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

Comparing libeio/eio.c (file contents):
Revision 1.27 by root, Wed Oct 22 18:15:36 2008 UTC vs.
Revision 1.104 by root, Mon Sep 26 16:54:25 2011 UTC

1/* 1/*
2 * libeio implementation 2 * libeio implementation
3 * 3 *
4 * Copyright (c) 2007,2008 Marc Alexander Lehmann <libeio@schmorp.de> 4 * Copyright (c) 2007,2008,2009,2010,2011 Marc Alexander Lehmann <libeio@schmorp.de>
5 * All rights reserved. 5 * All rights reserved.
6 * 6 *
7 * Redistribution and use in source and binary forms, with or without modifica- 7 * Redistribution and use in source and binary forms, with or without modifica-
8 * tion, are permitted provided that the following conditions are met: 8 * tion, are permitted provided that the following conditions are met:
9 * 9 *
… …
35 * and other provisions required by the GPL. If you do not delete the 35 * and other provisions required by the GPL. If you do not delete the
36 * provisions above, a recipient may use your version of this file under 36 * provisions above, a recipient may use your version of this file under
37 * either the BSD or the GPL. 37 * either the BSD or the GPL.
38 */ 38 */
39 39
40#ifndef _WIN32
41# include "config.h"
42#endif
43
40#include "eio.h" 44#include "eio.h"
45#include "ecb.h"
46
47#ifdef EIO_STACKSIZE
48# define X_STACKSIZE EIO_STACKSIZE
49#endif
41#include "xthread.h" 50#include "xthread.h"
42 51
43#include <errno.h> 52#include <errno.h>
44#include <stddef.h> 53#include <stddef.h>
45#include <stdlib.h> 54#include <stdlib.h>
… …
49#include <sys/stat.h> 58#include <sys/stat.h>
50#include <limits.h> 59#include <limits.h>
51#include <fcntl.h> 60#include <fcntl.h>
52#include <assert.h> 61#include <assert.h>
53 62
63/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
64/* intptr_t only comes from stdint.h, says idiot openbsd coder */
65#if HAVE_STDINT_H
66# include <stdint.h>
67#endif
68
69#ifndef ECANCELED
70# define ECANCELED EDOM
71#endif
72#ifndef ELOOP
73# define ELOOP EDOM
74#endif
75
76#if !defined(ENOTSOCK) && defined(WSAENOTSOCK)
77# define ENOTSOCK WSAENOTSOCK
78#endif
79
80static void eio_destroy (eio_req *req);
81
54#ifndef EIO_FINISH 82#ifndef EIO_FINISH
55# 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
56#endif 84#endif
57 85
58#ifndef EIO_DESTROY 86#ifndef EIO_DESTROY
… …
61 89
62#ifndef EIO_FEED 90#ifndef EIO_FEED
63# 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)
64#endif 92#endif
65 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
66#ifdef _WIN32 105#ifdef _WIN32
67 106
68 /*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)
69 196
70#else 197#else
71 198
72# include "config.h"
73# include <sys/time.h> 199 #include <sys/time.h>
74# include <sys/select.h> 200 #include <sys/select.h>
201 #include <sys/statvfs.h>
75# 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
76# include <utime.h> 237# include <utime.h>
77# include <signal.h>
78# include <dirent.h>
79
80# ifndef EIO_STRUCT_DIRENT
81# define EIO_STRUCT_DIRENT struct dirent
82# endif 238#endif
83 239
240#if HAVE_SYS_SYSCALL_H
241# include <sys/syscall.h>
242#endif
243
244#if HAVE_SYS_PRCTL_H
245# include <sys/prctl.h>
84#endif 246#endif
85 247
86#if HAVE_SENDFILE 248#if HAVE_SENDFILE
87# if __linux 249# if __linux
88# include <sys/sendfile.h> 250# include <sys/sendfile.h>
89# elif __freebsd 251# elif __FreeBSD__ || defined __APPLE__
90# include <sys/socket.h> 252# include <sys/socket.h>
91# include <sys/uio.h> 253# include <sys/uio.h>
92# elif __hpux 254# elif __hpux
93# include <sys/socket.h> 255# include <sys/socket.h>
94# elif __solaris /* not yet */ 256# elif __solaris
95# include <sys/sendfile.h> 257# include <sys/sendfile.h>
96# else 258# else
97# error sendfile support requested but not available 259# error sendfile support requested but not available
98# endif 260# endif
99#endif 261#endif
100 262
101/* number of seconds after which an idle threads exit */ 263#ifndef D_TYPE
102#define IDLE_TIMEOUT 10 264# define D_TYPE(de) 0
265#endif
266#ifndef D_INO
267# define D_INO(de) 0
268#endif
269#ifndef D_NAMLEN
270# define D_NAMLEN(entp) strlen (D_NAME (entp))
271#endif
103 272
104/* used for struct dirent, AIX doesn't provide it */ 273/* used for struct dirent, AIX doesn't provide it */
105#ifndef NAME_MAX 274#ifndef NAME_MAX
106# define NAME_MAX 4096 275# define NAME_MAX 4096
107#endif 276#endif
108 277
278/* used for readlink etc. */
279#ifndef PATH_MAX
280# define PATH_MAX 4096
281#endif
282
109/* buffer size for various temporary buffers */ 283/* buffer size for various temporary buffers */
110#define EIO_BUFSIZE 65536 284#define EIO_BUFSIZE 65536
111 285
112#define dBUF \ 286#define dBUF \
113 char *eio_buf; \
114 ETP_WORKER_LOCK (self); \
115 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 287 char *eio_buf = malloc (EIO_BUFSIZE); \
116 ETP_WORKER_UNLOCK (self); \
117 errno = ENOMEM; \ 288 errno = ENOMEM; \
118 if (!eio_buf) \ 289 if (!eio_buf) \
119 return -1; 290 return -1
291
292#define FUBd \
293 free (eio_buf)
120 294
121#define EIO_TICKS ((1000000 + 1023) >> 10) 295#define EIO_TICKS ((1000000 + 1023) >> 10)
122 296
123/*****************************************************************************/ 297/*****************************************************************************/
124 298
125#if __GNUC__ >= 3 299struct tmpbuf
126# define expect(expr,value) __builtin_expect ((expr),(value)) 300{
127#else 301 void *ptr;
128# define expect(expr,value) (expr) 302 int len;
129#endif 303};
130 304
131#define expect_false(expr) expect ((expr) != 0, 0) 305static void *
132#define expect_true(expr) expect ((expr) != 0, 1) 306tmpbuf_get (struct tmpbuf *buf, int len)
307{
308 if (buf->len < len)
309 {
310 free (buf->ptr);
311 buf->ptr = malloc (buf->len = len);
312 }
313
314 return buf->ptr;
315}
133 316
134/*****************************************************************************/ 317/*****************************************************************************/
135 318
136#define ETP_PRI_MIN EIO_PRI_MIN 319#define ETP_PRI_MIN EIO_PRI_MIN
137#define ETP_PRI_MAX EIO_PRI_MAX 320#define ETP_PRI_MAX EIO_PRI_MAX
… …
143static int eio_finish (eio_req *req); 326static int eio_finish (eio_req *req);
144#define ETP_FINISH(req) eio_finish (req) 327#define ETP_FINISH(req) eio_finish (req)
145static void eio_execute (struct etp_worker *self, eio_req *req); 328static void eio_execute (struct etp_worker *self, eio_req *req);
146#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) 329#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
147 330
148#define ETP_WORKER_CLEAR(req) \
149 if (wrk->dbuf) \
150 { \
151 free (wrk->dbuf); \
152 wrk->dbuf = 0; \
153 } \
154 \
155 if (wrk->dirp) \
156 { \
157 closedir (wrk->dirp); \
158 wrk->dirp = 0; \
159 }
160#define ETP_WORKER_COMMON \
161 void *dbuf; \
162 DIR *dirp;
163
164/*****************************************************************************/ 331/*****************************************************************************/
165 332
166#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) 333#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
167 334
168/* calculcate time difference in ~1/EIO_TICKS of a second */ 335/* calculate time difference in ~1/EIO_TICKS of a second */
336ecb_inline int
169static int tvdiff (struct timeval *tv1, struct timeval *tv2) 337tvdiff (struct timeval *tv1, struct timeval *tv2)
170{ 338{
171 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 339 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
172 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 340 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
173} 341}
174 342
… …
178static void (*done_poll_cb) (void); 346static void (*done_poll_cb) (void);
179 347
180static unsigned int max_poll_time; /* reslock */ 348static unsigned int max_poll_time; /* reslock */
181static unsigned int max_poll_reqs; /* reslock */ 349static unsigned int max_poll_reqs; /* reslock */
182 350
183static volatile unsigned int nreqs; /* reqlock */ 351static unsigned int nreqs; /* reqlock */
184static volatile unsigned int nready; /* reqlock */ 352static unsigned int nready; /* reqlock */
185static volatile unsigned int npending; /* reqlock */ 353static unsigned int npending; /* reqlock */
186static volatile unsigned int max_idle = 4; 354static unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
355static unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
187 356
188static mutex_t wrklock = X_MUTEX_INIT; 357static xmutex_t wrklock;
189static mutex_t reslock = X_MUTEX_INIT; 358static xmutex_t reslock;
190static mutex_t reqlock = X_MUTEX_INIT; 359static xmutex_t reqlock;
191static cond_t reqwait = X_COND_INIT; 360static xcond_t reqwait;
192 361
193#if !HAVE_PREADWRITE 362#if !HAVE_PREADWRITE
194/* 363/*
195 * make our pread/pwrite emulation safe against themselves, but not against 364 * make our pread/pwrite emulation safe against themselves, but not against
196 * normal read/write by using a mutex. slows down execution a lot, 365 * normal read/write by using a mutex. slows down execution a lot,
197 * but that's your problem, not mine. 366 * but that's your problem, not mine.
