1482 X_UNLOCK (wrklock);
1483 goto quit;
1484 }
1485 1928
1486 /* we are allowed to idle, so do so without any timeout */
1487 X_COND_WAIT (reqwait, reqlock); 1929 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
1930 ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1488 } 1931 }
1489 1932
1490 --idle; 1933 --idle;
1491 } 1934 }
1492 1935
… …
1495 X_UNLOCK (reqlock); 1938 X_UNLOCK (reqlock);
1496 1939
1497 if (req->type < 0) 1940 if (req->type < 0)
1498 goto quit; 1941 goto quit;
1499 1942
1500 if (!EIO_CANCELLED (req))
1501 ETP_EXECUTE (self, req); 1943 ETP_EXECUTE (self, req);
1502 1944
1503 X_LOCK (reslock); 1945 X_LOCK (reslock);
1504 1946
1505 ++npending; 1947 ++npending;
1506 1948
… …
1521 return 0; 1963 return 0;
1522} 1964}
1523 1965
1524/*****************************************************************************/ 1966/*****************************************************************************/
1525 1967
1968int ecb_cold
1526int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1969eio_init (void (*want_poll)(void), void (*done_poll)(void))
1527{ 1970{
1971#if !HAVE_PREADWRITE
1972 X_MUTEX_CREATE (preadwritelock);
1973#endif
1974
1528 return etp_init (want_poll, done_poll); 1975 return etp_init (want_poll, done_poll);
1529} 1976}
1530 1977
1978ecb_inline void
1531static void eio_api_destroy (eio_req *req) 1979eio_api_destroy (eio_req *req)
1532{ 1980{
1533 free (req); 1981 free (req);
1534} 1982}
1535 1983
1536#define REQ(rtype) \ 1984#define REQ(rtype) \
… …
1555 { \ 2003 { \
1556 eio_api_destroy (req); \ 2004 eio_api_destroy (req); \
1557 return 0; \ 2005 return 0; \
1558 } 2006 }
1559 2007
2008static void
1560static void eio_execute (etp_worker *self, eio_req *req) 2009eio_execute (etp_worker *self, eio_req *req)
1561{ 2010{
1562 errno = 0; 2011 if (ecb_expect_false (EIO_CANCELLED (req)))
2012 {
2013 req->result = -1;
2014 req->errorno = ECANCELED;
2015 return;
2016 }
1563 2017
1564 switch (req->type) 2018 switch (req->type)
1565 { 2019 {
1566 case EIO_READ: ALLOC (req->size); 2020 case EIO_READ: ALLOC (req->size);
1567 req->result = req->offs >= 0 2021 req->result = req->offs >= 0
… …
1570 case EIO_WRITE: req->result = req->offs >= 0 2024 case EIO_WRITE: req->result = req->offs >= 0
1571 ? pwrite (req->int1, req->ptr2, req->size, req->offs) 2025 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1572 : write (req->int1, req->ptr2, req->size); break; 2026 : write (req->int1, req->ptr2, req->size); break;
1573 2027
1574 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break; 2028 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1575 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break; 2029 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
1576 2030
1577 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2031 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1578 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2032 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1579 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2033 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1580 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2034 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
… …
1600 case EIO_RMDIR: req->result = rmdir (req->ptr1); break; 2054 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1601 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; 2055 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1602 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; 2056 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1603 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; 2057 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1604 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; 2058 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1605 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break; 2059 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1606 2060
2061 case EIO_REALPATH: eio__realpath (req, self); break;
2062
1607 case EIO_READLINK: ALLOC (NAME_MAX); 2063 case EIO_READLINK: ALLOC (PATH_MAX);
1608 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; 2064 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1609 2065
1610 case EIO_SYNC: req->result = 0; sync (); break; 2066 case EIO_SYNC: req->result = 0; sync (); break;
1611 case EIO_FSYNC: req->result = fsync (req->int1); break; 2067 case EIO_FSYNC: req->result = fsync (req->int1); break;
1612 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 2068 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1613 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break; 2069 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
2070 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1614 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break; 2071 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
2072 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1615 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break; 2073 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
2074 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
1616 2075
1617 case EIO_READDIR: eio__scandir (req, self); break; 2076 case EIO_READDIR: eio__scandir (req, self); break;
1618 2077
1619 case EIO_BUSY: 2078 case EIO_BUSY:
1620#ifdef _WIN32 2079#ifdef _WIN32
1621 Sleep (req->nv1 * 1000.); 2080 Sleep (req->nv1 * 1e3);
1622#else 2081#else
1623 { 2082 {
1624 struct timeval tv; 2083 struct timeval tv;
1625 2084
1626 tv.tv_sec = req->nv1; 2085 tv.tv_sec = req->nv1;
1627 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; 2086 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1628 2087
1629 req->result = select (0, 0, 0, 0, &tv); 2088 req->result = select (0, 0, 0, 0, &tv);
1630 } 2089 }
1631#endif 2090#endif
1632 break; 2091 break;
… …
1647 times = tv; 2106 times = tv;
1648 } 2107 }
1649 else 2108 else
1650 times = 0; 2109 times = 0;
1651 2110
1652
1653 req->result = req->type == EIO_FUTIME 2111 req->result = req->type == EIO_FUTIME
1654 ? futimes (req->int1, times) 2112 ? futimes (req->int1, times)
1655 : utimes (req->ptr1, times); 2113 : utimes (req->ptr1, times);
1656 } 2114 }
1657 break; 2115 break;
… …
1662 case EIO_NOP: 2120 case EIO_NOP:
1663 req->result = 0; 2121 req->result = 0;
1664 break; 2122 break;
1665 2123
1666 case EIO_CUSTOM: 2124 case EIO_CUSTOM:
1667 ((void (*)(eio_req *))req->feed) (req); 2125 req->feed (req);
1668 break; 2126 break;
1669 2127
1670 default: 2128 default:
2129 errno = ENOSYS;
