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

Comparing libeio/eio.c (file contents):
Revision 1.53 by root, Sat Jan 9 10:03:09 2010 UTC vs.
Revision 1.103 by root, Thu Sep 15 13:20:54 2011 UTC

1/* 1/*
2 * libeio implementation 2 * libeio implementation
3 * 3 *
4 * Copyright (c) 2007,2008,2009,2010 Marc Alexander Lehmann <libeio@schmorp.de> 4 * Copyright (c) 2007,2008,2009,2010,2011 Marc Alexander Lehmann <libeio@schmorp.de>
5 * All rights reserved. 5 * All rights reserved.
6 * 6 *
7 * Redistribution and use in source and binary forms, with or without modifica- 7 * Redistribution and use in source and binary forms, with or without modifica-
8 * tion, are permitted provided that the following conditions are met: 8 * tion, are permitted provided that the following conditions are met:
9 * 9 *
… …
35 * and other provisions required by the GPL. If you do not delete the 35 * and other provisions required by the GPL. If you do not delete the
36 * provisions above, a recipient may use your version of this file under 36 * provisions above, a recipient may use your version of this file under
37 * either the BSD or the GPL. 37 * either the BSD or the GPL.
38 */ 38 */
39 39
40#ifndef _WIN32
41# include "config.h"
42#endif
43
40#include "eio.h" 44#include "eio.h"
45#include "ecb.h"
41 46
42#ifdef EIO_STACKSIZE 47#ifdef EIO_STACKSIZE
43# define XTHREAD_STACKSIZE EIO_STACKSIZE 48# define X_STACKSIZE EIO_STACKSIZE
44#endif 49#endif
45#include "xthread.h" 50#include "xthread.h"
46 51
47#include <errno.h> 52#include <errno.h>
48#include <stddef.h> 53#include <stddef.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 /* TODO: look at how perl does stat (non-sloppy), unlink (ro-files), utime, link */
111
112 #ifdef EIO_STRUCT_STATI64
113 /* look at perl's non-sloppy stat */
114 #define stat(path,buf) _stati64 (path,buf)
115 #define fstat(fd,buf) _fstati64 (fd,buf)
116 #endif
117 #define lstat(path,buf) stat (path,buf)
118 #define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
119 #define mkdir(path,mode) _mkdir (path)
120 #define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
121
122 #define chmod(path,mode) _chmod (path, mode)
123 #define dup(fd) _dup (fd)
124 #define dup2(fd1,fd2) _dup2 (fd1, fd2)
125
126 #define fchmod(fd,mode) EIO_ENOSYS ()
127 #define chown(path,uid,gid) EIO_ENOSYS ()
128 #define fchown(fd,uid,gid) EIO_ENOSYS ()
129 #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
130 #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
131 #define mknod(path,mode,dev) EIO_ENOSYS ()
132 #define sync() EIO_ENOSYS ()
133 #define readlink(path,buf,s) EIO_ENOSYS ()
134 #define statvfs(path,buf) EIO_ENOSYS ()
135 #define fstatvfs(fd,buf) EIO_ENOSYS ()
136
137 /* rename() uses MoveFile, which fails to overwrite */
138 #define rename(old,neu) eio__rename (old, neu)
139
140 static int
141 eio__rename (const char *old, const char *neu)
142 {
143 if (MoveFileEx (old, neu, MOVEFILE_REPLACE_EXISTING))
144 return 0;
145
146 /* should steal _dosmaperr */
147 switch (GetLastError ())
148 {
149 case ERROR_FILE_NOT_FOUND:
150 case ERROR_PATH_NOT_FOUND:
151 case ERROR_INVALID_DRIVE:
152 case ERROR_NO_MORE_FILES:
153 case ERROR_BAD_NETPATH:
154 case ERROR_BAD_NET_NAME:
155 case ERROR_BAD_PATHNAME:
156 case ERROR_FILENAME_EXCED_RANGE:
157 errno = ENOENT;
158 break;
159
160 default:
161 errno = EACCES;
162 break;
163 }
164
165 return -1;
166 }
167
168 /* we could even stat and see if it exists */
169 static int
170 symlink (const char *old, const char *neu)
171 {
172 #if WINVER >= 0x0600
173 if (CreateSymbolicLink (neu, old, 1))
174 return 0;
175
176 if (CreateSymbolicLink (neu, old, 0))
177 return 0;
178 #endif
179
180 return EIO_ERRNO (ENOENT, -1);
181 }
182
183 /* POSIX API only */
184 #define CreateHardLink(neu,old,flags) 0
185 #define CreateSymbolicLink(neu,old,flags) 0
186
187 struct statvfs
188 {
189 int dummy;
190 };
191
192 #define DT_DIR EIO_DT_DIR
193 #define DT_REG EIO_DT_REG
194 #define D_NAME(entp) entp.cFileName
195 #define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
196
74#else 197#else
75 198
76# include "config.h"
77# include <sys/time.h> 199 #include <sys/time.h>
78# include <sys/select.h> 200 #include <sys/select.h>
79# include <sys/mman.h> 201 #include <sys/statvfs.h>
80# include <unistd.h> 202 #include <unistd.h>
203 #include <signal.h>
204 #include <dirent.h>
205
206 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
207 #include <sys/mman.h>
208 #endif
209
210 #define D_NAME(entp) entp->d_name
211
212 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
213 #if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
214 #define _DIRENT_HAVE_D_TYPE /* sigh */
215 #define D_INO(de) (de)->d_fileno
216 #define D_NAMLEN(de) (de)->d_namlen
217 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
218 #define D_INO(de) (de)->d_ino
219 #endif
220
221 #ifdef _D_EXACT_NAMLEN
222 #undef D_NAMLEN
223 #define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
224 #endif
225
226 #ifdef _DIRENT_HAVE_D_TYPE
227 #define D_TYPE(de) (de)->d_type
228 #endif
229
230 #ifndef EIO_STRUCT_DIRENT
231 #define EIO_STRUCT_DIRENT struct dirent
232 #endif
233
234#endif
235
236#if HAVE_UTIMES
81# include <utime.h> 237# include <utime.h>
82# include <signal.h>
83# include <dirent.h>
84
85/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
86# if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
87# define _DIRENT_HAVE_D_TYPE /* sigh */
88# define D_INO(de) (de)->d_fileno
89# define D_NAMLEN(de) (de)->d_namlen
90# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
91# define D_INO(de) (de)->d_ino
92# endif 238#endif
93 239
94#ifdef _D_EXACT_NAMLEN 240#if HAVE_SYS_SYSCALL_H
95# undef D_NAMLEN 241# include <sys/syscall.h>
96# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
97#endif
98
99# ifdef _DIRENT_HAVE_D_TYPE
100# define D_TYPE(de) (de)->d_type
101# endif 242#endif
102 243
103# ifndef EIO_STRUCT_DIRENT 244#if HAVE_SYS_PRCTL_H
104# define EIO_STRUCT_DIRENT struct dirent 245# include <sys/prctl.h>
105# endif
106
107#endif 246#endif
108 247
109#if HAVE_SENDFILE 248#if HAVE_SENDFILE
110# if __linux 249# if __linux
111# include <sys/sendfile.h> 250# include <sys/sendfile.h>
… …
126#endif 265#endif
127#ifndef D_INO 266#ifndef D_INO
128# define D_INO(de) 0 267# define D_INO(de) 0
129#endif 268#endif
130#ifndef D_NAMLEN 269#ifndef D_NAMLEN
131# define D_NAMLEN(de) strlen ((de)->d_name) 270# define D_NAMLEN(entp) strlen (D_NAME (entp))
132#endif 271#endif
133
134/* number of seconds after which an idle threads exit */
135#define IDLE_TIMEOUT 10
136 272
137/* used for struct dirent, AIX doesn't provide it */ 273/* used for struct dirent, AIX doesn't provide it */
138#ifndef NAME_MAX 274#ifndef NAME_MAX
139# define NAME_MAX 4096 275# define NAME_MAX 4096
140#endif 276#endif
141 277
278/* used for readlink etc. */
279#ifndef PATH_MAX
280# define PATH_MAX 4096
281#endif
282
142/* buffer size for various temporary buffers */ 283/* buffer size for various temporary buffers */
143#define EIO_BUFSIZE 65536 284#define EIO_BUFSIZE 65536
144 285
145#define dBUF \ 286#define dBUF \
146 char *eio_buf; \
147 ETP_WORKER_LOCK (self); \
148 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 287 char *eio_buf = malloc (EIO_BUFSIZE); \
149 ETP_WORKER_UNLOCK (self); \
150 errno = ENOMEM; \ 288 errno = ENOMEM; \
151 if (!eio_buf) \ 289 if (!eio_buf) \
152 return -1; 290 return -1
291
292#define FUBd \
293 free (eio_buf)
153 294
154#define EIO_TICKS ((1000000 + 1023) >> 10) 295#define EIO_TICKS ((1000000 + 1023) >> 10)
155
156/*****************************************************************************/
157
158#if __GNUC__ >= 3
159# define expect(expr,value) __builtin_expect ((expr),(value))
160#else
161# define expect(expr,value) (expr)
162#endif
163
164#define expect_false(expr) expect ((expr) != 0, 0)
165#define expect_true(expr) expect ((expr) != 0, 1)
166
167/*****************************************************************************/
168 296
169#define ETP_PRI_MIN EIO_PRI_MIN 297#define ETP_PRI_MIN EIO_PRI_MIN
170#define ETP_PRI_MAX EIO_PRI_MAX 298#define ETP_PRI_MAX EIO_PRI_MAX
171 299
172struct etp_worker; 300struct etp_worker;
… …
176static int eio_finish (eio_req *req); 304static int eio_finish (eio_req *req);
177#define ETP_FINISH(req) eio_finish (req) 305#define ETP_FINISH(req) eio_finish (req)
178static void eio_execute (struct etp_worker *self, eio_req *req); 306static void eio_execute (struct etp_worker *self, eio_req *req);
179#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) 307#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
180 308
181#define ETP_WORKER_CLEAR(req) \
182 if (wrk->dbuf) \
183 { \
184 free (wrk->dbuf); \
185 wrk->dbuf = 0; \
186 } \
187 \
188 if (wrk->dirp) \
189 { \
190 closedir (wrk->dirp); \
191 wrk->dirp = 0; \
192 }
193
194#define ETP_WORKER_COMMON \
195 void *dbuf; \
196 DIR *dirp;
197
198/*****************************************************************************/ 309/*****************************************************************************/
199 310
200#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) 311#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
201 312
202/* calculate time difference in ~1/EIO_TICKS of a second */ 313/* calculate time difference in ~1/EIO_TICKS of a second */
314ecb_inline int
203static int tvdiff (struct timeval *tv1, struct timeval *tv2) 315tvdiff (struct timeval *tv1, struct timeval *tv2)
204{ 316{
205 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 317 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
206 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 318 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
207} 319}
208 320
… …
212static void (*done_poll_cb) (void); 324static void (*done_poll_cb) (void);
213 325
214static unsigned int max_poll_time; /* reslock */ 326static unsigned int max_poll_time; /* reslock */
215static unsigned int max_poll_reqs; /* reslock */ 327static unsigned int max_poll_reqs; /* reslock */
216 328
217static volatile unsigned int nreqs; /* reqlock */ 329static unsigned int nreqs; /* reqlock */
218static volatile unsigned int nready; /* reqlock */ 330static unsigned int nready; /* reqlock */
219static volatile unsigned int npending; /* reqlock */ 331static unsigned int npending; /* reqlock */
220static volatile unsigned int max_idle = 4; 332static unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
333static unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
221 334
222static mutex_t wrklock = X_MUTEX_INIT; 335static xmutex_t wrklock;
223static mutex_t reslock = X_MUTEX_INIT; 336static xmutex_t reslock;
224static mutex_t reqlock = X_MUTEX_INIT; 337static xmutex_t reqlock;
225static cond_t reqwait = X_COND_INIT; 338static xcond_t reqwait;
226 339
227#if !HAVE_PREADWRITE 340#if !HAVE_PREADWRITE
228/* 341/*
229 * make our pread/pwrite emulation safe against themselves, but not against 342 * make our pread/pwrite emulation safe against themselves, but not against
230 * normal read/write by using a mutex. slows down execution a lot, 343 * normal read/write by using a mutex. slows down execution a lot,
231 * but that's your problem, not mine. 344 * but that's your problem, not mine.
