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 1.38 by root, Fri Jun 12 20:01:42 2009 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 Marc Alexander Lehmann <libeio@schmorp.de> 4 * Copyright (c) 2007,2008,2009,2010,2011 Marc Alexander Lehmann <libeio@schmorp.de>
5 * All rights reserved. 5 * All rights reserved.
6 * 6 *
7 * Redistribution and use in source and binary forms, with or without modifica- 7 * Redistribution and use in source and binary forms, with or without modifica-
8 * tion, are permitted provided that the following conditions are met: 8 * tion, are permitted provided that the following conditions are met:
9 * 9 *
… …
35 * and other provisions required by the GPL. If you do not delete the 35 * and other provisions required by the GPL. If you do not delete the
36 * provisions above, a recipient may use your version of this file under 36 * provisions above, a recipient may use your version of this file under
37 * either the BSD or the GPL. 37 * either the BSD or the GPL.
38 */ 38 */
39 39
40#ifndef _WIN32
41# include "config.h"
42#endif
43
40#include "eio.h" 44#include "eio.h"
45#include "ecb.h"
41 46
42#ifdef EIO_STACKSIZE 47#ifdef EIO_STACKSIZE
43# define XTHREAD_STACKSIZE EIO_STACKSIZE 48# define XTHREAD_STACKSIZE EIO_STACKSIZE
44#endif 49#endif
45#include "xthread.h" 50#include "xthread.h"
… …
53#include <sys/stat.h> 58#include <sys/stat.h>
54#include <limits.h> 59#include <limits.h>
55#include <fcntl.h> 60#include <fcntl.h>
56#include <assert.h> 61#include <assert.h>
57 62
63/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
64/* intptr_t only comes from stdint.h, says idiot openbsd coder */
65#if HAVE_STDINT_H
66# include <stdint.h>
67#endif
68
69#ifndef ECANCELED
70# define ECANCELED EDOM
71#endif
72#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
58#ifndef EIO_FINISH 82#ifndef EIO_FINISH
59# 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
60#endif 84#endif
61 85
62#ifndef EIO_DESTROY 86#ifndef EIO_DESTROY
… …
65 89
66#ifndef EIO_FEED 90#ifndef EIO_FEED
67# 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)
68#endif 92#endif
69 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
70#ifdef _WIN32 105#ifdef _WIN32
71 106
72 /*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
73#else 163#else
74 164
75# include "config.h"
76# include <sys/time.h> 165 #include <sys/time.h>
77# include <sys/select.h> 166 #include <sys/select.h>
78# include <sys/mman.h> 167 #include <sys/statvfs.h>
79# include <unistd.h> 168 #include <unistd.h>
80# include <utime.h> 169 #include <utime.h>
81# include <signal.h> 170 #include <signal.h>
82# include <dirent.h> 171 #include <dirent.h>
83 172
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
84/* 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 */
85# if defined(__FreeBSD__) || defined(__NetBSD__) || defined(__OpenBSD__) 180 #if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
86# define _DIRENT_HAVE_D_TYPE /* sigh */ 181 #define _DIRENT_HAVE_D_TYPE /* sigh */
87# define D_INO(de) (de)->d_fileno 182 #define D_INO(de) (de)->d_fileno
88# define D_NAMLEN(de) (de)->d_namlen 183 #define D_NAMLEN(de) (de)->d_namlen
89# elif defined(__linux) || defined(d_ino) || _XOPEN_SOURCE >= 600 184 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
90# define D_INO(de) (de)->d_ino 185 #define D_INO(de) (de)->d_ino
91# endif 186 #endif
92 187
93#ifdef _D_EXACT_NAMLEN 188 #ifdef _D_EXACT_NAMLEN
94# undef D_NAMLEN 189 #undef D_NAMLEN
95# define D_NAMLEN(de) _D_EXACT_NAMLEN (de) 190 #define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
96#endif 191 #endif
97 192
98# ifdef _DIRENT_HAVE_D_TYPE 193 #ifdef _DIRENT_HAVE_D_TYPE
99# define D_TYPE(de) (de)->d_type 194 #define D_TYPE(de) (de)->d_type
100# endif 195 #endif
101 196
102# ifndef EIO_STRUCT_DIRENT 197 #ifndef EIO_STRUCT_DIRENT
103# define EIO_STRUCT_DIRENT struct dirent 198 #define EIO_STRUCT_DIRENT struct dirent
104# endif 199 #endif
105 200
106#endif 201#endif
107 202
108#if HAVE_SENDFILE 203#if HAVE_SENDFILE
109# if __linux 204# if __linux
110# include <sys/sendfile.h> 205# include <sys/sendfile.h>
111# elif __freebsd 206# elif __FreeBSD__ || defined __APPLE__
112# include <sys/socket.h> 207# include <sys/socket.h>
113# include <sys/uio.h> 208# include <sys/uio.h>
114# elif __hpux 209# elif __hpux
115# include <sys/socket.h> 210# include <sys/socket.h>
116# elif __solaris /* not yet */ 211# elif __solaris
117# include <sys/sendfile.h> 212# include <sys/sendfile.h>
118# else 213# else
119# error sendfile support requested but not available 214# error sendfile support requested but not available
120# endif 215# endif
121#endif 216#endif
… …
125#endif 220#endif
126#ifndef D_INO 221#ifndef D_INO
127# define D_INO(de) 0 222# define D_INO(de) 0
128#endif 223#endif
129#ifndef D_NAMLEN 224#ifndef D_NAMLEN
130# define D_NAMLEN(de) strlen ((de)->d_name) 225# define D_NAMLEN(entp) strlen (D_NAME (entp))
131#endif 226#endif
132
133/* number of seconds after which an idle threads exit */
134#define IDLE_TIMEOUT 10
135 227
136/* used for struct dirent, AIX doesn't provide it */ 228/* used for struct dirent, AIX doesn't provide it */
137#ifndef NAME_MAX 229#ifndef NAME_MAX
138# define NAME_MAX 4096 230# define NAME_MAX 4096
139#endif 231#endif
140 232
233/* used for readlink etc. */
234#ifndef PATH_MAX
235# define PATH_MAX 4096
236#endif
237
141/* buffer size for various temporary buffers */ 238/* buffer size for various temporary buffers */
142#define EIO_BUFSIZE 65536 239#define EIO_BUFSIZE 65536
143 240
144#define dBUF \ 241#define dBUF \
145 char *eio_buf; \
146 ETP_WORKER_LOCK (self); \
147 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 242 char *eio_buf = malloc (EIO_BUFSIZE); \
148 ETP_WORKER_UNLOCK (self); \
149 errno = ENOMEM; \ 243 errno = ENOMEM; \
150 if (!eio_buf) \ 244 if (!eio_buf) \
151 return -1; 245 return -1
246
247#define FUBd \
248 free (eio_buf)
152 249
153#define EIO_TICKS ((1000000 + 1023) >> 10) 250#define EIO_TICKS ((1000000 + 1023) >> 10)
154
155/*****************************************************************************/
156
157#if __GNUC__ >= 3
158# define expect(expr,value) __builtin_expect ((expr),(value))
159#else
160# define expect(expr,value) (expr)
161#endif
162
163#define expect_false(expr) expect ((expr) != 0, 0)
164#define expect_true(expr) expect ((expr) != 0, 1)
165
166/*****************************************************************************/
167 251
168#define ETP_PRI_MIN EIO_PRI_MIN 252#define ETP_PRI_MIN EIO_PRI_MIN
169#define ETP_PRI_MAX EIO_PRI_MAX 253#define ETP_PRI_MAX EIO_PRI_MAX
170 254
171struct etp_worker; 255struct etp_worker;
… …
175static int eio_finish (eio_req *req); 259static int eio_finish (eio_req *req);
176#define ETP_FINISH(req) eio_finish (req) 260#define ETP_FINISH(req) eio_finish (req)
177static void eio_execute (struct etp_worker *self, eio_req *req); 261static void eio_execute (struct etp_worker *self, eio_req *req);
178#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) 262#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
179 263
180#define ETP_WORKER_CLEAR(req) \
181 if (wrk->dbuf) \
182 { \
183 free (wrk->dbuf); \
184 wrk->dbuf = 0; \
185 } \
186 \
187 if (wrk->dirp) \
188 { \
189 closedir (wrk->dirp); \
190 wrk->dirp = 0; \
191 }
192
193#define ETP_WORKER_COMMON \
194 void *dbuf; \
195 DIR *dirp;
196
197/*****************************************************************************/ 264/*****************************************************************************/
198 265
199#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) 266#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
200 267
201/* calculcate time difference in ~1/EIO_TICKS of a second */ 268/* calculate time difference in ~1/EIO_TICKS of a second */
269ecb_inline int
202static int tvdiff (struct timeval *tv1, struct timeval *tv2) 270tvdiff (struct timeval *tv1, struct timeval *tv2)
203{ 271{
204 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 272 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
205 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 273 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
206} 274}
207 275
… …
214static unsigned int max_poll_reqs; /* reslock */ 282static unsigned int max_poll_reqs; /* reslock */
215 283
216static volatile unsigned int nreqs; /* reqlock */ 284static volatile unsigned int nreqs; /* reqlock */
217static volatile unsigned int nready; /* reqlock */ 285static volatile unsigned int nready; /* reqlock */
218static volatile unsigned int npending; /* reqlock */ 286static volatile unsigned int npending; /* reqlock */
219static 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 */
220 289
221static mutex_t wrklock = X_MUTEX_INIT; 290static xmutex_t wrklock;
222static mutex_t reslock = X_MUTEX_INIT; 291static xmutex_t reslock;
223static mutex_t reqlock = X_MUTEX_INIT; 292static xmutex_t reqlock;
224static cond_t reqwait = X_COND_INIT; 293static xcond_t reqwait;
225 294
226#if !HAVE_PREADWRITE 295#if !HAVE_PREADWRITE
227/* 296/*
228 * make our pread/pwrite emulation safe against themselves, but not against 297 * make our pread/pwrite emulation safe against themselves, but not against
229 * 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,
230 * but that's your problem, not mine. 299 * but that's your problem, not mine.
231 */ 300 */
232static mutex_t preadwritelock = X_MUTEX_INIT; 301static xmutex_t preadwritelock;
233#endif 302#endif
234 303
235typedef struct etp_worker 304typedef struct etp_worker
236{ 305{
237 /* locked by wrklock */ 306 /* locked by wrklock */
238 struct etp_worker *prev, *next; 307 struct etp_worker *prev, *next;
239 308
240 thread_t tid; 309 xthread_t tid;
241 310
242 /* locked by reslock, reqlock or wrklock */ 311 /* locked by reslock, reqlock or wrklock */
243 ETP_REQ *req; /* currently processed request */ 312 ETP_REQ *req; /* currently processed request */
244 313
314#ifdef ETP_WORKER_COMMON
245 ETP_WORKER_COMMON 315 ETP_WORKER_COMMON
316#endif
246} etp_worker; 317} etp_worker;
247 318
248static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */ 319static etp_worker wrk_first; /* NOT etp */
249 320
250#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock) 321#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
251#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock) 322#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
252 323
253/* worker threads management */ 324/* worker threads management */
254 325
326static void ecb_cold
255static void etp_worker_clear (etp_worker *wrk) 327etp_worker_clear (etp_worker *wrk)
256{ 328{
257 ETP_WORKER_CLEAR (wrk);
258} 329}
259 330
331static void ecb_cold
260static void etp_worker_free (etp_worker *wrk) 332etp_worker_free (etp_worker *wrk)
261{ 333{
262 wrk->next->prev = wrk->prev; 334 wrk->next->prev = wrk->prev;
263 wrk->prev->next = wrk->next; 335 wrk->prev->next = wrk->next;
264 336
265 free (wrk); 337 free (wrk);
266} 338}
267 339
268static unsigned int etp_nreqs (void) 340static unsigned int
341etp_nreqs (void)
269{ 342{
270 int retval; 343 int retval;
271 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 344 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
272 retval = nreqs; 345 retval = nreqs;
273 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 346 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
274 return retval; 347 return retval;
275} 348}
276 349
277static unsigned int etp_nready (void) 350static unsigned int
351etp_nready (void)
278{ 352{
279 unsigned int retval; 353 unsigned int retval;
280 354
281 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 355 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
282 retval = nready; 356 retval = nready;
283 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 357 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
284 358
285 return retval; 359 return retval;
286} 360}
287 361
288static unsigned int etp_npending (void) 362static unsigned int
363etp_npending (void)
289{ 364{
290 unsigned int retval; 365 unsigned int retval;
291 366
292 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 367 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
293 retval = npending; 368 retval = npending;
294 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 369 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
295 370
296 return retval; 371 return retval;
297} 372}
298 373
299static unsigned int etp_nthreads (void) 374static unsigned int
375etp_nthreads (void)
300{ 376{
301 unsigned int retval; 377 unsigned int retval;
302 378
303 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 379 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
304 retval = started; 380 retval = started;
… …
318} etp_reqq; 394} etp_reqq;
319 395
320static etp_reqq req_queue; 396static etp_reqq req_queue;
321static etp_reqq res_queue; 397static etp_reqq res_queue;
322 398
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
323static int reqq_push (etp_reqq *q, ETP_REQ *req) 411reqq_push (etp_reqq *q, ETP_REQ *req)
324{ 412{
325 int pri = req->pri; 413 int pri = req->pri;
326 req->next = 0; 414 req->next = 0;
327 415
328 if (q->qe[pri]) 416 if (q->qe[pri])
… …
334 q->qe[pri] = q->qs[pri] = req; 422 q->qe[pri] = q->qs[pri] = req;
335 423
336 return q->size++; 424 return q->size++;
337} 425}
338 426
427static ETP_REQ * ecb_noinline
339static ETP_REQ *reqq_shift (etp_reqq *q) 428reqq_shift (etp_reqq *q)
340{ 429{
341 int pri; 430 int pri;
342 431
343 if (!q->size) 432 if (!q->size)
344 return 0; 433 return 0;
… …
359 } 448 }
360 449
361 abort (); 450 abort ();
362} 451}
363 452
364static void etp_atfork_prepare (void) 453static int ecb_cold
454etp_init (void (*want_poll)(void), void (*done_poll)(void))
365{ 455{
366 X_LOCK (wrklock); 456 X_MUTEX_CREATE (wrklock);
367 X_LOCK (reqlock); 457 X_MUTEX_CREATE (reslock);
368 X_LOCK (reslock); 458 X_MUTEX_CREATE (reqlock);
369#if !HAVE_PREADWRITE 459 X_COND_CREATE (reqwait);
370 X_LOCK (preadwritelock);
371#endif
372}
373 460
374static void etp_atfork_parent (void) 461 reqq_init (&req_queue);
375{ 462 reqq_init (&res_queue);
376#if !HAVE_PREADWRITE
377 X_UNLOCK (preadwritelock);
378#endif
379 X_UNLOCK (reslock);
380 X_UNLOCK (reqlock);
381 X_UNLOCK (wrklock);
382}
383 463
384static void etp_atfork_child (void) 464 wrk_first.next =
385{ 465 wrk_first.prev = &wrk_first;
386 ETP_REQ *prv;
387
388 while ((prv = reqq_shift (&req_queue)))
389 ETP_DESTROY (prv);
390
391 while ((prv = reqq_shift (&res_queue)))
392 ETP_DESTROY (prv);
393
394 while (wrk_first.next != &wrk_first)
395 {
396 etp_worker *wrk = wrk_first.next;
397
398 if (wrk->req)
399 ETP_DESTROY (wrk->req);
400
401 etp_worker_clear (wrk);
402 etp_worker_free (wrk);
403 }
404 466
405 started = 0; 467 started = 0;
406 idle = 0; 468 idle = 0;
407 nreqs = 0; 469 nreqs = 0;
408 nready = 0; 470 nready = 0;
409 npending = 0; 471 npending = 0;
410 472
411 etp_atfork_parent ();
412}
413
414static void
415etp_once_init (void)
416{
417 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
418}
419
420static int
421etp_init (void (*want_poll)(void), void (*done_poll)(void))
422{
423 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
424
425 pthread_once (&doinit, etp_once_init);
426
427 want_poll_cb = want_poll; 473 want_poll_cb = want_poll;
428 done_poll_cb = done_poll; 474 done_poll_cb = done_poll;
429 475
430 return 0; 476 return 0;
431} 477}
432 478
433X_THREAD_PROC (etp_proc); 479X_THREAD_PROC (etp_proc);
434 480
481static void ecb_cold
435static void etp_start_thread (void) 482etp_start_thread (void)
436{ 483{
437 etp_worker *wrk = calloc (1, sizeof (etp_worker)); 484 etp_worker *wrk = calloc (1, sizeof (etp_worker));
438 485
439 /*TODO*/ 486 /*TODO*/
440 assert (("unable to allocate worker thread data", wrk)); 487 assert (("unable to allocate worker thread data", wrk));
… …
453 free (wrk); 500 free (wrk);
454 501
455 X_UNLOCK (wrklock); 502 X_UNLOCK (wrklock);
456} 503}
457 504
505static void
458static void etp_maybe_start_thread (void) 506etp_maybe_start_thread (void)
459{ 507{
460 if (expect_true (etp_nthreads () >= wanted)) 508 if (ecb_expect_true (etp_nthreads () >= wanted))
461 return; 509 return;
462 510
463 /* todo: maybe use idle here, but might be less exact */ 511 /* todo: maybe use idle here, but might be less exact */
464 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 ()))
465 return; 513 return;
466 514
467 etp_start_thread (); 515 etp_start_thread ();
468} 516}
469 517
518static void ecb_cold
470static void etp_end_thread (void) 519etp_end_thread (void)
471{ 520{
472 eio_req *req = calloc (1, sizeof (eio_req)); 521 eio_req *req = calloc (1, sizeof (eio_req));
473 522
474 req->type = -1; 523 req->type = -1;
475 req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 524 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
… …
482 X_LOCK (wrklock); 531 X_LOCK (wrklock);
483 --started; 532 --started;
484 X_UNLOCK (wrklock); 533 X_UNLOCK (wrklock);
485} 534}
486 535
487static int etp_poll (void) 536static int
537etp_poll (void)
488{ 538{
489 unsigned int maxreqs; 539 unsigned int maxreqs;
490 unsigned int maxtime; 540 unsigned int maxtime;
491 struct timeval tv_start, tv_now; 541 struct timeval tv_start, tv_now;
492 542
… …
522 572
523 X_LOCK (reqlock); 573 X_LOCK (reqlock);
524 --nreqs; 574 --nreqs;
525 X_UNLOCK (reqlock); 575 X_UNLOCK (reqlock);
526 576
527 if (expect_false (req->type == EIO_GROUP && req->size)) 577 if (ecb_expect_false (req->type == EIO_GROUP && req->size))
528 { 578 {
529 req->int1 = 1; /* mark request as delayed */ 579 req->int1 = 1; /* mark request as delayed */
530 continue; 580 continue;
531 } 581 }
532 else 582 else
533 { 583 {
534 int res = ETP_FINISH (req); 584 int res = ETP_FINISH (req);
535 if (expect_false (res)) 585 if (ecb_expect_false (res))
536 return res; 586 return res;
537 } 587 }
538 588
539 if (expect_false (maxreqs && !--maxreqs)) 589 if (ecb_expect_false (maxreqs && !--maxreqs))
540 break; 590 break;
541 591
542 if (maxtime) 592 if (maxtime)
543 { 593 {
544 gettimeofday (&tv_now, 0); 594 gettimeofday (&tv_now, 0);
… …
550 600
551 errno = EAGAIN; 601 errno = EAGAIN;
552 return -1; 602 return -1;
553} 603}
554 604
605static void
555static void etp_cancel (ETP_REQ *req) 606etp_cancel (ETP_REQ *req)
556{ 607{
557 X_LOCK (wrklock); 608 req->cancelled = 1;
558 req->flags |= EIO_FLAG_CANCELLED;
559 X_UNLOCK (wrklock);
560 609
561 eio_grp_cancel (req); 610 eio_grp_cancel (req);
562} 611}
563 612
613static void
564static void etp_submit (ETP_REQ *req) 614etp_submit (ETP_REQ *req)
565{ 615{
566 req->pri -= ETP_PRI_MIN; 616 req->pri -= ETP_PRI_MIN;
567 617
568 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;
569 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;
570 620
571 if (expect_false (req->type == EIO_GROUP)) 621 if (ecb_expect_false (req->type == EIO_GROUP))
572 { 622 {
573 /* I hope this is worth it :/ */ 623 /* I hope this is worth it :/ */
574 X_LOCK (reqlock); 624 X_LOCK (reqlock);
575 ++nreqs; 625 ++nreqs;
576 X_UNLOCK (reqlock); 626 X_UNLOCK (reqlock);
… …
595 645
596 etp_maybe_start_thread (); 646 etp_maybe_start_thread ();
597 } 647 }
598} 648}
599 649
650static void ecb_cold
600static void etp_set_max_poll_time (double nseconds) 651etp_set_max_poll_time (double nseconds)
601{ 652{
602 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 653 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
603 max_poll_time = nseconds; 654 max_poll_time = nseconds * EIO_TICKS;
604 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 655 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
605} 656}
606 657
658static void ecb_cold
607static void etp_set_max_poll_reqs (unsigned int maxreqs) 659etp_set_max_poll_reqs (unsigned int maxreqs)
608{ 660{
609 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 661 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
610 max_poll_reqs = maxreqs; 662 max_poll_reqs = maxreqs;
611 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 663 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
612} 664}
613 665
666static void ecb_cold
614static void etp_set_max_idle (unsigned int nthreads) 667etp_set_max_idle (unsigned int nthreads)
615{ 668{
616 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 669 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
617 max_idle = nthreads <= 0 ? 1 : nthreads; 670 max_idle = nthreads;
618 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 671 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
619} 672}
620 673
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
621static void etp_set_min_parallel (unsigned int nthreads) 683etp_set_min_parallel (unsigned int nthreads)
622{ 684{
623 if (wanted < nthreads) 685 if (wanted < nthreads)
624 wanted = nthreads; 686 wanted = nthreads;
625} 687}
626 688
689static void ecb_cold
627static void etp_set_max_parallel (unsigned int nthreads) 690etp_set_max_parallel (unsigned int nthreads)
628{ 691{
629 if (wanted > nthreads) 692 if (wanted > nthreads)
630 wanted = nthreads; 693 wanted = nthreads;
631 694
632 while (started > wanted) 695 while (started > wanted)
633 etp_end_thread (); 696 etp_end_thread ();
634} 697}
635 698
636/*****************************************************************************/ 699/*****************************************************************************/
637 700
701static void
638static void grp_try_feed (eio_req *grp) 702grp_try_feed (eio_req *grp)
639{ 703{
640 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 704 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
641 { 705 {
642 grp->flags &= ~EIO_FLAG_GROUPADD; 706 grp->flags &= ~EIO_FLAG_GROUPADD;
643 707
… …
650 break; 714 break;
651 } 715 }
652 } 716 }
653} 717}
654 718
719static int
655static int grp_dec (eio_req *grp) 720grp_dec (eio_req *grp)
656{ 721{
657 --grp->size; 722 --grp->size;
658 723
659 /* call feeder, if applicable */ 724 /* call feeder, if applicable */
660 grp_try_feed (grp); 725 grp_try_feed (grp);
… …
664 return eio_finish (grp); 729 return eio_finish (grp);
665 else 730 else
666 return 0; 731 return 0;
667} 732}
668 733
734static void
669void eio_destroy (eio_req *req) 735eio_destroy (eio_req *req)
670{ 736{
671 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); 737 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
672 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); 738 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
673 739
674 EIO_DESTROY (req); 740 EIO_DESTROY (req);
675} 741}
676 742
743static int
677static int eio_finish (eio_req *req) 744eio_finish (eio_req *req)
678{ 745{
679 int res = EIO_FINISH (req); 746 int res = EIO_FINISH (req);
680 747
681 if (req->grp) 748 if (req->grp)
682 { 749 {
… …
690 if (grp->grp_first == req) 757 if (grp->grp_first == req)
691 grp->grp_first = req->grp_next; 758 grp->grp_first = req->grp_next;
692 759
693 res2 = grp_dec (grp); 760 res2 = grp_dec (grp);
694 761
695 if (!res && res2) 762 if (!res)
696 res = res2; 763 res = res2;
697 } 764 }
698 765
699 eio_destroy (req); 766 eio_destroy (req);
700 767
701 return res; 768 return res;
702} 769}
703 770
771void
704void eio_grp_cancel (eio_req *grp) 772eio_grp_cancel (eio_req *grp)
705{ 773{
706 for (grp = grp->grp_first; grp; grp = grp->grp_next) 774 for (grp = grp->grp_first; grp; grp = grp->grp_next)
707 eio_cancel (grp); 775 eio_cancel (grp);
708} 776}
709 777
778void
710void eio_cancel (eio_req *req) 779eio_cancel (eio_req *req)
711{ 780{
712 etp_cancel (req); 781 etp_cancel (req);
713} 782}
714 783
784void
715void eio_submit (eio_req *req) 785eio_submit (eio_req *req)
716{ 786{
717 etp_submit (req); 787 etp_submit (req);
718} 788}
719 789
720unsigned int eio_nreqs (void) 790unsigned int
791eio_nreqs (void)
721{ 792{
722 return etp_nreqs (); 793 return etp_nreqs ();
723} 794}
724 795
725unsigned int eio_nready (void) 796unsigned int
797eio_nready (void)
726{ 798{
727 return etp_nready (); 799 return etp_nready ();
728} 800}
729 801
730unsigned int eio_npending (void) 802unsigned int
803eio_npending (void)
731{ 804{
732 return etp_npending (); 805 return etp_npending ();
733} 806}
734 807
735unsigned int eio_nthreads (void) 808unsigned int ecb_cold
809eio_nthreads (void)
736{ 810{
737 return etp_nthreads (); 811 return etp_nthreads ();
738} 812}
739 813
814void ecb_cold
740void eio_set_max_poll_time (double nseconds) 815eio_set_max_poll_time (double nseconds)
741{ 816{
742 etp_set_max_poll_time (nseconds); 817 etp_set_max_poll_time (nseconds);
743} 818}
744 819
820void ecb_cold
745void eio_set_max_poll_reqs (unsigned int maxreqs) 821eio_set_max_poll_reqs (unsigned int maxreqs)
746{ 822{
747 etp_set_max_poll_reqs (maxreqs); 823 etp_set_max_poll_reqs (maxreqs);
748} 824}
749 825
826void ecb_cold
750void eio_set_max_idle (unsigned int nthreads) 827eio_set_max_idle (unsigned int nthreads)
751{ 828{
752 etp_set_max_idle (nthreads); 829 etp_set_max_idle (nthreads);
753} 830}
754 831
832void ecb_cold
833eio_set_idle_timeout (unsigned int seconds)
834{
835 etp_set_idle_timeout (seconds);
836}
837
838void ecb_cold
755void eio_set_min_parallel (unsigned int nthreads) 839eio_set_min_parallel (unsigned int nthreads)
756{ 840{
757 etp_set_min_parallel (nthreads); 841 etp_set_min_parallel (nthreads);
758} 842}
759 843
844void ecb_cold
760void eio_set_max_parallel (unsigned int nthreads) 845eio_set_max_parallel (unsigned int nthreads)
761{ 846{
762 etp_set_max_parallel (nthreads); 847 etp_set_max_parallel (nthreads);
763} 848}
764 849
765int eio_poll (void) 850int eio_poll (void)
… …
774# undef pread 859# undef pread
775# undef pwrite 860# undef pwrite
776# define pread eio__pread 861# define pread eio__pread
777# define pwrite eio__pwrite 862# define pwrite eio__pwrite
778 863
779static ssize_t 864static eio_ssize_t
780eio__pread (int fd, void *buf, size_t count, off_t offset) 865eio__pread (int fd, void *buf, size_t count, off_t offset)
781{ 866{
782 ssize_t res; 867 eio_ssize_t res;
783 off_t ooffset; 868 off_t ooffset;
784 869
785 X_LOCK (preadwritelock); 870 X_LOCK (preadwritelock);
786 ooffset = lseek (fd, 0, SEEK_CUR); 871 ooffset = lseek (fd, 0, SEEK_CUR);
787 lseek (fd, offset, SEEK_SET); 872 lseek (fd, offset, SEEK_SET);
… …
790 X_UNLOCK (preadwritelock); 875 X_UNLOCK (preadwritelock);
791 876
792 return res; 877 return res;
793} 878}
794 879
795static ssize_t 880static eio_ssize_t
796eio__pwrite (int fd, void *buf, size_t count, off_t offset) 881eio__pwrite (int fd, void *buf, size_t count, off_t offset)
797{ 882{
798 ssize_t res; 883 eio_ssize_t res;
799 off_t ooffset; 884 off_t ooffset;
800 885
801 X_LOCK (preadwritelock); 886 X_LOCK (preadwritelock);
802 ooffset = lseek (fd, 0, SEEK_CUR); 887 ooffset = lseek (fd, 0, SEEK_CUR);
803 lseek (fd, offset, SEEK_SET); 888 lseek (fd, offset, SEEK_SET);
… …
807 892
808 return res; 893 return res;
809} 894}
810#endif 895#endif
811 896
812#ifndef HAVE_FUTIMES 897#ifndef HAVE_UTIMES
813 898
814# undef utimes 899# undef utimes
815# undef futimes
816# define utimes(path,times) eio__utimes (path, times) 900# define utimes(path,times) eio__utimes (path, times)
817# define futimes(fd,times) eio__futimes (fd, times)
818 901
819static int 902static int
820eio__utimes (const char *filename, const struct timeval times[2]) 903eio__utimes (const char *filename, const struct timeval times[2])
821{ 904{
822 if (times) 905 if (times)
… …
830 } 913 }
831 else 914 else
832 return utime (filename, 0); 915 return utime (filename, 0);
833} 916}
834 917
918#endif
919
920#ifndef HAVE_FUTIMES
921
922# undef futimes
923# define futimes(fd,times) eio__futimes (fd, times)
924
925static int
835static int eio__futimes (int fd, const struct timeval tv[2]) 926eio__futimes (int fd, const struct timeval tv[2])
836{ 927{
837 errno = ENOSYS; 928 errno = ENOSYS;
838 return -1; 929 return -1;
839} 930}
840 931
… …
844# undef fdatasync 935# undef fdatasync
845# define fdatasync(fd) fsync (fd) 936# define fdatasync(fd) fsync (fd)
846#endif 937#endif
847 938
848/* sync_file_range always needs emulation */ 939/* sync_file_range always needs emulation */
849int 940static int
850eio__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)
851{ 942{
852#if HAVE_SYNC_FILE_RANGE 943#if HAVE_SYNC_FILE_RANGE
853 int res; 944 int res;
854 945
… …
867 if (!res || errno != ENOSYS) 958 if (!res || errno != ENOSYS)
868 return res; 959 return res;
869#endif 960#endif
870 961
871 /* 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
872 * call fdatasync, as thta matches the expectation of it's users best */ 963 * call fdatasync, as that matches the expectation of its users best */
873 return fdatasync (fd); 964 return fdatasync (fd);
965}
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
874} 976}
875 977
876#if !HAVE_READAHEAD 978#if !HAVE_READAHEAD
877# undef readahead 979# undef readahead
878# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 980# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
879 981
880static ssize_t 982static eio_ssize_t
881eio__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)
882{ 984{
883 size_t todo = count; 985 size_t todo = count;
884 dBUF; 986 dBUF;
885 987
… …
890 pread (fd, eio_buf, len, offset); 992 pread (fd, eio_buf, len, offset);
891 offset += len; 993 offset += len;
892 todo -= len; 994 todo -= len;
893 } 995 }
894 996
997 FUBd;
998
895 errno = 0; 999 errno = 0;
896 return count; 1000 return count;
897} 1001}
898 1002
899#endif 1003#endif
900 1004
901/* sendfile always needs emulation */ 1005/* sendfile always needs emulation */
902static ssize_t 1006static eio_ssize_t
903eio__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)
904{ 1008{
1009 eio_ssize_t written = 0;
905 ssize_t res; 1010 eio_ssize_t res;
906 1011
907 if (!count) 1012 if (!count)
908 return 0; 1013 return 0;
909 1014
1015 for (;;)
1016 {
1017#ifdef __APPLE__
1018# undef HAVE_SENDFILE /* broken, as everything on os x */
1019#endif
910#if HAVE_SENDFILE 1020#if HAVE_SENDFILE
911# if __linux 1021# if __linux
1022 off_t soffset = offset;
912 res = sendfile (ofd, ifd, &offset, count); 1023 res = sendfile (ofd, ifd, &soffset, count);
913 1024
914# elif __freebsd 1025# elif __FreeBSD__
915 /* 1026 /*
916 * 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
917 * wasted on making it similar to other I/O functions. 1028 * wasted on making it similar to other I/O functions.
918 */ 1029 */
919 {
920 off_t sbytes; 1030 off_t sbytes;
921 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 1031 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
922 1032
923 if (res < 0 && sbytes) 1033 #if 0 /* according to the manpage, this is correct, but broken behaviour */
924 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 1034 /* freebsd' sendfile will return 0 on success */
1035 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
1036 /* not on e.g. EIO or EPIPE - sounds broken */
1037 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
925 res = sbytes; 1038 res = sbytes;
926 } 1039 #endif
1040
1041 /* according to source inspection, this is correct, and useful behaviour */
1042 if (sbytes)
1043 res = sbytes;
1044
1045# elif defined (__APPLE__)
1046 off_t sbytes = count;
1047 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
1048
1049 /* according to the manpage, sbytes is always valid */
1050 if (sbytes)
1051 res = sbytes;
927 1052
928# elif __hpux 1053# elif __hpux
929 res = sendfile (ofd, ifd, offset, count, 0, 0); 1054 res = sendfile (ofd, ifd, offset, count, 0, 0);
930 1055
931# elif __solaris 1056# elif __solaris
932 {
933 struct sendfilevec vec; 1057 struct sendfilevec vec;
934 size_t sbytes; 1058 size_t sbytes;
935 1059
936 vec.sfv_fd = ifd; 1060 vec.sfv_fd = ifd;
937 vec.sfv_flag = 0; 1061 vec.sfv_flag = 0;
938 vec.sfv_off = offset; 1062 vec.sfv_off = offset;
939 vec.sfv_len = count; 1063 vec.sfv_len = count;
940 1064
941 res = sendfilev (ofd, &vec, 1, &sbytes); 1065 res = sendfilev (ofd, &vec, 1, &sbytes);
942 1066
943 if (res < 0 && sbytes) 1067 if (res < 0 && sbytes)
944 res = sbytes; 1068 res = sbytes;
945 }
946 1069
947# endif 1070# endif
1071
1072#elif defined (_WIN32) && 0
1073 /* does not work, just for documentation of what would need to be done */
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 */
1077 HANDLE h = TO_SOCKET (ifd);
1078 SetFilePointer (h, offset, 0, FILE_BEGIN);
1079 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1080
948#else 1081#else
949 res = -1; 1082 res = -1;
950 errno = ENOSYS; 1083 errno = ENOSYS;
951#endif 1084#endif
952 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
953 if (res < 0 1109 if (res < 0
954 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1110 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1111 /* BSDs */
1112#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1113 || errno == ENOTSUP
1114#endif
1115#ifdef EOPNOTSUPP /* windows */
1116 || errno == EOPNOTSUPP /* BSDs */
1117#endif
955#if __solaris 1118#if __solaris
956 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1119 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
957#endif 1120#endif
958 ) 1121 )
959 ) 1122 )
… …
963 1126
964 res = 0; 1127 res = 0;
965 1128
966 while (count) 1129 while (count)
967 { 1130 {
968 ssize_t cnt; 1131 eio_ssize_t cnt;
969 1132
970 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);
971 1134
972 if (cnt <= 0) 1135 if (cnt <= 0)
973 { 1136 {
… …
985 1148
986 offset += cnt; 1149 offset += cnt;
987 res += cnt; 1150 res += cnt;
988 count -= cnt; 1151 count -= cnt;
989 } 1152 }
1153
1154 FUBd;
990 } 1155 }
991 1156
992 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 */
993} 1437}
994 1438
995static signed char 1439static signed char
996eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) 1440eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
997{ 1441{
998 return b->score - a->score ? b->score - a->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 */
999 : 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;
1000} 1446}
1001 1447
1448#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1449
1002#define EIO_QSORT_CUTOFF 20 /* quite high, but performs well on many filesystems */ 1450#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1451#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1003 1452
1004static void 1453static void
1005eio_dent_sort (eio_dirent *dents, int size) 1454eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1006{ 1455{
1456 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1457 unsigned char *bit = bits;
1458
1459 assert (CHAR_BIT == 8);
1460 assert (sizeof (eio_dirent) * 8 < 256);
1461 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1462 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1463
1464 if (size <= EIO_SORT_FAST)
1465 return;
1466
1467 /* first prepare an array of bits to test in our radix sort */
1468 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1469 /* inode_bits must contain all inodes ORed together */
1470 /* which is used to skip bits that are 0 everywhere, which is very common */
1471 {
1472 eio_ino_t endianness;
1007 int i, j; 1473 int i, j;
1008 1474
1009 if (size > EIO_QSORT_CUTOFF * 3) /* skip quicksort for small directories */ 1475 /* we store the byte offset of byte n into byte n of "endianness" */
1476 for (i = 0; i < sizeof (eio_ino_t); ++i)
1477 ((unsigned char *)&endianness)[i] = i;
1478
1479 *bit++ = 0;
1480
1481 for (i = 0; i < sizeof (eio_ino_t); ++i)
1482 {
1483 /* shifting off the byte offsets out of "endianness" */
1484 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1485 endianness >>= 8;
1486
1487 for (j = 0; j < 8; ++j)
1488 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1489 *bit++ = offs + j;
1490 }
1491
1492 for (j = 0; j < 8; ++j)
1493 if (score_bits & (1 << j))
1494 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1495 }
1496
1497 /* now actually do the sorting (a variant of MSD radix sort) */
1498 {
1499 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1500 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1501 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1502 int stk_idx = 0;
1503
1504 base_stk [stk_idx] = dents;
1505 end_stk [stk_idx] = dents + size;
1506 bit_stk [stk_idx] = bit - 1;
1507
1508 do
1509 {
1510 base = base_stk [stk_idx];
1511 end = end_stk [stk_idx];
1512 bit = bit_stk [stk_idx];
1513
1514 for (;;)
1515 {
1516 unsigned char O = *bit >> 3;
1517 unsigned char M = 1 << (*bit & 7);
1518
1519 eio_dirent *a = base;
1520 eio_dirent *b = end;
1521
1522 if (b - a < EIO_SORT_CUTOFF)
1523 break;
1524
1525 /* now bit-partition the array on the bit */
1526 /* this ugly asymmetric loop seems to perform much better than typical */
1527 /* partition algos found in the literature */
1528 do
1529 if (!(((unsigned char *)a)[O] & M))
1530 ++a;
1531 else if (!(((unsigned char *)--b)[O] & M))
1532 {
1533 eio_dirent tmp = *a; *a = *b; *b = tmp;
1534 ++a;
1535 }
1536 while (b > a);
1537
1538 /* next bit, or stop, if no bits left in this path */
1539 if (!*--bit)
1540 break;
1541
1542 base_stk [stk_idx] = a;
1543 end_stk [stk_idx] = end;
1544 bit_stk [stk_idx] = bit;
1545 ++stk_idx;
1546
1547 end = a;
1548 }
1549 }
1550 while (stk_idx--);
1551 }
1552}
1553
1554static void
1555eio_dent_insertion_sort (eio_dirent *dents, int size)
1556{
1557 /* first move the smallest element to the front, to act as a sentinel */
1558 {
1559 int i;
1560 eio_dirent *min = dents;
1561
1562 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1563 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1564 if (EIO_DENT_CMP (dents [i], <, *min))
1565 min = &dents [i];
1566
1567 /* swap elements 0 and j (minimum) */
1010 { 1568 {
1011 /* first, use quicksort */ 1569 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1012 /* should be good for 2**31 entries */
1013 struct rng { int l, r; } rng [32];
1014
1015 i = 0;
1016 rng[0].l = 0;
1017 rng[0].r = size;
1018
1019 while (expect_true (i >= 0))
1020 {
1021 int L = rng [i].l;
1022 int R = rng [i].r - 1;
1023
1024 if (expect_false (L + EIO_QSORT_CUTOFF < R))
1025 {
1026 eio_dirent piv = dents [L];
1027
1028 while (L < R)
1029 {
1030 while (eio_dent_cmp (&dents [R], &piv) >= 0 && L < R)
1031 --R;
1032
1033 if (L < R)
1034 dents [L++] = dents [R];
1035
1036 while (eio_dent_cmp (&dents [L], &piv) <= 0 && L < R)
1037 ++L;
1038
1039 if (L < R)
1040 dents [R--] = dents [L];
1041 }
1042
1043 dents [L] = piv;
1044
1045 ++i;
1046 rng [i].l = L + 1;
1047 rng [i].r = rng [i - 1].r;
1048 rng [i - 1].r = L;
1049
1050 if (rng [i].r - rng [i].l > rng [i - 1].r - rng [i - 1].l)
1051 {
1052 struct rng t;
1053
1054 t = rng [i]; rng [i] = rng [i - 1]; rng [i - 1] = t;
1055 }
1056 }
1057 else
1058 --i;
1059 }
1060 } 1570 }
1571 }
1061 1572
1062 /* use a simple insertion sort at the end */ 1573 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1574 {
1575 eio_dirent *i, *j;
1576
1063 for (i = 1; i < size; ++i) 1577 for (i = dents + 1; i < dents + size; ++i)
1064 { 1578 {
1065 eio_dirent value = dents [i]; 1579 eio_dirent value = *i;
1066 1580
1067 for (j = i - 1; j >= 0 && eio_dent_cmp (&dents [j], &value) > 0; --j) 1581 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1068 dents [j + 1] = dents [j]; 1582 j [1] = j [0];
1069 1583
1070 dents [j + 1] = value; 1584 j [1] = value;
1071 } 1585 }
1586 }
1587}
1588
1589static void
1590eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1591{
1592 if (size <= 1)
1593 return; /* our insertion sort relies on size > 0 */
1594
1595 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1596 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1597 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1598
1599 /* use an insertion sort at the end, or for small arrays, */
1600 /* as insertion sort is more efficient for small partitions */
1601 eio_dent_insertion_sort (dents, size);
1072} 1602}
1073 1603
1074/* read a full directory */ 1604/* read a full directory */
1075static void 1605static void
1076eio__scandir (eio_req *req, etp_worker *self) 1606eio__scandir (eio_req *req, etp_worker *self)
1077{ 1607{
1078 DIR *dirp;
1079 EIO_STRUCT_DIRENT *entp;
1080 char *name, *names; 1608 char *name, *names;
1081 int namesalloc = 4096; 1609 int namesalloc = 4096 - sizeof (void *) * 4;
1082 int namesoffs = 0; 1610 int namesoffs = 0;
1083 int flags = req->int1; 1611 int flags = req->int1;
1084 eio_dirent *dents = 0; 1612 eio_dirent *dents = 0;
1085 int dentalloc = 128; 1613 int dentalloc = 128;
1086 int dentoffs = 0; 1614 int dentoffs = 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
1087 1623
1088 req->result = -1; 1624 req->result = -1;
1089 1625
1090 if (!(flags & EIO_READDIR_DENTS)) 1626 if (!(flags & EIO_READDIR_DENTS))
1091 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); 1627 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1092 1628
1093 X_LOCK (wrklock); 1629#ifdef _WIN32
1094 /* 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
1095 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
1096 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1679 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1097 req->ptr1 = names = malloc (namesalloc);
1098 req->ptr2 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; 1680 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1099 X_UNLOCK (wrklock); 1681 req->ptr2 = names = malloc (namesalloc);
1100 1682
1101 if (dirp && names && (!flags || dents)) 1683 if (dirp && names && (!flags || dents))
1102 for (;;) 1684 for (;;)
1103 { 1685 {
1686 int done;
1687
1688#ifdef _WIN32
1689 done = !dirp;
1690#else
1104 errno = 0; 1691 errno = 0;
1105 entp = readdir (dirp); 1692 entp = readdir (dirp);
1693 done = !entp;
1694#endif
1106 1695
1107 if (!entp) 1696 if (done)
1108 { 1697 {
1698#ifndef _WIN32
1699 int old_errno = errno;
1700 closedir (dirp);
1701 errno = old_errno;
1702
1109 if (errno) 1703 if (errno)
1110 break; 1704 break;
1705#endif
1111 1706
1112 /* sort etc. */ 1707 /* sort etc. */
1113 req->int1 = flags; 1708 req->int1 = flags;
1114 req->result = dentoffs; 1709 req->result = dentoffs;
1115 1710
1116 if (dents)
1117 {
1118 eio_dirent *ent = dents + dentoffs;
1119
1120 while (ent > dents)
1121 (--ent)->name = names + (size_t)ent->name;
1122 }
1123
1124 if (flags & EIO_READDIR_STAT_ORDER 1711 if (flags & EIO_READDIR_STAT_ORDER)
1125 || !(~flags & (EIO_READDIR_DIRS_FIRST | EIO_READDIR_FOUND_UNKNOWN))) 1712 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1126 eio_dent_sort (dents, dentoffs); /* score depends of DIRS_FIRST */
1127 else if (flags & EIO_READDIR_DIRS_FIRST) 1713 else if (flags & EIO_READDIR_DIRS_FIRST)
1714 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1715 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1716 else
1128 { 1717 {
1129 /* in this case, all is known, and we just put dirs first and sort them */ 1718 /* in this case, all is known, and we just put dirs first and sort them */
1130 eio_dirent *ent = dents + dentoffs; 1719 eio_dirent *oth = dents + dentoffs;
1131 eio_dirent *dir = dents; 1720 eio_dirent *dir = dents;
1132 1721
1133 /* now move dirs to the front, and non-dirs to the back */ 1722 /* now partition dirs to the front, and non-dirs to the back */
1134 /* by walking from both sides and swapping if necessary */ 1723 /* by walking from both sides and swapping if necessary */
1135 while (ent > dir) 1724 while (oth > dir)
1136 { 1725 {
1137 if (dir->type == DT_DIR) 1726 if (dir->type == EIO_DT_DIR)
1138 ++dir; 1727 ++dir;
1139 else
1140 {
1141 --ent;
1142
1143 if (ent->type == DT_DIR) 1728 else if ((--oth)->type == EIO_DT_DIR)
1144 { 1729 {
1145 eio_dirent tmp = *dir; 1730 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1146 *dir = *ent;
1147 *ent = tmp;
1148 1731
1149 ++dir; 1732 ++dir;
1150 } 1733 }
1151 } 1734 }
1735
1736 /* now sort the dirs only (dirs all have the same score) */
1737 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1152 } 1738 }
1153
1154 /* now sort the dirs only */
1155 eio_dent_sort (dents, dir - dents);
1156 }
1157
1158 /* only provide the names array unless DENTS is specified */
1159 if (!(flags & EIO_READDIR_DENTS))
1160 {
1161 X_LOCK (wrklock);
1162 assert (!dents);
1163 req->ptr1 = 0;
1164 req->ptr2 = names;
1165 X_UNLOCK (wrklock);
1166 }
1167 1739
1168 break; 1740 break;
1169 } 1741 }
1170 1742
1171 /* now add the entry to our list(s) */ 1743 /* now add the entry to our list(s) */
1172 name = entp->d_name; 1744 name = D_NAME (entp);
1173 1745
1174 /* skip . and .. entries */ 1746 /* skip . and .. entries */
1175 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1747 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1176 { 1748 {
1177 int len = D_NAMLEN (entp) + 1; 1749 int len = D_NAMLEN (entp) + 1;
1178 1750
1179 while (expect_false (namesoffs + len > namesalloc)) 1751 while (ecb_expect_false (namesoffs + len > namesalloc))
1180 { 1752 {
1181 namesalloc *= 2; 1753 namesalloc *= 2;
1182 X_LOCK (wrklock);
1183 req->ptr1 = names = realloc (names, namesalloc); 1754 req->ptr2 = names = realloc (names, namesalloc);
1184 X_UNLOCK (wrklock);
1185 1755
1186 if (!names) 1756 if (!names)
1187 break; 1757 break;
1188 } 1758 }
1189 1759
… …
1191 1761
1192 if (dents) 1762 if (dents)
1193 { 1763 {
1194 struct eio_dirent *ent; 1764 struct eio_dirent *ent;
1195 1765
1196 if (expect_false (dentoffs == dentalloc)) 1766 if (ecb_expect_false (dentoffs == dentalloc))
1197 { 1767 {
1198 dentalloc *= 2; 1768 dentalloc *= 2;
1199 X_LOCK (wrklock);
1200 req->ptr2 = dents = realloc (dents, dentalloc * sizeof (eio_dirent)); 1769 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1201 X_UNLOCK (wrklock);
1202 1770
1203 if (!dents) 1771 if (!dents)
1204 break; 1772 break;
1205 } 1773 }
1206 1774
1207 ent = dents + dentoffs; 1775 ent = dents + dentoffs;
1208 1776
1209 ent->name = (char *)(size_t)namesoffs; /* rather dirtily we store the offset in the pointer */ 1777 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1210 ent->namelen = len - 1; 1778 ent->namelen = len - 1;
1211 ent->inode = D_INO (entp); 1779 ent->inode = D_INO (entp);
1780
1781 inode_bits |= ent->inode;
1212 1782
1213 switch (D_TYPE (entp)) 1783 switch (D_TYPE (entp))
1214 { 1784 {
1215 default: 1785 default:
1216 ent->type = EIO_DT_UNKNOWN; 1786 ent->type = EIO_DT_UNKNOWN;
… …
1259 #ifdef DT_WHT 1829 #ifdef DT_WHT
1260 case DT_WHT: ent->type = EIO_DT_WHT; break; 1830 case DT_WHT: ent->type = EIO_DT_WHT; break;
1261 #endif 1831 #endif
1262 } 1832 }
1263 1833
1264 ent->score = 0; 1834 ent->score = 7;
1265 1835
1266 if (flags & EIO_READDIR_DIRS_FIRST) 1836 if (flags & EIO_READDIR_DIRS_FIRST)
1267 { 1837 {
1268 if (ent->type == EIO_DT_UNKNOWN) 1838 if (ent->type == EIO_DT_UNKNOWN)
1269 { 1839 {
1270 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */ 1840 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1271 ent->score = 98; 1841 ent->score = 1;
1272 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */ 1842 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1273 ent->score = len <= 2 ? len + 6 : len <= 4 ? 5 : len <= 7 ? 4 : 1; /* shorter == more likely dir, but avoid too many classes */ 1843 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1274 } 1844 }
1275 else if (ent->type == EIO_DT_DIR) 1845 else if (ent->type == EIO_DT_DIR)
1276 ent->score = 100; 1846 ent->score = 0;
1277 } 1847 }
1278 } 1848 }
1279 1849
1280 namesoffs += len; 1850 namesoffs += len;
1281 ++dentoffs; 1851 ++dentoffs;
… …
1284 if (EIO_CANCELLED (req)) 1854 if (EIO_CANCELLED (req))
1285 { 1855 {
1286 errno = ECANCELED; 1856 errno = ECANCELED;
1287 break; 1857 break;
1288 } 1858 }
1859
1860#ifdef _WIN32
1861 if (!FindNextFile (dirp, &entp))
1862 {
1863 FindClose (dirp);
1864 dirp = 0;
1865 }
1866#endif
1289 } 1867 }
1290}
1291
1292#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1293# undef msync
1294# define msync(a,b,c) ((errno = ENOSYS), -1)
1295#endif
1296
1297int
1298eio__mtouch (void *mem, size_t len, int flags)
1299{
1300 intptr_t addr = (intptr_t)mem;
1301 intptr_t end = addr + len;
1302#ifdef PAGESIZE
1303 const intptr_t page = PAGESIZE;
1304#else
1305 static intptr_t page;
1306
1307 if (!page)
1308 page = sysconf (_SC_PAGESIZE);
1309#endif
1310
1311 addr &= ~(page - 1); /* assume page size is always a power of two */
1312
1313 if (addr < end)
1314 if (flags) /* modify */
1315 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
1316 else
1317 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
1318
1319 return 0;
1320} 1868}
1321 1869
1322/*****************************************************************************/ 1870/*****************************************************************************/
1323 1871
1324#define ALLOC(len) \ 1872#define ALLOC(len) \
… …
1340{ 1888{
1341 ETP_REQ *req; 1889 ETP_REQ *req;
1342 struct timespec ts; 1890 struct timespec ts;
1343 etp_worker *self = (etp_worker *)thr_arg; 1891 etp_worker *self = (etp_worker *)thr_arg;
1344 1892
1345 /* try to distribute timeouts somewhat randomly */ 1893 /* try to distribute timeouts somewhat evenly */
1346 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 1894 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1347 1895
1348 for (;;) 1896 for (;;)
1349 { 1897 {
1898 ts.tv_sec = 0;
1899
1350 X_LOCK (reqlock); 1900 X_LOCK (reqlock);
1351 1901
1352 for (;;) 1902 for (;;)
1353 { 1903 {
1354 self->req = req = reqq_shift (&req_queue); 1904 self->req = req = reqq_shift (&req_queue);
1355 1905
1356 if (req) 1906 if (req)
1357 break; 1907 break;
1358 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
1359 ++idle; 1918 ++idle;
1360 1919
1361 ts.tv_sec = time (0) + IDLE_TIMEOUT; 1920 if (idle <= max_idle)
1362 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
1363 { 1924 {
1364 if (idle > max_idle) 1925 /* initialise timeout once */
1365 { 1926 if (!ts.tv_sec)
1366 --idle; 1927 ts.tv_sec = time (0) + idle_timeout;
1367 X_UNLOCK (reqlock);
1368 X_LOCK (wrklock);
1369 --started;
1370 X_UNLOCK (wrklock);
1371 goto quit;
1372 }
1373 1928
1374 /* we are allowed to idle, so do so without any timeout */
1375 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.,.. */
1376 } 1931 }
1377 1932
1378 --idle; 1933 --idle;
1379 } 1934 }
1380 1935
… …
1383 X_UNLOCK (reqlock); 1938 X_UNLOCK (reqlock);
1384 1939
1385 if (req->type < 0) 1940 if (req->type < 0)
1386 goto quit; 1941 goto quit;
1387 1942
1388 if (!EIO_CANCELLED (req))
1389 ETP_EXECUTE (self, req); 1943 ETP_EXECUTE (self, req);
1390 1944
1391 X_LOCK (reslock); 1945 X_LOCK (reslock);
1392 1946
1393 ++npending; 1947 ++npending;
1394 1948
… …
1409 return 0; 1963 return 0;
1410} 1964}
1411 1965
1412/*****************************************************************************/ 1966/*****************************************************************************/
1413 1967
1968int ecb_cold
1414int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1969eio_init (void (*want_poll)(void), void (*done_poll)(void))
1415{ 1970{
1971#if !HAVE_PREADWRITE
1972 X_MUTEX_CREATE (preadwritelock);
1973#endif
1974
1416 return etp_init (want_poll, done_poll); 1975 return etp_init (want_poll, done_poll);
1417} 1976}
1418 1977
1978ecb_inline void
1419static void eio_api_destroy (eio_req *req) 1979eio_api_destroy (eio_req *req)
1420{ 1980{
1421 free (req); 1981 free (req);
1422} 1982}
1423 1983
1424#define REQ(rtype) \ 1984#define REQ(rtype) \
… …
1443 { \ 2003 { \
1444 eio_api_destroy (req); \ 2004 eio_api_destroy (req); \
1445 return 0; \ 2005 return 0; \
1446 } 2006 }
1447 2007
2008static void
1448static void eio_execute (etp_worker *self, eio_req *req) 2009eio_execute (etp_worker *self, eio_req *req)
1449{ 2010{
1450 errno = 0; 2011 if (ecb_expect_false (EIO_CANCELLED (req)))
2012 {
2013 req->result = -1;
2014 req->errorno = ECANCELED;
2015 return;
2016 }
1451 2017
1452 switch (req->type) 2018 switch (req->type)
1453 { 2019 {
1454 case EIO_READ: ALLOC (req->size); 2020 case EIO_READ: ALLOC (req->size);
1455 req->result = req->offs >= 0 2021 req->result = req->offs >= 0
… …
1458 case EIO_WRITE: req->result = req->offs >= 0 2024 case EIO_WRITE: req->result = req->offs >= 0
1459 ? pwrite (req->int1, req->ptr2, req->size, req->offs) 2025 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1460 : write (req->int1, req->ptr2, req->size); break; 2026 : write (req->int1, req->ptr2, req->size); break;
1461 2027
1462 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;
1463 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;
1464 2030
1465 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2031 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1466 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2032 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1467 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2033 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1468 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2034 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1469 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2035 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1470 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; 2036 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
2037
2038 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2039 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2040 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2041 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1471 2042
1472 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; 2043 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1473 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; 2044 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1474 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break; 2045 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1475 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break; 2046 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
… …
1483 case EIO_RMDIR: req->result = rmdir (req->ptr1); break; 2054 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1484 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;
1485 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; 2056 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1486 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; 2057 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1487 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; 2058 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1488 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;
1489 2060
2061 case EIO_REALPATH: eio__realpath (req, self); break;
2062
1490 case EIO_READLINK: ALLOC (NAME_MAX); 2063 case EIO_READLINK: ALLOC (PATH_MAX);
1491 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; 2064 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1492 2065
1493 case EIO_SYNC: req->result = 0; sync (); break; 2066 case EIO_SYNC: req->result = 0; sync (); break;
1494 case EIO_FSYNC: req->result = fsync (req->int1); break; 2067 case EIO_FSYNC: req->result = fsync (req->int1); break;
1495 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 2068 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1496 case EIO_MSYNC: req->result = 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;
1497 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;
1498 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;
1499 2075
1500 case EIO_READDIR: eio__scandir (req, self); break; 2076 case EIO_READDIR: eio__scandir (req, self); break;
1501 2077
1502 case EIO_BUSY: 2078 case EIO_BUSY:
1503#ifdef _WIN32 2079#ifdef _WIN32
1504 Sleep (req->nv1 * 1000.); 2080 Sleep (req->nv1 * 1e3);
1505#else 2081#else
1506 { 2082 {
1507 struct timeval tv; 2083 struct timeval tv;
1508 2084
1509 tv.tv_sec = req->nv1; 2085 tv.tv_sec = req->nv1;
1510 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; 2086 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1511 2087
1512 req->result = select (0, 0, 0, 0, &tv); 2088 req->result = select (0, 0, 0, 0, &tv);
1513 } 2089 }
1514#endif 2090#endif
1515 break; 2091 break;
… …
1530 times = tv; 2106 times = tv;
1531 } 2107 }
1532 else 2108 else
1533 times = 0; 2109 times = 0;
1534 2110
1535
1536 req->result = req->type == EIO_FUTIME 2111 req->result = req->type == EIO_FUTIME
1537 ? futimes (req->int1, times) 2112 ? futimes (req->int1, times)
1538 : utimes (req->ptr1, times); 2113 : utimes (req->ptr1, times);
1539 } 2114 }
1540 break; 2115 break;
… …
1545 case EIO_NOP: 2120 case EIO_NOP:
1546 req->result = 0; 2121 req->result = 0;
1547 break; 2122 break;
1548 2123
1549 case EIO_CUSTOM: 2124 case EIO_CUSTOM:
1550 ((void (*)(eio_req *))req->feed) (req); 2125 req->feed (req);
1551 break; 2126 break;
1552 2127
1553 default: 2128 default:
2129 errno = ENOSYS;
1554 req->result = -1; 2130 req->result = -1;
1555 break; 2131 break;
1556 } 2132 }
1557 2133
1558 req->errorno = errno; 2134 req->errorno = errno;
… …
1588eio_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)
1589{ 2165{
1590 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;
1591} 2167}
1592 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
1593eio_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)
1594{ 2180{
1595 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;
1596} 2182}
1597 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
1598eio_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)
1599{ 2190{
1600 REQ (EIO_FDATASYNC); req->int1 = fd; SEND; 2191 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1601} 2192}
1602 2193
… …
1623eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) 2214eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1624{ 2215{
1625 REQ (EIO_FSTAT); req->int1 = fd; SEND; 2216 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1626} 2217}
1627 2218
2219eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
2220{
2221 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
2222}
2223
1628eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) 2224eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1629{ 2225{
1630 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; 2226 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1631} 2227}
1632 2228
… …
1638eio_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)
1639{ 2235{
1640 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2236 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1641} 2237}
1642 2238
1643eio_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)
1644{ 2240{
1645 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;
1646} 2242}
1647 2243
1648eio_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)
… …
1668eio_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)
1669{ 2265{
1670 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2266 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1671} 2267}
1672 2268
1673eio_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)
1674{ 2270{
1675 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;
1676} 2272}
1677 2273
1678eio_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)
… …
1694eio_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)
1695{ 2291{
1696 return eio__1path (EIO_READLINK, path, pri, cb, data); 2292 return eio__1path (EIO_READLINK, path, pri, cb, data);
1697} 2293}
1698 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
1699eio_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)
1700{ 2301{
1701 return eio__1path (EIO_STAT, path, pri, cb, data); 2302 return eio__1path (EIO_STAT, path, pri, cb, data);
1702} 2303}
1703 2304
1704eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) 2305eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1705{ 2306{
1706 return eio__1path (EIO_LSTAT, path, pri, cb, data); 2307 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1707} 2308}
1708 2309
2310eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
2311{
2312 return eio__1path (EIO_STATVFS, path, pri, cb, data);
2313}
2314
1709eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) 2315eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1710{ 2316{
1711 return eio__1path (EIO_UNLINK, path, pri, cb, data); 2317 return eio__1path (EIO_UNLINK, path, pri, cb, data);
1712} 2318}
1713 2319
… …
1721 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND; 2327 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1722} 2328}
1723 2329
1724eio_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)
1725{ 2331{
1726 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;
1727} 2333}
1728 2334
1729static eio_req * 2335static eio_req *
1730eio__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)
1731{ 2337{
… …
1755eio_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)
1756{ 2362{
1757 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2363 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1758} 2364}
1759 2365
1760eio_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)
1761{ 2367{
1762 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND; 2368 REQ (EIO_CUSTOM); req->feed = execute; SEND;
1763} 2369}
1764 2370
1765#endif 2371#endif
1766 2372
1767eio_req *eio_grp (eio_cb cb, void *data) 2373eio_req *eio_grp (eio_cb cb, void *data)
… …
1776#undef SEND 2382#undef SEND
1777 2383
1778/*****************************************************************************/ 2384/*****************************************************************************/
1779/* grp functions */ 2385/* grp functions */
1780 2386
2387void
1781void 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)
1782{ 2389{
1783 grp->int2 = limit; 2390 grp->int2 = limit;
1784 grp->feed = feed; 2391 grp->feed = feed;
1785 2392
1786 grp_try_feed (grp); 2393 grp_try_feed (grp);
1787} 2394}
1788 2395
2396void
1789void eio_grp_limit (eio_req *grp, int limit) 2397eio_grp_limit (eio_req *grp, int limit)
1790{ 2398{
1791 grp->int2 = limit; 2399 grp->int2 = limit;
1792 2400
1793 grp_try_feed (grp); 2401 grp_try_feed (grp);
1794} 2402}
1795 2403
2404void
1796void eio_grp_add (eio_req *grp, eio_req *req) 2405eio_grp_add (eio_req *grp, eio_req *req)
1797{ 2406{
1798 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));
1799 2408
1800 grp->flags |= EIO_FLAG_GROUPADD; 2409 grp->flags |= EIO_FLAG_GROUPADD;
1801 2410
… …
1812} 2421}
1813 2422
1814/*****************************************************************************/ 2423/*****************************************************************************/
1815/* misc garbage */ 2424/* misc garbage */
1816 2425
2426eio_ssize_t
1817ssize_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)
1818{ 2428{
1819 etp_worker wrk;
1820 ssize_t ret;
1821
1822 wrk.dbuf = 0;
1823
1824 ret = eio__sendfile (ofd, ifd, offset, count, &wrk); 2429 return eio__sendfile (ofd, ifd, offset, count);
1825
1826 if (wrk.dbuf)
1827 free (wrk.dbuf);
1828
1829 return ret;
1830} 2430}
1831 2431

Diff Legend

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