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 1.61 by root, Fri Feb 11 00:53:24 2011 UTC vs.
Revision 1.98 by root, Sun Jul 24 05:53:34 2011 UTC

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>
167 #include <sys/statvfs.h>
79# include <unistd.h> 168 #include <unistd.h>
169 #include <signal.h>
170 #include <dirent.h>
171
172 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
173 #include <sys/mman.h>
174 #endif
175
176 #define D_NAME(entp) entp->d_name
177
178 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
179 #if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
180 #define _DIRENT_HAVE_D_TYPE /* sigh */
181 #define D_INO(de) (de)->d_fileno
182 #define D_NAMLEN(de) (de)->d_namlen
183 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
184 #define D_INO(de) (de)->d_ino
185 #endif
186
187 #ifdef _D_EXACT_NAMLEN
188 #undef D_NAMLEN
189 #define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
190 #endif
191
192 #ifdef _DIRENT_HAVE_D_TYPE
193 #define D_TYPE(de) (de)->d_type
194 #endif
195
196 #ifndef EIO_STRUCT_DIRENT
197 #define EIO_STRUCT_DIRENT struct dirent
198 #endif
199
200#endif
201
202#if HAVE_UTIMES
80# include <utime.h> 203# include <utime.h>
81# include <signal.h>
82# include <dirent.h>
83
84#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
85# include <sys/mman.h>
86#endif
87
88/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
89# if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
90# define _DIRENT_HAVE_D_TYPE /* sigh */
91# define D_INO(de) (de)->d_fileno
92# define D_NAMLEN(de) (de)->d_namlen
93# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
94# define D_INO(de) (de)->d_ino
95# endif 204#endif
96 205
97#ifdef _D_EXACT_NAMLEN 206#if HAVE_SYS_SYSCALL_H
98# undef D_NAMLEN 207# include <sys/syscall.h>
99# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
100#endif
101
102# ifdef _DIRENT_HAVE_D_TYPE
103# define D_TYPE(de) (de)->d_type
104# endif 208#endif
105 209
106# ifndef EIO_STRUCT_DIRENT 210#if HAVE_SYS_PRCTL_H
107# define EIO_STRUCT_DIRENT struct dirent 211# include <sys/prctl.h>
108# endif
109
110#endif 212#endif
111 213
112#if HAVE_SENDFILE 214#if HAVE_SENDFILE
113# if __linux 215# if __linux
114# include <sys/sendfile.h> 216# include <sys/sendfile.h>
129#endif 231#endif
130#ifndef D_INO 232#ifndef D_INO
131# define D_INO(de) 0 233# define D_INO(de) 0
132#endif 234#endif
133#ifndef D_NAMLEN 235#ifndef D_NAMLEN
134# define D_NAMLEN(de) strlen ((de)->d_name) 236# define D_NAMLEN(entp) strlen (D_NAME (entp))
135#endif 237#endif
136
137/* number of seconds after which an idle threads exit */
138#define IDLE_TIMEOUT 10
139 238
140/* used for struct dirent, AIX doesn't provide it */ 239/* used for struct dirent, AIX doesn't provide it */
141#ifndef NAME_MAX 240#ifndef NAME_MAX
142# define NAME_MAX 4096 241# define NAME_MAX 4096
143#endif 242#endif
149 248
150/* buffer size for various temporary buffers */ 249/* buffer size for various temporary buffers */
151#define EIO_BUFSIZE 65536 250#define EIO_BUFSIZE 65536
152 251
153#define dBUF \ 252#define dBUF \
154 char *eio_buf; \
155 ETP_WORKER_LOCK (self); \
156 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 253 char *eio_buf = malloc (EIO_BUFSIZE); \
157 ETP_WORKER_UNLOCK (self); \
158 errno = ENOMEM; \ 254 errno = ENOMEM; \
159 if (!eio_buf) \ 255 if (!eio_buf) \
160 return -1; 256 return -1
257
258#define FUBd \
259 free (eio_buf)
161 260
162#define EIO_TICKS ((1000000 + 1023) >> 10) 261#define EIO_TICKS ((1000000 + 1023) >> 10)
163
164/*****************************************************************************/
165
166#if __GNUC__ >= 3
167# define expect(expr,value) __builtin_expect ((expr),(value))
168#else
169# define expect(expr,value) (expr)
170#endif
171
172#define expect_false(expr) expect ((expr) != 0, 0)
173#define expect_true(expr) expect ((expr) != 0, 1)
174
175/*****************************************************************************/
176 262
177#define ETP_PRI_MIN EIO_PRI_MIN 263#define ETP_PRI_MIN EIO_PRI_MIN
178#define ETP_PRI_MAX EIO_PRI_MAX 264#define ETP_PRI_MAX EIO_PRI_MAX
179 265
180struct etp_worker; 266struct etp_worker;
184static int eio_finish (eio_req *req); 270static int eio_finish (eio_req *req);
185#define ETP_FINISH(req) eio_finish (req) 271#define ETP_FINISH(req) eio_finish (req)
186static void eio_execute (struct etp_worker *self, eio_req *req); 272static void eio_execute (struct etp_worker *self, eio_req *req);
187#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) 273#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
188 274
189#define ETP_WORKER_CLEAR(req) \
190 if (wrk->dbuf) \
191 { \
192 free (wrk->dbuf); \
193 wrk->dbuf = 0; \
194 } \
195 \
196 if (wrk->dirp) \
197 { \
198 closedir (wrk->dirp); \
199 wrk->dirp = 0; \
200 }
201
202#define ETP_WORKER_COMMON \
203 void *dbuf; \
204 DIR *dirp;
205
206/*****************************************************************************/ 275/*****************************************************************************/
207 276
208#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) 277#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
209 278
210/* calculate time difference in ~1/EIO_TICKS of a second */ 279/* calculate time difference in ~1/EIO_TICKS of a second */
280ecb_inline int
211static int tvdiff (struct timeval *tv1, struct timeval *tv2) 281tvdiff (struct timeval *tv1, struct timeval *tv2)
212{ 282{
213 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 283 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
214 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 284 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
215} 285}
216 286
223static unsigned int max_poll_reqs; /* reslock */ 293static unsigned int max_poll_reqs; /* reslock */
224 294
225static volatile unsigned int nreqs; /* reqlock */ 295static volatile unsigned int nreqs; /* reqlock */
226static volatile unsigned int nready; /* reqlock */ 296static volatile unsigned int nready; /* reqlock */
227static volatile unsigned int npending; /* reqlock */ 297static volatile unsigned int npending; /* reqlock */
228static volatile unsigned int max_idle = 4; 298static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
299static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
229 300
230static xmutex_t wrklock = X_MUTEX_INIT; 301static xmutex_t wrklock;
231static xmutex_t reslock = X_MUTEX_INIT; 302static xmutex_t reslock;
232static xmutex_t reqlock = X_MUTEX_INIT; 303static xmutex_t reqlock;
233static xcond_t reqwait = X_COND_INIT; 304static xcond_t reqwait;
234 305
235#if !HAVE_PREADWRITE 306#if !HAVE_PREADWRITE
236/* 307/*
237 * make our pread/pwrite emulation safe against themselves, but not against 308 * make our pread/pwrite emulation safe against themselves, but not against
238 * normal read/write by using a mutex. slows down execution a lot, 309 * normal read/write by using a mutex. slows down execution a lot,
239 * but that's your problem, not mine. 310 * but that's your problem, not mine.