232 */ 345 */
233static mutex_t preadwritelock = X_MUTEX_INIT; 346static xmutex_t preadwritelock;
234#endif 347#endif
235 348
236typedef struct etp_worker 349typedef struct etp_worker
237{ 350{
238 /* locked by wrklock */ 351 /* locked by wrklock */
239 struct etp_worker *prev, *next; 352 struct etp_worker *prev, *next;
240 353
241 thread_t tid; 354 xthread_t tid;
242 355
243 /* locked by reslock, reqlock or wrklock */ 356 /* locked by reslock, reqlock or wrklock */
244 ETP_REQ *req; /* currently processed request */ 357 ETP_REQ *req; /* currently processed request */
245 358
359#ifdef ETP_WORKER_COMMON
246 ETP_WORKER_COMMON 360 ETP_WORKER_COMMON
361#endif
247} etp_worker; 362} etp_worker;
248 363
249static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */ 364static etp_worker wrk_first; /* NOT etp */
250 365
251#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock) 366#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
252#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock) 367#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
253 368
254/* worker threads management */ 369/* worker threads management */
255 370
371static void ecb_cold
256static void etp_worker_clear (etp_worker *wrk) 372etp_worker_clear (etp_worker *wrk)
257{ 373{
258 ETP_WORKER_CLEAR (wrk);
259} 374}
260 375
376static void ecb_cold
261static void etp_worker_free (etp_worker *wrk) 377etp_worker_free (etp_worker *wrk)
262{ 378{
263 wrk->next->prev = wrk->prev; 379 wrk->next->prev = wrk->prev;
264 wrk->prev->next = wrk->next; 380 wrk->prev->next = wrk->next;
265 381
266 free (wrk); 382 free (wrk);
267} 383}
268 384
269static unsigned int etp_nreqs (void) 385static unsigned int
386etp_nreqs (void)
270{ 387{
271 int retval; 388 int retval;
272 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 389 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
273 retval = nreqs; 390 retval = nreqs;
274 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 391 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
275 return retval; 392 return retval;
276} 393}
277 394
278static unsigned int etp_nready (void) 395static unsigned int
396etp_nready (void)
279{ 397{
280 unsigned int retval; 398 unsigned int retval;
281 399
282 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 400 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
283 retval = nready; 401 retval = nready;
284 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 402 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
285 403
286 return retval; 404 return retval;
287} 405}
288 406
289static unsigned int etp_npending (void) 407static unsigned int
408etp_npending (void)
290{ 409{
291 unsigned int retval; 410 unsigned int retval;
292 411
293 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 412 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
294 retval = npending; 413 retval = npending;
295 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 414 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
296 415
297 return retval; 416 return retval;
298} 417}
299 418
300static unsigned int etp_nthreads (void) 419static unsigned int
420etp_nthreads (void)
301{ 421{
302 unsigned int retval; 422 unsigned int retval;
303 423
304 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 424 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
305 retval = started; 425 retval = started;
… …
319} etp_reqq; 439} etp_reqq;
320 440
321static etp_reqq req_queue; 441static etp_reqq req_queue;
322static etp_reqq res_queue; 442static etp_reqq res_queue;
323 443
444static void ecb_noinline ecb_cold
445reqq_init (etp_reqq *q)
446{
447 int pri;
448
449 for (pri = 0; pri < ETP_NUM_PRI; ++pri)
450 q->qs[pri] = q->qe[pri] = 0;
451
452 q->size = 0;
453}
454
455static int ecb_noinline
324static int reqq_push (etp_reqq *q, ETP_REQ *req) 456reqq_push (etp_reqq *q, ETP_REQ *req)
325{ 457{
326 int pri = req->pri; 458 int pri = req->pri;
327 req->next = 0; 459 req->next = 0;
328 460
329 if (q->qe[pri]) 461 if (q->qe[pri])
… …
335 q->qe[pri] = q->qs[pri] = req; 467 q->qe[pri] = q->qs[pri] = req;
336 468
337 return q->size++; 469 return q->size++;
338} 470}
339 471
472static ETP_REQ * ecb_noinline
340static ETP_REQ *reqq_shift (etp_reqq *q) 473reqq_shift (etp_reqq *q)
341{ 474{
342 int pri; 475 int pri;
343 476
344 if (!q->size) 477 if (!q->size)
345 return 0; 478 return 0;
… …
360 } 493 }
361 494
362 abort (); 495 abort ();
363} 496}
364 497
365static void etp_atfork_prepare (void) 498static int ecb_cold
499etp_init (void (*want_poll)(void), void (*done_poll)(void))
366{ 500{
367 X_LOCK (wrklock); 501 X_MUTEX_CREATE (wrklock);
368 X_LOCK (reqlock); 502 X_MUTEX_CREATE (reslock);
369 X_LOCK (reslock); 503 X_MUTEX_CREATE (reqlock);
370#if !HAVE_PREADWRITE 504 X_COND_CREATE (reqwait);
371 X_LOCK (preadwritelock);
372#endif
373}
374 505
375static void etp_atfork_parent (void) 506 reqq_init (&req_queue);
376{ 507 reqq_init (&res_queue);
377#if !HAVE_PREADWRITE
378 X_UNLOCK (preadwritelock);
379#endif
380 X_UNLOCK (reslock);
381 X_UNLOCK (reqlock);
382 X_UNLOCK (wrklock);
383}
384 508
385static void etp_atfork_child (void) 509 wrk_first.next =
386{ 510 wrk_first.prev = &wrk_first;
387 ETP_REQ *prv;
388
389 while ((prv = reqq_shift (&req_queue)))
390 ETP_DESTROY (prv);
391
392 while ((prv = reqq_shift (&res_queue)))
393 ETP_DESTROY (prv);
394
395 while (wrk_first.next != &wrk_first)
396 {
397 etp_worker *wrk = wrk_first.next;
398
399 if (wrk->req)
400 ETP_DESTROY (wrk->req);
401
402 etp_worker_clear (wrk);
403 etp_worker_free (wrk);
404 }
405 511
406 started = 0; 512 started = 0;
407 idle = 0; 513 idle = 0;
408 nreqs = 0; 514 nreqs = 0;
409 nready = 0; 515 nready = 0;
410 npending = 0; 516 npending = 0;
411 517
412 etp_atfork_parent ();
413}
414
415static void
416etp_once_init (void)
417{
418 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
419}
420
421static int
422etp_init (void (*want_poll)(void), void (*done_poll)(void))
423{
424 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
425
426 pthread_once (&doinit, etp_once_init);
427
428 want_poll_cb = want_poll; 518 want_poll_cb = want_poll;
429 done_poll_cb = done_poll; 519 done_poll_cb = done_poll;
430 520
431 return 0; 521 return 0;
432} 522}
433 523
434X_THREAD_PROC (etp_proc); 524X_THREAD_PROC (etp_proc);
435 525
526static void ecb_cold
436static void etp_start_thread (void) 527etp_start_thread (void)
437{ 528{
438 etp_worker *wrk = calloc (1, sizeof (etp_worker)); 529 etp_worker *wrk = calloc (1, sizeof (etp_worker));
439 530
440 /*TODO*/ 531 /*TODO*/
441 assert (("unable to allocate worker thread data", wrk)); 532 assert (("unable to allocate worker thread data", wrk));
… …
454 free (wrk); 545 free (wrk);
455 546
456 X_UNLOCK (wrklock); 547 X_UNLOCK (wrklock);
457} 548}
458 549
550static void
459static void etp_maybe_start_thread (void) 551etp_maybe_start_thread (void)
460{ 552{
461 if (expect_true (etp_nthreads () >= wanted)) 553 if (ecb_expect_true (etp_nthreads () >= wanted))
462 return; 554 return;
463 555
464 /* todo: maybe use idle here, but might be less exact */ 556 /* todo: maybe use idle here, but might be less exact */
465 if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ())) 557 if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
466 return; 558 return;
467 559
468 etp_start_thread (); 560 etp_start_thread ();
469} 561}
470 562
563static void ecb_cold
471static void etp_end_thread (void) 564etp_end_thread (void)
472{ 565{
473 eio_req *req = calloc (1, sizeof (eio_req)); 566 eio_req *req = calloc (1, sizeof (eio_req));
474 567
475 req->type = -1; 568 req->type = -1;
476 req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 569 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
… …
483 X_LOCK (wrklock); 576 X_LOCK (wrklock);
484 --started; 577 --started;
485 X_UNLOCK (wrklock); 578 X_UNLOCK (wrklock);
486} 579}
487 580
488static int etp_poll (void) 581static int
582etp_poll (void)
489{ 583{
490 unsigned int maxreqs; 584 unsigned int maxreqs;
491 unsigned int maxtime; 585 unsigned int maxtime;
492 struct timeval tv_start, tv_now; 586 struct timeval tv_start, tv_now;
493 587
… …
523 617
524 X_LOCK (reqlock); 618 X_LOCK (reqlock);
525 --nreqs; 619 --nreqs;
526 X_UNLOCK (reqlock); 620 X_UNLOCK (reqlock);
527 621
528 if (expect_false (req->type == EIO_GROUP && req->size)) 622 if (ecb_expect_false (req->type == EIO_GROUP && req->size))
529 { 623 {
530 req->int1 = 1; /* mark request as delayed */ 624 req->int1 = 1; /* mark request as delayed */
531 continue; 625 continue;
532 } 626 }
533 else 627 else
534 { 628 {
535 int res = ETP_FINISH (req); 629 int res = ETP_FINISH (req);
536 if (expect_false (res)) 630 if (ecb_expect_false (res))
537 return res; 631 return res;
538 } 632 }
539 633
540 if (expect_false (maxreqs && !--maxreqs)) 634 if (ecb_expect_false (maxreqs && !--maxreqs))
541 break; 635 break;
542 636
543 if (maxtime) 637 if (maxtime)
544 { 638 {
545 gettimeofday (&tv_now, 0); 639 gettimeofday (&tv_now, 0);
… …
551 645
552 errno = EAGAIN; 646 errno = EAGAIN;
553 return -1; 647 return -1;
554} 648}
555 649
650static void
556static void etp_cancel (ETP_REQ *req) 651etp_cancel (ETP_REQ *req)
557{ 652{
558 X_LOCK (wrklock); 653 req->cancelled = 1;
559 req->flags |= EIO_FLAG_CANCELLED;
560 X_UNLOCK (wrklock);
561 654
562 eio_grp_cancel (req); 655 eio_grp_cancel (req);
563} 656}
564 657
658static void