198 */ 367 */
199static mutex_t preadwritelock = X_MUTEX_INIT; 368static xmutex_t preadwritelock;
200#endif 369#endif
201 370
202typedef struct etp_worker 371typedef struct etp_worker
203{ 372{
373 struct tmpbuf tmpbuf;
374
204 /* locked by wrklock */ 375 /* locked by wrklock */
205 struct etp_worker *prev, *next; 376 struct etp_worker *prev, *next;
206 377
207 thread_t tid; 378 xthread_t tid;
208 379
209 /* locked by reslock, reqlock or wrklock */ 380#ifdef ETP_WORKER_COMMON
210 ETP_REQ *req; /* currently processed request */
211
212 ETP_WORKER_COMMON 381 ETP_WORKER_COMMON
382#endif
213} etp_worker; 383} etp_worker;
214 384
215static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */ 385static etp_worker wrk_first; /* NOT etp */
216 386
217#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock) 387#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
218#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock) 388#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
219 389
220/* worker threads management */ 390/* worker threads management */
221 391
392static void ecb_cold
222static void etp_worker_clear (etp_worker *wrk) 393etp_worker_clear (etp_worker *wrk)
223{ 394{
224 ETP_WORKER_CLEAR (wrk);
225} 395}
226 396
397static void ecb_cold
227static void etp_worker_free (etp_worker *wrk) 398etp_worker_free (etp_worker *wrk)
228{ 399{
400 free (wrk->tmpbuf.ptr);
401
229 wrk->next->prev = wrk->prev; 402 wrk->next->prev = wrk->prev;
230 wrk->prev->next = wrk->next; 403 wrk->prev->next = wrk->next;
231 404
232 free (wrk); 405 free (wrk);
233} 406}
234 407
235static unsigned int etp_nreqs (void) 408static unsigned int
409etp_nreqs (void)
236{ 410{
237 int retval; 411 int retval;
238 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 412 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
239 retval = nreqs; 413 retval = nreqs;
240 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 414 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
241 return retval; 415 return retval;
242} 416}
243 417
244static unsigned int etp_nready (void) 418static unsigned int
419etp_nready (void)
245{ 420{
246 unsigned int retval; 421 unsigned int retval;
247 422
248 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 423 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
249 retval = nready; 424 retval = nready;
250 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 425 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
251 426
252 return retval; 427 return retval;
253} 428}
254 429
255static unsigned int etp_npending (void) 430static unsigned int
431etp_npending (void)
256{ 432{
257 unsigned int retval; 433 unsigned int retval;
258 434
259 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 435 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
260 retval = npending; 436 retval = npending;
261 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 437 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
262 438
263 return retval; 439 return retval;
264} 440}
265 441
266static unsigned int etp_nthreads (void) 442static unsigned int
443etp_nthreads (void)
267{ 444{
268 unsigned int retval; 445 unsigned int retval;
269 446
270 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 447 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
271 retval = started; 448 retval = started;
… …
285} etp_reqq; 462} etp_reqq;
286 463
287static etp_reqq req_queue; 464static etp_reqq req_queue;
288static etp_reqq res_queue; 465static etp_reqq res_queue;
289 466
467static void ecb_noinline ecb_cold
468reqq_init (etp_reqq *q)
469{
470 int pri;
471
472 for (pri = 0; pri < ETP_NUM_PRI; ++pri)
473 q->qs[pri] = q->qe[pri] = 0;
474
475 q->size = 0;
476}
477
478static int ecb_noinline
290static int reqq_push (etp_reqq *q, ETP_REQ *req) 479reqq_push (etp_reqq *q, ETP_REQ *req)
291{ 480{
292 int pri = req->pri; 481 int pri = req->pri;
293 req->next = 0; 482 req->next = 0;
294 483
295 if (q->qe[pri]) 484 if (q->qe[pri])
… …
301 q->qe[pri] = q->qs[pri] = req; 490 q->qe[pri] = q->qs[pri] = req;
302 491
303 return q->size++; 492 return q->size++;
304} 493}
305 494
495static ETP_REQ * ecb_noinline
306static ETP_REQ *reqq_shift (etp_reqq *q) 496reqq_shift (etp_reqq *q)
307{ 497{
308 int pri; 498 int pri;
309 499
310 if (!q->size) 500 if (!q->size)
311 return 0; 501 return 0;
… …
326 } 516 }
327 517
328 abort (); 518 abort ();
329} 519}
330 520
331static void etp_atfork_prepare (void) 521static int ecb_cold
522etp_init (void (*want_poll)(void), void (*done_poll)(void))
332{ 523{
333 X_LOCK (wrklock); 524 X_MUTEX_CREATE (wrklock);
334 X_LOCK (reqlock); 525 X_MUTEX_CREATE (reslock);
335 X_LOCK (reslock); 526 X_MUTEX_CREATE (reqlock);
336#if !HAVE_PREADWRITE 527 X_COND_CREATE (reqwait);
337 X_LOCK (preadwritelock);
338#endif
339}
340 528
341static void etp_atfork_parent (void) 529 reqq_init (&req_queue);
342{ 530 reqq_init (&res_queue);
343#if !HAVE_PREADWRITE
344 X_UNLOCK (preadwritelock);
345#endif
346 X_UNLOCK (reslock);
347 X_UNLOCK (reqlock);
348 X_UNLOCK (wrklock);
349}
350 531
351static void etp_atfork_child (void) 532 wrk_first.next =
352{ 533 wrk_first.prev = &wrk_first;
353 ETP_REQ *prv;
354
355 while ((prv = reqq_shift (&req_queue)))
356 ETP_DESTROY (prv);
357
358 while ((prv = reqq_shift (&res_queue)))
359 ETP_DESTROY (prv);
360
361 while (wrk_first.next != &wrk_first)
362 {
363 etp_worker *wrk = wrk_first.next;
364
365 if (wrk->req)
366 ETP_DESTROY (wrk->req);
367
368 etp_worker_clear (wrk);
369 etp_worker_free (wrk);
370 }
371 534
372 started = 0; 535 started = 0;
373 idle = 0; 536 idle = 0;
374 nreqs = 0; 537 nreqs = 0;
375 nready = 0; 538 nready = 0;
376 npending = 0; 539 npending = 0;
377 540
378 etp_atfork_parent ();
379}
380
381static void
382etp_once_init (void)
383{
384 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
385}
386
387static int
388etp_init (void (*want_poll)(void), void (*done_poll)(void))
389{
390 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
391
392 pthread_once (&doinit, etp_once_init);
393
394 want_poll_cb = want_poll; 541 want_poll_cb = want_poll;
395 done_poll_cb = done_poll; 542 done_poll_cb = done_poll;
396 543
397 return 0; 544 return 0;
398} 545}
399 546
400X_THREAD_PROC (etp_proc); 547X_THREAD_PROC (etp_proc);
401 548
549static void ecb_cold
402static void etp_start_thread (void) 550etp_start_thread (void)
403{ 551{
404 etp_worker *wrk = calloc (1, sizeof (etp_worker)); 552 etp_worker *wrk = calloc (1, sizeof (etp_worker));
405 553
406 /*TODO*/ 554 /*TODO*/
407 assert (("unable to allocate worker thread data", wrk)); 555 assert (("unable to allocate worker thread data", wrk));
… …
420 free (wrk); 568 free (wrk);
421 569
422 X_UNLOCK (wrklock); 570 X_UNLOCK (wrklock);
423} 571}
424 572
573static void
425static void etp_maybe_start_thread (void) 574etp_maybe_start_thread (void)
426{ 575{
427 if (expect_true (etp_nthreads () >= wanted)) 576 if (ecb_expect_true (etp_nthreads () >= wanted))
428 return; 577 return;
429 578
430 /* todo: maybe use idle here, but might be less exact */ 579 /* todo: maybe use idle here, but might be less exact */
431 if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ())) 580 if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
432 return; 581 return;
433 582
434 etp_start_thread (); 583 etp_start_thread ();
435} 584}
436 585
586static void ecb_cold
437static void etp_end_thread (void) 587etp_end_thread (void)
438{ 588{
439 eio_req *req = calloc (1, sizeof (eio_req)); 589 eio_req *req = calloc (1, sizeof (eio_req));
440 590
441 req->type = -1; 591 req->type = -1;
442 req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 592 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
… …
449 X_LOCK (wrklock); 599 X_LOCK (wrklock);
450 --started; 600 --started;
451 X_UNLOCK (wrklock); 601 X_UNLOCK (wrklock);
452} 602}
453 603
454static int etp_poll (void) 604static int
605etp_poll (void)
455{ 606{
456 unsigned int maxreqs; 607 unsigned int maxreqs;
457 unsigned int maxtime; 608 unsigned int maxtime;
458 struct timeval tv_start, tv_now; 609 struct timeval tv_start, tv_now;
459 610
… …
489 640
490 X_LOCK (reqlock); 641 X_LOCK (reqlock);
491 --nreqs; 642 --nreqs;
492 X_UNLOCK (reqlock); 643 X_UNLOCK (reqlock);
493 644
494 if (expect_false (req->type == EIO_GROUP && req->size)) 645 if (ecb_expect_false (req->type == EIO_GROUP && req->size))
495 { 646 {
496 req->int1 = 1; /* mark request as delayed */ 647 req->int1 = 1; /* mark request as delayed */
497 continue; 648 continue;
498 } 649 }
499 else 650 else
500 { 651 {
501 int res = ETP_FINISH (req); 652 int res = ETP_FINISH (req);
502 if (expect_false (res)) 653 if (ecb_expect_false (res))
503 return res; 654 return res;
504 } 655 }
505 656
506 if (expect_false (maxreqs && !--maxreqs)) 657 if (ecb_expect_false (maxreqs && !--maxreqs))
507 break; 658 break;
508 659
509 if (maxtime) 660 if (maxtime)
510 { 661 {
511 gettimeofday (&tv_now, 0); 662 gettimeofday (&tv_now, 0);
… …
517 668
518 errno = EAGAIN; 669 errno = EAGAIN;
519 return -1; 670 return -1;
520} 671}
521 672
673static void
522static void etp_cancel (ETP_REQ *req) 674etp_cancel (ETP_REQ *req)
523{ 675{
524 X_LOCK (wrklock); 676 req->cancelled = 1;
525 req->flags |= EIO_FLAG_CANCELLED;
526 X_UNLOCK (wrklock);
527 677
528 eio_grp_cancel (req); 678 eio_grp_cancel (req);
529} 679}
530 680
681static void
531static void etp_submit (ETP_REQ *req) 682etp_submit (ETP_REQ *req)
532{ 683{
533 req->pri -= ETP_PRI_MIN; 684 req->pri -= ETP_PRI_MIN;
534 685
535 if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN; 686 if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
536 if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 687 if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
537 688
538 if (expect_false (req->type == EIO_GROUP)) 689 if (ecb_expect_false (req->type == EIO_GROUP))
539 { 690 {
540 /* I hope this is worth it :/ */ 691 /* I hope this is worth it :/ */
541 X_LOCK (reqlock); 692 X_LOCK (reqlock);
542 ++nreqs; 693 ++nreqs;
543 X_UNLOCK (reqlock); 694 X_UNLOCK (reqlock);
… …
562 713
563 etp_maybe_start_thread (); 714 etp_maybe_start_thread ();
564 } 715 }
565} 716}
566 717
718static void ecb_cold
567static void etp_set_max_poll_time (double nseconds) 719etp_set_max_poll_time (double nseconds)
568{ 720{
569 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 721 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
570 max_poll_time = nseconds; 722 max_poll_time = nseconds * EIO_TICKS;
571 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 723 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
572} 724}
573 725
726static void ecb_cold
574static void etp_set_max_poll_reqs (unsigned int maxreqs) 727etp_set_max_poll_reqs (unsigned int maxreqs)
575{ 728{
576 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 729 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
577 max_poll_reqs = maxreqs; 730 max_poll_reqs = maxreqs;
578 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 731 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
579} 732}
580 733
734static void ecb_cold
581static void etp_set_max_idle (unsigned int nthreads) 735etp_set_max_idle (unsigned int nthreads)
582{ 736{
583 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 737 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
584 max_idle = nthreads <= 0 ? 1 : nthreads; 738 max_idle = nthreads;
585 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 739 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
586} 740}
587 741
742static void ecb_cold
743etp_set_idle_timeout (unsigned int seconds)
744{
745 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
746 idle_timeout = seconds;
747 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
748}
749
750static void ecb_cold
588static void etp_set_min_parallel (unsigned int nthreads) 751etp_set_min_parallel (unsigned int nthreads)
589{ 752{
590 if (wanted < nthreads) 753 if (wanted < nthreads)
591 wanted = nthreads; 754 wanted = nthreads;
592} 755}
593 756
757static void ecb_cold
594static void etp_set_max_parallel (unsigned int nthreads) 758etp_set_max_parallel (unsigned int nthreads)
595{ 759{
596 if (wanted > nthreads) 760 if (wanted > nthreads)
597 wanted = nthreads; 761 wanted = nthreads;
598 762
599 while (started > wanted) 763 while (started > wanted)
600 etp_end_thread (); 764 etp_end_thread ();
601} 765}
602 766
603/*****************************************************************************/ 767/*****************************************************************************/
604 768
769static void
605static void grp_try_feed (eio_req *grp) 770grp_try_feed (eio_req *grp)
606{ 771{
607 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 772 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
608 { 773 {
609 grp->flags &= ~EIO_FLAG_GROUPADD; 774 grp->flags &= ~EIO_FLAG_GROUPADD;
610 775
… …
617 break; 782 break;
618 } 783 }
619 } 784 }
620} 785}
621 786
787static int
622static int grp_dec (eio_req *grp) 788grp_dec (eio_req *grp)
623{ 789{
624 --grp->size; 790 --grp->size;
625 791
626 /* call feeder, if applicable */ 792 /* call feeder, if applicable */
627 grp_try_feed (grp); 793 grp_try_feed (grp);
… …
631 return eio_finish (grp); 797 return eio_finish (grp);
632 else 798 else
633 return 0; 799 return 0;
634} 800}
635 801
802static void
636void eio_destroy (eio_req *req) 803eio_destroy (eio_req *req)
637{ 804{
638 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); 805 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
639 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); 806 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
640 807
641 EIO_DESTROY (req); 808 EIO_DESTROY (req);
642} 809}
643 810
811static int
644static int eio_finish (eio_req *req) 812eio_finish (eio_req *req)
645{ 813{
646 int res = EIO_FINISH (req); 814 int res = EIO_FINISH (req);
647 815
648 if (req->grp) 816 if (req->grp)
649 { 817 {
… …
657 if (grp->grp_first == req) 825 if (grp->grp_first == req)
658 grp->grp_first = req->grp_next; 826 grp->grp_first = req->grp_next;
659 827
660 res2 = grp_dec (grp); 828 res2 = grp_dec (grp);
661 829
662 if (!res && res2) 830 if (!res)
663 res = res2; 831 res = res2;
664 } 832 }
665 833
666 eio_destroy (req); 834 eio_destroy (req);
667 835
668 return res; 836 return res;
669} 837}
670 838
839void
671void eio_grp_cancel (eio_req *grp) 840eio_grp_cancel (eio_req *grp)
672{ 841{
673 for (grp = grp->grp_first; grp; grp = grp->grp_next) 842 for (grp = grp->grp_first; grp; grp = grp->grp_next)
674 eio_cancel (grp); 843 eio_cancel (grp);
675} 844}
676 845
846void
677void eio_cancel (eio_req *req) 847eio_cancel (eio_req *req)
678{ 848{
679 etp_cancel (req); 849 etp_cancel (req);
680} 850}
681 851
852void
682void eio_submit (eio_req *req) 853eio_submit (eio_req *req)
683{ 854{
684 etp_submit (req); 855 etp_submit (req);
685} 856}
686 857
687unsigned int eio_nreqs (void) 858unsigned int
859eio_nreqs (void)
688{ 860{
689 return etp_nreqs (); 861 return etp_nreqs ();
690} 862}
691 863
692unsigned int eio_nready (void) 864unsigned int
865eio_nready (void)
693{ 866{
694 return etp_nready (); 867 return etp_nready ();
695} 868}
696 869
697unsigned int eio_npending (void) 870unsigned int
871eio_npending (void)
698{ 872{
699 return etp_npending (); 873 return etp_npending ();
700} 874}
701 875
702unsigned int eio_nthreads (void) 876unsigned int ecb_cold
877eio_nthreads (void)
703{ 878{
704 return etp_nthreads (); 879 return etp_nthreads ();
705} 880}
706 881
882void ecb_cold
707void eio_set_max_poll_time (double nseconds) 883eio_set_max_poll_time (double nseconds)
708{ 884{
709 etp_set_max_poll_time (nseconds); 885 etp_set_max_poll_time (nseconds);
710} 886}
711 887
888void ecb_cold
712void eio_set_max_poll_reqs (unsigned int maxreqs) 889eio_set_max_poll_reqs (unsigned int maxreqs)
713{ 890{
714 etp_set_max_poll_reqs (maxreqs); 891 etp_set_max_poll_reqs (maxreqs);
715} 892}
716 893
894void ecb_cold
717void eio_set_max_idle (unsigned int nthreads) 895eio_set_max_idle (unsigned int nthreads)
718{ 896{
719 etp_set_max_idle (nthreads); 897 etp_set_max_idle (nthreads);
720} 898}
721 899
900void ecb_cold
901eio_set_idle_timeout (unsigned int seconds)
902{
903 etp_set_idle_timeout (seconds);
904}
905
906void ecb_cold
722void eio_set_min_parallel (unsigned int nthreads) 907eio_set_min_parallel (unsigned int nthreads)
723{ 908{
724 etp_set_min_parallel (nthreads); 909 etp_set_min_parallel (nthreads);
725} 910}
726 911
912void ecb_cold
727void eio_set_max_parallel (unsigned int nthreads) 913eio_set_max_parallel (unsigned int nthreads)
728{ 914{
729 etp_set_max_parallel (nthreads); 915 etp_set_max_parallel (nthreads);
730} 916}
731 917
732int eio_poll (void) 918int eio_poll (void)
… …
741# undef pread 927# undef pread
742# undef pwrite 928# undef pwrite
743# define pread eio__pread 929# define pread eio__pread
744# define pwrite eio__pwrite 930# define pwrite eio__pwrite
745 931
746static ssize_t 932static eio_ssize_t
747eio__pread (int fd, void *buf, size_t count, off_t offset) 933eio__pread (int fd, void *buf, size_t count, off_t offset)
748{ 934{
749 ssize_t res; 935 eio_ssize_t res;
750 off_t ooffset; 936 off_t ooffset;
751 937
752 X_LOCK (preadwritelock); 938 X_LOCK (preadwritelock);
753 ooffset = lseek (fd, 0, SEEK_CUR); 939 ooffset = lseek (fd, 0, SEEK_CUR);
754 lseek (fd, offset, SEEK_SET); 940 lseek (fd, offset, SEEK_SET);
… …
757 X_UNLOCK (preadwritelock); 943 X_UNLOCK (preadwritelock);
758 944
759 return res; 945 return res;
760} 946}
761 947
762static ssize_t 948static eio_ssize_t
763eio__pwrite (int fd, void *buf, size_t count, off_t offset) 949eio__pwrite (int fd, void *buf, size_t count, off_t offset)
764{ 950{
765 ssize_t res; 951 eio_ssize_t res;
766 off_t ooffset; 952 off_t ooffset;
767 953
768 X_LOCK (preadwritelock); 954 X_LOCK (preadwritelock);
769 ooffset = lseek (fd, 0, SEEK_CUR); 955 ooffset = lseek (fd, 0, SEEK_CUR);
770 lseek (fd, offset, SEEK_SET); 956 lseek (fd, offset, SEEK_SET);
771 res = write (fd, buf, count); 957 res = write (fd, buf, count);
772 lseek (fd, offset, SEEK_SET); 958 lseek (fd, ooffset, SEEK_SET);
773 X_UNLOCK (preadwritelock); 959 X_UNLOCK (preadwritelock);
774 960
775 return res; 961 return res;
776} 962}
777#endif 963#endif
778 964
779#ifndef HAVE_FUTIMES 965#ifndef HAVE_UTIMES
780 966
781# undef utimes 967# undef utimes
782# undef futimes
783# define utimes(path,times) eio__utimes (path, times) 968# define utimes(path,times) eio__utimes (path, times)
784# define futimes(fd,times) eio__futimes (fd, times)
785 969
786static int 970static int
787eio__utimes (const char *filename, const struct timeval times[2]) 971eio__utimes (const char *filename, const struct timeval times[2])
788{ 972{
789 if (times) 973 if (times)
… …
797 } 981 }
798 else 982 else
799 return utime (filename, 0); 983 return utime (filename, 0);
800} 984}
801 985
986#endif
987
988#ifndef HAVE_FUTIMES
989
990# undef futimes
991# define futimes(fd,times) eio__futimes (fd, times)
992
993static int
802static int eio__futimes (int fd, const struct timeval tv[2]) 994eio__futimes (int fd, const struct timeval tv[2])
803{ 995{
804 errno = ENOSYS; 996 errno = ENOSYS;
805 return -1; 997 return -1;
806} 998}
807 999
… …
810#if !HAVE_FDATASYNC 1002#if !HAVE_FDATASYNC
811# undef fdatasync 1003# undef fdatasync
812# define fdatasync(fd) fsync (fd) 1004# define fdatasync(fd) fsync (fd)
813#endif 1005#endif
814 1006
1007static int
1008eio__syncfs (int fd)
1009{
1010 int res;
1011
1012#if HAVE_SYS_SYNCFS
1013 res = (int)syscall (__NR_syncfs, (int)(fd));
1014#else
1015 res = -1;
1016 errno = ENOSYS;
1017#endif
1018
1019 if (res < 0 && errno == ENOSYS && fd >= 0)
1020 sync ();
1021
1022 return res;
1023}
1024
815/* sync_file_range always needs emulation */ 1025/* sync_file_range always needs emulation */
816int 1026static int
817eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags) 1027eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
818{ 1028{
819#if HAVE_SYNC_FILE_RANGE 1029#if HAVE_SYNC_FILE_RANGE
820 int res; 1030 int res;
821 1031
… …
829 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0); 1039 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
830 } 1040 }
831 1041
832 res = sync_file_range (fd, offset, nbytes, flags); 1042 res = sync_file_range (fd, offset, nbytes, flags);
833 1043
834 if (res != ENOSYS) 1044 if (!res || errno != ENOSYS)
835 return res; 1045 return res;
836#endif 1046#endif
837 1047
838 /* even though we could play tricks with the flags, it's better to always 1048 /* even though we could play tricks with the flags, it's better to always
839 * call fdatasync, as thta matches the expectation of it's users best */ 1049 * call fdatasync, as that matches the expectation of its users best */
840 return fdatasync (fd); 1050 return fdatasync (fd);
1051}
1052
1053static int
1054eio__fallocate (int fd, int mode, off_t offset, size_t len)
1055{
1056#if HAVE_FALLOCATE
1057 return fallocate (fd, mode, offset, len);
1058#else
1059 errno = ENOSYS;
1060 return -1;
1061#endif
841} 1062}
842 1063
843#if !HAVE_READAHEAD 1064#if !HAVE_READAHEAD
844# undef readahead 1065# undef readahead
845# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 1066# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
846 1067
847static ssize_t 1068static eio_ssize_t
848eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) 1069eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
849{ 1070{
850 size_t todo = count; 1071 size_t todo = count;
851 dBUF; 1072 dBUF;
852 1073
… …
857 pread (fd, eio_buf, len, offset); 1078 pread (fd, eio_buf, len, offset);
858 offset += len; 1079 offset += len;
859 todo -= len; 1080 todo -= len;
860 } 1081 }
861 1082
1083 FUBd;
1084
862 errno = 0; 1085 errno = 0;
863 return count; 1086 return count;
864} 1087}
865 1088
866#endif 1089#endif
867 1090
868/* sendfile always needs emulation */ 1091/* sendfile always needs emulation */
869static ssize_t 1092static eio_ssize_t
870eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) 1093eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
871{ 1094{
1095 eio_ssize_t written = 0;
872 ssize_t res; 1096 eio_ssize_t res;
873 1097
874 if (!count) 1098 if (!count)
875 return 0; 1099 return 0;
876 1100
1101 for (;;)
1102 {
1103#ifdef __APPLE__
1104# undef HAVE_SENDFILE /* broken, as everything on os x */
1105#endif
877#if HAVE_SENDFILE 1106#if HAVE_SENDFILE
878# if __linux 1107# if __linux
1108 off_t soffset = offset;
879 res = sendfile (ofd, ifd, &offset, count); 1109 res = sendfile (ofd, ifd, &soffset, count);
880 1110
881# elif __freebsd 1111# elif __FreeBSD__
882 /* 1112 /*
883 * Of course, the freebsd sendfile is a dire hack with no thoughts 1113 * Of course, the freebsd sendfile is a dire hack with no thoughts
884 * wasted on making it similar to other I/O functions. 1114 * wasted on making it similar to other I/O functions.
885 */ 1115 */
886 {
887 off_t sbytes; 1116 off_t sbytes;
888 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 1117 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
889 1118
890 if (res < 0 && sbytes) 1119 #if 0 /* according to the manpage, this is correct, but broken behaviour */
891 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 1120 /* freebsd' sendfile will return 0 on success */
1121 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
1122 /* not on e.g. EIO or EPIPE - sounds broken */
1123 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
892 res = sbytes; 1124 res = sbytes;
893 } 1125 #endif
1126
1127 /* according to source inspection, this is correct, and useful behaviour */
1128 if (sbytes)
1129 res = sbytes;
1130
1131# elif defined (__APPLE__)
1132 off_t sbytes = count;
1133 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
1134
1135 /* according to the manpage, sbytes is always valid */
1136 if (sbytes)
1137 res = sbytes;
894 1138
895# elif __hpux 1139# elif __hpux
896 res = sendfile (ofd, ifd, offset, count, 0, 0); 1140 res = sendfile (ofd, ifd, offset, count, 0, 0);
897 1141
898# elif __solaris 1142# elif __solaris
899 {
900 struct sendfilevec vec; 1143 struct sendfilevec vec;
901 size_t sbytes; 1144 size_t sbytes;
902 1145
903 vec.sfv_fd = ifd; 1146 vec.sfv_fd = ifd;
904 vec.sfv_flag = 0; 1147 vec.sfv_flag = 0;
905 vec.sfv_off = offset; 1148 vec.sfv_off = offset;
906 vec.sfv_len = count; 1149 vec.sfv_len = count;
907 1150
908 res = sendfilev (ofd, &vec, 1, &sbytes); 1151 res = sendfilev (ofd, &vec, 1, &sbytes);
909 1152
910 if (res < 0 && sbytes) 1153 if (res < 0 && sbytes)
911 res = sbytes; 1154 res = sbytes;
912 }
913 1155
914# endif 1156# endif
1157
1158#elif defined (_WIN32) && 0
1159 /* does not work, just for documentation of what would need to be done */
1160 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
1161 /* libeio guarantees that the file offset does not change, and windows */
1162 /* has no way to get an independent handle to the same file description */
1163 HANDLE h = TO_SOCKET (ifd);
1164 SetFilePointer (h, offset, 0, FILE_BEGIN);
1165 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1166
915#else 1167#else
916 res = -1; 1168 res = -1;
917 errno = ENOSYS; 1169 errno = ENOSYS;
918#endif 1170#endif
919 1171
1172 /* we assume sendfile can copy at least 128mb in one go */
1173 if (res <= 128 * 1024 * 1024)
1174 {
1175 if (res > 0)
1176 written += res;
1177
1178 if (written)
1179 return written;
1180
1181 break;
1182 }
1183 else
1184 {
1185 /* if we requested more, then probably the kernel was lazy */
1186 written += res;
1187 offset += res;
1188 count -= res;
1189
1190 if (!count)
1191 return written;
1192 }
1193 }
1194
920 if (res < 0 1195 if (res < 0
921 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1196 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1197 /* BSDs */
1198#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1199 || errno == ENOTSUP
1200#endif
1201#ifdef EOPNOTSUPP /* windows */
1202 || errno == EOPNOTSUPP /* BSDs */
1203#endif
922#if __solaris 1204#if __solaris
923 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1205 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
924#endif 1206#endif
925 ) 1207 )
926 ) 1208 )
… …
930 1212
931 res = 0; 1213 res = 0;
932 1214
933 while (count) 1215 while (count)
934 { 1216 {
935 ssize_t cnt; 1217 eio_ssize_t cnt;
936 1218
937 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 1219 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
938 1220
939 if (cnt <= 0) 1221 if (cnt <= 0)
940 { 1222 {
… …
952 1234
953 offset += cnt; 1235 offset += cnt;
954 res += cnt; 1236 res += cnt;
955 count -= cnt; 1237 count -= cnt;
956 } 1238 }
1239
1240 FUBd;
957 } 1241 }
958 1242
959 return res; 1243 return res;
1244}
1245
1246#ifdef PAGESIZE
1247# define eio_pagesize() PAGESIZE
1248#else
1249static intptr_t
1250eio_pagesize (void)
1251{
1252 static intptr_t page;
1253
1254 if (!page)
1255 page = sysconf (_SC_PAGESIZE);
1256
1257 return page;
1258}
1259#endif
1260
1261static void
1262eio_page_align (void **addr, size_t *length)
1263{
1264 intptr_t mask = eio_pagesize () - 1;
1265
1266 /* round down addr */
1267 intptr_t adj = mask & (intptr_t)*addr;
1268
1269 *addr = (void *)((intptr_t)*addr - adj);
1270 *length += adj;
1271
1272 /* round up length */
1273 *length = (*length + mask) & ~mask;
1274}
1275
1276#if !_POSIX_MEMLOCK
1277# define eio__mlockall(a) EIO_ENOSYS ()
1278#else
1279
1280static int
1281eio__mlockall (int flags)
1282{
1283 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1284 extern int mallopt (int, int);
1285 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1286 #endif
1287
1288 if (EIO_MCL_CURRENT != MCL_CURRENT
1289 || EIO_MCL_FUTURE != MCL_FUTURE)
1290 {
1291 flags = 0
1292 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1293 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1294 }
1295
1296 return mlockall (flags);
1297}
1298#endif
1299
1300#if !_POSIX_MEMLOCK_RANGE
1301# define eio__mlock(a,b) EIO_ENOSYS ()
1302#else
1303
1304static int
1305eio__mlock (void *addr, size_t length)
1306{
1307 eio_page_align (&addr, &length);
1308
1309 return mlock (addr, length);
1310}
1311
1312#endif
1313
1314#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1315# define eio__msync(a,b,c) EIO_ENOSYS ()
1316#else
1317
1318static int
1319eio__msync (void *mem, size_t len, int flags)
1320{
1321 eio_page_align (&mem, &len);
1322
1323 if (EIO_MS_ASYNC != MS_SYNC
1324 || EIO_MS_INVALIDATE != MS_INVALIDATE
1325 || EIO_MS_SYNC != MS_SYNC)
1326 {
1327 flags = 0
1328 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1329 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1330 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1331 }
1332
1333 return msync (mem, len, flags);
1334}
1335
1336#endif
1337
1338static int
1339eio__mtouch (eio_req *req)
1340{
1341 void *mem = req->ptr2;
1342 size_t len = req->size;
1343 int flags = req->int1;
1344
1345 eio_page_align (&mem, &len);
1346
1347 {
1348 intptr_t addr = (intptr_t)mem;
1349 intptr_t end = addr + len;
1350 intptr_t page = eio_pagesize ();
1351
1352 if (addr < end)
1353 if (flags & EIO_MT_MODIFY) /* modify */
1354 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1355 else
1356 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1357 }
1358
1359 return 0;
1360}
1361
1362/*****************************************************************************/
1363/* requests implemented outside eio_execute, because they are so large */
1364
1365/* copies some absolute path to tmpbuf */
1366static char *
1367eio__getwd (struct tmpbuf *tmpbuf, eio_wd wd)
1368{
1369 if (wd == EIO_CWD)
1370 return getcwd (tmpbuf->ptr, PATH_MAX);
1371
1372#if HAVE_AT
1373 abort (); /*TODO*/
1374#else
1375 strcpy (tmpbuf->ptr, wd);
1376#endif
1377 return tmpbuf->ptr;
1378}
1379
1380/* result will always end up in tmpbuf, there is always space for adding a 0-byte */
1381static int
1382eio__realpath (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
1383{
1384 const char *rel = path;
1385 char *res;
1386 char *tmp1, *tmp2;
1387#if SYMLOOP_MAX > 32
1388 int symlinks = SYMLOOP_MAX;
1389#else
1390 int symlinks = 32;
1391#endif
1392
1393 /*D*/ /*TODO: wd ignored */
1394
1395 errno = EINVAL;
1396 if (!rel)
1397 return -1;
1398
1399 errno = ENOENT;
1400 if (!*rel)
1401 return -1;
1402
1403 res = tmpbuf_get (tmpbuf, PATH_MAX * 3);
1404 tmp1 = res + PATH_MAX;
1405 tmp2 = tmp1 + PATH_MAX;
1406
1407#if 0 /* disabled, the musl way to do things is just too racy */
1408#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1409 /* on linux we may be able to ask the kernel */
1410 {
1411 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1412
1413 if (fd >= 0)
1414 {
1415 sprintf (tmp1, "/proc/self/fd/%d", fd);
1416 req->result = readlink (tmp1, res, PATH_MAX);
1417 close (fd);
1418
1419 /* here we should probably stat the open file and the disk file, to make sure they still match */
1420
1421 if (req->result > 0)
1422 goto done;
1423 }
1424 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1425 return;
1426 }
1427#endif
1428#endif
1429
1430 if (*rel != '/')
1431 {
1432 if (!eio__getwd (tmpbuf, wd))
1433 return -1;
1434
1435 if (res [1]) /* only use if not / */
1436 res += strlen (res);
1437 }
1438
1439 while (*rel)
1440 {
1441 eio_ssize_t len, linklen;
1442 const char *beg = rel;
1443
1444 while (*rel && *rel != '/')
1445 ++rel;
1446
1447 len = rel - beg;
1448
1449 if (!len) /* skip slashes */
1450 {
1451 ++rel;
1452 continue;
1453 }
1454
1455 if (beg [0] == '.')
1456 {
1457 if (len == 1)
1458 continue; /* . - nop */
1459
1460 if (beg [1] == '.' && len == 2)
1461 {
1462 /* .. - back up one component, if possible */
1463
1464 while (res != tmpbuf->ptr)
1465 if (*--res == '/')
1466 break;
1467
1468 continue;
1469 }
1470 }
1471
1472 errno = ENAMETOOLONG;
1473 if (res + 1 + len + 1 >= tmp1)
1474 return;
1475
1476 /* copy one component */
1477 *res = '/';
1478 memcpy (res + 1, beg, len);
1479
1480 /* zero-terminate, for readlink */
1481 res [len + 1] = 0;
1482
1483 /* now check if it's a symlink */
1484 linklen = readlink (tmpbuf->ptr, tmp1, PATH_MAX);
1485
1486 if (linklen < 0)
1487 {
1488 if (errno != EINVAL)
1489 return -1;
1490
1491 /* it's a normal directory. hopefully */
1492 res += len + 1;
1493 }
1494 else
1495 {
1496 /* yay, it was a symlink - build new path in tmp2 */
1497 int rellen = strlen (rel);
1498
1499 errno = ENAMETOOLONG;
1500 if (linklen + 1 + rellen >= PATH_MAX)
1501 return -1;
1502
1503 errno = ELOOP;
1504 if (!--symlinks)
1505 return -1;
1506
1507 if (*tmp1 == '/')
1508 res = tmpbuf->ptr; /* symlink resolves to an absolute path */
1509
1510 /* we need to be careful, as rel might point into tmp2 already */
1511 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1512 tmp2 [linklen] = '/';
1513 memcpy (tmp2, tmp1, linklen);
1514
1515 rel = tmp2;
1516 }
1517 }
1518
1519 /* special case for the lone root path */
1520 if (res == tmpbuf->ptr)
1521 *res++ = '/';
1522
1523 return res - (char *)tmpbuf->ptr;
1524}
1525
1526static signed char
1527eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1528{
1529 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1530 : a->inode < b->inode ? -1
1531 : a->inode > b->inode ? 1
1532 : 0;
1533}
1534
1535#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1536
1537#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1538#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1539
1540static void
1541eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1542{
1543 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1544 unsigned char *bit = bits;
1545
1546 assert (CHAR_BIT == 8);
1547 assert (sizeof (eio_dirent) * 8 < 256);
1548 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1549 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1550
1551 if (size <= EIO_SORT_FAST)
1552 return;
1553
1554 /* first prepare an array of bits to test in our radix sort */
1555 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1556 /* inode_bits must contain all inodes ORed together */
1557 /* which is used to skip bits that are 0 everywhere, which is very common */
1558 {
1559 eio_ino_t endianness;
1560 int i, j;
1561
1562 /* we store the byte offset of byte n into byte n of "endianness" */
1563 for (i = 0; i < sizeof (eio_ino_t); ++i)
1564 ((unsigned char *)&endianness)[i] = i;
1565
1566 *bit++ = 0;
1567
1568 for (i = 0; i < sizeof (eio_ino_t); ++i)
1569 {
1570 /* shifting off the byte offsets out of "endianness" */
1571 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1572 endianness >>= 8;
1573
1574 for (j = 0; j < 8; ++j)
1575 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1576 *bit++ = offs + j;
1577 }
1578
1579 for (j = 0; j < 8; ++j)
1580 if (score_bits & (1 << j))
1581 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1582 }
1583
1584 /* now actually do the sorting (a variant of MSD radix sort) */
1585 {
1586 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1587 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1588 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1589 int stk_idx = 0;
1590
1591 base_stk [stk_idx] = dents;
1592 end_stk [stk_idx] = dents + size;
1593 bit_stk [stk_idx] = bit - 1;
1594
1595 do
1596 {
1597 base = base_stk [stk_idx];
1598 end = end_stk [stk_idx];
1599 bit = bit_stk [stk_idx];
1600
1601 for (;;)
1602 {
1603 unsigned char O = *bit >> 3;
1604 unsigned char M = 1 << (*bit & 7);
1605
1606 eio_dirent *a = base;
1607 eio_dirent *b = end;
1608
1609 if (b - a < EIO_SORT_CUTOFF)
1610 break;
1611
1612 /* now bit-partition the array on the bit */
1613 /* this ugly asymmetric loop seems to perform much better than typical */
1614 /* partition algos found in the literature */
1615 do
1616 if (!(((unsigned char *)a)[O] & M))
1617 ++a;
1618 else if (!(((unsigned char *)--b)[O] & M))
1619 {
1620 eio_dirent tmp = *a; *a = *b; *b = tmp;
1621 ++a;
1622 }
1623 while (b > a);
1624
1625 /* next bit, or stop, if no bits left in this path */
1626 if (!*--bit)
1627 break;
1628
1629 base_stk [stk_idx] = a;
1630 end_stk [stk_idx] = end;
1631 bit_stk [stk_idx] = bit;
1632 ++stk_idx;
1633
1634 end = a;
1635 }
1636 }
1637 while (stk_idx--);
1638 }
1639}
1640
1641static void
1642eio_dent_insertion_sort (eio_dirent *dents, int size)
1643{
1644 /* first move the smallest element to the front, to act as a sentinel */
1645 {
1646 int i;
1647 eio_dirent *min = dents;
1648
1649 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1650 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1651 if (EIO_DENT_CMP (dents [i], <, *min))
1652 min = &dents [i];
1653
1654 /* swap elements 0 and j (minimum) */
1655 {
1656 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1657 }
1658 }
1659
1660 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1661 {
1662 eio_dirent *i, *j;
1663
1664 for (i = dents + 1; i < dents + size; ++i)
1665 {
1666 eio_dirent value = *i;
1667
1668 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1669 j [1] = j [0];
1670
1671 j [1] = value;
1672 }
1673 }
1674}
1675
1676static void
1677eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1678{
1679 if (size <= 1)
1680 return; /* our insertion sort relies on size > 0 */
1681
1682 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1683 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1684 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1685
1686 /* use an insertion sort at the end, or for small arrays, */
1687 /* as insertion sort is more efficient for small partitions */
1688 eio_dent_insertion_sort (dents, size);
960} 1689}
961 1690
962/* read a full directory */ 1691/* read a full directory */
963static void 1692static void
964eio__scandir (eio_req *req, etp_worker *self) 1693eio__scandir (eio_req *req)
965{ 1694{
1695 char *name, *names;
1696 int namesalloc = 4096 - sizeof (void *) * 4;
1697 int namesoffs = 0;
1698 int flags = req->int1;
1699 eio_dirent *dents = 0;
1700 int dentalloc = 128;
1701 int dentoffs = 0;
1702 eio_ino_t inode_bits = 0;
1703#ifdef _WIN32
1704 HANDLE dirp;
1705 WIN32_FIND_DATA entp;
1706#else
966 DIR *dirp; 1707 DIR *dirp;
967 EIO_STRUCT_DIRENT *entp; 1708 EIO_STRUCT_DIRENT *entp;
968 char *name, *names; 1709#endif
969 int memlen = 4096;
970 int memofs = 0;
971 int res = 0;
972 1710
973 X_LOCK (wrklock); 1711 req->result = -1;
974 /* the corresponding closedir is in ETP_WORKER_CLEAR */
975 self->dirp = dirp = opendir (req->ptr1);
976 req->flags |= EIO_FLAG_PTR2_FREE;
977 req->ptr2 = names = malloc (memlen);
978 X_UNLOCK (wrklock);
979 1712
980 if (dirp && names) 1713 if (!(flags & EIO_READDIR_DENTS))
981 for (;;) 1714 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1715
1716#ifdef _WIN32
1717 {
1718 int len = strlen ((const char *)req->ptr1);
1719 char *path = malloc (MAX_PATH);
1720 const char *fmt;
1721
1722 if (!len)
1723 fmt = "./*";
1724 else if (((const char *)req->ptr1)[len - 1] == '/' || ((const char *)req->ptr1)[len - 1] == '\\')
1725 fmt = "%s*";
1726 else
1727 fmt = "%s/*";
1728
1729 _snprintf (path, MAX_PATH, fmt, (const char *)req->ptr1);
1730 dirp = FindFirstFile (path, &entp);
1731 free (path);
1732
1733 if (dirp == INVALID_HANDLE_VALUE)
982 { 1734 {
1735 /* should steal _dosmaperr */
1736 switch (GetLastError ())
1737 {
1738 case ERROR_FILE_NOT_FOUND:
1739 req->result = 0;
1740 break;
1741
1742 case ERROR_INVALID_NAME:
1743 case ERROR_PATH_NOT_FOUND:
1744 case ERROR_NO_MORE_FILES:
1745 errno = ENOENT;
1746 break;
1747
1748 case ERROR_NOT_ENOUGH_MEMORY:
1749 errno = ENOMEM;
1750 break;
1751
1752 default:
1753 errno = EINVAL;
1754 break;
1755 }
1756
1757 return;
1758 }
1759 }
1760#else
1761 dirp = opendir (req->ptr1);
1762
1763 if (!dirp)
1764 return;
1765#endif
1766
1767 if (req->flags & EIO_FLAG_PTR1_FREE)
1768 free (req->ptr1);
1769
1770 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1771 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1772 req->ptr2 = names = malloc (namesalloc);
1773
1774 if (!names || (flags && !dents))
1775 return;
1776
1777 for (;;)
1778 {
1779 int done;
1780
1781#ifdef _WIN32
1782 done = !dirp;
1783#else
983 errno = 0; 1784 errno = 0;
984 entp = readdir (dirp); 1785 entp = readdir (dirp);
1786 done = !entp;
1787#endif
985 1788
1789 if (done)
1790 {
1791#ifndef _WIN32
1792 int old_errno = errno;
1793 closedir (dirp);
1794 errno = old_errno;
1795
986 if (!entp) 1796 if (errno)
1797 break;
1798#endif
1799
1800 /* sort etc. */
1801 req->int1 = flags;
1802 req->result = dentoffs;
1803
1804 if (flags & EIO_READDIR_STAT_ORDER)
1805 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1806 else if (flags & EIO_READDIR_DIRS_FIRST)
1807 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1808 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1809 else
1810 {
1811 /* in this case, all is known, and we just put dirs first and sort them */
1812 eio_dirent *oth = dents + dentoffs;
1813 eio_dirent *dir = dents;
1814
1815 /* now partition dirs to the front, and non-dirs to the back */
1816 /* by walking from both sides and swapping if necessary */
1817 while (oth > dir)
1818 {
1819 if (dir->type == EIO_DT_DIR)
1820 ++dir;
1821 else if ((--oth)->type == EIO_DT_DIR)
1822 {
1823 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1824
1825 ++dir;
1826 }
1827 }
1828
1829 /* now sort the dirs only (dirs all have the same score) */
1830 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1831 }
1832
987 break; 1833 break;
1834 }
988 1835
989 name = entp->d_name; 1836 /* now add the entry to our list(s) */
1837 name = D_NAME (entp);
990 1838
1839 /* skip . and .. entries */
991 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1840 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
992 { 1841 {
993 int len = strlen (name) + 1; 1842 int len = D_NAMLEN (entp) + 1;
994 1843
995 res++; 1844 while (ecb_expect_false (namesoffs + len > namesalloc))
996
997 while (memofs + len > memlen)
998 { 1845 {
999 memlen *= 2; 1846 namesalloc *= 2;
1000 X_LOCK (wrklock);
1001 req->ptr2 = names = realloc (names, memlen); 1847 req->ptr2 = names = realloc (names, namesalloc);
1002 X_UNLOCK (wrklock);
1003 1848
1004 if (!names) 1849 if (!names)
1850 break;
1851 }
1852
1853 memcpy (names + namesoffs, name, len);
1854
1855 if (dents)
1856 {
1857 struct eio_dirent *ent;
1858
1859 if (ecb_expect_false (dentoffs == dentalloc))
1860 {
1861 dentalloc *= 2;
1862 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1863
1864 if (!dents)
1005 break; 1865 break;
1006 } 1866 }
1007 1867
1008 memcpy (names + memofs, name, len); 1868 ent = dents + dentoffs;
1009 memofs += len; 1869
1870 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1871 ent->namelen = len - 1;
1872 ent->inode = D_INO (entp);
1873
1874 inode_bits |= ent->inode;
1875
1876 switch (D_TYPE (entp))
1877 {
1878 default:
1879 ent->type = EIO_DT_UNKNOWN;
1880 flags |= EIO_READDIR_FOUND_UNKNOWN;
1881 break;
1882
1883 #ifdef DT_FIFO
1884 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1885 #endif
1886 #ifdef DT_CHR
1887 case DT_CHR: ent->type = EIO_DT_CHR; break;
1888 #endif
1889 #ifdef DT_MPC
1890 case DT_MPC: ent->type = EIO_DT_MPC; break;
1891 #endif
1892 #ifdef DT_DIR
1893 case DT_DIR: ent->type = EIO_DT_DIR; break;
1894 #endif
1895 #ifdef DT_NAM
1896 case DT_NAM: ent->type = EIO_DT_NAM; break;
1897 #endif
1898 #ifdef DT_BLK
1899 case DT_BLK: ent->type = EIO_DT_BLK; break;
1900 #endif
1901 #ifdef DT_MPB
1902 case DT_MPB: ent->type = EIO_DT_MPB; break;
1903 #endif
1904 #ifdef DT_REG
1905 case DT_REG: ent->type = EIO_DT_REG; break;
1906 #endif
1907 #ifdef DT_NWK
1908 case DT_NWK: ent->type = EIO_DT_NWK; break;
1909 #endif
1910 #ifdef DT_CMP
1911 case DT_CMP: ent->type = EIO_DT_CMP; break;
1912 #endif
1913 #ifdef DT_LNK
1914 case DT_LNK: ent->type = EIO_DT_LNK; break;
1915 #endif
1916 #ifdef DT_SOCK
1917 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1918 #endif
1919 #ifdef DT_DOOR
1920 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1921 #endif
1922 #ifdef DT_WHT
1923 case DT_WHT: ent->type = EIO_DT_WHT; break;
1924 #endif
1925 }
1926
1927 ent->score = 7;
1928
1929 if (flags & EIO_READDIR_DIRS_FIRST)
1930 {
1931 if (ent->type == EIO_DT_UNKNOWN)
1932 {
1933 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1934 ent->score = 1;
1935 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1936 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1937 }
1938 else if (ent->type == EIO_DT_DIR)
1939 ent->score = 0;
1940 }
1010 } 1941 }
1942
1943 namesoffs += len;
1944 ++dentoffs;
1011 } 1945 }
1012 1946
1013 if (errno) 1947 if (EIO_CANCELLED (req))
1014 res = -1; 1948 {
1015 1949 errno = ECANCELED;
1016 req->result = res; 1950 break;
1017} 1951 }
1018 1952
1019#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) 1953#ifdef _WIN32
1020# undef msync 1954 if (!FindNextFile (dirp, &entp))
1021# define msync(a,b,c) ENOSYS 1955 {
1956 FindClose (dirp);
1957 dirp = 0;
1958 }
1022#endif 1959#endif
1960 }
1961}
1023 1962
1024int 1963/*****************************************************************************/
1025eio__mtouch (void *mem, size_t len, int flags) 1964/* working directory stuff */
1965
1966#if HAVE_AT
1967
1968#define WD2FD(wd) (wd ? ((int)wd) - 1 : AT_FDCWD)
1969
1970eio_wd
1971eio_wd_open_sync (eio_wd wd, const char *path)
1026{ 1972{
1027 intptr_t addr = (intptr_t)mem; 1973 int fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1028 intptr_t end = addr + len; 1974
1029#ifdef PAGESIZE 1975 return fd >= 0 ? (eio_wd)(fd + 1) : EIO_INVALID_WD;
1030 const intptr_t page = PAGESIZE; 1976}
1977
1978void
1979eio_wd_close_sync (eio_wd wd)
1980{
1981 int fd = WD2FD (wd);
1982
1983 if (fd >= 0)
1984 close (fd);
1985}
1986
1031#else 1987#else
1032 static intptr_t page;
1033 1988
1034 if (!page) 1989/* on legacy systems, we represent the working directories simply by their path strings */
1035 page = sysconf (_SC_PAGESIZE);
1036#endif
1037 1990
1038 addr &= ~(page - 1); /* assume page size is always a power of two */ 1991static const char *
1992wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
1993{
1994 if (!wd || *path == '/')
1995 return path;
1039 1996
1040 if (addr < end) 1997 {
1041 if (flags) /* modify */ 1998 int l1 = strlen ((const char *)wd);
1042 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len); 1999 int l2 = strlen (path);
2000
2001 char *res = tmpbuf_get (tmpbuf, l1 + l2 + 2);
2002
2003 memcpy (res, wd, l1);
2004 res [l1] = '/';
2005 memcpy (res + l1 + 1, path, l2 + 1);
2006
2007 return res;
2008 }
2009}
2010
2011eio_wd
2012eio__wd_open_sync (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
2013{
2014 if (*path == '/') /* absolute paths ignore wd */
2015 path = strdup (path);
2016 else if (path [0] == '.' && !path [1]) /* special case '.', as it is common */
2017 return wd;
1043 else 2018 else
1044 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len); 2019 {
2020 int len = eio__realpath (tmpbuf, wd, path);
1045 2021
2022 path = EIO_INVALID_WD;
2023
2024 if (len >= 0)
2025 {
2026 ((char *)tmpbuf->ptr)[len] = 0;
2027 path = strdup (tmpbuf->ptr);
2028 }
2029 }
2030
2031 if (!path)
2032 path = EIO_INVALID_WD;
2033
2034 return (eio_wd)path;
2035}
2036
2037eio_wd
2038eio_wd_open_sync (eio_wd wd, const char *path)
2039{
2040 struct tmpbuf tmpbuf = { 0 };
2041 wd = eio__wd_open_sync (&tmpbuf, wd, path);
2042 free (tmpbuf.ptr);
2043
1046 return 0; 2044 return wd;
1047} 2045}
2046
2047void
2048eio_wd_close_sync (eio_wd wd)
2049{
2050 if (wd != EIO_INVALID_WD)
2051 free (wd);
2052}
2053
2054#endif
1048 2055
1049/*****************************************************************************/ 2056/*****************************************************************************/
1050 2057
1051#define ALLOC(len) \ 2058#define ALLOC(len) \
1052 if (!req->ptr2) \ 2059 if (!req->ptr2) \
… …
1067{ 2074{
1068 ETP_REQ *req; 2075 ETP_REQ *req;
1069 struct timespec ts; 2076 struct timespec ts;
1070 etp_worker *self = (etp_worker *)thr_arg; 2077 etp_worker *self = (etp_worker *)thr_arg;
1071 2078
2079#if HAVE_PRCTL_SET_NAME
2080 prctl (PR_SET_NAME, (unsigned long)"eio_thread", 0, 0, 0);
2081#endif
2082
1072 /* try to distribute timeouts somewhat randomly */ 2083 /* try to distribute timeouts somewhat evenly */
1073 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 2084 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1074 2085
1075 for (;;) 2086 for (;;)
1076 { 2087 {
2088 ts.tv_sec = 0;
2089
1077 X_LOCK (reqlock); 2090 X_LOCK (reqlock);
1078 2091
1079 for (;;) 2092 for (;;)
1080 { 2093 {
1081 self->req = req = reqq_shift (&req_queue); 2094 req = reqq_shift (&req_queue);
1082 2095
1083 if (req) 2096 if (req)
1084 break; 2097 break;
1085 2098
2099 if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
2100 {
2101 X_UNLOCK (reqlock);
2102 X_LOCK (wrklock);
2103 --started;
2104 X_UNLOCK (wrklock);
2105 goto quit;
2106 }
2107
1086 ++idle; 2108 ++idle;
1087 2109
1088 ts.tv_sec = time (0) + IDLE_TIMEOUT; 2110 if (idle <= max_idle)
1089 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 2111 /* we are allowed to idle, so do so without any timeout */
2112 X_COND_WAIT (reqwait, reqlock);
2113 else
1090 { 2114 {
1091 if (idle > max_idle) 2115 /* initialise timeout once */
1092 { 2116 if (!ts.tv_sec)
1093 --idle; 2117 ts.tv_sec = time (0) + idle_timeout;
1094 X_UNLOCK (reqlock);
1095 X_LOCK (wrklock);
1096 --started;
1097 X_UNLOCK (wrklock);
1098 goto quit;
1099 }
1100 2118
1101 /* we are allowed to idle, so do so without any timeout */
1102 X_COND_WAIT (reqwait, reqlock); 2119 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
2120 ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1103 } 2121 }
1104 2122
1105 --idle; 2123 --idle;
1106 } 2124 }
1107 2125
… …
1110 X_UNLOCK (reqlock); 2128 X_UNLOCK (reqlock);
1111 2129
1112 if (req->type < 0) 2130 if (req->type < 0)
1113 goto quit; 2131 goto quit;
1114 2132
1115 if (!EIO_CANCELLED (req))
1116 ETP_EXECUTE (self, req); 2133 ETP_EXECUTE (self, req);
1117 2134
1118 X_LOCK (reslock); 2135 X_LOCK (reslock);
1119 2136
1120 ++npending; 2137 ++npending;
1121 2138
1122 if (!reqq_push (&res_queue, req) && want_poll_cb) 2139 if (!reqq_push (&res_queue, req) && want_poll_cb)
1123 want_poll_cb (); 2140 want_poll_cb ();
1124 2141
1125 self->req = 0;
1126 etp_worker_clear (self); 2142 etp_worker_clear (self);
1127 2143
1128 X_UNLOCK (reslock); 2144 X_UNLOCK (reslock);
1129 } 2145 }
1130 2146
1131quit: 2147quit:
2148 free (req);
2149
1132 X_LOCK (wrklock); 2150 X_LOCK (wrklock);
1133 etp_worker_free (self); 2151 etp_worker_free (self);
1134 X_UNLOCK (wrklock); 2152 X_UNLOCK (wrklock);
1135 2153
1136 return 0; 2154 return 0;
1137} 2155}
1138 2156
1139/*****************************************************************************/ 2157/*****************************************************************************/
1140 2158
2159int ecb_cold
1141int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 2160eio_init (void (*want_poll)(void), void (*done_poll)(void))
1142{ 2161{
2162#if !HAVE_PREADWRITE
2163 X_MUTEX_CREATE (preadwritelock);
2164#endif
2165
1143 return etp_init (want_poll, done_poll); 2166 return etp_init (want_poll, done_poll);
1144} 2167}
1145 2168
2169ecb_inline void
1146static void eio_api_destroy (eio_req *req) 2170eio_api_destroy (eio_req *req)
1147{ 2171{
1148 free (req); 2172 free (req);
1149} 2173}
1150 2174
1151#define REQ(rtype) \ 2175#define REQ(rtype) \
… …
1170 { \ 2194 { \
1171 eio_api_destroy (req); \ 2195 eio_api_destroy (req); \
1172 return 0; \ 2196 return 0; \
1173 } 2197 }
1174 2198
2199static void
1175static void eio_execute (etp_worker *self, eio_req *req) 2200eio_execute (etp_worker *self, eio_req *req)
1176{ 2201{
1177 errno = 0; 2202#if HAVE_AT
2203 int dirfd;
2204#else
2205 const char *path;
2206#endif
2207
2208 if (ecb_expect_false (EIO_CANCELLED (req)))
2209 {
2210 req->result = -1;
2211 req->errorno = ECANCELED;
2212 return;
2213 }
2214
2215 if (ecb_expect_false (req->wd == EIO_INVALID_WD))
2216 {
2217 req->result = -1;
2218 req->errorno = ENOENT;
2219 return;
2220 }
2221
2222 if (req->type >= EIO_OPEN)
2223 {
2224 #if HAVE_AT
2225 dirfd = WD2FD (req->wd);
2226 #else
2227 path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
2228 #endif
2229 }
1178 2230
1179 switch (req->type) 2231 switch (req->type)
1180 { 2232 {
2233 case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
2234 req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
2235 break;
2236 case EIO_WD_CLOSE: eio_wd_close_sync (req->wd); break;
2237
1181 case EIO_READ: ALLOC (req->size); 2238 case EIO_READ: ALLOC (req->size);
1182 req->result = req->offs >= 0 2239 req->result = req->offs >= 0
1183 ? pread (req->int1, req->ptr2, req->size, req->offs) 2240 ? pread (req->int1, req->ptr2, req->size, req->offs)
1184 : read (req->int1, req->ptr2, req->size); break; 2241 : read (req->int1, req->ptr2, req->size); break;
1185 case EIO_WRITE: req->result = req->offs >= 0 2242 case EIO_WRITE: req->result = req->offs >= 0
1186 ? pwrite (req->int1, req->ptr2, req->size, req->offs) 2243 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1187 : write (req->int1, req->ptr2, req->size); break; 2244 : write (req->int1, req->ptr2, req->size); break;
1188 2245
1189 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break; 2246 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1190 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break; 2247 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
1191 2248
2249#if HAVE_AT
2250 case EIO_GETPATH: abort ();
1192 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2251 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1193 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2252 req->result = fstatat (dirfd, path, (EIO_STRUCT_STAT *)req->ptr2); break;
1194 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2253 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1195 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2254 req->result = lstat (dirfd, path, (EIO_STRUCT_STAT *)req->ptr2); break;
2255#if 0/*D*/
2256 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2257 req->result = statvfs (dirfd, path, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2258#endif
2259 case EIO_CHOWN: req->result = chown (dirfd, path, req->int2, req->int3); break;
2260 case EIO_CHMOD: req->result = chmod (dirfd, path, (mode_t)req->int2); break;
2261 case EIO_TRUNCATE: req->result = truncate (dirfd, path, req->offs); break;
2262 case EIO_OPEN: req->result = open (dirfd, path, req->int1, (mode_t)req->int2); break;
2263
2264 case EIO_UNLINK: req->result = unlink (dirfd, path); break;
2265 case EIO_RMDIR: req->result = rmdir (dirfd, path); break;
2266 case EIO_MKDIR: req->result = mkdir (dirfd, path, (mode_t)req->int2); break;
2267 case EIO_RENAME: req->result = rename (dirfd, path, req->ptr2); break;
2268 case EIO_LINK: req->result = link (dirfd, path, req->ptr2); break;
2269 case EIO_SYMLINK: req->result = symlink (dirfd, path, req->ptr2); break;
2270 case EIO_MKNOD: req->result = mknod (dirfd, path, (mode_t)req->int2, (dev_t)req->offs); break;
2271#else
2272 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2273 req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
2274 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2275 req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
2276 case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
2277 case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
2278 case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
2279 case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
2280
2281 case EIO_UNLINK: req->result = unlink (path ); break;
2282 case EIO_RMDIR: req->result = rmdir (path ); break;
2283 case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
2284 case EIO_RENAME: req->result = rename (path , req->ptr2); break;
2285 case EIO_LINK: req->result = link (path , req->ptr2); break;
2286 case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
2287 case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
2288
2289 case EIO_READLINK: ALLOC (PATH_MAX);
2290 req->result = readlink (path, req->ptr2, PATH_MAX); break;
2291#endif
2292
2293 case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
2294 {
2295 ALLOC (req->result);
2296 memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
2297 }
2298 break;
2299
2300 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS)); /*D*/
2301 req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
2302
1196 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2303 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1197 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; 2304 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1198 2305
1199 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; 2306 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2307 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2308
1200 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; 2309 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1201 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1202 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break; 2310 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1203 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1204 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break; 2311 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1205 2312
1206 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
1207 case EIO_CLOSE: req->result = close (req->int1); break; 2313 case EIO_CLOSE: req->result = close (req->int1); break;
1208 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break; 2314 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1209 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
1210 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1211 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1212 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1213 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1214 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1215 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break;
1216
1217 case EIO_READLINK: ALLOC (NAME_MAX);
1218 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
1219
1220 case EIO_SYNC: req->result = 0; sync (); break; 2315 case EIO_SYNC: req->result = 0; sync (); break;
1221 case EIO_FSYNC: req->result = fsync (req->int1); break; 2316 case EIO_FSYNC: req->result = fsync (req->int1); break;
1222 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 2317 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1223 case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break; 2318 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
1224 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;
1225 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break; 2319 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
2320 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
2321 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
2322 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
2323 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
2324 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
1226 2325
1227 case EIO_READDIR: eio__scandir (req, self); break; 2326 case EIO_READDIR: eio__scandir (req); break;
1228 2327
1229 case EIO_BUSY: 2328 case EIO_BUSY:
1230#ifdef _WIN32 2329#ifdef _WIN32
1231 Sleep (req->nv1 * 1000.); 2330 Sleep (req->nv1 * 1e3);
1232#else 2331#else
1233 { 2332 {
1234 struct timeval tv; 2333 struct timeval tv;
1235 2334
1236 tv.tv_sec = req->nv1; 2335 tv.tv_sec = req->nv1;
1237 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; 2336 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1238 2337
1239 req->result = select (0, 0, 0, 0, &tv); 2338 req->result = select (0, 0, 0, 0, &tv);
1240 } 2339 }
1241#endif 2340#endif
1242 break; 2341 break;
… …
1257 times = tv; 2356 times = tv;
1258 } 2357 }
1259 else 2358 else
1260 times = 0; 2359 times = 0;
1261 2360
1262
1263 req->result = req->type == EIO_FUTIME 2361 req->result = req->type == EIO_FUTIME
1264 ? futimes (req->int1, times) 2362 ? futimes (req->int1, times)
1265 : utimes (req->ptr1, times); 2363 : utimes (req->ptr1, times);
1266 } 2364 }
1267 break; 2365 break;
… …
1272 case EIO_NOP: 2370 case EIO_NOP:
1273 req->result = 0; 2371 req->result = 0;
1274 break; 2372 break;
1275 2373
1276 case EIO_CUSTOM: 2374 case EIO_CUSTOM:
1277 ((void (*)(eio_req *))req->feed) (req); 2375 req->feed (req);
1278 break; 2376 break;
1279 2377
1280 default: 2378 default:
2379 errno = ENOSYS;
1281 req->result = -1; 2380 req->result = -1;
1282 break; 2381 break;
1283 } 2382 }
1284 2383
1285 req->errorno = errno; 2384 req->errorno = errno;
… …
1310eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2409eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1311{ 2410{
1312 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2411 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1313} 2412}
1314 2413
2414eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2415{
2416 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2417}
2418
2419eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
2420{
2421 REQ (EIO_SYNCFS); req->int1 = fd; SEND;
2422}
2423
2424eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2425{
2426 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2427}
2428
1315eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2429eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1316{ 2430{
1317 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2431 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1318} 2432}
1319 2433
1320eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) 2434eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
1321{ 2435{
1322 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; 2436 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
1323} 2437}
1324 2438
1325eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) 2439eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
1326{ 2440{
1327 REQ (EIO_FDATASYNC); req->int1 = fd; SEND; 2441 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2442}
2443
2444eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2445{
2446 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
1328} 2447}
1329 2448
1330eio_req *eio_close (int fd, int pri, eio_cb cb, void *data) 2449eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
1331{ 2450{
1332 REQ (EIO_CLOSE); req->int1 = fd; SEND; 2451 REQ (EIO_CLOSE); req->int1 = fd; SEND;
… …
1350eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) 2469eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1351{ 2470{
1352 REQ (EIO_FSTAT); req->int1 = fd; SEND; 2471 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1353} 2472}
1354 2473
2474eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
2475{
2476 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
2477}
2478
1355eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) 2479eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1356{ 2480{
1357 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; 2481 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1358} 2482}
1359 2483
… …
1365eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) 2489eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1366{ 2490{
1367 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2491 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1368} 2492}
1369 2493
1370eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2494eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1371{ 2495{
1372 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2496 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1373} 2497}
1374 2498
1375eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) 2499eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
… …
1395eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data) 2519eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1396{ 2520{
1397 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2521 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1398} 2522}
1399 2523
1400eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2524eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1401{ 2525{
1402 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2526 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1403} 2527}
1404 2528
1405eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) 2529eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
… …
1421eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data) 2545eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1422{ 2546{
1423 return eio__1path (EIO_READLINK, path, pri, cb, data); 2547 return eio__1path (EIO_READLINK, path, pri, cb, data);
1424} 2548}
1425 2549
2550eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2551{
2552 return eio__1path (EIO_REALPATH, path, pri, cb, data);
2553}
2554
1426eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) 2555eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1427{ 2556{
1428 return eio__1path (EIO_STAT, path, pri, cb, data); 2557 return eio__1path (EIO_STAT, path, pri, cb, data);
1429} 2558}
1430 2559
1431eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) 2560eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1432{ 2561{
1433 return eio__1path (EIO_LSTAT, path, pri, cb, data); 2562 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1434} 2563}
1435 2564
2565eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
2566{
2567 return eio__1path (EIO_STATVFS, path, pri, cb, data);
2568}
2569
1436eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) 2570eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1437{ 2571{
1438 return eio__1path (EIO_UNLINK, path, pri, cb, data); 2572 return eio__1path (EIO_UNLINK, path, pri, cb, data);
1439} 2573}
1440 2574
1441eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data) 2575eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1442{ 2576{
1443 return eio__1path (EIO_RMDIR, path, pri, cb, data); 2577 return eio__1path (EIO_RMDIR, path, pri, cb, data);
1444} 2578}
1445 2579
1446eio_req *eio_readdir (const char *path, int pri, eio_cb cb, void *data) 2580eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1447{ 2581{
1448 return eio__1path (EIO_READDIR, path, pri, cb, data); 2582 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1449} 2583}
1450 2584
1451eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) 2585eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1452{ 2586{
1453 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; 2587 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1454} 2588}
1455 2589
1456static eio_req * 2590static eio_req *
1457eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2591eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1458{ 2592{
… …
1482eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2616eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1483{ 2617{
1484 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2618 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1485} 2619}
1486 2620
1487eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) 2621eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
1488{ 2622{
1489 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND; 2623 REQ (EIO_CUSTOM); req->feed = execute; SEND;
1490} 2624}
1491 2625
1492#endif 2626#endif
1493 2627
1494eio_req *eio_grp (eio_cb cb, void *data) 2628eio_req *eio_grp (eio_cb cb, void *data)
… …
1503#undef SEND 2637#undef SEND
1504 2638
1505/*****************************************************************************/ 2639/*****************************************************************************/
1506/* grp functions */ 2640/* grp functions */
1507 2641
2642void
1508void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2643eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1509{ 2644{
1510 grp->int2 = limit; 2645 grp->int2 = limit;
1511 grp->feed = feed; 2646 grp->feed = feed;
1512 2647
1513 grp_try_feed (grp); 2648 grp_try_feed (grp);
1514} 2649}
1515 2650
2651void
1516void eio_grp_limit (eio_req *grp, int limit) 2652eio_grp_limit (eio_req *grp, int limit)
1517{ 2653{
1518 grp->int2 = limit; 2654 grp->int2 = limit;
1519 2655
1520 grp_try_feed (grp); 2656 grp_try_feed (grp);
1521} 2657}
1522 2658
2659void
1523void eio_grp_add (eio_req *grp, eio_req *req) 2660eio_grp_add (eio_req *grp, eio_req *req)
1524{ 2661{
1525 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2662 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1526 2663
1527 grp->flags |= EIO_FLAG_GROUPADD; 2664 grp->flags |= EIO_FLAG_GROUPADD;
1528 2665
… …
1539} 2676}
1540 2677
1541/*****************************************************************************/ 2678/*****************************************************************************/
1542/* misc garbage */ 2679/* misc garbage */
1543 2680
2681eio_ssize_t
1544ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2682eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1545{ 2683{
1546 etp_worker wrk;
1547
1548 wrk.dbuf = 0;
1549
1550 eio__sendfile (ofd, ifd, offset, count, &wrk); 2684 return eio__sendfile (ofd, ifd, offset, count);
1551
1552 if (wrk.dbuf)
1553 free (wrk.dbuf);
1554} 2685}
1555 2686

Diff Legend

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