240 */ 311 */
241static xmutex_t preadwritelock = X_MUTEX_INIT; 312static xmutex_t preadwritelock;
242#endif 313#endif
243 314
244typedef struct etp_worker 315typedef struct etp_worker
245{ 316{
246 /* locked by wrklock */ 317 /* locked by wrklock */
249 xthread_t tid; 320 xthread_t tid;
250 321
251 /* locked by reslock, reqlock or wrklock */ 322 /* locked by reslock, reqlock or wrklock */
252 ETP_REQ *req; /* currently processed request */ 323 ETP_REQ *req; /* currently processed request */
253 324
325#ifdef ETP_WORKER_COMMON
254 ETP_WORKER_COMMON 326 ETP_WORKER_COMMON
327#endif
255} etp_worker; 328} etp_worker;
256 329
257static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */ 330static etp_worker wrk_first; /* NOT etp */
258 331
259#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock) 332#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
260#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock) 333#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
261 334
262/* worker threads management */ 335/* worker threads management */
263 336
337static void ecb_cold
264static void etp_worker_clear (etp_worker *wrk) 338etp_worker_clear (etp_worker *wrk)
265{ 339{
266 ETP_WORKER_CLEAR (wrk);
267} 340}
268 341
342static void ecb_cold
269static void etp_worker_free (etp_worker *wrk) 343etp_worker_free (etp_worker *wrk)
270{ 344{
271 wrk->next->prev = wrk->prev; 345 wrk->next->prev = wrk->prev;
272 wrk->prev->next = wrk->next; 346 wrk->prev->next = wrk->next;
273 347
274 free (wrk); 348 free (wrk);
275} 349}
276 350
277static unsigned int etp_nreqs (void) 351static unsigned int
352etp_nreqs (void)
278{ 353{
279 int retval; 354 int retval;
280 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 355 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
281 retval = nreqs; 356 retval = nreqs;
282 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 357 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
283 return retval; 358 return retval;
284} 359}
285 360
286static unsigned int etp_nready (void) 361static unsigned int
362etp_nready (void)
287{ 363{
288 unsigned int retval; 364 unsigned int retval;
289 365
290 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 366 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
291 retval = nready; 367 retval = nready;
292 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 368 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
293 369
294 return retval; 370 return retval;
295} 371}
296 372
297static unsigned int etp_npending (void) 373static unsigned int
374etp_npending (void)
298{ 375{
299 unsigned int retval; 376 unsigned int retval;
300 377
301 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 378 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
302 retval = npending; 379 retval = npending;
303 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 380 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
304 381
305 return retval; 382 return retval;
306} 383}
307 384
308static unsigned int etp_nthreads (void) 385static unsigned int
386etp_nthreads (void)
309{ 387{
310 unsigned int retval; 388 unsigned int retval;
311 389
312 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 390 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
313 retval = started; 391 retval = started;
327} etp_reqq; 405} etp_reqq;
328 406
329static etp_reqq req_queue; 407static etp_reqq req_queue;
330static etp_reqq res_queue; 408static etp_reqq res_queue;
331 409
410static void ecb_noinline ecb_cold
411reqq_init (etp_reqq *q)
412{
413 int pri;
414
415 for (pri = 0; pri < ETP_NUM_PRI; ++pri)
416 q->qs[pri] = q->qe[pri] = 0;
417
418 q->size = 0;
419}
420
421static int ecb_noinline
332static int reqq_push (etp_reqq *q, ETP_REQ *req) 422reqq_push (etp_reqq *q, ETP_REQ *req)
333{ 423{
334 int pri = req->pri; 424 int pri = req->pri;
335 req->next = 0; 425 req->next = 0;
336 426
337 if (q->qe[pri]) 427 if (q->qe[pri])
343 q->qe[pri] = q->qs[pri] = req; 433 q->qe[pri] = q->qs[pri] = req;
344 434
345 return q->size++; 435 return q->size++;
346} 436}
347 437
438static ETP_REQ * ecb_noinline
348static ETP_REQ *reqq_shift (etp_reqq *q) 439reqq_shift (etp_reqq *q)
349{ 440{
350 int pri; 441 int pri;
351 442
352 if (!q->size) 443 if (!q->size)
353 return 0; 444 return 0;
368 } 459 }
369 460
370 abort (); 461 abort ();
371} 462}
372 463
373static void etp_atfork_prepare (void) 464static int ecb_cold
465etp_init (void (*want_poll)(void), void (*done_poll)(void))
374{ 466{
375 X_LOCK (wrklock); 467 X_MUTEX_CREATE (wrklock);
376 X_LOCK (reqlock); 468 X_MUTEX_CREATE (reslock);
377 X_LOCK (reslock); 469 X_MUTEX_CREATE (reqlock);
378#if !HAVE_PREADWRITE 470 X_COND_CREATE (reqwait);
379 X_LOCK (preadwritelock);
380#endif
381}
382 471
383static void etp_atfork_parent (void) 472 reqq_init (&req_queue);
384{ 473 reqq_init (&res_queue);
385#if !HAVE_PREADWRITE
386 X_UNLOCK (preadwritelock);
387#endif
388 X_UNLOCK (reslock);
389 X_UNLOCK (reqlock);
390 X_UNLOCK (wrklock);
391}
392 474
393static void etp_atfork_child (void) 475 wrk_first.next =
394{ 476 wrk_first.prev = &wrk_first;
395 ETP_REQ *prv;
396
397 while ((prv = reqq_shift (&req_queue)))
398 ETP_DESTROY (prv);
399
400 while ((prv = reqq_shift (&res_queue)))
401 ETP_DESTROY (prv);
402
403 while (wrk_first.next != &wrk_first)
404 {
405 etp_worker *wrk = wrk_first.next;
406
407 if (wrk->req)
408 ETP_DESTROY (wrk->req);
409
410 etp_worker_clear (wrk);
411 etp_worker_free (wrk);
412 }
413 477
414 started = 0; 478 started = 0;
415 idle = 0; 479 idle = 0;
416 nreqs = 0; 480 nreqs = 0;
417 nready = 0; 481 nready = 0;
418 npending = 0; 482 npending = 0;
419 483
420 etp_atfork_parent ();
421}
422
423static void
424etp_once_init (void)
425{
426 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
427}
428
429static int
430etp_init (void (*want_poll)(void), void (*done_poll)(void))
431{
432 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
433
434 pthread_once (&doinit, etp_once_init);
435
436 want_poll_cb = want_poll; 484 want_poll_cb = want_poll;
437 done_poll_cb = done_poll; 485 done_poll_cb = done_poll;
438 486
439 return 0; 487 return 0;
440} 488}
441 489
442X_THREAD_PROC (etp_proc); 490X_THREAD_PROC (etp_proc);
443 491
492static void ecb_cold
444static void etp_start_thread (void) 493etp_start_thread (void)
445{ 494{
446 etp_worker *wrk = calloc (1, sizeof (etp_worker)); 495 etp_worker *wrk = calloc (1, sizeof (etp_worker));
447 496
448 /*TODO*/ 497 /*TODO*/
449 assert (("unable to allocate worker thread data", wrk)); 498 assert (("unable to allocate worker thread data", wrk));
462 free (wrk); 511 free (wrk);
463 512
464 X_UNLOCK (wrklock); 513 X_UNLOCK (wrklock);
465} 514}
466 515
516static void
467static void etp_maybe_start_thread (void) 517etp_maybe_start_thread (void)
468{ 518{
469 if (expect_true (etp_nthreads () >= wanted)) 519 if (ecb_expect_true (etp_nthreads () >= wanted))
470 return; 520 return;
471 521
472 /* todo: maybe use idle here, but might be less exact */ 522 /* todo: maybe use idle here, but might be less exact */
473 if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ())) 523 if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
474 return; 524 return;
475 525
476 etp_start_thread (); 526 etp_start_thread ();
477} 527}
478 528
529static void ecb_cold
479static void etp_end_thread (void) 530etp_end_thread (void)
480{ 531{
481 eio_req *req = calloc (1, sizeof (eio_req)); 532 eio_req *req = calloc (1, sizeof (eio_req));
482 533
483 req->type = -1; 534 req->type = -1;
484 req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 535 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
491 X_LOCK (wrklock); 542 X_LOCK (wrklock);
492 --started; 543 --started;