565static void etp_submit (ETP_REQ *req) 659etp_submit (ETP_REQ *req)
566{ 660{
567 req->pri -= ETP_PRI_MIN; 661 req->pri -= ETP_PRI_MIN;
568 662
569 if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN; 663 if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
570 if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 664 if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
571 665
572 if (expect_false (req->type == EIO_GROUP)) 666 if (ecb_expect_false (req->type == EIO_GROUP))
573 { 667 {
574 /* I hope this is worth it :/ */ 668 /* I hope this is worth it :/ */
575 X_LOCK (reqlock); 669 X_LOCK (reqlock);
576 ++nreqs; 670 ++nreqs;
577 X_UNLOCK (reqlock); 671 X_UNLOCK (reqlock);
… …
596 690
597 etp_maybe_start_thread (); 691 etp_maybe_start_thread ();
598 } 692 }
599} 693}
600 694
695static void ecb_cold
601static void etp_set_max_poll_time (double nseconds) 696etp_set_max_poll_time (double nseconds)
602{ 697{
603 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 698 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
604 max_poll_time = nseconds * EIO_TICKS; 699 max_poll_time = nseconds * EIO_TICKS;
605 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 700 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
606} 701}
607 702
703static void ecb_cold
608static void etp_set_max_poll_reqs (unsigned int maxreqs) 704etp_set_max_poll_reqs (unsigned int maxreqs)
609{ 705{
610 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 706 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
611 max_poll_reqs = maxreqs; 707 max_poll_reqs = maxreqs;
612 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 708 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
613} 709}
614 710
711static void ecb_cold
615static void etp_set_max_idle (unsigned int nthreads) 712etp_set_max_idle (unsigned int nthreads)
616{ 713{
617 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 714 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
618 max_idle = nthreads <= 0 ? 1 : nthreads; 715 max_idle = nthreads;
619 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 716 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
620} 717}
621 718
719static void ecb_cold
720etp_set_idle_timeout (unsigned int seconds)
721{
722 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
723 idle_timeout = seconds;
724 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
725}
726
727static void ecb_cold
622static void etp_set_min_parallel (unsigned int nthreads) 728etp_set_min_parallel (unsigned int nthreads)
623{ 729{
624 if (wanted < nthreads) 730 if (wanted < nthreads)
625 wanted = nthreads; 731 wanted = nthreads;
626} 732}
627 733
734static void ecb_cold
628static void etp_set_max_parallel (unsigned int nthreads) 735etp_set_max_parallel (unsigned int nthreads)
629{ 736{
630 if (wanted > nthreads) 737 if (wanted > nthreads)
631 wanted = nthreads; 738 wanted = nthreads;
632 739
633 while (started > wanted) 740 while (started > wanted)
634 etp_end_thread (); 741 etp_end_thread ();
635} 742}
636 743
637/*****************************************************************************/ 744/*****************************************************************************/
638 745
746static void
639static void grp_try_feed (eio_req *grp) 747grp_try_feed (eio_req *grp)
640{ 748{
641 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 749 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
642 { 750 {
643 grp->flags &= ~EIO_FLAG_GROUPADD; 751 grp->flags &= ~EIO_FLAG_GROUPADD;
644 752
… …
651 break; 759 break;
652 } 760 }
653 } 761 }
654} 762}
655 763
764static int
656static int grp_dec (eio_req *grp) 765grp_dec (eio_req *grp)
657{ 766{
658 --grp->size; 767 --grp->size;
659 768
660 /* call feeder, if applicable */ 769 /* call feeder, if applicable */
661 grp_try_feed (grp); 770 grp_try_feed (grp);
… …
665 return eio_finish (grp); 774 return eio_finish (grp);
666 else 775 else
667 return 0; 776 return 0;
668} 777}
669 778
779static void
670void eio_destroy (eio_req *req) 780eio_destroy (eio_req *req)
671{ 781{
672 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); 782 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
673 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); 783 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
674 784
675 EIO_DESTROY (req); 785 EIO_DESTROY (req);
676} 786}
677 787
788static int
678static int eio_finish (eio_req *req) 789eio_finish (eio_req *req)
679{ 790{
680 int res = EIO_FINISH (req); 791 int res = EIO_FINISH (req);
681 792
682 if (req->grp) 793 if (req->grp)
683 { 794 {
… …
691 if (grp->grp_first == req) 802 if (grp->grp_first == req)
692 grp->grp_first = req->grp_next; 803 grp->grp_first = req->grp_next;
693 804
694 res2 = grp_dec (grp); 805 res2 = grp_dec (grp);
695 806
696 if (!res && res2) 807 if (!res)
697 res = res2; 808 res = res2;
698 } 809 }
699 810
700 eio_destroy (req); 811 eio_destroy (req);
701 812
702 return res; 813 return res;
703} 814}
704 815
816void
705void eio_grp_cancel (eio_req *grp) 817eio_grp_cancel (eio_req *grp)
706{ 818{
707 for (grp = grp->grp_first; grp; grp = grp->grp_next) 819 for (grp = grp->grp_first; grp; grp = grp->grp_next)
708 eio_cancel (grp); 820 eio_cancel (grp);
709} 821}
710 822
823void
711void eio_cancel (eio_req *req) 824eio_cancel (eio_req *req)
712{ 825{
713 etp_cancel (req); 826 etp_cancel (req);
714} 827}
715 828
829void
716void eio_submit (eio_req *req) 830eio_submit (eio_req *req)
717{ 831{
718 etp_submit (req); 832 etp_submit (req);
719} 833}
720 834
721unsigned int eio_nreqs (void) 835unsigned int
836eio_nreqs (void)
722{ 837{
723 return etp_nreqs (); 838 return etp_nreqs ();
724} 839}
725 840
726unsigned int eio_nready (void) 841unsigned int
842eio_nready (void)
727{ 843{
728 return etp_nready (); 844 return etp_nready ();
729} 845}
730 846
731unsigned int eio_npending (void) 847unsigned int
848eio_npending (void)
732{ 849{
733 return etp_npending (); 850 return etp_npending ();
734} 851}
735 852
736unsigned int eio_nthreads (void) 853unsigned int ecb_cold
854eio_nthreads (void)
737{ 855{
738 return etp_nthreads (); 856 return etp_nthreads ();
739} 857}
740 858
859void ecb_cold
741void eio_set_max_poll_time (double nseconds) 860eio_set_max_poll_time (double nseconds)
742{ 861{
743 etp_set_max_poll_time (nseconds); 862 etp_set_max_poll_time (nseconds);
744} 863}
745 864
865void ecb_cold
746void eio_set_max_poll_reqs (unsigned int maxreqs) 866eio_set_max_poll_reqs (unsigned int maxreqs)
747{ 867{
748 etp_set_max_poll_reqs (maxreqs); 868 etp_set_max_poll_reqs (maxreqs);
749} 869}
750 870
871void ecb_cold
751void eio_set_max_idle (unsigned int nthreads) 872eio_set_max_idle (unsigned int nthreads)
752{ 873{
753 etp_set_max_idle (nthreads); 874 etp_set_max_idle (nthreads);
754} 875}
755 876
877void ecb_cold
878eio_set_idle_timeout (unsigned int seconds)
879{
880 etp_set_idle_timeout (seconds);
881}
882
883void ecb_cold
756void eio_set_min_parallel (unsigned int nthreads) 884eio_set_min_parallel (unsigned int nthreads)
757{ 885{
758 etp_set_min_parallel (nthreads); 886 etp_set_min_parallel (nthreads);
759} 887}
760 888
889void ecb_cold
761void eio_set_max_parallel (unsigned int nthreads) 890eio_set_max_parallel (unsigned int nthreads)
762{ 891{
763 etp_set_max_parallel (nthreads); 892 etp_set_max_parallel (nthreads);
764} 893}
765 894
766int eio_poll (void) 895int eio_poll (void)
… …
775# undef pread 904# undef pread
776# undef pwrite 905# undef pwrite
777# define pread eio__pread 906# define pread eio__pread
778# define pwrite eio__pwrite 907# define pwrite eio__pwrite
779 908
780static ssize_t 909static eio_ssize_t
781eio__pread (int fd, void *buf, size_t count, off_t offset) 910eio__pread (int fd, void *buf, size_t count, off_t offset)
782{ 911{
783 ssize_t res; 912 eio_ssize_t res;
784 off_t ooffset; 913 off_t ooffset;
785 914
786 X_LOCK (preadwritelock); 915 X_LOCK (preadwritelock);
787 ooffset = lseek (fd, 0, SEEK_CUR); 916 ooffset = lseek (fd, 0, SEEK_CUR);
788 lseek (fd, offset, SEEK_SET); 917 lseek (fd, offset, SEEK_SET);
… …
791 X_UNLOCK (preadwritelock); 920 X_UNLOCK (preadwritelock);
792 921
793 return res; 922 return res;
794} 923}
795 924
796static ssize_t 925static eio_ssize_t
797eio__pwrite (int fd, void *buf, size_t count, off_t offset) 926eio__pwrite (int fd, void *buf, size_t count, off_t offset)
798{ 927{
799 ssize_t res; 928 eio_ssize_t res;
800 off_t ooffset; 929 off_t ooffset;
801 930
802 X_LOCK (preadwritelock); 931 X_LOCK (preadwritelock);
803 ooffset = lseek (fd, 0, SEEK_CUR); 932 ooffset = lseek (fd, 0, SEEK_CUR);
804 lseek (fd, offset, SEEK_SET); 933 lseek (fd, offset, SEEK_SET);
… …
808 937
809 return res; 938 return res;
810} 939}
811#endif 940#endif
812 941
813#ifndef HAVE_FUTIMES 942#ifndef HAVE_UTIMES
814 943
815# undef utimes 944# undef utimes
816# undef futimes
817# define utimes(path,times) eio__utimes (path, times) 945# define utimes(path,times) eio__utimes (path, times)
818# define futimes(fd,times) eio__futimes (fd, times)
819 946
820static int 947static int
821eio__utimes (const char *filename, const struct timeval times[2]) 948eio__utimes (const char *filename, const struct timeval times[2])
822{ 949{
823 if (times) 950 if (times)
… …
831 } 958 }
832 else 959 else
833 return utime (filename, 0); 960 return utime (filename, 0);
834} 961}
835 962
963#endif
964
965#ifndef HAVE_FUTIMES
966
967# undef futimes
968# define futimes(fd,times) eio__futimes (fd, times)