1671 req->result = -1; 2130 req->result = -1;
1672 break; 2131 break;
1673 } 2132 }
1674 2133
1675 req->errorno = errno; 2134 req->errorno = errno;
… …
1705eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2164eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1706{ 2165{
1707 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2166 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1708} 2167}
1709 2168
2169eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
2170{
2171 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
2172}
2173
2174eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
2175{
2176 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2177}
2178
1710eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) 2179eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
1711{ 2180{
1712 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; 2181 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
1713} 2182}
1714 2183
2184eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2185{
2186 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
2187}
2188
1715eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) 2189eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1716{ 2190{
1717 REQ (EIO_FDATASYNC); req->int1 = fd; SEND; 2191 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1718} 2192}
1719 2193
… …
1760eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) 2234eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1761{ 2235{
1762 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2236 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1763} 2237}
1764 2238
1765eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2239eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1766{ 2240{
1767 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2241 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1768} 2242}
1769 2243
1770eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) 2244eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
… …
1790eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data) 2264eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1791{ 2265{
1792 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2266 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1793} 2267}
1794 2268
1795eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2269eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1796{ 2270{
1797 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2271 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1798} 2272}
1799 2273
1800eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) 2274eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
… …
1816eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data) 2290eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1817{ 2291{
1818 return eio__1path (EIO_READLINK, path, pri, cb, data); 2292 return eio__1path (EIO_READLINK, path, pri, cb, data);
1819} 2293}
1820 2294
2295eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2296{
2297 return eio__1path (EIO_REALPATH, path, pri, cb, data);
2298}
2299
1821eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) 2300eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1822{ 2301{
1823 return eio__1path (EIO_STAT, path, pri, cb, data); 2302 return eio__1path (EIO_STAT, path, pri, cb, data);
1824} 2303}
1825 2304
… …
1848 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND; 2327 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1849} 2328}
1850 2329
1851eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) 2330eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1852{ 2331{
1853 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; 2332 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1854} 2333}
1855 2334
1856static eio_req * 2335static eio_req *
1857eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2336eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1858{ 2337{
… …
1882eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2361eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1883{ 2362{
1884 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2363 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1885} 2364}
1886 2365
1887eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) 2366eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
1888{ 2367{
1889 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND; 2368 REQ (EIO_CUSTOM); req->feed = execute; SEND;
1890} 2369}
1891 2370
1892#endif 2371#endif
1893 2372
1894eio_req *eio_grp (eio_cb cb, void *data) 2373eio_req *eio_grp (eio_cb cb, void *data)
… …
1903#undef SEND 2382#undef SEND
1904 2383
1905/*****************************************************************************/ 2384/*****************************************************************************/
1906/* grp functions */ 2385/* grp functions */
1907 2386
2387void
1908void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2388eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1909{ 2389{
1910 grp->int2 = limit; 2390 grp->int2 = limit;
1911 grp->feed = feed; 2391 grp->feed = feed;
1912 2392
1913 grp_try_feed (grp); 2393 grp_try_feed (grp);
1914} 2394}
1915 2395
2396void
1916void eio_grp_limit (eio_req *grp, int limit) 2397eio_grp_limit (eio_req *grp, int limit)
1917{ 2398{
1918 grp->int2 = limit; 2399 grp->int2 = limit;
1919 2400
1920 grp_try_feed (grp); 2401 grp_try_feed (grp);
1921} 2402}
1922 2403
2404void
1923void eio_grp_add (eio_req *grp, eio_req *req) 2405eio_grp_add (eio_req *grp, eio_req *req)
1924{ 2406{
1925 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2407 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1926 2408
1927 grp->flags |= EIO_FLAG_GROUPADD; 2409 grp->flags |= EIO_FLAG_GROUPADD;
1928 2410
… …
1939} 2421}
1940 2422
1941/*****************************************************************************/ 2423/*****************************************************************************/
1942/* misc garbage */ 2424/* misc garbage */
1943 2425
2426eio_ssize_t
1944ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2427eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1945{ 2428{
1946 etp_worker wrk;
1947 ssize_t ret;
1948
1949 wrk.dbuf = 0;
1950
1951 ret = eio__sendfile (ofd, ifd, offset, count, &wrk); 2429 return eio__sendfile (ofd, ifd, offset, count);
1952
1953 if (wrk.dbuf)
1954 free (wrk.dbuf);
1955
1956 return ret;
1957} 2430}
1958 2431

Diff Legend

– Removed lines
+ Added lines
< Changed lines
> Changed lines