493 X_UNLOCK (wrklock); 544 X_UNLOCK (wrklock);
494} 545}
495 546
496static int etp_poll (void) 547static int
548etp_poll (void)
497{ 549{
498 unsigned int maxreqs; 550 unsigned int maxreqs;
499 unsigned int maxtime; 551 unsigned int maxtime;
500 struct timeval tv_start, tv_now; 552 struct timeval tv_start, tv_now;
501 553
531 583
532 X_LOCK (reqlock); 584 X_LOCK (reqlock);
533 --nreqs; 585 --nreqs;
534 X_UNLOCK (reqlock); 586 X_UNLOCK (reqlock);
535 587
536 if (expect_false (req->type == EIO_GROUP && req->size)) 588 if (ecb_expect_false (req->type == EIO_GROUP && req->size))
537 { 589 {
538 req->int1 = 1; /* mark request as delayed */ 590 req->int1 = 1; /* mark request as delayed */
539 continue; 591 continue;
540 } 592 }
541 else 593 else
542 { 594 {
543 int res = ETP_FINISH (req); 595 int res = ETP_FINISH (req);
544 if (expect_false (res)) 596 if (ecb_expect_false (res))
545 return res; 597 return res;
546 } 598 }
547 599
548 if (expect_false (maxreqs && !--maxreqs)) 600 if (ecb_expect_false (maxreqs && !--maxreqs))
549 break; 601 break;
550 602
551 if (maxtime) 603 if (maxtime)
552 { 604 {
553 gettimeofday (&tv_now, 0); 605 gettimeofday (&tv_now, 0);
559 611
560 errno = EAGAIN; 612 errno = EAGAIN;
561 return -1; 613 return -1;
562} 614}
563 615
616static void
564static void etp_cancel (ETP_REQ *req) 617etp_cancel (ETP_REQ *req)
565{ 618{
566 X_LOCK (wrklock); 619 req->cancelled = 1;
567 req->flags |= EIO_FLAG_CANCELLED;
568 X_UNLOCK (wrklock);
569 620
570 eio_grp_cancel (req); 621 eio_grp_cancel (req);
571} 622}
572 623
624static void
573static void etp_submit (ETP_REQ *req) 625etp_submit (ETP_REQ *req)
574{ 626{
575 req->pri -= ETP_PRI_MIN; 627 req->pri -= ETP_PRI_MIN;
576 628
577 if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN; 629 if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
578 if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 630 if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
579 631
580 if (expect_false (req->type == EIO_GROUP)) 632 if (ecb_expect_false (req->type == EIO_GROUP))
581 { 633 {
582 /* I hope this is worth it :/ */ 634 /* I hope this is worth it :/ */
583 X_LOCK (reqlock); 635 X_LOCK (reqlock);
584 ++nreqs; 636 ++nreqs;
585 X_UNLOCK (reqlock); 637 X_UNLOCK (reqlock);
604 656
605 etp_maybe_start_thread (); 657 etp_maybe_start_thread ();
606 } 658 }
607} 659}
608 660
661static void ecb_cold
609static void etp_set_max_poll_time (double nseconds) 662etp_set_max_poll_time (double nseconds)
610{ 663{
611 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 664 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
612 max_poll_time = nseconds * EIO_TICKS; 665 max_poll_time = nseconds * EIO_TICKS;
613 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 666 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
614} 667}
615 668
669static void ecb_cold
616static void etp_set_max_poll_reqs (unsigned int maxreqs) 670etp_set_max_poll_reqs (unsigned int maxreqs)
617{ 671{
618 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 672 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
619 max_poll_reqs = maxreqs; 673 max_poll_reqs = maxreqs;
620 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 674 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
621} 675}
622 676
677static void ecb_cold
623static void etp_set_max_idle (unsigned int nthreads) 678etp_set_max_idle (unsigned int nthreads)
624{ 679{
625 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 680 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
626 max_idle = nthreads <= 0 ? 1 : nthreads; 681 max_idle = nthreads;
627 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 682 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
628} 683}
629 684
685static void ecb_cold
686etp_set_idle_timeout (unsigned int seconds)
687{
688 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
689 idle_timeout = seconds;
690 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
691}
692
693static void ecb_cold
630static void etp_set_min_parallel (unsigned int nthreads) 694etp_set_min_parallel (unsigned int nthreads)
631{ 695{
632 if (wanted < nthreads) 696 if (wanted < nthreads)
633 wanted = nthreads; 697 wanted = nthreads;
634} 698}
635 699
700static void ecb_cold
636static void etp_set_max_parallel (unsigned int nthreads) 701etp_set_max_parallel (unsigned int nthreads)
637{ 702{
638 if (wanted > nthreads) 703 if (wanted > nthreads)
639 wanted = nthreads; 704 wanted = nthreads;
640 705
641 while (started > wanted) 706 while (started > wanted)
642 etp_end_thread (); 707 etp_end_thread ();
643} 708}
644 709
645/*****************************************************************************/ 710/*****************************************************************************/
646 711
712static void
647static void grp_try_feed (eio_req *grp) 713grp_try_feed (eio_req *grp)
648{ 714{
649 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 715 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
650 { 716 {
651 grp->flags &= ~EIO_FLAG_GROUPADD; 717 grp->flags &= ~EIO_FLAG_GROUPADD;
652 718
659 break; 725 break;
660 } 726 }
661 } 727 }
662} 728}
663 729
730static int
664static int grp_dec (eio_req *grp) 731grp_dec (eio_req *grp)
665{ 732{
666 --grp->size; 733 --grp->size;
667 734
668 /* call feeder, if applicable */ 735 /* call feeder, if applicable */
669 grp_try_feed (grp); 736 grp_try_feed (grp);
673 return eio_finish (grp); 740 return eio_finish (grp);
674 else 741 else
675 return 0; 742 return 0;
676} 743}
677 744
745static void
678void eio_destroy (eio_req *req) 746eio_destroy (eio_req *req)
679{ 747{
680 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); 748 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
681 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); 749 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
682 750
683 EIO_DESTROY (req); 751 EIO_DESTROY (req);
684} 752}
685 753
754static int
686static int eio_finish (eio_req *req) 755eio_finish (eio_req *req)
687{ 756{
688 int res = EIO_FINISH (req); 757 int res = EIO_FINISH (req);
689 758
690 if (req->grp) 759 if (req->grp)
691 { 760 {
699 if (grp->grp_first == req) 768 if (grp->grp_first == req)
700 grp->grp_first = req->grp_next; 769 grp->grp_first = req->grp_next;
701 770
702 res2 = grp_dec (grp); 771 res2 = grp_dec (grp);
703 772
704 if (!res && res2) 773 if (!res)
705 res = res2; 774 res = res2;
706 } 775 }
707 776
708 eio_destroy (req); 777 eio_destroy (req);
709 778
710 return res; 779 return res;
711} 780}
712 781
782void
713void eio_grp_cancel (eio_req *grp) 783eio_grp_cancel (eio_req *grp)
714{ 784{
715 for (grp = grp->grp_first; grp; grp = grp->grp_next) 785 for (grp = grp->grp_first; grp; grp = grp->grp_next)
716 eio_cancel (grp); 786 eio_cancel (grp);
717} 787}
718 788
789void
719void eio_cancel (eio_req *req) 790eio_cancel (eio_req *req)
720{ 791{
721 etp_cancel (req); 792 etp_cancel (req);
722} 793}
723 794
795void
724void eio_submit (eio_req *req) 796eio_submit (eio_req *req)
725{ 797{
726 etp_submit (req); 798 etp_submit (req);
727} 799}
728 800
729unsigned int eio_nreqs (void) 801unsigned int
802eio_nreqs (void)
730{ 803{
731 return etp_nreqs (); 804 return etp_nreqs ();
732} 805}
733 806
734unsigned int eio_nready (void) 807unsigned int
808eio_nready (void)
735{ 809{
736 return etp_nready (); 810 return etp_nready ();
737} 811}
738 812
739unsigned int eio_npending (void) 813unsigned int
814eio_npending (void)
740{ 815{
741 return etp_npending (); 816 return etp_npending ();
742} 817}
743 818
744unsigned int eio_nthreads (void) 819unsigned int ecb_cold
820eio_nthreads (void)
745{ 821{
746 return etp_nthreads (); 822 return etp_nthreads ();
747} 823}
748 824
825void ecb_cold