969
970static int
836static int eio__futimes (int fd, const struct timeval tv[2]) 971eio__futimes (int fd, const struct timeval tv[2])
837{ 972{
838 errno = ENOSYS; 973 errno = ENOSYS;
839 return -1; 974 return -1;
840} 975}
841 976
… …
844#if !HAVE_FDATASYNC 979#if !HAVE_FDATASYNC
845# undef fdatasync 980# undef fdatasync
846# define fdatasync(fd) fsync (fd) 981# define fdatasync(fd) fsync (fd)
847#endif 982#endif
848 983
984static int
985eio__syncfs (int fd)
986{
987 int res;
988
989#if HAVE_SYS_SYNCFS
990 res = (int)syscall (__NR_syncfs, (int)(fd));
991#else
992 res = -1;
993 errno = ENOSYS;
994#endif
995
996 if (res < 0 && errno == ENOSYS && fd >= 0)
997 sync ();
998
999 return res;
1000}
1001
849/* sync_file_range always needs emulation */ 1002/* sync_file_range always needs emulation */
850int 1003static int
851eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags) 1004eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
852{ 1005{
853#if HAVE_SYNC_FILE_RANGE 1006#if HAVE_SYNC_FILE_RANGE
854 int res; 1007 int res;
855 1008
… …
872 /* even though we could play tricks with the flags, it's better to always 1025 /* even though we could play tricks with the flags, it's better to always
873 * call fdatasync, as that matches the expectation of its users best */ 1026 * call fdatasync, as that matches the expectation of its users best */
874 return fdatasync (fd); 1027 return fdatasync (fd);
875} 1028}
876 1029
1030static int
1031eio__fallocate (int fd, int mode, off_t offset, size_t len)
1032{
1033#if HAVE_FALLOCATE
1034 return fallocate (fd, mode, offset, len);
1035#else
1036 errno = ENOSYS;
1037 return -1;
1038#endif
1039}
1040
877#if !HAVE_READAHEAD 1041#if !HAVE_READAHEAD
878# undef readahead 1042# undef readahead
879# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 1043# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
880 1044
881static ssize_t 1045static eio_ssize_t
882eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) 1046eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
883{ 1047{
884 size_t todo = count; 1048 size_t todo = count;
885 dBUF; 1049 dBUF;
886 1050
… …
891 pread (fd, eio_buf, len, offset); 1055 pread (fd, eio_buf, len, offset);
892 offset += len; 1056 offset += len;
893 todo -= len; 1057 todo -= len;
894 } 1058 }
895 1059
1060 FUBd;
1061
896 errno = 0; 1062 errno = 0;
897 return count; 1063 return count;
898} 1064}
899 1065
900#endif 1066#endif
901 1067
902/* sendfile always needs emulation */ 1068/* sendfile always needs emulation */
903static ssize_t 1069static eio_ssize_t
904eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) 1070eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
905{ 1071{
1072 eio_ssize_t written = 0;
906 ssize_t res; 1073 eio_ssize_t res;
907 1074
908 if (!count) 1075 if (!count)
909 return 0; 1076 return 0;
910 1077
1078 for (;;)
1079 {
1080#ifdef __APPLE__
1081# undef HAVE_SENDFILE /* broken, as everything on os x */
1082#endif
911#if HAVE_SENDFILE 1083#if HAVE_SENDFILE
912# if __linux 1084# if __linux
1085 off_t soffset = offset;
913 res = sendfile (ofd, ifd, &offset, count); 1086 res = sendfile (ofd, ifd, &soffset, count);
914 1087
915# elif __FreeBSD__ 1088# elif __FreeBSD__
916 /* 1089 /*
917 * Of course, the freebsd sendfile is a dire hack with no thoughts 1090 * Of course, the freebsd sendfile is a dire hack with no thoughts
918 * wasted on making it similar to other I/O functions. 1091 * wasted on making it similar to other I/O functions.
919 */ 1092 */
920 {
921 off_t sbytes; 1093 off_t sbytes;
922 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 1094 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
923 1095
924 #if 0 /* according to the manpage, this is correct, but broken behaviour */ 1096 #if 0 /* according to the manpage, this is correct, but broken behaviour */
925 /* freebsd' sendfile will return 0 on success */ 1097 /* freebsd' sendfile will return 0 on success */
926 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */ 1098 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
927 /* not on e.g. EIO or EPIPE - sounds broken */ 1099 /* not on e.g. EIO or EPIPE - sounds broken */
928 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0) 1100 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
929 res = sbytes; 1101 res = sbytes;
930 #endif 1102 #endif
931 1103
932 /* according to source inspection, this is correct, and useful behaviour */ 1104 /* according to source inspection, this is correct, and useful behaviour */
933 if (sbytes) 1105 if (sbytes)
934 res = sbytes; 1106 res = sbytes;
935 }
936 1107
937# elif defined (__APPLE__) 1108# elif defined (__APPLE__)
938
939 {
940 off_t sbytes = count; 1109 off_t sbytes = count;
941 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0); 1110 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
942 1111
943 /* according to the manpage, sbytes is always valid */ 1112 /* according to the manpage, sbytes is always valid */
944 if (sbytes) 1113 if (sbytes)
945 res = sbytes; 1114 res = sbytes;
946 }
947 1115
948# elif __hpux 1116# elif __hpux
949 res = sendfile (ofd, ifd, offset, count, 0, 0); 1117 res = sendfile (ofd, ifd, offset, count, 0, 0);
950 1118
951# elif __solaris 1119# elif __solaris
952 {
953 struct sendfilevec vec; 1120 struct sendfilevec vec;
954 size_t sbytes; 1121 size_t sbytes;
955 1122
956 vec.sfv_fd = ifd; 1123 vec.sfv_fd = ifd;
957 vec.sfv_flag = 0; 1124 vec.sfv_flag = 0;
958 vec.sfv_off = offset; 1125 vec.sfv_off = offset;
959 vec.sfv_len = count; 1126 vec.sfv_len = count;
960 1127
961 res = sendfilev (ofd, &vec, 1, &sbytes); 1128 res = sendfilev (ofd, &vec, 1, &sbytes);
962 1129
963 if (res < 0 && sbytes) 1130 if (res < 0 && sbytes)
964 res = sbytes; 1131 res = sbytes;
965 }
966 1132
967# endif 1133# endif
968 1134
969#elif defined (_WIN32) 1135#elif defined (_WIN32) && 0
970
971 /* does not work, just for documentation of what would need to be done */ 1136 /* does not work, just for documentation of what would need to be done */
972 { 1137 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
1138 /* libeio guarantees that the file offset does not change, and windows */
1139 /* has no way to get an independent handle to the same file description */
973 HANDLE h = TO_SOCKET (ifd); 1140 HANDLE h = TO_SOCKET (ifd);
974 SetFilePointer (h, offset, 0, FILE_BEGIN); 1141 SetFilePointer (h, offset, 0, FILE_BEGIN);
975 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0); 1142 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
976 }
977 1143
978#else 1144#else
979 res = -1; 1145 res = -1;
980 errno = ENOSYS; 1146 errno = ENOSYS;
981#endif 1147#endif
982 1148
1149 /* we assume sendfile can copy at least 128mb in one go */
1150 if (res <= 128 * 1024 * 1024)
1151 {
1152 if (res > 0)
1153 written += res;
1154
1155 if (written)
1156 return written;
1157
1158 break;
1159 }
1160 else
1161 {
1162 /* if we requested more, then probably the kernel was lazy */
1163 written += res;
1164 offset += res;
1165 count -= res;
1166
1167 if (!count)
1168 return written;
1169 }
1170 }
1171
983 if (res < 0 1172 if (res < 0
984 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1173 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
985 /* BSDs */ 1174 /* BSDs */
986#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */ 1175#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
987 || errno == ENOTSUP 1176 || errno == ENOTSUP
988#endif 1177#endif
1178#ifdef EOPNOTSUPP /* windows */
989 || errno == EOPNOTSUPP /* BSDs */ 1179 || errno == EOPNOTSUPP /* BSDs */
1180#endif
990#if __solaris 1181#if __solaris
991 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1182 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
992#endif 1183#endif
993 ) 1184 )
994 ) 1185 )
… …
998 1189
999 res = 0; 1190 res = 0;
1000 1191
1001 while (count) 1192 while (count)
1002 { 1193 {
1003 ssize_t cnt; 1194 eio_ssize_t cnt;
1004 1195
1005 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 1196 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
1006 1197
1007 if (cnt <= 0) 1198 if (cnt <= 0)
1008 { 1199 {
… …
1020 1211
1021 offset += cnt; 1212 offset += cnt;
1022 res += cnt; 1213 res += cnt;
1023 count -= cnt; 1214 count -= cnt;
1024 } 1215 }
1216
1217 FUBd;
1025 } 1218 }
1026 1219
1027 return res; 1220 return res;
1221}
1222
1223#ifdef PAGESIZE
1224# define eio_pagesize() PAGESIZE
1225#else
1226static intptr_t
1227eio_pagesize (void)
1228{
1229 static intptr_t page;
1230
1231 if (!page)
1232 page = sysconf (_SC_PAGESIZE);
1233
1234 return page;
1235}
1236#endif
1237
1238static void
1239eio_page_align (void **addr, size_t *length)
1240{
1241 intptr_t mask = eio_pagesize () - 1;
1242
1243 /* round down addr */
1244 intptr_t adj = mask & (intptr_t)*addr;
1245
1246 *addr = (void *)((intptr_t)*addr - adj);
1247 *length += adj;
1248
1249 /* round up length */
1250 *length = (*length + mask) & ~mask;
1251}
1252
1253#if !_POSIX_MEMLOCK
1254# define eio__mlockall(a) EIO_ENOSYS ()
1255#else
1256
1257static int
1258eio__mlockall (int flags)
1259{
1260 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1261 extern int mallopt (int, int);
1262 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1263 #endif
1264
1265 if (EIO_MCL_CURRENT != MCL_CURRENT
1266 || EIO_MCL_FUTURE != MCL_FUTURE)
1267 {
1268 flags = 0
1269 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1270 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1271 }