749void eio_set_max_poll_time (double nseconds) 826eio_set_max_poll_time (double nseconds)
750{ 827{
751 etp_set_max_poll_time (nseconds); 828 etp_set_max_poll_time (nseconds);
752} 829}
753 830
831void ecb_cold
754void eio_set_max_poll_reqs (unsigned int maxreqs) 832eio_set_max_poll_reqs (unsigned int maxreqs)
755{ 833{
756 etp_set_max_poll_reqs (maxreqs); 834 etp_set_max_poll_reqs (maxreqs);
757} 835}
758 836
837void ecb_cold
759void eio_set_max_idle (unsigned int nthreads) 838eio_set_max_idle (unsigned int nthreads)
760{ 839{
761 etp_set_max_idle (nthreads); 840 etp_set_max_idle (nthreads);
762} 841}
763 842
843void ecb_cold
844eio_set_idle_timeout (unsigned int seconds)
845{
846 etp_set_idle_timeout (seconds);
847}
848
849void ecb_cold
764void eio_set_min_parallel (unsigned int nthreads) 850eio_set_min_parallel (unsigned int nthreads)
765{ 851{
766 etp_set_min_parallel (nthreads); 852 etp_set_min_parallel (nthreads);
767} 853}
768 854
855void ecb_cold
769void eio_set_max_parallel (unsigned int nthreads) 856eio_set_max_parallel (unsigned int nthreads)
770{ 857{
771 etp_set_max_parallel (nthreads); 858 etp_set_max_parallel (nthreads);
772} 859}
773 860
774int eio_poll (void) 861int eio_poll (void)
783# undef pread 870# undef pread
784# undef pwrite 871# undef pwrite
785# define pread eio__pread 872# define pread eio__pread
786# define pwrite eio__pwrite 873# define pwrite eio__pwrite
787 874
788static ssize_t 875static eio_ssize_t
789eio__pread (int fd, void *buf, size_t count, off_t offset) 876eio__pread (int fd, void *buf, size_t count, off_t offset)
790{ 877{
791 ssize_t res; 878 eio_ssize_t res;
792 off_t ooffset; 879 off_t ooffset;
793 880
794 X_LOCK (preadwritelock); 881 X_LOCK (preadwritelock);
795 ooffset = lseek (fd, 0, SEEK_CUR); 882 ooffset = lseek (fd, 0, SEEK_CUR);
796 lseek (fd, offset, SEEK_SET); 883 lseek (fd, offset, SEEK_SET);
799 X_UNLOCK (preadwritelock); 886 X_UNLOCK (preadwritelock);
800 887
801 return res; 888 return res;
802} 889}
803 890
804static ssize_t 891static eio_ssize_t
805eio__pwrite (int fd, void *buf, size_t count, off_t offset) 892eio__pwrite (int fd, void *buf, size_t count, off_t offset)
806{ 893{
807 ssize_t res; 894 eio_ssize_t res;
808 off_t ooffset; 895 off_t ooffset;
809 896
810 X_LOCK (preadwritelock); 897 X_LOCK (preadwritelock);
811 ooffset = lseek (fd, 0, SEEK_CUR); 898 ooffset = lseek (fd, 0, SEEK_CUR);
812 lseek (fd, offset, SEEK_SET); 899 lseek (fd, offset, SEEK_SET);
844#ifndef HAVE_FUTIMES 931#ifndef HAVE_FUTIMES
845 932
846# undef futimes 933# undef futimes
847# define futimes(fd,times) eio__futimes (fd, times) 934# define futimes(fd,times) eio__futimes (fd, times)
848 935
936static int
849static int eio__futimes (int fd, const struct timeval tv[2]) 937eio__futimes (int fd, const struct timeval tv[2])
850{ 938{
851 errno = ENOSYS; 939 errno = ENOSYS;
852 return -1; 940 return -1;
853} 941}
854 942
857#if !HAVE_FDATASYNC 945#if !HAVE_FDATASYNC
858# undef fdatasync 946# undef fdatasync
859# define fdatasync(fd) fsync (fd) 947# define fdatasync(fd) fsync (fd)
860#endif 948#endif
861 949
950static int
951eio__syncfs (int fd)
952{
953 int res;
954
955#if HAVE_SYS_SYNCFS
956 res = (int)syscall (__NR_syncfs, (int)(fd));
957#else
958 res = -1;
959 errno = ENOSYS;
960#endif
961
962 if (res < 0 && errno == ENOSYS && fd >= 0)
963 sync ();
964
965 return res;
966}
967
862/* sync_file_range always needs emulation */ 968/* sync_file_range always needs emulation */
863int 969static int
864eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags) 970eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
865{ 971{
866#if HAVE_SYNC_FILE_RANGE 972#if HAVE_SYNC_FILE_RANGE
867 int res; 973 int res;
868 974
885 /* even though we could play tricks with the flags, it's better to always 991 /* even though we could play tricks with the flags, it's better to always
886 * call fdatasync, as that matches the expectation of its users best */ 992 * call fdatasync, as that matches the expectation of its users best */
887 return fdatasync (fd); 993 return fdatasync (fd);
888} 994}
889 995
996static int
997eio__fallocate (int fd, int mode, off_t offset, size_t len)
998{
999#if HAVE_FALLOCATE
1000 return fallocate (fd, mode, offset, len);
1001#else
1002 errno = ENOSYS;
1003 return -1;
1004#endif
1005}
1006
890#if !HAVE_READAHEAD 1007#if !HAVE_READAHEAD
891# undef readahead 1008# undef readahead
892# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 1009# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
893 1010
894static ssize_t 1011static eio_ssize_t
895eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) 1012eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
896{ 1013{
897 size_t todo = count; 1014 size_t todo = count;
898 dBUF; 1015 dBUF;
899 1016
904 pread (fd, eio_buf, len, offset); 1021 pread (fd, eio_buf, len, offset);
905 offset += len; 1022 offset += len;
906 todo -= len; 1023 todo -= len;
907 } 1024 }
908 1025
1026 FUBd;
1027
909 errno = 0; 1028 errno = 0;
910 return count; 1029 return count;
911} 1030}
912 1031
913#endif 1032#endif
914 1033
915/* sendfile always needs emulation */ 1034/* sendfile always needs emulation */
916static ssize_t 1035static eio_ssize_t
917eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) 1036eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
918{ 1037{
1038 eio_ssize_t written = 0;
919 ssize_t res; 1039 eio_ssize_t res;
920 1040
921 if (!count) 1041 if (!count)
922 return 0; 1042 return 0;
923 1043
1044 for (;;)
1045 {
1046#ifdef __APPLE__
1047# undef HAVE_SENDFILE /* broken, as everything on os x */
1048#endif
924#if HAVE_SENDFILE 1049#if HAVE_SENDFILE
925# if __linux 1050# if __linux
1051 off_t soffset = offset;
926 res = sendfile (ofd, ifd, &offset, count); 1052 res = sendfile (ofd, ifd, &soffset, count);
927 1053
928# elif __FreeBSD__ 1054# elif __FreeBSD__
929 /* 1055 /*
930 * Of course, the freebsd sendfile is a dire hack with no thoughts 1056 * Of course, the freebsd sendfile is a dire hack with no thoughts
931 * wasted on making it similar to other I/O functions. 1057 * wasted on making it similar to other I/O functions.
932 */ 1058 */
933 {
934 off_t sbytes; 1059 off_t sbytes;
935 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 1060 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
936 1061
937 #if 0 /* according to the manpage, this is correct, but broken behaviour */ 1062 #if 0 /* according to the manpage, this is correct, but broken behaviour */
938 /* freebsd' sendfile will return 0 on success */ 1063 /* freebsd' sendfile will return 0 on success */
939 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */ 1064 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
940 /* not on e.g. EIO or EPIPE - sounds broken */ 1065 /* not on e.g. EIO or EPIPE - sounds broken */
941 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0) 1066 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
942 res = sbytes; 1067 res = sbytes;
943 #endif 1068 #endif
944 1069
945 /* according to source inspection, this is correct, and useful behaviour */ 1070 /* according to source inspection, this is correct, and useful behaviour */
946 if (sbytes) 1071 if (sbytes)
947 res = sbytes; 1072 res = sbytes;
948 }
949 1073
950# elif defined (__APPLE__) 1074# elif defined (__APPLE__)
951
952 {
953 off_t sbytes = count; 1075 off_t sbytes = count;
954 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0); 1076 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
955 1077
956 /* according to the manpage, sbytes is always valid */ 1078 /* according to the manpage, sbytes is always valid */
957 if (sbytes) 1079 if (sbytes)
958 res = sbytes; 1080 res = sbytes;
959 }
960 1081
961# elif __hpux 1082# elif __hpux