1272
1273 return mlockall (flags);
1274}
1275#endif
1276
1277#if !_POSIX_MEMLOCK_RANGE
1278# define eio__mlock(a,b) EIO_ENOSYS ()
1279#else
1280
1281static int
1282eio__mlock (void *addr, size_t length)
1283{
1284 eio_page_align (&addr, &length);
1285
1286 return mlock (addr, length);
1287}
1288
1289#endif
1290
1291#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1292# define eio__msync(a,b,c) EIO_ENOSYS ()
1293#else
1294
1295static int
1296eio__msync (void *mem, size_t len, int flags)
1297{
1298 eio_page_align (&mem, &len);
1299
1300 if (EIO_MS_ASYNC != MS_SYNC
1301 || EIO_MS_INVALIDATE != MS_INVALIDATE
1302 || EIO_MS_SYNC != MS_SYNC)
1303 {
1304 flags = 0
1305 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1306 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1307 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1308 }
1309
1310 return msync (mem, len, flags);
1311}
1312
1313#endif
1314
1315static int
1316eio__mtouch (eio_req *req)
1317{
1318 void *mem = req->ptr2;
1319 size_t len = req->size;
1320 int flags = req->int1;
1321
1322 eio_page_align (&mem, &len);
1323
1324 {
1325 intptr_t addr = (intptr_t)mem;
1326 intptr_t end = addr + len;
1327 intptr_t page = eio_pagesize ();
1328
1329 if (addr < end)
1330 if (flags & EIO_MT_MODIFY) /* modify */
1331 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1332 else
1333 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1334 }
1335
1336 return 0;
1337}
1338
1339/*****************************************************************************/
1340/* requests implemented outside eio_execute, because they are so large */
1341
1342static void
1343eio__realpath (eio_req *req, etp_worker *self)
1344{
1345 char *rel = req->ptr1;
1346 char *res;
1347 char *tmp1, *tmp2;
1348#if SYMLOOP_MAX > 32
1349 int symlinks = SYMLOOP_MAX;
1350#else
1351 int symlinks = 32;
1352#endif
1353
1354 req->result = -1;
1355
1356 errno = EINVAL;
1357 if (!rel)
1358 return;
1359
1360 errno = ENOENT;
1361 if (!*rel)
1362 return;
1363
1364 if (!req->ptr2)
1365 {
1366 X_LOCK (wrklock);
1367 req->flags |= EIO_FLAG_PTR2_FREE;
1368 X_UNLOCK (wrklock);
1369 req->ptr2 = malloc (PATH_MAX * 3);
1370
1371 errno = ENOMEM;
1372 if (!req->ptr2)
1373 return;
1374 }
1375
1376 res = req->ptr2;
1377 tmp1 = res + PATH_MAX;
1378 tmp2 = tmp1 + PATH_MAX;
1379
1380#if 0 /* disabled, the musl way to do things is just too racy */
1381#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1382 /* on linux we may be able to ask the kernel */
1383 {
1384 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1385
1386 if (fd >= 0)
1387 {
1388 sprintf (tmp1, "/proc/self/fd/%d", fd);
1389 req->result = readlink (tmp1, res, PATH_MAX);
1390 close (fd);
1391
1392 /* here we should probably stat the open file and the disk file, to make sure they still match */
1393
1394 if (req->result > 0)
1395 goto done;
1396 }
1397 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1398 return;
1399 }
1400#endif
1401#endif
1402
1403 if (*rel != '/')
1404 {
1405 if (!getcwd (res, PATH_MAX))
1406 return;
1407
1408 if (res [1]) /* only use if not / */
1409 res += strlen (res);
1410 }
1411
1412 while (*rel)
1413 {
1414 eio_ssize_t len, linklen;
1415 char *beg = rel;
1416
1417 while (*rel && *rel != '/')
1418 ++rel;
1419
1420 len = rel - beg;
1421
1422 if (!len) /* skip slashes */
1423 {
1424 ++rel;
1425 continue;
1426 }
1427
1428 if (beg [0] == '.')
1429 {
1430 if (len == 1)
1431 continue; /* . - nop */
1432
1433 if (beg [1] == '.' && len == 2)
1434 {
1435 /* .. - back up one component, if possible */
1436
1437 while (res != req->ptr2)
1438 if (*--res == '/')
1439 break;
1440
1441 continue;
1442 }
1443 }
1444
1445 errno = ENAMETOOLONG;
1446 if (res + 1 + len + 1 >= tmp1)
1447 return;
1448
1449 /* copy one component */
1450 *res = '/';
1451 memcpy (res + 1, beg, len);
1452
1453 /* zero-terminate, for readlink */
1454 res [len + 1] = 0;
1455
1456 /* now check if it's a symlink */
1457 linklen = readlink (req->ptr2, tmp1, PATH_MAX);
1458
1459 if (linklen < 0)
1460 {
1461 if (errno != EINVAL)
1462 return;
1463
1464 /* it's a normal directory. hopefully */
1465 res += len + 1;
1466 }
1467 else
1468 {
1469 /* yay, it was a symlink - build new path in tmp2 */
1470 int rellen = strlen (rel);
1471
1472 errno = ENAMETOOLONG;
1473 if (linklen + 1 + rellen >= PATH_MAX)
1474 return;
1475
1476 errno = ELOOP;
1477 if (!--symlinks)
1478 return;
1479
1480 if (*tmp1 == '/')
1481 res = req->ptr2; /* symlink resolves to an absolute path */
1482
1483 /* we need to be careful, as rel might point into tmp2 already */
1484 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1485 tmp2 [linklen] = '/';
1486 memcpy (tmp2, tmp1, linklen);
1487
1488 rel = tmp2;
1489 }
1490 }
1491
1492 /* special case for the lone root path */
1493 if (res == req->ptr2)
1494 *res++ = '/';
1495
1496 req->result = res - (char *)req->ptr2;
1497
1498done:
1499 req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
1028} 1500}
1029 1501
1030static signed char 1502static signed char
1031eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) 1503eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1032{ 1504{
1033 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */ 1505 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1034 : a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0; 1506 : a->inode < b->inode ? -1
1507 : a->inode > b->inode ? 1
1508 : 0;
1035} 1509}
1036 1510
1037#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0 1511#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1038 1512
1039#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */ 1513#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1040#define EIO_SORT_FAST 60 /* when to only use insertion sort */ 1514#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1041 1515
1042static void 1516static void
1043eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1517eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1044{ 1518{
1045 unsigned char bits [9 + sizeof (ino_t) * 8]; 1519 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1046 unsigned char *bit = bits; 1520 unsigned char *bit = bits;
1047 1521
1048 assert (CHAR_BIT == 8); 1522 assert (CHAR_BIT == 8);
1049 assert (sizeof (eio_dirent) * 8 < 256); 1523 assert (sizeof (eio_dirent) * 8 < 256);
1050 assert (offsetof (eio_dirent, inode)); /* we use 0 as sentinel */ 1524 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1051 assert (offsetof (eio_dirent, score)); /* we use 0 as sentinel */ 1525 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1052 1526
1053 if (size <= EIO_SORT_FAST) 1527 if (size <= EIO_SORT_FAST)
1054 return; 1528 return;
1055 1529
1056 /* first prepare an array of bits to test in our radix sort */ 1530 /* first prepare an array of bits to test in our radix sort */
1057 /* try to take endianness into account, as well as differences in ino_t sizes */ 1531 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1058 /* inode_bits must contain all inodes ORed together */ 1532 /* inode_bits must contain all inodes ORed together */
1059 /* which is used to skip bits that are 0 everywhere, which is very common */ 1533 /* which is used to skip bits that are 0 everywhere, which is very common */
1060 { 1534 {
1061 ino_t endianness; 1535 eio_ino_t endianness;
1062 int i, j; 1536 int i, j;
1063 1537
1064 /* we store the byte offset of byte n into byte n of "endianness" */ 1538 /* we store the byte offset of byte n into byte n of "endianness" */
1065 for (i = 0; i < sizeof (ino_t); ++i) 1539 for (i = 0; i < sizeof (eio_ino_t); ++i)
1066 ((unsigned char *)&endianness)[i] = i; 1540 ((unsigned char *)&endianness)[i] = i;
1067 1541
1068 *bit++ = 0; 1542 *bit++ = 0;
1069 1543
1070 for (i = 0; i < sizeof (ino_t); ++i) 1544 for (i = 0; i < sizeof (eio_ino_t); ++i)
1071 { 1545 {
1072 /* shifting off the byte offsets out of "endianness" */ 1546 /* shifting off the byte offsets out of "endianness" */
1073 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8; 1547 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1074 endianness >>= 8; 1548 endianness >>= 8;
1075 1549
1076 for (j = 0; j < 8; ++j) 1550 for (j = 0; j < 8; ++j)
1077 if (inode_bits & (((ino_t)1) << (i * 8 + j))) 1551 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1078 *bit++ = offs + j; 1552 *bit++ = offs + j;
1079 } 1553 }
1080 1554
1081 for (j = 0; j < 8; ++j) 1555 for (j = 0; j < 8; ++j)
1082 if (score_bits & (1 << j)) 1556 if (score_bits & (1 << j))
1083 *bit++ = offsetof (eio_dirent, score) * 8 + j; 1557 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1084 } 1558 }
1085 1559
1086 /* now actually do the sorting (a variant of MSD radix sort) */ 1560 /* now actually do the sorting (a variant of MSD radix sort) */
1087 { 1561 {
1088 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base; 1562 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1089 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end; 1563 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1090 unsigned char *bit_stk [9 + sizeof (ino_t) * 8]; 1564 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1091 int stk_idx = 0; 1565 int stk_idx = 0;
1092 1566
1093 base_stk [stk_idx] = dents; 1567 base_stk [stk_idx] = dents;
1094 end_stk [stk_idx] = dents + size; 1568 end_stk [stk_idx] = dents + size;
1095 bit_stk [stk_idx] = bit - 1; 1569 bit_stk [stk_idx] = bit - 1;
… …