962 res = sendfile (ofd, ifd, offset, count, 0, 0); 1083 res = sendfile (ofd, ifd, offset, count, 0, 0);
963 1084
964# elif __solaris 1085# elif __solaris
965 {
966 struct sendfilevec vec; 1086 struct sendfilevec vec;
967 size_t sbytes; 1087 size_t sbytes;
968 1088
969 vec.sfv_fd = ifd; 1089 vec.sfv_fd = ifd;
970 vec.sfv_flag = 0; 1090 vec.sfv_flag = 0;
971 vec.sfv_off = offset; 1091 vec.sfv_off = offset;
972 vec.sfv_len = count; 1092 vec.sfv_len = count;
973 1093
974 res = sendfilev (ofd, &vec, 1, &sbytes); 1094 res = sendfilev (ofd, &vec, 1, &sbytes);
975 1095
976 if (res < 0 && sbytes) 1096 if (res < 0 && sbytes)
977 res = sbytes; 1097 res = sbytes;
978 }
979 1098
980# endif 1099# endif
981 1100
982#elif defined (_WIN32) 1101#elif defined (_WIN32) && 0
983
984 /* does not work, just for documentation of what would need to be done */ 1102 /* does not work, just for documentation of what would need to be done */
985 { 1103 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
1104 /* libeio guarantees that the file offset does not change, and windows */
1105 /* has no way to get an independent handle to the same file description */
986 HANDLE h = TO_SOCKET (ifd); 1106 HANDLE h = TO_SOCKET (ifd);
987 SetFilePointer (h, offset, 0, FILE_BEGIN); 1107 SetFilePointer (h, offset, 0, FILE_BEGIN);
988 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0); 1108 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
989 }
990 1109
991#else 1110#else
992 res = -1; 1111 res = -1;
993 errno = ENOSYS; 1112 errno = ENOSYS;
994#endif 1113#endif
995 1114
1115 /* we assume sendfile can copy at least 128mb in one go */
1116 if (res <= 128 * 1024 * 1024)
1117 {
1118 if (res > 0)
1119 written += res;
1120
1121 if (written)
1122 return written;
1123
1124 break;
1125 }
1126 else
1127 {
1128 /* if we requested more, then probably the kernel was lazy */
1129 written += res;
1130 offset += res;
1131 count -= res;
1132
1133 if (!count)
1134 return written;
1135 }
1136 }
1137
996 if (res < 0 1138 if (res < 0
997 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1139 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
998 /* BSDs */ 1140 /* BSDs */
999#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */ 1141#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1000 || errno == ENOTSUP 1142 || errno == ENOTSUP
1001#endif 1143#endif
1144#ifdef EOPNOTSUPP /* windows */
1002 || errno == EOPNOTSUPP /* BSDs */ 1145 || errno == EOPNOTSUPP /* BSDs */
1146#endif
1003#if __solaris 1147#if __solaris
1004 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1148 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
1005#endif 1149#endif
1006 ) 1150 )
1007 ) 1151 )
1011 1155
1012 res = 0; 1156 res = 0;
1013 1157
1014 while (count) 1158 while (count)
1015 { 1159 {
1016 ssize_t cnt; 1160 eio_ssize_t cnt;
1017 1161
1018 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 1162 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
1019 1163
1020 if (cnt <= 0) 1164 if (cnt <= 0)
1021 { 1165 {
1033 1177
1034 offset += cnt; 1178 offset += cnt;
1035 res += cnt; 1179 res += cnt;
1036 count -= cnt; 1180 count -= cnt;
1037 } 1181 }
1182
1183 FUBd;
1038 } 1184 }
1039 1185
1040 return res; 1186 return res;
1187}
1188
1189#ifdef PAGESIZE
1190# define eio_pagesize() PAGESIZE
1191#else
1192static intptr_t
1193eio_pagesize (void)
1194{
1195 static intptr_t page;
1196
1197 if (!page)
1198 page = sysconf (_SC_PAGESIZE);
1199
1200 return page;
1201}
1202#endif
1203
1204static void
1205eio_page_align (void **addr, size_t *length)
1206{
1207 intptr_t mask = eio_pagesize () - 1;
1208
1209 /* round down addr */
1210 intptr_t adj = mask & (intptr_t)*addr;
1211
1212 *addr = (void *)((intptr_t)*addr - adj);
1213 *length += adj;
1214
1215 /* round up length */
1216 *length = (*length + mask) & ~mask;
1217}
1218
1219#if !_POSIX_MEMLOCK
1220# define eio__mlockall(a) EIO_ENOSYS ()
1221#else
1222
1223static int
1224eio__mlockall (int flags)
1225{
1226 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1227 extern int mallopt (int, int);
1228 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1229 #endif
1230
1231 if (EIO_MCL_CURRENT != MCL_CURRENT
1232 || EIO_MCL_FUTURE != MCL_FUTURE)
1233 {
1234 flags = 0
1235 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1236 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1237 }
1238
1239 return mlockall (flags);
1240}
1241#endif
1242
1243#if !_POSIX_MEMLOCK_RANGE
1244# define eio__mlock(a,b) EIO_ENOSYS ()
1245#else
1246
1247static int
1248eio__mlock (void *addr, size_t length)
1249{
1250 eio_page_align (&addr, &length);
1251
1252 return mlock (addr, length);
1253}
1254
1255#endif
1256
1257#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1258# define eio__msync(a,b,c) EIO_ENOSYS ()
1259#else
1260
1261static int
1262eio__msync (void *mem, size_t len, int flags)
1263{
1264 eio_page_align (&mem, &len);
1265
1266 if (EIO_MS_ASYNC != MS_SYNC
1267 || EIO_MS_INVALIDATE != MS_INVALIDATE
1268 || EIO_MS_SYNC != MS_SYNC)
1269 {
1270 flags = 0
1271 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1272 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1273 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1274 }
1275
1276 return msync (mem, len, flags);
1277}
1278
1279#endif
1280
1281static int
1282eio__mtouch (eio_req *req)
1283{
1284 void *mem = req->ptr2;
1285 size_t len = req->size;
1286 int flags = req->int1;
1287
1288 eio_page_align (&mem, &len);
1289
1290 {
1291 intptr_t addr = (intptr_t)mem;
1292 intptr_t end = addr + len;
1293 intptr_t page = eio_pagesize ();
1294
1295 if (addr < end)
1296 if (flags & EIO_MT_MODIFY) /* modify */
1297 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1298 else
1299 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1300 }
1301
1302 return 0;
1303}
1304
1305/*****************************************************************************/
1306/* requests implemented outside eio_execute, because they are so large */
1307
1308static void
1309eio__realpath (eio_req *req, etp_worker *self)
1310{
1311 char *rel = req->ptr1;
1312 char *res;
1313 char *tmp1, *tmp2;
1314#if SYMLOOP_MAX > 32
1315 int symlinks = SYMLOOP_MAX;
1316#else
1317 int symlinks = 32;
1318#endif
1319
1320 req->result = -1;
1321
1322 errno = EINVAL;
1323 if (!rel)
1324 return;
1325
1326 errno = ENOENT;
1327 if (!*rel)
1328 return;
1329
1330 if (!req->ptr2)
1331 {
1332 X_LOCK (wrklock);
1333 req->flags |= EIO_FLAG_PTR2_FREE;
1334 X_UNLOCK (wrklock);
1335 req->ptr2 = malloc (PATH_MAX * 3);
1336
1337 errno = ENOMEM;
1338 if (!req->ptr2)
1339 return;
1340 }
1341
1342 res = req->ptr2;
1343 tmp1 = res + PATH_MAX;
1344 tmp2 = tmp1 + PATH_MAX;
1345
1346#if 0 /* disabled, the musl way to do things is just too racy */
1347#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1348 /* on linux we may be able to ask the kernel */
1349 {
1350 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1351
1352 if (fd >= 0)
1353 {
1354 sprintf (tmp1, "/proc/self/fd/%d", fd);
1355 req->result = readlink (tmp1, res, PATH_MAX);
1356 close (fd);
1357
1358 /* here we should probably stat the open file and the disk file, to make sure they still match */
1359
1360 if (req->result > 0)
1361 goto done;
1362 }
1363 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1364 return;
1365 }
1366#endif
1367#endif
1368
1369 if (*rel != '/')
1370 {
1371 if (!getcwd (res, PATH_MAX))
1372 return;
1373
1374 if (res [1]) /* only use if not / */
1375 res += strlen (res);
1376 }
1377
1378 while (*rel)
1379 {
1380 eio_ssize_t len, linklen;
1381 char *beg = rel;
1382
1383 while (*rel && *rel != '/')
1384 ++rel;
1385
1386 len = rel - beg;
1387
1388 if (!len) /* skip slashes */
1389 {
1390 ++rel;
1391 continue;
1392 }
1393
1394 if (beg [0] == '.')
1395 {
1396 if (len == 1)
1397 continue; /* . - nop */
1398
1399 if (beg [1] == '.' && len == 2)
1400 {
1401 /* .. - back up one component, if possible */
1402
1403 while (res != req->ptr2)
1404 if (*--res == '/')
1405 break;
1406
1407 continue;
1408 }
1409 }
1410
1411 errno = ENAMETOOLONG;
1412 if (res + 1 + len + 1 >= tmp1)
1413 return;
1414
1415 /* copy one component */
1416 *res = '/';
1417 memcpy (res + 1, beg, len);
1418
1419 /* zero-terminate, for readlink */
1420 res [len + 1] = 0;
1421
1422 /* now check if it's a symlink */
1423 linklen = readlink (req->ptr2, tmp1, PATH_MAX);
1424
1425 if (linklen < 0)
1426 {
1427 if (errno != EINVAL)
1428 return;
1429
1430 /* it's a normal directory. hopefully */
1431 res += len + 1;
1432 }
1433 else
1434 {
1435 /* yay, it was a symlink - build new path in tmp2 */
1436 int rellen = strlen (rel);
1437
1438 errno = ENAMETOOLONG;
1439 if (linklen + 1 + rellen >= PATH_MAX)
1440 return;
1441
1442 errno = ELOOP;
1443 if (!--symlinks)
1444 return;
1445
1446 if (*tmp1 == '/')
1447 res = req->ptr2; /* symlink resolves to an absolute path */
1448
1449 /* we need to be careful, as rel might point into tmp2 already */
1450 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1451 tmp2 [linklen] = '/';
1452 memcpy (tmp2, tmp1, linklen);
1453
1454 rel = tmp2;
1455 }
1456 }
1457
1458 /* special case for the lone root path */
1459 if (res == req->ptr2)
1460 *res++ = '/';
1461
1462 req->result = res - (char *)req->ptr2;
1463
1464done:
1465 req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
1041} 1466}
1042 1467
1043static signed char 1468static signed char
1044eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) 1469eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1045{ 1470{
1046 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */ 1471 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1047 : a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0; 1472 : a->inode < b->inode ? -1
1473 : a->inode > b->inode ? 1
1474 : 0;
1048} 1475}
1049 1476
1050#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0 1477#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1051 1478
1052#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */ 1479#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1053#define EIO_SORT_FAST 60 /* when to only use insertion sort */ 1480#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1054 1481
1055static void 1482static void
1056eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1483eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1057{ 1484{
1058 unsigned char bits [9 + sizeof (ino_t) * 8]; 1485 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1059 unsigned char *bit = bits; 1486 unsigned char *bit = bits;
1060 1487
1061 assert (CHAR_BIT == 8); 1488 assert (CHAR_BIT == 8);
1062 assert (sizeof (eio_dirent) * 8 < 256); 1489 assert (sizeof (eio_dirent) * 8 < 256);
1063 assert (offsetof (eio_dirent, inode)); /* we use 0 as sentinel */ 1490 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1064 assert (offsetof (eio_dirent, score)); /* we use 0 as sentinel */ 1491 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1065 1492
1066 if (size <= EIO_SORT_FAST) 1493 if (size <= EIO_SORT_FAST)
1067 return; 1494 return;
1068 1495
1069 /* first prepare an array of bits to test in our radix sort */ 1496 /* first prepare an array of bits to test in our radix sort */
1070 /* try to take endianness into account, as well as differences in ino_t sizes */ 1497 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1071 /* inode_bits must contain all inodes ORed together */ 1498 /* inode_bits must contain all inodes ORed together */
1072 /* which is used to skip bits that are 0 everywhere, which is very common */ 1499 /* which is used to skip bits that are 0 everywhere, which is very common */
1073 { 1500 {
1074 ino_t endianness; 1501 eio_ino_t endianness;
1075 int i, j; 1502 int i, j;
1076 1503
1077 /* we store the byte offset of byte n into byte n of "endianness" */ 1504 /* we store the byte offset of byte n into byte n of "endianness" */
1078 for (i = 0; i < sizeof (ino_t); ++i) 1505 for (i = 0; i < sizeof (eio_ino_t); ++i)
1079 ((unsigned char *)&endianness)[i] = i; 1506 ((unsigned char *)&endianness)[i] = i;
1080 1507
1081 *bit++ = 0; 1508 *bit++ = 0;
1082 1509
1083 for (i = 0; i < sizeof (ino_t); ++i) 1510 for (i = 0; i < sizeof (eio_ino_t); ++i)
1084 { 1511 {
1085 /* shifting off the byte offsets out of "endianness" */ 1512 /* shifting off the byte offsets out of "endianness" */
1086 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8; 1513 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1087 endianness >>= 8; 1514 endianness >>= 8;
1088 1515
1089 for (j = 0; j < 8; ++j) 1516 for (j = 0; j < 8; ++j)
1090 if (inode_bits & (((ino_t)1) << (i * 8 + j))) 1517 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1091 *bit++ = offs + j; 1518 *bit++ = offs + j;
1092 } 1519 }
1093 1520
1094 for (j = 0; j < 8; ++j) 1521 for (j = 0; j < 8; ++j)
1095 if (score_bits & (1 << j)) 1522 if (score_bits & (1 << j))
1096 *bit++ = offsetof (eio_dirent, score) * 8 + j; 1523 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1097 } 1524 }
1098 1525
1099 /* now actually do the sorting (a variant of MSD radix sort) */ 1526 /* now actually do the sorting (a variant of MSD radix sort) */
1100 { 1527 {
1101 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base; 1528 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1102 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end; 1529 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1103 unsigned char *bit_stk [9 + sizeof (ino_t) * 8]; 1530 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1104 int stk_idx = 0; 1531 int stk_idx = 0;
1105 1532
1106 base_stk [stk_idx] = dents; 1533 base_stk [stk_idx] = dents;
1107 end_stk [stk_idx] = dents + size; 1534 end_stk [stk_idx] = dents + size;
1108 bit_stk [stk_idx] = bit - 1; 1535 bit_stk [stk_idx] = bit - 1;
1187 } 1614 }
1188 } 1615 }
1189} 1616}
1190 1617
1191static void 1618static void
1192eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1619eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1193{ 1620{
1194 if (size <= 1) 1621 if (size <= 1)
1195 return; /* our insertion sort relies on size > 0 */ 1622 return; /* our insertion sort relies on size > 0 */
1196 1623
1197 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */ 1624 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1205 1632
1206/* read a full directory */ 1633/* read a full directory */
1207static void 1634static void
1208eio__scandir (eio_req *req, etp_worker *self) 1635eio__scandir (eio_req *req, etp_worker *self)
1209{ 1636{
1210 DIR *dirp;
1211 EIO_STRUCT_DIRENT *entp;
1212 char *name, *names; 1637 char *name, *names;
1213 int namesalloc = 4096; 1638 int namesalloc = 4096 - sizeof (void *) * 4;
1214 int namesoffs = 0; 1639 int namesoffs = 0;
1215 int flags = req->int1; 1640 int flags = req->int1;
1216 eio_dirent *dents = 0; 1641 eio_dirent *dents = 0;
1217 int dentalloc = 128; 1642 int dentalloc = 128;
1218 int dentoffs = 0; 1643 int dentoffs = 0;
1219 ino_t inode_bits = 0; 1644 eio_ino_t inode_bits = 0;
1645#ifdef _WIN32
1646 HANDLE dirp;
1647 WIN32_FIND_DATA entp;
1648#else
1649 DIR *dirp;
1650 EIO_STRUCT_DIRENT *entp;
1651#endif
1220 1652
1221 req->result = -1; 1653 req->result = -1;
1222 1654
1223 if (!(flags & EIO_READDIR_DENTS)) 1655 if (!(flags & EIO_READDIR_DENTS))
1224 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); 1656 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1225 1657
1226 X_LOCK (wrklock); 1658#ifdef _WIN32
1227 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 1659 {
1660 int len = strlen ((const char *)req->ptr1);
1661 char *path = malloc (MAX_PATH);
1662 const char *fmt;
1663
1664 if (!len)
1665 fmt = "./*";
1666 else if (((const char *)req->ptr1)[len - 1] == '/' || ((const char *)req->ptr1)[len - 1] == '\\')
1667 fmt = "%s*";
1668 else
1669 fmt = "%s/*";
1670
1671 _snprintf (path, MAX_PATH, fmt, (const char *)req->ptr1);
1672 dirp = FindFirstFile (path, &entp);
1673 free (path);
1674
1675 if (dirp == INVALID_HANDLE_VALUE)
1676 {
1677 dirp = 0;
1678
1679 switch (GetLastError ())
1680 {
1681 case ERROR_FILE_NOT_FOUND:
1682 req->result = 0;
1683 break;
1684
1685 case ERROR_INVALID_NAME:
1686 case ERROR_PATH_NOT_FOUND:
1687 case ERROR_NO_MORE_FILES:
1688 errno = ENOENT;
1689 break;
1690
1691 case ERROR_NOT_ENOUGH_MEMORY:
1692 errno = ENOMEM;
1693 break;
1694
1695 default:
1696 errno = EINVAL;
1697 break;
1698 }
1699 }
1700 }
1701#else
1228 self->dirp = dirp = opendir (req->ptr1); 1702 dirp = opendir (req->ptr1);
1703#endif
1704
1705 if (req->flags & EIO_FLAG_PTR1_FREE)
1706 free (req->ptr1);
1707
1229 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1708 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1230 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; 1709 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1231 req->ptr2 = names = malloc (namesalloc); 1710 req->ptr2 = names = malloc (namesalloc);
1232 X_UNLOCK (wrklock);
1233 1711
1234 if (dirp && names && (!flags || dents)) 1712 if (dirp && names && (!flags || dents))
1235 for (;;) 1713 for (;;)
1236 { 1714 {
1715 int done;
1716
1717#ifdef _WIN32
1718 done = !dirp;
1719#else
1237 errno = 0; 1720 errno = 0;
1238 entp = readdir (dirp); 1721 entp = readdir (dirp);
1722 done = !entp;
1723#endif
1239 1724
1240 if (!entp) 1725 if (done)
1241 { 1726 {
1727#ifndef _WIN32
1728 int old_errno = errno;
1729 closedir (dirp);
1730 errno = old_errno;
1731
1242 if (errno) 1732 if (errno)
1243 break; 1733 break;
1734#endif
1244 1735
1245 /* sort etc. */ 1736 /* sort etc. */
1246 req->int1 = flags; 1737 req->int1 = flags;
1247 req->result = dentoffs; 1738 req->result = dentoffs;
1248 1739