1174 } 1648 }
1175 } 1649 }
1176} 1650}
1177 1651
1178static void 1652static void
1179eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1653eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1180{ 1654{
1181 if (size <= 1) 1655 if (size <= 1)
1182 return; /* our insertion sort relies on size > 0 */ 1656 return; /* our insertion sort relies on size > 0 */
1183 1657
1184 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */ 1658 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
… …
1192 1666
1193/* read a full directory */ 1667/* read a full directory */
1194static void 1668static void
1195eio__scandir (eio_req *req, etp_worker *self) 1669eio__scandir (eio_req *req, etp_worker *self)
1196{ 1670{
1197 DIR *dirp;
1198 EIO_STRUCT_DIRENT *entp;
1199 char *name, *names; 1671 char *name, *names;
1200 int namesalloc = 4096; 1672 int namesalloc = 4096 - sizeof (void *) * 4;
1201 int namesoffs = 0; 1673 int namesoffs = 0;
1202 int flags = req->int1; 1674 int flags = req->int1;
1203 eio_dirent *dents = 0; 1675 eio_dirent *dents = 0;
1204 int dentalloc = 128; 1676 int dentalloc = 128;
1205 int dentoffs = 0; 1677 int dentoffs = 0;
1206 ino_t inode_bits = 0; 1678 eio_ino_t inode_bits = 0;
1679#ifdef _WIN32
1680 HANDLE dirp;
1681 WIN32_FIND_DATA entp;
1682#else
1683 DIR *dirp;
1684 EIO_STRUCT_DIRENT *entp;
1685#endif
1207 1686
1208 req->result = -1; 1687 req->result = -1;
1209 1688
1210 if (!(flags & EIO_READDIR_DENTS)) 1689 if (!(flags & EIO_READDIR_DENTS))
1211 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); 1690 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1212 1691
1213 X_LOCK (wrklock); 1692#ifdef _WIN32
1214 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 1693 {
1694 int len = strlen ((const char *)req->ptr1);
1695 char *path = malloc (MAX_PATH);
1696 const char *fmt;
1697
1698 if (!len)
1699 fmt = "./*";
1700 else if (((const char *)req->ptr1)[len - 1] == '/' || ((const char *)req->ptr1)[len - 1] == '\\')
1701 fmt = "%s*";
1702 else
1703 fmt = "%s/*";
1704
1705 _snprintf (path, MAX_PATH, fmt, (const char *)req->ptr1);
1706 dirp = FindFirstFile (path, &entp);
1707 free (path);
1708
1709 if (dirp == INVALID_HANDLE_VALUE)
1710 {
1711 /* should steal _dosmaperr */
1712 switch (GetLastError ())
1713 {
1714 case ERROR_FILE_NOT_FOUND:
1715 req->result = 0;
1716 break;
1717
1718 case ERROR_INVALID_NAME:
1719 case ERROR_PATH_NOT_FOUND:
1720 case ERROR_NO_MORE_FILES:
1721 errno = ENOENT;
1722 break;
1723
1724 case ERROR_NOT_ENOUGH_MEMORY:
1725 errno = ENOMEM;
1726 break;
1727
1728 default:
1729 errno = EINVAL;
1730 break;
1731 }
1732
1733 return;
1734 }
1735 }
1736#else
1215 self->dirp = dirp = opendir (req->ptr1); 1737 dirp = opendir (req->ptr1);
1738
1739 if (!dirp)
1740 return;
1741#endif
1742
1743 if (req->flags & EIO_FLAG_PTR1_FREE)
1744 free (req->ptr1);
1745
1216 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1746 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1217 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; 1747 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1218 req->ptr2 = names = malloc (namesalloc); 1748 req->ptr2 = names = malloc (namesalloc);
1219 X_UNLOCK (wrklock);
1220 1749
1221 if (dirp && names && (!flags || dents)) 1750 if (!names || (flags && !dents))
1751 return;
1752
1222 for (;;) 1753 for (;;)
1223 { 1754 {
1755 int done;
1756
1757#ifdef _WIN32
1758 done = !dirp;
1759#else
1224 errno = 0; 1760 errno = 0;
1225 entp = readdir (dirp); 1761 entp = readdir (dirp);
1762 done = !entp;
1763#endif
1226 1764
1227 if (!entp) 1765 if (done)
1228 { 1766 {
1767#ifndef _WIN32
1768 int old_errno = errno;
1769 closedir (dirp);
1770 errno = old_errno;
1771
1229 if (errno) 1772 if (errno)
1230 break; 1773 break;
1774#endif
1231 1775
1232 /* sort etc. */ 1776 /* sort etc. */
1233 req->int1 = flags; 1777 req->int1 = flags;
1234 req->result = dentoffs; 1778 req->result = dentoffs;
1235 1779
1236 if (flags & EIO_READDIR_STAT_ORDER) 1780 if (flags & EIO_READDIR_STAT_ORDER)
1237 eio_dent_sort (dents, dentoffs, 0, inode_bits); /* sort by inode exclusively */ 1781 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1238 else if (flags & EIO_READDIR_DIRS_FIRST) 1782 else if (flags & EIO_READDIR_DIRS_FIRST)
1239 if (flags & EIO_READDIR_FOUND_UNKNOWN) 1783 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1240 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */ 1784 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1241 else 1785 else
1786 {
1787 /* in this case, all is known, and we just put dirs first and sort them */
1788 eio_dirent *oth = dents + dentoffs;
1789 eio_dirent *dir = dents;
1790
1791 /* now partition dirs to the front, and non-dirs to the back */
1792 /* by walking from both sides and swapping if necessary */
1793 while (oth > dir)
1794 {
1795 if (dir->type == EIO_DT_DIR)
1796 ++dir;
1797 else if ((--oth)->type == EIO_DT_DIR)
1798 {
1799 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1800
1801 ++dir;
1802 }
1803 }
1804
1805 /* now sort the dirs only (dirs all have the same score) */
1806 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1807 }
1808
1809 break;
1810 }
1811
1812 /* now add the entry to our list(s) */
1813 name = D_NAME (entp);
1814
1815 /* skip . and .. entries */
1816 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1817 {
1818 int len = D_NAMLEN (entp) + 1;
1819
1820 while (ecb_expect_false (namesoffs + len > namesalloc))
1821 {
1822 namesalloc *= 2;
1823 req->ptr2 = names = realloc (names, namesalloc);
1824
1825 if (!names)
1826 break;
1827 }
1828
1829 memcpy (names + namesoffs, name, len);
1830
1831 if (dents)
1832 {
1833 struct eio_dirent *ent;
1834
1835 if (ecb_expect_false (dentoffs == dentalloc))
1242 { 1836 {
1243 /* in this case, all is known, and we just put dirs first and sort them */ 1837 dentalloc *= 2;
1838 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1839
1840 if (!dents)
1841 break;
1842 }
1843
1244 eio_dirent *oth = dents + dentoffs; 1844 ent = dents + dentoffs;
1245 eio_dirent *dir = dents;
1246 1845
1247 /* now partition dirs to the front, and non-dirs to the back */ 1846 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1248 /* by walking from both sides and swapping if necessary */ 1847 ent->namelen = len - 1;
1249 /* also clear score, so it doesn't influence sorting */ 1848 ent->inode = D_INO (entp);
1250 while (oth > dir) 1849
1850 inode_bits |= ent->inode;
1851
1852 switch (D_TYPE (entp))
1853 {
1854 default:
1855 ent->type = EIO_DT_UNKNOWN;
1856 flags |= EIO_READDIR_FOUND_UNKNOWN;
1857 break;
1858
1859 #ifdef DT_FIFO
1860 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1861 #endif
1862 #ifdef DT_CHR
1863 case DT_CHR: ent->type = EIO_DT_CHR; break;
1864 #endif
1865 #ifdef DT_MPC
1866 case DT_MPC: ent->type = EIO_DT_MPC; break;
1867 #endif
1868 #ifdef DT_DIR
1869 case DT_DIR: ent->type = EIO_DT_DIR; break;
1870 #endif
1871 #ifdef DT_NAM
1872 case DT_NAM: ent->type = EIO_DT_NAM; break;
1873 #endif
1874 #ifdef DT_BLK
1875 case DT_BLK: ent->type = EIO_DT_BLK; break;
1876 #endif
1877 #ifdef DT_MPB
1878 case DT_MPB: ent->type = EIO_DT_MPB; break;
1879 #endif
1880 #ifdef DT_REG
1881 case DT_REG: ent->type = EIO_DT_REG; break;
1882 #endif
1883 #ifdef DT_NWK
1884 case DT_NWK: ent->type = EIO_DT_NWK; break;
1885 #endif
1886 #ifdef DT_CMP
1887 case DT_CMP: ent->type = EIO_DT_CMP; break;
1888 #endif
1889 #ifdef DT_LNK
1890 case DT_LNK: ent->type = EIO_DT_LNK; break;
1891 #endif
1892 #ifdef DT_SOCK
1893 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1894 #endif
1895 #ifdef DT_DOOR
1896 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1897 #endif
1898 #ifdef DT_WHT
1899 case DT_WHT: ent->type = EIO_DT_WHT; break;
1900 #endif
1901 }
1902
1903 ent->score = 7;
1904
1905 if (flags & EIO_READDIR_DIRS_FIRST)
1906 {
1907 if (ent->type == EIO_DT_UNKNOWN)
1251 { 1908 {
1252 if (dir->type == EIO_DT_DIR) 1909 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1253 ++dir; 1910 ent->score = 1;
1254 else if ((--oth)->type == EIO_DT_DIR) 1911 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1255 { 1912 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1256 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1257
1258 ++dir;
1259 }
1260 } 1913 }
1261 1914 else if (ent->type == EIO_DT_DIR)
1262 /* now sort the dirs only */ 1915 ent->score = 0;
1263 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1264 } 1916 }
1265
1266 break;
1267 }
1268
1269 /* now add the entry to our list(s) */
1270 name = entp->d_name;
1271
1272 /* skip . and .. entries */
1273 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1274 {
1275 int len = D_NAMLEN (entp) + 1;
1276
1277 while (expect_false (namesoffs + len > namesalloc))
1278 {
1279 namesalloc *= 2;
1280 X_LOCK (wrklock);
1281 req->ptr2 = names = realloc (names, namesalloc);
1282 X_UNLOCK (wrklock);
1283
1284 if (!names)
1285 break;
1286 } 1917 }
1287 1918
1288 memcpy (names + namesoffs, name, len);
1289
1290 if (dents)
1291 {
1292 struct eio_dirent *ent;
1293
1294 if (expect_false (dentoffs == dentalloc))
1295 {
1296 dentalloc *= 2;
1297 X_LOCK (wrklock);
1298 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1299 X_UNLOCK (wrklock);
1300
1301 if (!dents)
1302 break;
1303 }
1304
1305 ent = dents + dentoffs;
1306
1307 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1308 ent->namelen = len - 1;
1309 ent->inode = D_INO (entp);
1310
1311 inode_bits |= ent->inode;
1312
1313 switch (D_TYPE (entp))
1314 {
1315 default:
1316 ent->type = EIO_DT_UNKNOWN;
1317 flags |= EIO_READDIR_FOUND_UNKNOWN;
1318 break;
1319
1320 #ifdef DT_FIFO
1321 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1322 #endif
1323 #ifdef DT_CHR
1324 case DT_CHR: ent->type = EIO_DT_CHR; break;
1325 #endif
1326 #ifdef DT_MPC
1327 case DT_MPC: ent->type = EIO_DT_MPC; break;
1328 #endif
1329 #ifdef DT_DIR
1330 case DT_DIR: ent->type = EIO_DT_DIR; break;
1331 #endif
1332 #ifdef DT_NAM
1333 case DT_NAM: ent->type = EIO_DT_NAM; break;
1334 #endif
1335 #ifdef DT_BLK
1336 case DT_BLK: ent->type = EIO_DT_BLK; break;
1337 #endif
1338 #ifdef DT_MPB
1339 case DT_MPB: ent->type = EIO_DT_MPB; break;
1340 #endif
1341 #ifdef DT_REG
1342 case DT_REG: ent->type = EIO_DT_REG; break;
1343 #endif
1344 #ifdef DT_NWK
1345 case DT_NWK: ent->type = EIO_DT_NWK; break;
1346 #endif
1347 #ifdef DT_CMP
1348 case DT_CMP: ent->type = EIO_DT_CMP; break;
1349 #endif
1350 #ifdef DT_LNK
1351 case DT_LNK: ent->type = EIO_DT_LNK; break;
1352 #endif
1353 #ifdef DT_SOCK
1354 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1355 #endif
1356 #ifdef DT_DOOR
1357 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1358 #endif
1359 #ifdef DT_WHT
1360 case DT_WHT: ent->type = EIO_DT_WHT; break;
1361 #endif
1362 }
1363
1364 ent->score = 7;
1365
1366 if (flags & EIO_READDIR_DIRS_FIRST)
1367 {
1368 if (ent->type == EIO_DT_UNKNOWN)
1369 {
1370 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1371 ent->score = 1;
1372 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1373 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1374 }
1375 else if (ent->type == EIO_DT_DIR)
1376 ent->score = 0;
1377 }
1378 }
1379
1380 namesoffs += len; 1919 namesoffs += len;
1381 ++dentoffs; 1920 ++dentoffs;
1382 } 1921 }
1383 1922
1384 if (EIO_CANCELLED (req)) 1923 if (EIO_CANCELLED (req))
1385 { 1924 {
1386 errno = ECANCELED; 1925 errno = ECANCELED;
1387 break; 1926 break;
1388 } 1927 }
1928
1929#ifdef _WIN32
1930 if (!FindNextFile (dirp, &entp))
1931 {
1932 FindClose (dirp);
1933 dirp = 0;
1389 } 1934 }
1390} 1935#endif
1391
1392#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1393# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
1394#else
1395
1396int
1397eio__msync (void *mem, size_t len, int flags)
1398{
1399 if (EIO_MS_ASYNC != MS_SYNC
1400 || EIO_MS_INVALIDATE != MS_INVALIDATE
1401 || EIO_MS_SYNC != MS_SYNC)
1402 {
1403 flags = 0
1404 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1405 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1406 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1407 } 1936 }
1408
1409 return msync (mem, len, flags);
1410}
1411
1412#endif
1413
1414int
1415eio__mtouch (void *mem, size_t len, int flags)
1416{
1417 intptr_t addr = (intptr_t)mem;
1418 intptr_t end = addr + len;
1419#ifdef PAGESIZE
1420 const intptr_t page = PAGESIZE;
1421#else
1422 static intptr_t page;
1423
1424 if (!page)
1425 page = sysconf (_SC_PAGESIZE);
1426#endif
1427
1428 /* round down to start of page, although this is probably useless */
1429 addr &= ~(page - 1); /* assume page size is always a power of two */
1430
1431 if (addr < end)
1432 if (flags & EIO_MT_MODIFY) /* modify */
1433 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
1434 else
1435 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
1436
1437 return 0;
1438} 1937}
1439 1938
1440/*****************************************************************************/ 1939/*****************************************************************************/
1441 1940
1442#define ALLOC(len) \ 1941#define ALLOC(len) \
… …
1458{ 1957{
1459 ETP_REQ *req; 1958 ETP_REQ *req;
1460 struct timespec ts; 1959 struct timespec ts;
1461 etp_worker *self = (etp_worker *)thr_arg; 1960 etp_worker *self = (etp_worker *)thr_arg;
1462 1961
1962#if HAVE_PRCTL_SET_NAME
1963 prctl (PR_SET_NAME, (unsigned long)"eio_thread", 0, 0, 0);
1964#endif
1965
1463 /* try to distribute timeouts somewhat randomly */ 1966 /* try to distribute timeouts somewhat evenly */
1464 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 1967 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1465 1968
1466 for (;;) 1969 for (;;)
1467 { 1970 {
1971 ts.tv_sec = 0;
1972
1468 X_LOCK (reqlock); 1973 X_LOCK (reqlock);
1469 1974
1470 for (;;) 1975 for (;;)
1471 { 1976 {
1472 self->req = req = reqq_shift (&req_queue); 1977 self->req = req = reqq_shift (&req_queue);
1473 1978
1474 if (req) 1979 if (req)
1475 break; 1980 break;
1476 1981
1982 if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
1983 {
1984 X_UNLOCK (reqlock);
1985 X_LOCK (wrklock);
1986 --started;
1987 X_UNLOCK (wrklock);
1988 goto quit;
1989 }
1990
1477 ++idle; 1991 ++idle;
1478 1992
1479 ts.tv_sec = time (0) + IDLE_TIMEOUT; 1993 if (idle <= max_idle)
1480 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 1994 /* we are allowed to idle, so do so without any timeout */
1995 X_COND_WAIT (reqwait, reqlock);
1996 else
1481 { 1997 {
1482 if (idle > max_idle) 1998 /* initialise timeout once */
1483 { 1999 if (!ts.tv_sec)
1484 --idle; 2000 ts.tv_sec = time (0) + idle_timeout;
1485 X_UNLOCK (reqlock);
1486 X_LOCK (wrklock);
1487 --started;
1488 X_UNLOCK (wrklock);
1489 goto quit;
1490 }
1491 2001
1492 /* we are allowed to idle, so do so without any timeout */
1493 X_COND_WAIT (reqwait, reqlock); 2002 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
2003 ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1494 } 2004 }
1495 2005
1496 --idle; 2006 --idle;
1497 } 2007 }
1498 2008
… …
1501 X_UNLOCK (reqlock); 2011 X_UNLOCK (reqlock);
1502 2012
1503 if (req->type < 0) 2013 if (req->type < 0)
1504 goto quit; 2014 goto quit;
1505 2015
1506 if (!EIO_CANCELLED (req))
1507 ETP_EXECUTE (self, req); 2016 ETP_EXECUTE (self, req);
1508 2017
1509 X_LOCK (reslock); 2018 X_LOCK (reslock);
1510 2019
1511 ++npending; 2020 ++npending;
1512 2021
… …
1527 return 0; 2036 return 0;
1528} 2037}
1529 2038
1530/*****************************************************************************/ 2039/*****************************************************************************/
1531 2040
2041int ecb_cold
1532int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 2042eio_init (void (*want_poll)(void), void (*done_poll)(void))
1533{ 2043{
2044#if !HAVE_PREADWRITE
2045 X_MUTEX_CREATE (preadwritelock);
2046#endif
2047
1534 return etp_init (want_poll, done_poll); 2048 return etp_init (want_poll, done_poll);
1535} 2049}
1536 2050
2051ecb_inline void
1537static void eio_api_destroy (eio_req *req) 2052eio_api_destroy (eio_req *req)
1538{ 2053{
1539 free (req); 2054 free (req);
1540} 2055}
1541 2056
1542#define REQ(rtype) \ 2057#define REQ(rtype) \
… …
1561 { \ 2076 { \
1562 eio_api_destroy (req); \ 2077 eio_api_destroy (req); \
1563 return 0; \ 2078 return 0; \
1564 } 2079 }
1565 2080
2081static void
1566static void eio_execute (etp_worker *self, eio_req *req) 2082eio_execute (etp_worker *self, eio_req *req)
1567{ 2083{
1568 errno = 0; 2084 if (ecb_expect_false (EIO_CANCELLED (req)))
2085 {
2086 req->result = -1;
2087 req->errorno = ECANCELED;
2088 return;
2089 }
1569 2090
1570 switch (req->type) 2091 switch (req->type)
1571 { 2092 {
1572 case EIO_READ: ALLOC (req->size); 2093 case EIO_READ: ALLOC (req->size);
1573 req->result = req->offs >= 0 2094 req->result = req->offs >= 0
… …
1576 case EIO_WRITE: req->result = req->offs >= 0 2097 case EIO_WRITE: req->result = req->offs >= 0
1577 ? pwrite (req->int1, req->ptr2, req->size, req->offs) 2098 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1578 : write (req->int1, req->ptr2, req->size); break; 2099 : write (req->int1, req->ptr2, req->size); break;
1579 2100
1580 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break; 2101 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1581 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break; 2102 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
1582 2103
1583 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2104 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1584 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2105 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1585 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2106 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1586 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2107 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
… …
1606 case EIO_RMDIR: req->result = rmdir (req->ptr1); break; 2127 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1607 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; 2128 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1608 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; 2129 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1609 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; 2130 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1610 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; 2131 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1611 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break; 2132 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1612 2133
2134 case EIO_REALPATH: eio__realpath (req, self); break;
2135
1613 case EIO_READLINK: ALLOC (NAME_MAX); 2136 case EIO_READLINK: ALLOC (PATH_MAX);
1614 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; 2137 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1615 2138