1249 if (flags & EIO_READDIR_STAT_ORDER) 1740 if (flags & EIO_READDIR_STAT_ORDER)
1250 eio_dent_sort (dents, dentoffs, 0, inode_bits); /* sort by inode exclusively */ 1741 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1251 else if (flags & EIO_READDIR_DIRS_FIRST) 1742 else if (flags & EIO_READDIR_DIRS_FIRST)
1252 if (flags & EIO_READDIR_FOUND_UNKNOWN) 1743 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1253 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */ 1744 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1254 else 1745 else
1255 { 1746 {
1257 eio_dirent *oth = dents + dentoffs; 1748 eio_dirent *oth = dents + dentoffs;
1258 eio_dirent *dir = dents; 1749 eio_dirent *dir = dents;
1259 1750
1260 /* now partition dirs to the front, and non-dirs to the back */ 1751 /* now partition dirs to the front, and non-dirs to the back */
1261 /* by walking from both sides and swapping if necessary */ 1752 /* by walking from both sides and swapping if necessary */
1262 /* also clear score, so it doesn't influence sorting */
1263 while (oth > dir) 1753 while (oth > dir)
1264 { 1754 {
1265 if (dir->type == EIO_DT_DIR) 1755 if (dir->type == EIO_DT_DIR)
1266 ++dir; 1756 ++dir;
1267 else if ((--oth)->type == EIO_DT_DIR) 1757 else if ((--oth)->type == EIO_DT_DIR)
1270 1760
1271 ++dir; 1761 ++dir;
1272 } 1762 }
1273 } 1763 }
1274 1764
1275 /* now sort the dirs only */ 1765 /* now sort the dirs only (dirs all have the same score) */
1276 eio_dent_sort (dents, dir - dents, 0, inode_bits); 1766 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1277 } 1767 }
1278 1768
1279 break; 1769 break;
1280 } 1770 }
1281 1771
1282 /* now add the entry to our list(s) */ 1772 /* now add the entry to our list(s) */
1283 name = entp->d_name; 1773 name = D_NAME (entp);
1284 1774
1285 /* skip . and .. entries */ 1775 /* skip . and .. entries */
1286 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1776 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1287 { 1777 {
1288 int len = D_NAMLEN (entp) + 1; 1778 int len = D_NAMLEN (entp) + 1;
1289 1779
1290 while (expect_false (namesoffs + len > namesalloc)) 1780 while (ecb_expect_false (namesoffs + len > namesalloc))
1291 { 1781 {
1292 namesalloc *= 2; 1782 namesalloc *= 2;
1293 X_LOCK (wrklock);
1294 req->ptr2 = names = realloc (names, namesalloc); 1783 req->ptr2 = names = realloc (names, namesalloc);
1295 X_UNLOCK (wrklock);
1296 1784
1297 if (!names) 1785 if (!names)
1298 break; 1786 break;
1299 } 1787 }
1300 1788
1302 1790
1303 if (dents) 1791 if (dents)
1304 { 1792 {
1305 struct eio_dirent *ent; 1793 struct eio_dirent *ent;
1306 1794
1307 if (expect_false (dentoffs == dentalloc)) 1795 if (ecb_expect_false (dentoffs == dentalloc))
1308 { 1796 {
1309 dentalloc *= 2; 1797 dentalloc *= 2;
1310 X_LOCK (wrklock);
1311 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent)); 1798 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1312 X_UNLOCK (wrklock);
1313 1799
1314 if (!dents) 1800 if (!dents)
1315 break; 1801 break;
1316 } 1802 }
1317 1803
1397 if (EIO_CANCELLED (req)) 1883 if (EIO_CANCELLED (req))
1398 { 1884 {
1399 errno = ECANCELED; 1885 errno = ECANCELED;
1400 break; 1886 break;
1401 } 1887 }
1888
1889#ifdef _WIN32
1890 if (!FindNextFile (dirp, &entp))
1891 {
1892 FindClose (dirp);
1893 dirp = 0;
1894 }
1895#endif
1402 } 1896 }
1403}
1404
1405#ifdef PAGESIZE
1406# define eio_pagesize() PAGESIZE
1407#else
1408static intptr_t
1409eio_pagesize (void)
1410{
1411 static intptr_t page;
1412
1413 if (!page)
1414 page = sysconf (_SC_PAGESIZE);
1415
1416 return page;
1417}
1418#endif
1419
1420static void
1421eio_page_align (void **addr, size_t *length)
1422{
1423 intptr_t mask = eio_pagesize () - 1;
1424
1425 /* round down addr */
1426 intptr_t adj = mask & (intptr_t)*addr;
1427
1428 *addr = (void *)((intptr_t)*addr - adj);
1429 *length += adj;
1430
1431 /* round up length */
1432 *length = (*length + mask) & ~mask;
1433}
1434
1435#if !_POSIX_MEMLOCK
1436# define eio__mlockall(a) ((errno = ENOSYS), -1)
1437#else
1438
1439static int
1440eio__mlockall (int flags)
1441{
1442 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1443 extern int mallopt (int, int);
1444 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1445 #endif
1446
1447 if (EIO_MCL_CURRENT != MCL_CURRENT
1448 || EIO_MCL_FUTURE != MCL_FUTURE)
1449 {
1450 flags = 0
1451 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1452 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1453 }
1454
1455 return mlockall (flags);
1456}
1457#endif
1458
1459#if !_POSIX_MEMLOCK_RANGE
1460# define eio__mlock(a,b) ((errno = ENOSYS), -1)
1461#else
1462
1463static int
1464eio__mlock (void *addr, size_t length)
1465{
1466 eio_page_align (&addr, &length);
1467
1468 return mlock (addr, length);
1469}
1470
1471#endif
1472
1473#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1474# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
1475#else
1476
1477int
1478eio__msync (void *mem, size_t len, int flags)
1479{
1480 eio_page_align (&mem, &len);
1481
1482 if (EIO_MS_ASYNC != MS_SYNC
1483 || EIO_MS_INVALIDATE != MS_INVALIDATE
1484 || EIO_MS_SYNC != MS_SYNC)
1485 {
1486 flags = 0
1487 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1488 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1489 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1490 }
1491
1492 return msync (mem, len, flags);
1493}
1494
1495#endif
1496
1497int
1498eio__mtouch (void *mem, size_t len, int flags)
1499{
1500 eio_page_align (&mem, &len);
1501
1502 {
1503 intptr_t addr = (intptr_t)mem;
1504 intptr_t end = addr + len;
1505 intptr_t page = eio_pagesize ();
1506
1507 if (addr < end)
1508 if (flags & EIO_MT_MODIFY) /* modify */
1509 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
1510 else
1511 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
1512 }
1513
1514 return 0;
1515} 1897}
1516 1898
1517/*****************************************************************************/ 1899/*****************************************************************************/
1518 1900
1519#define ALLOC(len) \ 1901#define ALLOC(len) \
1535{ 1917{
1536 ETP_REQ *req; 1918 ETP_REQ *req;
1537 struct timespec ts; 1919 struct timespec ts;
1538 etp_worker *self = (etp_worker *)thr_arg; 1920 etp_worker *self = (etp_worker *)thr_arg;
1539 1921
1922#if HAVE_PRCTL_SET_NAME
1923 prctl (PR_SET_NAME, (unsigned long)"eio_thread", 0, 0, 0);
1924#endif
1925
1540 /* try to distribute timeouts somewhat randomly */ 1926 /* try to distribute timeouts somewhat evenly */
1541 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 1927 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1542 1928
1543 for (;;) 1929 for (;;)
1544 { 1930 {
1931 ts.tv_sec = 0;
1932
1545 X_LOCK (reqlock); 1933 X_LOCK (reqlock);
1546 1934
1547 for (;;) 1935 for (;;)
1548 { 1936 {
1549 self->req = req = reqq_shift (&req_queue); 1937 self->req = req = reqq_shift (&req_queue);
1550 1938
1551 if (req) 1939 if (req)
1552 break; 1940 break;
1553 1941
1942 if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
1943 {
1944 X_UNLOCK (reqlock);
1945 X_LOCK (wrklock);
1946 --started;
1947 X_UNLOCK (wrklock);
1948 goto quit;
1949 }
1950
1554 ++idle; 1951 ++idle;
1555 1952
1556 ts.tv_sec = time (0) + IDLE_TIMEOUT; 1953 if (idle <= max_idle)
1557 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 1954 /* we are allowed to idle, so do so without any timeout */
1955 X_COND_WAIT (reqwait, reqlock);
1956 else
1558 { 1957 {
1559 if (idle > max_idle) 1958 /* initialise timeout once */
1560 { 1959 if (!ts.tv_sec)
1561 --idle; 1960 ts.tv_sec = time (0) + idle_timeout;
1562 X_UNLOCK (reqlock);
1563 X_LOCK (wrklock);
1564 --started;
1565 X_UNLOCK (wrklock);
1566 goto quit;
1567 }
1568 1961
1569 /* we are allowed to idle, so do so without any timeout */
1570 X_COND_WAIT (reqwait, reqlock); 1962 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
1963 ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1571 } 1964 }
1572 1965
1573 --idle; 1966 --idle;
1574 } 1967 }
1575 1968
1578 X_UNLOCK (reqlock); 1971 X_UNLOCK (reqlock);
1579 1972
1580 if (req->type < 0) 1973 if (req->type < 0)
1581 goto quit; 1974 goto quit;
1582 1975
1583 if (!EIO_CANCELLED (req))
1584 ETP_EXECUTE (self, req); 1976 ETP_EXECUTE (self, req);
1585 1977
1586 X_LOCK (reslock); 1978 X_LOCK (reslock);
1587 1979
1588 ++npending; 1980 ++npending;
1589 1981
1604 return 0; 1996 return 0;
1605} 1997}
1606 1998
1607/*****************************************************************************/ 1999/*****************************************************************************/
1608 2000
2001int ecb_cold
1609int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 2002eio_init (void (*want_poll)(void), void (*done_poll)(void))
1610{ 2003{
2004#if !HAVE_PREADWRITE
2005 X_MUTEX_CREATE (preadwritelock);
2006#endif
2007
1611 return etp_init (want_poll, done_poll); 2008 return etp_init (want_poll, done_poll);
1612} 2009}
1613 2010
2011ecb_inline void
1614static void eio_api_destroy (eio_req *req) 2012eio_api_destroy (eio_req *req)
1615{ 2013{
1616 free (req); 2014 free (req);
1617} 2015}
1618 2016
1619#define REQ(rtype) \ 2017#define REQ(rtype) \
1638 { \ 2036 { \
1639 eio_api_destroy (req); \ 2037 eio_api_destroy (req); \
1640 return 0; \ 2038 return 0; \
1641 } 2039 }
1642 2040
2041static void
1643static void eio_execute (etp_worker *self, eio_req *req) 2042eio_execute (etp_worker *self, eio_req *req)
1644{ 2043{
2044 if (ecb_expect_false (EIO_CANCELLED (req)))
2045 {
2046 req->result = -1;
2047 req->errorno = ECANCELED;
2048 return;
2049 }
2050
1645 switch (req->type) 2051 switch (req->type)
1646 { 2052 {
1647 case EIO_READ: ALLOC (req->size); 2053 case EIO_READ: ALLOC (req->size);
1648 req->result = req->offs >= 0 2054 req->result = req->offs >= 0
1649 ? pread (req->int1, req->ptr2, req->size, req->offs) 2055 ? pread (req->int1, req->ptr2, req->size, req->offs)