1616 case EIO_SYNC: req->result = 0; sync (); break; 2139 case EIO_SYNC: req->result = 0; sync (); break;
1617 case EIO_FSYNC: req->result = fsync (req->int1); break; 2140 case EIO_FSYNC: req->result = fsync (req->int1); break;
1618 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 2141 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
2142 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
2143 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1619 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break; 2144 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
2145 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1620 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break; 2146 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1621 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break; 2147 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
2148 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
1622 2149
1623 case EIO_READDIR: eio__scandir (req, self); break; 2150 case EIO_READDIR: eio__scandir (req, self); break;
1624 2151
1625 case EIO_BUSY: 2152 case EIO_BUSY:
1626#ifdef _WIN32 2153#ifdef _WIN32
1627 Sleep (req->nv1 * 1000.); 2154 Sleep (req->nv1 * 1e3);
1628#else 2155#else
1629 { 2156 {
1630 struct timeval tv; 2157 struct timeval tv;
1631 2158
1632 tv.tv_sec = req->nv1; 2159 tv.tv_sec = req->nv1;
1633 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; 2160 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1634 2161
1635 req->result = select (0, 0, 0, 0, &tv); 2162 req->result = select (0, 0, 0, 0, &tv);
1636 } 2163 }
1637#endif 2164#endif
1638 break; 2165 break;
… …
1653 times = tv; 2180 times = tv;
1654 } 2181 }
1655 else 2182 else
1656 times = 0; 2183 times = 0;
1657 2184
1658
1659 req->result = req->type == EIO_FUTIME 2185 req->result = req->type == EIO_FUTIME
1660 ? futimes (req->int1, times) 2186 ? futimes (req->int1, times)
1661 : utimes (req->ptr1, times); 2187 : utimes (req->ptr1, times);
1662 } 2188 }
1663 break; 2189 break;
… …
1668 case EIO_NOP: 2194 case EIO_NOP:
1669 req->result = 0; 2195 req->result = 0;
1670 break; 2196 break;
1671 2197
1672 case EIO_CUSTOM: 2198 case EIO_CUSTOM:
1673 ((void (*)(eio_req *))req->feed) (req); 2199 req->feed (req);
1674 break; 2200 break;
1675 2201
1676 default: 2202 default:
2203 errno = ENOSYS;
1677 req->result = -1; 2204 req->result = -1;
1678 break; 2205 break;
1679 } 2206 }
1680 2207
1681 req->errorno = errno; 2208 req->errorno = errno;
… …
1706eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2233eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1707{ 2234{
1708 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2235 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1709} 2236}
1710 2237
2238eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2239{
2240 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2241}
2242
2243eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
2244{
2245 REQ (EIO_SYNCFS); req->int1 = fd; SEND;
2246}
2247
2248eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2249{
2250 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2251}
2252
1711eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2253eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1712{ 2254{
1713 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2255 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1714} 2256}
1715 2257
1716eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) 2258eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
1717{ 2259{
1718 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; 2260 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
1719} 2261}
1720 2262
1721eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) 2263eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
1722{ 2264{
1723 REQ (EIO_FDATASYNC); req->int1 = fd; SEND; 2265 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2266}
2267
2268eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2269{
2270 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
1724} 2271}
1725 2272
1726eio_req *eio_close (int fd, int pri, eio_cb cb, void *data) 2273eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
1727{ 2274{
1728 REQ (EIO_CLOSE); req->int1 = fd; SEND; 2275 REQ (EIO_CLOSE); req->int1 = fd; SEND;
… …
1766eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) 2313eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1767{ 2314{
1768 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2315 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1769} 2316}
1770 2317
1771eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2318eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1772{ 2319{
1773 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2320 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1774} 2321}
1775 2322
1776eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) 2323eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
… …
1796eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data) 2343eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1797{ 2344{
1798 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2345 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1799} 2346}
1800 2347
1801eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2348eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1802{ 2349{
1803 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2350 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1804} 2351}
1805 2352
1806eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) 2353eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
… …
1822eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data) 2369eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1823{ 2370{
1824 return eio__1path (EIO_READLINK, path, pri, cb, data); 2371 return eio__1path (EIO_READLINK, path, pri, cb, data);
1825} 2372}
1826 2373
2374eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2375{
2376 return eio__1path (EIO_REALPATH, path, pri, cb, data);
2377}
2378
1827eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) 2379eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1828{ 2380{
1829 return eio__1path (EIO_STAT, path, pri, cb, data); 2381 return eio__1path (EIO_STAT, path, pri, cb, data);
1830} 2382}
1831 2383
… …
1854 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND; 2406 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1855} 2407}
1856 2408
1857eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) 2409eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1858{ 2410{
1859 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; 2411 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1860} 2412}
1861 2413
1862static eio_req * 2414static eio_req *
1863eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2415eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1864{ 2416{
… …
1888eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2440eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1889{ 2441{
1890 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2442 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1891} 2443}
1892 2444
1893eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) 2445eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
1894{ 2446{
1895 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND; 2447 REQ (EIO_CUSTOM); req->feed = execute; SEND;
1896} 2448}
1897 2449
1898#endif 2450#endif
1899 2451
1900eio_req *eio_grp (eio_cb cb, void *data) 2452eio_req *eio_grp (eio_cb cb, void *data)
… …
1909#undef SEND 2461#undef SEND
1910 2462
1911/*****************************************************************************/ 2463/*****************************************************************************/
1912/* grp functions */ 2464/* grp functions */
1913 2465
2466void
1914void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2467eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1915{ 2468{
1916 grp->int2 = limit; 2469 grp->int2 = limit;
1917 grp->feed = feed; 2470 grp->feed = feed;
1918 2471
1919 grp_try_feed (grp); 2472 grp_try_feed (grp);
1920} 2473}
1921 2474
2475void
1922void eio_grp_limit (eio_req *grp, int limit) 2476eio_grp_limit (eio_req *grp, int limit)
1923{ 2477{
1924 grp->int2 = limit; 2478 grp->int2 = limit;
1925 2479
1926 grp_try_feed (grp); 2480 grp_try_feed (grp);
1927} 2481}
1928 2482
2483void
1929void eio_grp_add (eio_req *grp, eio_req *req) 2484eio_grp_add (eio_req *grp, eio_req *req)
1930{ 2485{
1931 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2486 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1932 2487
1933 grp->flags |= EIO_FLAG_GROUPADD; 2488 grp->flags |= EIO_FLAG_GROUPADD;
1934 2489
… …
1945} 2500}
1946 2501
1947/*****************************************************************************/ 2502/*****************************************************************************/
1948/* misc garbage */ 2503/* misc garbage */
1949 2504
2505eio_ssize_t
1950ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2506eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1951{ 2507{
1952 etp_worker wrk;
1953 ssize_t ret;
1954
1955 wrk.dbuf = 0;
1956
1957 ret = eio__sendfile (ofd, ifd, offset, count, &wrk); 2508 return eio__sendfile (ofd, ifd, offset, count);
1958
1959 if (wrk.dbuf)
1960 free (wrk.dbuf);
1961
1962 return ret;
1963} 2509}
1964 2510

Diff Legend

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