1651 case EIO_WRITE: req->result = req->offs >= 0 2057 case EIO_WRITE: req->result = req->offs >= 0
1652 ? pwrite (req->int1, req->ptr2, req->size, req->offs) 2058 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1653 : write (req->int1, req->ptr2, req->size); break; 2059 : write (req->int1, req->ptr2, req->size); break;
1654 2060
1655 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break; 2061 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1656 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break; 2062 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
1657 2063
1658 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2064 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1659 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2065 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1660 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2066 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1661 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2067 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1683 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; 2089 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1684 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; 2090 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1685 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; 2091 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1686 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break; 2092 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1687 2093
2094 case EIO_REALPATH: eio__realpath (req, self); break;
2095
1688 case EIO_READLINK: ALLOC (PATH_MAX); 2096 case EIO_READLINK: ALLOC (PATH_MAX);
1689 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break; 2097 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1690 2098
1691 case EIO_SYNC: req->result = 0; sync (); break; 2099 case EIO_SYNC: req->result = 0; sync (); break;
1692 case EIO_FSYNC: req->result = fsync (req->int1); break; 2100 case EIO_FSYNC: req->result = fsync (req->int1); break;
1693 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 2101 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
2102 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
2103 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1694 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break; 2104 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1695 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break; 2105 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1696 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break; 2106 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1697 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break; 2107 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1698 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break; 2108 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
1699 2109
1700 case EIO_READDIR: eio__scandir (req, self); break; 2110 case EIO_READDIR: eio__scandir (req, self); break;
1701 2111
1702 case EIO_BUSY: 2112 case EIO_BUSY:
1703#ifdef _WIN32 2113#ifdef _WIN32
1744 case EIO_NOP: 2154 case EIO_NOP:
1745 req->result = 0; 2155 req->result = 0;
1746 break; 2156 break;
1747 2157
1748 case EIO_CUSTOM: 2158 case EIO_CUSTOM:
1749 ((void (*)(eio_req *))req->feed) (req); 2159 req->feed (req);
1750 break; 2160 break;
1751 2161
1752 default: 2162 default:
1753 errno = ENOSYS; 2163 errno = ENOSYS;
1754 req->result = -1; 2164 req->result = -1;
1783eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2193eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1784{ 2194{
1785 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2195 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1786} 2196}
1787 2197
2198eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2199{
2200 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2201}
2202
2203eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
2204{
2205 REQ (EIO_SYNCFS); req->int1 = fd; SEND;
2206}
2207
2208eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2209{
2210 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2211}
2212
1788eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2213eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1789{ 2214{
1790 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2215 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1791} 2216}
1792 2217
1798eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data) 2223eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
1799{ 2224{
1800 REQ (EIO_MLOCKALL); req->int1 = flags; SEND; 2225 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
1801} 2226}
1802 2227
1803eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) 2228eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
1804{ 2229{
1805 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; 2230 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
1806}
1807
1808eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1809{
1810 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1811} 2231}
1812 2232
1813eio_req *eio_close (int fd, int pri, eio_cb cb, void *data) 2233eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
1814{ 2234{
1815 REQ (EIO_CLOSE); req->int1 = fd; SEND; 2235 REQ (EIO_CLOSE); req->int1 = fd; SEND;
1853eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) 2273eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1854{ 2274{
1855 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2275 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1856} 2276}
1857 2277
1858eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2278eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1859{ 2279{
1860 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2280 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1861} 2281}
1862 2282
1863eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) 2283eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
1883eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data) 2303eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1884{ 2304{
1885 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2305 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1886} 2306}
1887 2307
1888eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2308eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1889{ 2309{
1890 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2310 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1891} 2311}
1892 2312
1893eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) 2313eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
1907} 2327}
1908 2328
1909eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data) 2329eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1910{ 2330{
1911 return eio__1path (EIO_READLINK, path, pri, cb, data); 2331 return eio__1path (EIO_READLINK, path, pri, cb, data);
2332}
2333
2334eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2335{
2336 return eio__1path (EIO_REALPATH, path, pri, cb, data);
1912} 2337}
1913 2338
1914eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) 2339eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1915{ 2340{
1916 return eio__1path (EIO_STAT, path, pri, cb, data); 2341 return eio__1path (EIO_STAT, path, pri, cb, data);
1975eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2400eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1976{ 2401{
1977 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2402 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1978} 2403}
1979 2404
1980eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) 2405eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
1981{ 2406{
1982 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND; 2407 REQ (EIO_CUSTOM); req->feed = execute; SEND;
1983} 2408}
1984 2409
1985#endif 2410#endif
1986 2411
1987eio_req *eio_grp (eio_cb cb, void *data) 2412eio_req *eio_grp (eio_cb cb, void *data)
1996#undef SEND 2421#undef SEND
1997 2422
1998/*****************************************************************************/ 2423/*****************************************************************************/
1999/* grp functions */ 2424/* grp functions */
2000 2425
2426void
2001void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2427eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
2002{ 2428{
2003 grp->int2 = limit; 2429 grp->int2 = limit;
2004 grp->feed = feed; 2430 grp->feed = feed;
2005 2431
2006 grp_try_feed (grp); 2432 grp_try_feed (grp);
2007} 2433}
2008 2434
2435void
2009void eio_grp_limit (eio_req *grp, int limit) 2436eio_grp_limit (eio_req *grp, int limit)
2010{ 2437{
2011 grp->int2 = limit; 2438 grp->int2 = limit;
2012 2439
2013 grp_try_feed (grp); 2440 grp_try_feed (grp);
2014} 2441}
2015 2442
2443void
2016void eio_grp_add (eio_req *grp, eio_req *req) 2444eio_grp_add (eio_req *grp, eio_req *req)
2017{ 2445{
2018 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2446 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
2019 2447
2020 grp->flags |= EIO_FLAG_GROUPADD; 2448 grp->flags |= EIO_FLAG_GROUPADD;
2021 2449
2032} 2460}
2033 2461
2034/*****************************************************************************/ 2462/*****************************************************************************/
2035/* misc garbage */ 2463/* misc garbage */
2036 2464
2465eio_ssize_t
2037ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2466eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2038{ 2467{
2039 etp_worker wrk;
2040 ssize_t ret;
2041
2042 wrk.dbuf = 0;
2043
2044 ret = eio__sendfile (ofd, ifd, offset, count, &wrk); 2468 return eio__sendfile (ofd, ifd, offset, count);
2045
2046 if (wrk.dbuf)
2047 free (wrk.dbuf);
2048
2049 return ret;
2050} 2469}
2051 2470

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines