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

Comparing libeio/eio.c (file contents):
Revision 1.49 by root, Sat Jan 2 14:24:32 2010 UTC vs.
Revision 1.127 by root, Mon Feb 18 03:15:19 2013 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) 2007,2008,2009,2010,2011,2012 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 *
10 * 1. Redistributions of source code must retain the above copyright notice, 10 * 1. Redistributions of source code must retain the above copyright notice,
11 * this list of conditions and the following disclaimer. 11 * this list of conditions and the following disclaimer.
12 * 12 *
13 * 2. Redistributions in binary form must reproduce the above copyright 13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the 14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution. 15 * documentation and/or other materials provided with the distribution.
16 * 16 *
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR IMPLIED 17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR IMPLIED
18 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MER- 18 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MER-
19 * CHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO 19 * CHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO
20 * EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPE- 20 * EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPE-
21 * CIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, 21 * CIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
… …
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>
… …
53#include <sys/stat.h> 58#include <sys/stat.h>
54#include <limits.h> 59#include <limits.h>
55#include <fcntl.h> 60#include <fcntl.h>
56#include <assert.h> 61#include <assert.h>
57 62
63/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
64/* intptr_t only comes from stdint.h, says idiot openbsd coder */
65#if HAVE_STDINT_H
66# include <stdint.h>
67#endif
68
69#ifndef ECANCELED
70# define ECANCELED EDOM
71#endif
72#ifndef ELOOP
73# define ELOOP EDOM
74#endif
75
76#if !defined(ENOTSOCK) && defined(WSAENOTSOCK)
77# define ENOTSOCK WSAENOTSOCK
78#endif
79
80static void eio_destroy (eio_req *req);
81
58#ifndef EIO_FINISH 82#ifndef EIO_FINISH
59# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0 83# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
60#endif 84#endif
61 85
62#ifndef EIO_DESTROY 86#ifndef EIO_DESTROY
… …
65 89
66#ifndef EIO_FEED 90#ifndef EIO_FEED
67# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0) 91# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)
68#endif 92#endif
69 93
94#ifndef EIO_FD_TO_WIN32_HANDLE
95# define EIO_FD_TO_WIN32_HANDLE(fd) _get_osfhandle (fd)
96#endif
97#ifndef EIO_WIN32_HANDLE_TO_FD
98# define EIO_WIN32_HANDLE_TO_FD(handle) _open_osfhandle (handle, 0)
99#endif
100
101#define EIO_ERRNO(errval,retval) ((errno = errval), retval)
102
103#define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1)
104
70#ifdef _WIN32 105#ifdef _WIN32
71 106
72 /*doh*/ 107 #undef PAGESIZE
108 #define PAGESIZE 4096 /* GetSystemInfo? */
109
110 /* 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 #define pread(fd,buf,count,offset) eio__pread (fd, buf, count, offset)
138 #define pwrite(fd,buf,count,offset) eio__pwrite (fd, buf, count, offset)
139
140 #if __GNUC__
141 typedef long long eio_off_t; /* signed for compatibility to msvc */
142 #else
143 typedef __int64 eio_off_t; /* unsigned not supported by msvc */
144 #endif
145
146 static eio_ssize_t
147 eio__pread (int fd, void *buf, eio_ssize_t count, eio_off_t offset)
148 {
149 OVERLAPPED o = { 0 };
150 DWORD got;
151
152 o.Offset = offset;
153 o.OffsetHigh = offset >> 32;
154
155 return ReadFile ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd), buf, count, &got, &o)
156 ? got : -1;
157 }
158
159 static eio_ssize_t
160 eio__pwrite (int fd, void *buf, eio_ssize_t count, eio_off_t offset)
161 {
162 OVERLAPPED o = { 0 };
163 DWORD got;
164
165 o.Offset = offset;
166 o.OffsetHigh = offset >> 32;
167
168 return WriteFile ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd), buf, count, &got, &o)
169 ? got : -1;
170 }
171
172 /* rename() uses MoveFile, which fails to overwrite */
173 #define rename(old,neu) eio__rename (old, neu)
174
175 static int
176 eio__rename (const char *old, const char *neu)
177 {
178 if (MoveFileEx (old, neu, MOVEFILE_REPLACE_EXISTING))
179 return 0;
180
181 /* should steal _dosmaperr */
182 switch (GetLastError ())
183 {
184 case ERROR_FILE_NOT_FOUND:
185 case ERROR_PATH_NOT_FOUND:
186 case ERROR_INVALID_DRIVE:
187 case ERROR_NO_MORE_FILES:
188 case ERROR_BAD_NETPATH:
189 case ERROR_BAD_NET_NAME:
190 case ERROR_BAD_PATHNAME:
191 case ERROR_FILENAME_EXCED_RANGE:
192 errno = ENOENT;
193 break;
194
195 default:
196 errno = EACCES;
197 break;
198 }
199
200 return -1;
201 }
202
203 /* we could even stat and see if it exists */
204 static int
205 symlink (const char *old, const char *neu)
206 {
207 #if WINVER >= 0x0600
208 if (CreateSymbolicLink (neu, old, 1))
209 return 0;
210
211 if (CreateSymbolicLink (neu, old, 0))
212 return 0;
213 #endif
214
215 return EIO_ERRNO (ENOENT, -1);
216 }
217
218 /* POSIX API only */
219 #define CreateHardLink(neu,old,flags) 0
220 #define CreateSymbolicLink(neu,old,flags) 0
221
222 struct statvfs
223 {
224 int dummy;
225 };
226
227 #define DT_DIR EIO_DT_DIR
228 #define DT_REG EIO_DT_REG
229 #define D_NAME(entp) entp.cFileName
230 #define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
231
73#else 232#else
74 233
75# include "config.h"
76# include <sys/time.h> 234 #include <sys/time.h>
77# include <sys/select.h> 235 #include <sys/select.h>
78# include <sys/mman.h>
79# include <unistd.h> 236 #include <unistd.h>
237 #include <signal.h>
238 #include <dirent.h>
239
240 #if ANDROID
241 #include <sys/vfs.h>
242 #define statvfs statfs
243 #define fstatvfs fstatfs
244 #include <asm/page.h> /* supposedly limits.h does #define PAGESIZE PAGESIZE */
245 #else
246 #include <sys/statvfs.h>
247 #endif
248
249 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
250 #include <sys/mman.h>
251 #endif
252
253 #define D_NAME(entp) entp->d_name
254
255 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
256 #if __FreeBSD__ || __NetBSD__ || __OpenBSD__
257 #define _DIRENT_HAVE_D_TYPE /* sigh */
258 #define D_INO(de) (de)->d_fileno
259 #define D_NAMLEN(de) (de)->d_namlen
260 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
261 #define D_INO(de) (de)->d_ino
262 #endif
263
264 #ifdef _D_EXACT_NAMLEN
265 #undef D_NAMLEN
266 #define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
267 #endif
268
269 #ifdef _DIRENT_HAVE_D_TYPE
270 #define D_TYPE(de) (de)->d_type
271 #endif
272
273 #ifndef EIO_STRUCT_DIRENT
274 #define EIO_STRUCT_DIRENT struct dirent
275 #endif
276
277#endif
278
279#if HAVE_UTIMES
80# include <utime.h> 280# include <utime.h>
81# include <signal.h>
82# include <dirent.h>
83
84/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
85# if __freebsd || defined __NetBSD__ || defined __OpenBSD__
86# define _DIRENT_HAVE_D_TYPE /* sigh */
87# define D_INO(de) (de)->d_fileno
88# define D_NAMLEN(de) (de)->d_namlen
89# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
90# define D_INO(de) (de)->d_ino
91# endif 281#endif
92 282
93#ifdef _D_EXACT_NAMLEN 283#if HAVE_SYS_SYSCALL_H
94# undef D_NAMLEN 284# include <sys/syscall.h>
95# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
96#endif
97
98# ifdef _DIRENT_HAVE_D_TYPE
99# define D_TYPE(de) (de)->d_type
100# endif 285#endif
101 286
102# ifndef EIO_STRUCT_DIRENT 287#if HAVE_SYS_PRCTL_H
103# define EIO_STRUCT_DIRENT struct dirent 288# include <sys/prctl.h>
104# endif
105
106#endif 289#endif
107 290
108#if HAVE_SENDFILE 291#if HAVE_SENDFILE
109# if __linux 292# if __linux
110# include <sys/sendfile.h> 293# include <sys/sendfile.h>
111# elif __freebsd || defined __APPLE__ 294# elif __FreeBSD__ || defined __APPLE__
112# include <sys/socket.h> 295# include <sys/socket.h>
113# include <sys/uio.h> 296# include <sys/uio.h>
114# elif __hpux 297# elif __hpux
115# include <sys/socket.h> 298# include <sys/socket.h>
116# elif __solaris 299# elif __solaris
… …
125#endif 308#endif
126#ifndef D_INO 309#ifndef D_INO
127# define D_INO(de) 0 310# define D_INO(de) 0
128#endif 311#endif
129#ifndef D_NAMLEN 312#ifndef D_NAMLEN
130# define D_NAMLEN(de) strlen ((de)->d_name) 313# define D_NAMLEN(entp) strlen (D_NAME (entp))
131#endif 314#endif
132
133/* number of seconds after which an idle threads exit */
134#define IDLE_TIMEOUT 10
135 315
136/* used for struct dirent, AIX doesn't provide it */ 316/* used for struct dirent, AIX doesn't provide it */
137#ifndef NAME_MAX 317#ifndef NAME_MAX
138# define NAME_MAX 4096 318# define NAME_MAX 4096
139#endif 319#endif
140 320
321/* used for readlink etc. */
322#ifndef PATH_MAX
323# define PATH_MAX 4096
324#endif
325
141/* buffer size for various temporary buffers */ 326/* buffer size for various temporary buffers */
142#define EIO_BUFSIZE 65536 327#define EIO_BUFSIZE 65536
143 328
144#define dBUF \ 329#define dBUF \
145 char *eio_buf; \
146 ETP_WORKER_LOCK (self); \
147 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 330 char *eio_buf = malloc (EIO_BUFSIZE); \
148 ETP_WORKER_UNLOCK (self); \
149 errno = ENOMEM; \ 331 errno = ENOMEM; \
150 if (!eio_buf) \ 332 if (!eio_buf) \
151 return -1; 333 return -1
334
335#define FUBd \
336 free (eio_buf)
152 337
153#define EIO_TICKS ((1000000 + 1023) >> 10) 338#define EIO_TICKS ((1000000 + 1023) >> 10)
154 339
155/*****************************************************************************/ 340/*****************************************************************************/
156 341
157#if __GNUC__ >= 3 342struct tmpbuf
158# define expect(expr,value) __builtin_expect ((expr),(value)) 343{
344 void *ptr;
345 int len;
346};
347
348static void *
349tmpbuf_get (struct tmpbuf *buf, int len)
350{
351 if (buf->len < len)
352 {
353 free (buf->ptr);
354 buf->ptr = malloc (buf->len = len);
355 }
356
357 return buf->ptr;
358}
359
360struct tmpbuf;
361
362#if _POSIX_VERSION >= 200809L
363 #define HAVE_AT 1
364 #define WD2FD(wd) ((wd) ? (wd)->fd : AT_FDCWD)
365 #ifndef O_SEARCH
366 #define O_SEARCH O_RDONLY
367 #endif
159#else 368#else
160# define expect(expr,value) (expr) 369 #define HAVE_AT 0
370 static const char *wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path);
161#endif 371#endif
162 372
163#define expect_false(expr) expect ((expr) != 0, 0) 373struct eio_pwd
164#define expect_true(expr) expect ((expr) != 0, 1) 374{
375#if HAVE_AT
376 int fd;
377#endif
378 int len;
379 char str[1]; /* actually, a 0-terminated canonical path */
380};
165 381
166/*****************************************************************************/ 382/*****************************************************************************/
167 383
168#define ETP_PRI_MIN EIO_PRI_MIN 384#define ETP_PRI_MIN EIO_PRI_MIN
169#define ETP_PRI_MAX EIO_PRI_MAX 385#define ETP_PRI_MAX EIO_PRI_MAX
… …
175static int eio_finish (eio_req *req); 391static int eio_finish (eio_req *req);
176#define ETP_FINISH(req) eio_finish (req) 392#define ETP_FINISH(req) eio_finish (req)
177static void eio_execute (struct etp_worker *self, eio_req *req); 393static void eio_execute (struct etp_worker *self, eio_req *req);
178#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) 394#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
179 395
180#define ETP_WORKER_CLEAR(req) \
181 if (wrk->dbuf) \
182 { \
183 free (wrk->dbuf); \
184 wrk->dbuf = 0; \
185 } \
186 \
187 if (wrk->dirp) \
188 { \
189 closedir (wrk->dirp); \
190 wrk->dirp = 0; \
191 }
192
193#define ETP_WORKER_COMMON \
194 void *dbuf; \
195 DIR *dirp;
196
197/*****************************************************************************/ 396/*****************************************************************************/
198 397
199#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) 398#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
200 399
201/* calculate time difference in ~1/EIO_TICKS of a second */ 400/* calculate time difference in ~1/EIO_TICKS of a second */
401ecb_inline int
202static int tvdiff (struct timeval *tv1, struct timeval *tv2) 402tvdiff (struct timeval *tv1, struct timeval *tv2)
203{ 403{
204 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 404 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
205 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 405 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
206} 406}
207 407
… …
211static void (*done_poll_cb) (void); 411static void (*done_poll_cb) (void);
212 412
213static unsigned int max_poll_time; /* reslock */ 413static unsigned int max_poll_time; /* reslock */
214static unsigned int max_poll_reqs; /* reslock */ 414static unsigned int max_poll_reqs; /* reslock */
215 415
216static volatile unsigned int nreqs; /* reqlock */ 416static unsigned int nreqs; /* reqlock */
217static volatile unsigned int nready; /* reqlock */ 417static unsigned int nready; /* reqlock */
218static volatile unsigned int npending; /* reqlock */ 418static unsigned int npending; /* reqlock */
219static volatile unsigned int max_idle = 4; 419static unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
420static unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
220 421
221static mutex_t wrklock = X_MUTEX_INIT; 422static xmutex_t wrklock;
222static mutex_t reslock = X_MUTEX_INIT; 423static xmutex_t reslock;
223static mutex_t reqlock = X_MUTEX_INIT; 424static xmutex_t reqlock;
224static cond_t reqwait = X_COND_INIT; 425static xcond_t reqwait;
225
226#if !HAVE_PREADWRITE
227/*
228 * make our pread/pwrite emulation safe against themselves, but not against
229 * normal read/write by using a mutex. slows down execution a lot,
230 * but that's your problem, not mine.
231 */
232static mutex_t preadwritelock = X_MUTEX_INIT;
233#endif
234 426
235typedef struct etp_worker 427typedef struct etp_worker
236{ 428{
429 struct tmpbuf tmpbuf;
430
237 /* locked by wrklock */ 431 /* locked by wrklock */
238 struct etp_worker *prev, *next; 432 struct etp_worker *prev, *next;
239 433
240 thread_t tid; 434 xthread_t tid;
241 435
242 /* locked by reslock, reqlock or wrklock */ 436#ifdef ETP_WORKER_COMMON
243 ETP_REQ *req; /* currently processed request */
244
245 ETP_WORKER_COMMON 437 ETP_WORKER_COMMON
438#endif
246} etp_worker; 439} etp_worker;
247 440
248static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */ 441static etp_worker wrk_first; /* NOT etp */
249 442
250#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock) 443#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
251#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock) 444#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
252 445
253/* worker threads management */ 446/* worker threads management */
254 447
448static void
255static void etp_worker_clear (etp_worker *wrk) 449etp_worker_clear (etp_worker *wrk)
256{ 450{
257 ETP_WORKER_CLEAR (wrk);
258} 451}
259 452
453static void ecb_cold
260static void etp_worker_free (etp_worker *wrk) 454etp_worker_free (etp_worker *wrk)
261{ 455{
456 free (wrk->tmpbuf.ptr);
457
262 wrk->next->prev = wrk->prev; 458 wrk->next->prev = wrk->prev;
263 wrk->prev->next = wrk->next; 459 wrk->prev->next = wrk->next;
264 460
265 free (wrk); 461 free (wrk);
266} 462}
267 463
268static unsigned int etp_nreqs (void) 464static unsigned int
465etp_nreqs (void)
269{ 466{
270 int retval; 467 int retval;
271 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 468 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
272 retval = nreqs; 469 retval = nreqs;
273 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 470 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
274 return retval; 471 return retval;
275} 472}
276 473
277static unsigned int etp_nready (void) 474static unsigned int
475etp_nready (void)
278{ 476{
279 unsigned int retval; 477 unsigned int retval;
280 478
281 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 479 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
282 retval = nready; 480 retval = nready;
283 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 481 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
284 482
285 return retval; 483 return retval;
286} 484}
287 485
288static unsigned int etp_npending (void) 486static unsigned int
487etp_npending (void)
289{ 488{
290 unsigned int retval; 489 unsigned int retval;
291 490
292 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 491 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
293 retval = npending; 492 retval = npending;
294 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 493 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
295 494
296 return retval; 495 return retval;
297} 496}
298 497
299static unsigned int etp_nthreads (void) 498static unsigned int
499etp_nthreads (void)
300{ 500{
301 unsigned int retval; 501 unsigned int retval;
302 502
303 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 503 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
304 retval = started; 504 retval = started;
… …
318} etp_reqq; 518} etp_reqq;
319 519
320static etp_reqq req_queue; 520static etp_reqq req_queue;
321static etp_reqq res_queue; 521static etp_reqq res_queue;
322 522
523static void ecb_noinline ecb_cold
524reqq_init (etp_reqq *q)
525{
526 int pri;
527
528 for (pri = 0; pri < ETP_NUM_PRI; ++pri)
529 q->qs[pri] = q->qe[pri] = 0;
530
531 q->size = 0;
532}
533
534static int ecb_noinline
323static int reqq_push (etp_reqq *q, ETP_REQ *req) 535reqq_push (etp_reqq *q, ETP_REQ *req)
324{ 536{
325 int pri = req->pri; 537 int pri = req->pri;
326 req->next = 0; 538 req->next = 0;
327 539
328 if (q->qe[pri]) 540 if (q->qe[pri])
… …
334 q->qe[pri] = q->qs[pri] = req; 546 q->qe[pri] = q->qs[pri] = req;
335 547
336 return q->size++; 548 return q->size++;
337} 549}
338 550
551static ETP_REQ * ecb_noinline
339static ETP_REQ *reqq_shift (etp_reqq *q) 552reqq_shift (etp_reqq *q)
340{ 553{
341 int pri; 554 int pri;
342 555
343 if (!q->size) 556 if (!q->size)
344 return 0; 557 return 0;
… …
359 } 572 }
360 573
361 abort (); 574 abort ();
362} 575}
363 576
364static void etp_atfork_prepare (void) 577static int ecb_cold
578etp_init (void (*want_poll)(void), void (*done_poll)(void))
365{ 579{
366 X_LOCK (wrklock); 580 X_MUTEX_CREATE (wrklock);
367 X_LOCK (reqlock); 581 X_MUTEX_CREATE (reslock);
368 X_LOCK (reslock); 582 X_MUTEX_CREATE (reqlock);
369#if !HAVE_PREADWRITE 583 X_COND_CREATE (reqwait);
370 X_LOCK (preadwritelock);
371#endif
372}
373 584
374static void etp_atfork_parent (void) 585 reqq_init (&req_queue);
375{ 586 reqq_init (&res_queue);
376#if !HAVE_PREADWRITE
377 X_UNLOCK (preadwritelock);
378#endif
379 X_UNLOCK (reslock);
380 X_UNLOCK (reqlock);
381 X_UNLOCK (wrklock);
382}
383 587
384static void etp_atfork_child (void) 588 wrk_first.next =
385{ 589 wrk_first.prev = &wrk_first;
386 ETP_REQ *prv;
387
388 while ((prv = reqq_shift (&req_queue)))
389 ETP_DESTROY (prv);
390
391 while ((prv = reqq_shift (&res_queue)))
392 ETP_DESTROY (prv);
393
394 while (wrk_first.next != &wrk_first)
395 {
396 etp_worker *wrk = wrk_first.next;
397
398 if (wrk->req)
399 ETP_DESTROY (wrk->req);
400
401 etp_worker_clear (wrk);
402 etp_worker_free (wrk);
403 }
404 590
405 started = 0; 591 started = 0;
406 idle = 0; 592 idle = 0;
407 nreqs = 0; 593 nreqs = 0;
408 nready = 0; 594 nready = 0;
409 npending = 0; 595 npending = 0;
410 596
411 etp_atfork_parent ();
412}
413
414static void
415etp_once_init (void)
416{
417 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
418}
419
420static int
421etp_init (void (*want_poll)(void), void (*done_poll)(void))
422{
423 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
424
425 pthread_once (&doinit, etp_once_init);
426
427 want_poll_cb = want_poll; 597 want_poll_cb = want_poll;
428 done_poll_cb = done_poll; 598 done_poll_cb = done_poll;
429 599
430 return 0; 600 return 0;
431} 601}
432 602
433X_THREAD_PROC (etp_proc); 603X_THREAD_PROC (etp_proc);
434 604
605static void ecb_cold
435static void etp_start_thread (void) 606etp_start_thread (void)
436{ 607{
437 etp_worker *wrk = calloc (1, sizeof (etp_worker)); 608 etp_worker *wrk = calloc (1, sizeof (etp_worker));
438 609
439 /*TODO*/ 610 /*TODO*/
440 assert (("unable to allocate worker thread data", wrk)); 611 assert (("unable to allocate worker thread data", wrk));
441 612
442 X_LOCK (wrklock); 613 X_LOCK (wrklock);
443 614
444 if (thread_create (&wrk->tid, etp_proc, (void *)wrk)) 615 if (xthread_create (&wrk->tid, etp_proc, (void *)wrk))
445 { 616 {
446 wrk->prev = &wrk_first; 617 wrk->prev = &wrk_first;
447 wrk->next = wrk_first.next; 618 wrk->next = wrk_first.next;
448 wrk_first.next->prev = wrk; 619 wrk_first.next->prev = wrk;
449 wrk_first.next = wrk; 620 wrk_first.next = wrk;
… …
453 free (wrk); 624 free (wrk);
454 625
455 X_UNLOCK (wrklock); 626 X_UNLOCK (wrklock);
456} 627}
457 628
629static void
458static void etp_maybe_start_thread (void) 630etp_maybe_start_thread (void)
459{ 631{
460 if (expect_true (etp_nthreads () >= wanted)) 632 if (ecb_expect_true (etp_nthreads () >= wanted))
461 return; 633 return;
462 634
463 /* todo: maybe use idle here, but might be less exact */ 635 /* todo: maybe use idle here, but might be less exact */
464 if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ())) 636 if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
465 return; 637 return;
466 638
467 etp_start_thread (); 639 etp_start_thread ();
468} 640}
469 641
642static void ecb_cold
470static void etp_end_thread (void) 643etp_end_thread (void)
471{ 644{
472 eio_req *req = calloc (1, sizeof (eio_req)); 645 eio_req *req = calloc (1, sizeof (eio_req)); /* will be freed by worker */
473 646
474 req->type = -1; 647 req->type = -1;
475 req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 648 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
476 649
477 X_LOCK (reqlock); 650 X_LOCK (reqlock);
… …
482 X_LOCK (wrklock); 655 X_LOCK (wrklock);
483 --started; 656 --started;
484 X_UNLOCK (wrklock); 657 X_UNLOCK (wrklock);
485} 658}
486 659
487static int etp_poll (void) 660static int
661etp_poll (void)
488{ 662{
489 unsigned int maxreqs; 663 unsigned int maxreqs;
490 unsigned int maxtime; 664 unsigned int maxtime;
491 struct timeval tv_start, tv_now; 665 struct timeval tv_start, tv_now;
492 666
… …
522 696
523 X_LOCK (reqlock); 697 X_LOCK (reqlock);
524 --nreqs; 698 --nreqs;
525 X_UNLOCK (reqlock); 699 X_UNLOCK (reqlock);
526 700
527 if (expect_false (req->type == EIO_GROUP && req->size)) 701 if (ecb_expect_false (req->type == EIO_GROUP && req->size))
528 { 702 {
529 req->int1 = 1; /* mark request as delayed */ 703 req->int1 = 1; /* mark request as delayed */
530 continue; 704 continue;
531 } 705 }
532 else 706 else
533 { 707 {
534 int res = ETP_FINISH (req); 708 int res = ETP_FINISH (req);
535 if (expect_false (res)) 709 if (ecb_expect_false (res))
536 return res; 710 return res;
537 } 711 }
538 712
539 if (expect_false (maxreqs && !--maxreqs)) 713 if (ecb_expect_false (maxreqs && !--maxreqs))
540 break; 714 break;
541 715
542 if (maxtime) 716 if (maxtime)
543 { 717 {
544 gettimeofday (&tv_now, 0); 718 gettimeofday (&tv_now, 0);
… …
550 724
551 errno = EAGAIN; 725 errno = EAGAIN;
552 return -1; 726 return -1;
553} 727}
554 728
729static void
555static void etp_cancel (ETP_REQ *req) 730etp_cancel (ETP_REQ *req)
556{ 731{
557 X_LOCK (wrklock); 732 req->cancelled = 1;
558 req->flags |= EIO_FLAG_CANCELLED;
559 X_UNLOCK (wrklock);
560 733
561 eio_grp_cancel (req); 734 eio_grp_cancel (req);
562} 735}
563 736
737static void
564static void etp_submit (ETP_REQ *req) 738etp_submit (ETP_REQ *req)
565{ 739{
566 req->pri -= ETP_PRI_MIN; 740 req->pri -= ETP_PRI_MIN;
567 741
568 if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN; 742 if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
569 if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 743 if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
570 744
571 if (expect_false (req->type == EIO_GROUP)) 745 if (ecb_expect_false (req->type == EIO_GROUP))
572 { 746 {
573 /* I hope this is worth it :/ */ 747 /* I hope this is worth it :/ */
574 X_LOCK (reqlock); 748 X_LOCK (reqlock);
575 ++nreqs; 749 ++nreqs;
576 X_UNLOCK (reqlock); 750 X_UNLOCK (reqlock);
… …
595 769
596 etp_maybe_start_thread (); 770 etp_maybe_start_thread ();
597 } 771 }
598} 772}
599 773
774static void ecb_cold
600static void etp_set_max_poll_time (double nseconds) 775etp_set_max_poll_time (double nseconds)
601{ 776{
602 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 777 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
603 max_poll_time = nseconds * EIO_TICKS; 778 max_poll_time = nseconds * EIO_TICKS;
604 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 779 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
605} 780}
606 781
782static void ecb_cold
607static void etp_set_max_poll_reqs (unsigned int maxreqs) 783etp_set_max_poll_reqs (unsigned int maxreqs)
608{ 784{
609 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 785 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
610 max_poll_reqs = maxreqs; 786 max_poll_reqs = maxreqs;
611 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 787 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
612} 788}
613 789
790static void ecb_cold
614static void etp_set_max_idle (unsigned int nthreads) 791etp_set_max_idle (unsigned int nthreads)
615{ 792{
616 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 793 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
617 max_idle = nthreads <= 0 ? 1 : nthreads; 794 max_idle = nthreads;
618 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 795 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
619} 796}
620 797
798static void ecb_cold
799etp_set_idle_timeout (unsigned int seconds)
800{
801 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
802 idle_timeout = seconds;
803 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
804}
805
806static void ecb_cold
621static void etp_set_min_parallel (unsigned int nthreads) 807etp_set_min_parallel (unsigned int nthreads)
622{ 808{
623 if (wanted < nthreads) 809 if (wanted < nthreads)
624 wanted = nthreads; 810 wanted = nthreads;
625} 811}
626 812
813static void ecb_cold
627static void etp_set_max_parallel (unsigned int nthreads) 814etp_set_max_parallel (unsigned int nthreads)
628{ 815{
629 if (wanted > nthreads) 816 if (wanted > nthreads)
630 wanted = nthreads; 817 wanted = nthreads;
631 818
632 while (started > wanted) 819 while (started > wanted)
633 etp_end_thread (); 820 etp_end_thread ();
634} 821}
635 822
636/*****************************************************************************/ 823/*****************************************************************************/
637 824
825static void
638static void grp_try_feed (eio_req *grp) 826grp_try_feed (eio_req *grp)
639{ 827{
640 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 828 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
641 { 829 {
642 grp->flags &= ~EIO_FLAG_GROUPADD; 830 grp->flags &= ~EIO_FLAG_GROUPADD;
643 831
… …
650 break; 838 break;
651 } 839 }
652 } 840 }
653} 841}
654 842
843static int
655static int grp_dec (eio_req *grp) 844grp_dec (eio_req *grp)
656{ 845{
657 --grp->size; 846 --grp->size;
658 847
659 /* call feeder, if applicable */ 848 /* call feeder, if applicable */
660 grp_try_feed (grp); 849 grp_try_feed (grp);
… …
664 return eio_finish (grp); 853 return eio_finish (grp);
665 else 854 else
666 return 0; 855 return 0;
667} 856}
668 857
858static void
669void eio_destroy (eio_req *req) 859eio_destroy (eio_req *req)
670{ 860{
671 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); 861 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
672 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); 862 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
673 863
674 EIO_DESTROY (req); 864 EIO_DESTROY (req);
675} 865}
676 866
867static int
677static int eio_finish (eio_req *req) 868eio_finish (eio_req *req)
678{ 869{
679 int res = EIO_FINISH (req); 870 int res = EIO_FINISH (req);
680 871
681 if (req->grp) 872 if (req->grp)
682 { 873 {
… …
690 if (grp->grp_first == req) 881 if (grp->grp_first == req)
691 grp->grp_first = req->grp_next; 882 grp->grp_first = req->grp_next;
692 883
693 res2 = grp_dec (grp); 884 res2 = grp_dec (grp);
694 885
695 if (!res && res2) 886 if (!res)
696 res = res2; 887 res = res2;
697 } 888 }
698 889
699 eio_destroy (req); 890 eio_destroy (req);
700 891
701 return res; 892 return res;
702} 893}
703 894
895void
704void eio_grp_cancel (eio_req *grp) 896eio_grp_cancel (eio_req *grp)
705{ 897{
706 for (grp = grp->grp_first; grp; grp = grp->grp_next) 898 for (grp = grp->grp_first; grp; grp = grp->grp_next)
707 eio_cancel (grp); 899 eio_cancel (grp);
708} 900}
709 901
902void
710void eio_cancel (eio_req *req) 903eio_cancel (eio_req *req)
711{ 904{
712 etp_cancel (req); 905 etp_cancel (req);
713} 906}
714 907
908void
715void eio_submit (eio_req *req) 909eio_submit (eio_req *req)
716{ 910{
717 etp_submit (req); 911 etp_submit (req);
718} 912}
719 913
720unsigned int eio_nreqs (void) 914unsigned int
915eio_nreqs (void)
721{ 916{
722 return etp_nreqs (); 917 return etp_nreqs ();
723} 918}
724 919
725unsigned int eio_nready (void) 920unsigned int
921eio_nready (void)
726{ 922{
727 return etp_nready (); 923 return etp_nready ();
728} 924}
729 925
730unsigned int eio_npending (void) 926unsigned int
927eio_npending (void)
731{ 928{
732 return etp_npending (); 929 return etp_npending ();
733} 930}
734 931
735unsigned int eio_nthreads (void) 932unsigned int ecb_cold
933eio_nthreads (void)
736{ 934{
737 return etp_nthreads (); 935 return etp_nthreads ();
738} 936}
739 937
938void ecb_cold
740void eio_set_max_poll_time (double nseconds) 939eio_set_max_poll_time (double nseconds)
741{ 940{
742 etp_set_max_poll_time (nseconds); 941 etp_set_max_poll_time (nseconds);
743} 942}
744 943
944void ecb_cold
745void eio_set_max_poll_reqs (unsigned int maxreqs) 945eio_set_max_poll_reqs (unsigned int maxreqs)
746{ 946{
747 etp_set_max_poll_reqs (maxreqs); 947 etp_set_max_poll_reqs (maxreqs);
748} 948}
749 949
950void ecb_cold
750void eio_set_max_idle (unsigned int nthreads) 951eio_set_max_idle (unsigned int nthreads)
751{ 952{
752 etp_set_max_idle (nthreads); 953 etp_set_max_idle (nthreads);
753} 954}
754 955
956void ecb_cold
957eio_set_idle_timeout (unsigned int seconds)
958{
959 etp_set_idle_timeout (seconds);
960}
961
962void ecb_cold
755void eio_set_min_parallel (unsigned int nthreads) 963eio_set_min_parallel (unsigned int nthreads)
756{ 964{
757 etp_set_min_parallel (nthreads); 965 etp_set_min_parallel (nthreads);
758} 966}
759 967
968void ecb_cold
760void eio_set_max_parallel (unsigned int nthreads) 969eio_set_max_parallel (unsigned int nthreads)
761{ 970{
762 etp_set_max_parallel (nthreads); 971 etp_set_max_parallel (nthreads);
763} 972}
764 973
765int eio_poll (void) 974int eio_poll (void)
… …
768} 977}
769 978
770/*****************************************************************************/ 979/*****************************************************************************/
771/* work around various missing functions */ 980/* work around various missing functions */
772 981
773#if !HAVE_PREADWRITE
774# undef pread
775# undef pwrite
776# define pread eio__pread
777# define pwrite eio__pwrite
778
779static ssize_t
780eio__pread (int fd, void *buf, size_t count, off_t offset)
781{
782 ssize_t res;
783 off_t ooffset;
784
785 X_LOCK (preadwritelock);
786 ooffset = lseek (fd, 0, SEEK_CUR);
787 lseek (fd, offset, SEEK_SET);
788 res = read (fd, buf, count);
789 lseek (fd, ooffset, SEEK_SET);
790 X_UNLOCK (preadwritelock);
791
792 return res;
793}
794
795static ssize_t
796eio__pwrite (int fd, void *buf, size_t count, off_t offset)
797{
798 ssize_t res;
799 off_t ooffset;
800
801 X_LOCK (preadwritelock);
802 ooffset = lseek (fd, 0, SEEK_CUR);
803 lseek (fd, offset, SEEK_SET);
804 res = write (fd, buf, count);
805 lseek (fd, ooffset, SEEK_SET);
806 X_UNLOCK (preadwritelock);
807
808 return res;
809}
810#endif
811
812#ifndef HAVE_FUTIMES 982#ifndef HAVE_UTIMES
813 983
814# undef utimes 984# undef utimes
815# undef futimes
816# define utimes(path,times) eio__utimes (path, times) 985# define utimes(path,times) eio__utimes (path, times)
817# define futimes(fd,times) eio__futimes (fd, times)
818 986
819static int 987static int
820eio__utimes (const char *filename, const struct timeval times[2]) 988eio__utimes (const char *filename, const struct timeval times[2])
821{ 989{
822 if (times) 990 if (times)
… …
830 } 998 }
831 else 999 else
832 return utime (filename, 0); 1000 return utime (filename, 0);
833} 1001}
834 1002
1003#endif
1004
1005#ifndef HAVE_FUTIMES
1006
1007# undef futimes
1008# define futimes(fd,times) eio__futimes (fd, times)
1009
1010static int
835static int eio__futimes (int fd, const struct timeval tv[2]) 1011eio__futimes (int fd, const struct timeval tv[2])
836{ 1012{
837 errno = ENOSYS; 1013 errno = ENOSYS;
838 return -1; 1014 return -1;
839} 1015}
840 1016
… …
843#if !HAVE_FDATASYNC 1019#if !HAVE_FDATASYNC
844# undef fdatasync 1020# undef fdatasync
845# define fdatasync(fd) fsync (fd) 1021# define fdatasync(fd) fsync (fd)
846#endif 1022#endif
847 1023
1024static int
1025eio__syncfs (int fd)
1026{
1027 int res;
1028
1029#if HAVE_SYS_SYNCFS
1030 res = (int)syscall (__NR_syncfs, (int)(fd));
1031#else
1032 res = EIO_ENOSYS ();
1033#endif
1034
1035 if (res < 0 && errno == ENOSYS && fd >= 0)
1036 sync ();
1037
1038 return res;
1039}
1040
848/* sync_file_range always needs emulation */ 1041/* sync_file_range always needs emulation */
849int 1042static int
850eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags) 1043eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
851{ 1044{
852#if HAVE_SYNC_FILE_RANGE 1045#if HAVE_SYNC_FILE_RANGE
853 int res; 1046 int res;
854 1047
… …
871 /* even though we could play tricks with the flags, it's better to always 1064 /* even though we could play tricks with the flags, it's better to always
872 * call fdatasync, as that matches the expectation of its users best */ 1065 * call fdatasync, as that matches the expectation of its users best */
873 return fdatasync (fd); 1066 return fdatasync (fd);
874} 1067}
875 1068
1069static int
1070eio__fallocate (int fd, int mode, off_t offset, size_t len)
1071{
1072#if HAVE_LINUX_FALLOCATE
1073 return fallocate (fd, mode, offset, len);
1074#else
1075 return EIO_ENOSYS ();
1076#endif
1077}
1078
876#if !HAVE_READAHEAD 1079#if !HAVE_READAHEAD
877# undef readahead 1080# undef readahead
878# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 1081# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
879 1082
880static ssize_t 1083static eio_ssize_t
881eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) 1084eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
882{ 1085{
883 size_t todo = count; 1086 size_t todo = count;
884 dBUF; 1087 dBUF;
885 1088
… …
890 pread (fd, eio_buf, len, offset); 1093 pread (fd, eio_buf, len, offset);
891 offset += len; 1094 offset += len;
892 todo -= len; 1095 todo -= len;
893 } 1096 }
894 1097
895 errno = 0; 1098 FUBd;
1099
1100 /* linux's readahead basically only fails for EBADF or EINVAL (not mmappable) */
1101 /* but not for e.g. EIO or eof, so we also never fail */
896 return count; 1102 return 0;
897} 1103}
898 1104
899#endif 1105#endif
900 1106
901/* sendfile always needs emulation */ 1107/* sendfile always needs emulation */
902static ssize_t 1108static eio_ssize_t
903eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) 1109eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
904{ 1110{
1111 eio_ssize_t written = 0;
905 ssize_t res; 1112 eio_ssize_t res;
906 1113
907 if (!count) 1114 if (!count)
908 return 0; 1115 return 0;
909 1116
1117 for (;;)
1118 {
1119#ifdef __APPLE__
1120# undef HAVE_SENDFILE /* broken, as everything on os x */
1121#endif
910#if HAVE_SENDFILE 1122#if HAVE_SENDFILE
911# if __linux 1123# if __linux
1124 off_t soffset = offset;
912 res = sendfile (ofd, ifd, &offset, count); 1125 res = sendfile (ofd, ifd, &soffset, count);
913 1126
914# elif __freebsd 1127# elif __FreeBSD__
915 /* 1128 /*
916 * Of course, the freebsd sendfile is a dire hack with no thoughts 1129 * Of course, the freebsd sendfile is a dire hack with no thoughts
917 * wasted on making it similar to other I/O functions. 1130 * wasted on making it similar to other I/O functions.
918 */ 1131 */
919 {
920 off_t sbytes; 1132 off_t sbytes;
921 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 1133 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
922 1134
923 if (res < 0 && sbytes) 1135 #if 0 /* according to the manpage, this is correct, but broken behaviour */
924 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 1136 /* freebsd' sendfile will return 0 on success */
1137 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
1138 /* not on e.g. EIO or EPIPE - sounds broken */
1139 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
925 res = sbytes; 1140 res = sbytes;
926 } 1141 #endif
927 1142
1143 /* according to source inspection, this is correct, and useful behaviour */
1144 if (sbytes)
1145 res = sbytes;
1146
928# elif defined (__APPLE__) 1147# elif defined __APPLE__
929
930 {
931 off_t sbytes = count; 1148 off_t sbytes = count;
932 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0); 1149 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
933 1150
934 if (res < 0 && errno == EAGAIN && sbytes) 1151 /* according to the manpage, sbytes is always valid */
1152 if (sbytes)
935 res = sbytes; 1153 res = sbytes;
936 }
937 1154
938# elif __hpux 1155# elif __hpux
939 res = sendfile (ofd, ifd, offset, count, 0, 0); 1156 res = sendfile (ofd, ifd, offset, count, 0, 0);
940 1157
941# elif __solaris 1158# elif __solaris
942 {
943 struct sendfilevec vec; 1159 struct sendfilevec vec;
944 size_t sbytes; 1160 size_t sbytes;
945 1161
946 vec.sfv_fd = ifd; 1162 vec.sfv_fd = ifd;
947 vec.sfv_flag = 0; 1163 vec.sfv_flag = 0;
948 vec.sfv_off = offset; 1164 vec.sfv_off = offset;
949 vec.sfv_len = count; 1165 vec.sfv_len = count;
950 1166
951 res = sendfilev (ofd, &vec, 1, &sbytes); 1167 res = sendfilev (ofd, &vec, 1, &sbytes);
952 1168
953 if (res < 0 && sbytes) 1169 if (res < 0 && sbytes)
954 res = sbytes; 1170 res = sbytes;
955 }
956 1171
957# endif 1172# endif
958 1173
959#elif defined (_WIN32) 1174#elif defined (_WIN32) && 0
960
961 /* does not work, just for documentation of what would need to be done */ 1175 /* does not work, just for documentation of what would need to be done */
962 { 1176 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
1177 /* libeio guarantees that the file offset does not change, and windows */
1178 /* has no way to get an independent handle to the same file description */
963 HANDLE h = TO_SOCKET (ifd); 1179 HANDLE h = TO_SOCKET (ifd);
964 SetFilePointer (h, offset, 0, FILE_BEGIN); 1180 SetFilePointer (h, offset, 0, FILE_BEGIN);
965 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0); 1181 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
966 }
967 1182
968#else 1183#else
969 res = -1; 1184 res = EIO_ENOSYS ();
970 errno = ENOSYS;
971#endif 1185#endif
972 1186
1187 /* we assume sendfile can copy at least 128mb in one go */
1188 if (res <= 128 * 1024 * 1024)
1189 {
1190 if (res > 0)
1191 written += res;
1192
1193 if (written)
1194 return written;
1195
1196 break;
1197 }
1198 else
1199 {
1200 /* if we requested more, then probably the kernel was lazy */
1201 written += res;
1202 offset += res;
1203 count -= res;
1204
1205 if (!count)
1206 return written;
1207 }
1208 }
1209
973 if (res < 0 1210 if (res < 0
974 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1211 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
975 /* BSDs */ 1212 /* BSDs */
976#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */ 1213#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
977 || errno == ENOTSUP 1214 || errno == ENOTSUP
978#endif 1215#endif
1216#ifdef EOPNOTSUPP /* windows */
979 || errno == EOPNOTSUPP /* BSDs */ 1217 || errno == EOPNOTSUPP /* BSDs */
1218#endif
980#if __solaris 1219#if __solaris
981 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1220 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
982#endif 1221#endif
983 ) 1222 )
984 ) 1223 )
… …
988 1227
989 res = 0; 1228 res = 0;
990 1229
991 while (count) 1230 while (count)
992 { 1231 {
993 ssize_t cnt; 1232 eio_ssize_t cnt;
994 1233
995 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 1234 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
996 1235
997 if (cnt <= 0) 1236 if (cnt <= 0)
998 { 1237 {
… …
1010 1249
1011 offset += cnt; 1250 offset += cnt;
1012 res += cnt; 1251 res += cnt;
1013 count -= cnt; 1252 count -= cnt;
1014 } 1253 }
1254
1255 FUBd;
1015 } 1256 }
1016 1257
1017 return res; 1258 return res;
1259}
1260
1261#ifdef PAGESIZE
1262# define eio_pagesize() PAGESIZE
1263#else
1264static intptr_t
1265eio_pagesize (void)
1266{
1267 static intptr_t page;
1268
1269 if (!page)
1270 page = sysconf (_SC_PAGESIZE);
1271
1272 return page;
1273}
1274#endif
1275
1276static void
1277eio_page_align (void **addr, size_t *length)
1278{
1279 intptr_t mask = eio_pagesize () - 1;
1280
1281 /* round down addr */
1282 intptr_t adj = mask & (intptr_t)*addr;
1283
1284 *addr = (void *)((intptr_t)*addr - adj);
1285 *length += adj;
1286
1287 /* round up length */
1288 *length = (*length + mask) & ~mask;
1289}
1290
1291#if !_POSIX_MEMLOCK
1292# define eio__mlockall(a) EIO_ENOSYS ()
1293#else
1294
1295static int
1296eio__mlockall (int flags)
1297{
1298 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1299 extern int mallopt (int, int);
1300 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1301 #endif
1302
1303 if (EIO_MCL_CURRENT != MCL_CURRENT
1304 || EIO_MCL_FUTURE != MCL_FUTURE)
1305 {
1306 flags = 0
1307 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1308 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1309 }
1310
1311 return mlockall (flags);
1312}
1313#endif
1314
1315#if !_POSIX_MEMLOCK_RANGE
1316# define eio__mlock(a,b) EIO_ENOSYS ()
1317#else
1318
1319static int
1320eio__mlock (void *addr, size_t length)
1321{
1322 eio_page_align (&addr, &length);
1323
1324 return mlock (addr, length);
1325}
1326
1327#endif
1328
1329#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1330# define eio__msync(a,b,c) EIO_ENOSYS ()
1331#else
1332
1333static int
1334eio__msync (void *mem, size_t len, int flags)
1335{
1336 eio_page_align (&mem, &len);
1337
1338 if (EIO_MS_ASYNC != MS_SYNC
1339 || EIO_MS_INVALIDATE != MS_INVALIDATE
1340 || EIO_MS_SYNC != MS_SYNC)
1341 {
1342 flags = 0
1343 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1344 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1345 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1346 }
1347
1348 return msync (mem, len, flags);
1349}
1350
1351#endif
1352
1353static int
1354eio__mtouch (eio_req *req)
1355{
1356 void *mem = req->ptr2;
1357 size_t len = req->size;
1358 int flags = req->int1;
1359
1360 eio_page_align (&mem, &len);
1361
1362 {
1363 intptr_t addr = (intptr_t)mem;
1364 intptr_t end = addr + len;
1365 intptr_t page = eio_pagesize ();
1366
1367 if (addr < end)
1368 if (flags & EIO_MT_MODIFY) /* modify */
1369 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1370 else
1371 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1372 }
1373
1374 return 0;
1375}
1376
1377/*****************************************************************************/
1378/* requests implemented outside eio_execute, because they are so large */
1379
1380static void
1381eio__lseek (eio_req *req)
1382{
1383 /* this usually gets optimised away completely, or your compiler sucks, */
1384 /* or the whence constants really are not 0, 1, 2 */
1385 int whence = req->int2 == EIO_SEEK_SET ? SEEK_SET
1386 : req->int2 == EIO_SEEK_CUR ? SEEK_CUR
1387 : req->int2 == EIO_SEEK_END ? SEEK_END
1388 : req->int2;
1389
1390 req->offs = lseek (req->int1, req->offs, whence);
1391 req->result = req->offs == (off_t)-1 ? -1 : 0;
1392}
1393
1394/* result will always end up in tmpbuf, there is always space for adding a 0-byte */
1395static int
1396eio__realpath (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
1397{
1398 const char *rel = path;
1399 char *res;
1400 char *tmp1, *tmp2;
1401#if SYMLOOP_MAX > 32
1402 int symlinks = SYMLOOP_MAX;
1403#else
1404 int symlinks = 32;
1405#endif
1406
1407 errno = EINVAL;
1408 if (!rel)
1409 return -1;
1410
1411 errno = ENOENT;
1412 if (!*rel)
1413 return -1;
1414
1415 res = tmpbuf_get (tmpbuf, PATH_MAX * 3);
1416 tmp1 = res + PATH_MAX;
1417 tmp2 = tmp1 + PATH_MAX;
1418
1419#if 0 /* disabled, the musl way to do things is just too racy */
1420#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1421 /* on linux we may be able to ask the kernel */
1422 {
1423 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1424
1425 if (fd >= 0)
1426 {
1427 sprintf (tmp1, "/proc/self/fd/%d", fd);
1428 req->result = readlink (tmp1, res, PATH_MAX);
1429 /* here we should probably stat the open file and the disk file, to make sure they still match */
1430 close (fd);
1431
1432 if (req->result > 0)
1433 goto done;
1434 }
1435 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1436 return -1;
1437 }
1438#endif
1439#endif
1440
1441 if (*rel != '/')
1442 {
1443 int len;
1444
1445 errno = ENOENT;
1446 if (wd == EIO_INVALID_WD)
1447 return -1;
1448
1449 if (wd == EIO_CWD)
1450 {
1451 if (!getcwd (res, PATH_MAX))
1452 return -1;
1453
1454 len = strlen (res);
1455 }
1456 else
1457 memcpy (res, wd->str, len = wd->len);
1458
1459 if (res [1]) /* only use if not / */
1460 res += len;
1461 }
1462
1463 while (*rel)
1464 {
1465 eio_ssize_t len, linklen;
1466 const char *beg = rel;
1467
1468 while (*rel && *rel != '/')
1469 ++rel;
1470
1471 len = rel - beg;
1472
1473 if (!len) /* skip slashes */
1474 {
1475 ++rel;
1476 continue;
1477 }
1478
1479 if (beg [0] == '.')
1480 {
1481 if (len == 1)
1482 continue; /* . - nop */
1483
1484 if (beg [1] == '.' && len == 2)
1485 {
1486 /* .. - back up one component, if possible */
1487
1488 while (res != tmpbuf->ptr)
1489 if (*--res == '/')
1490 break;
1491
1492 continue;
1493 }
1494 }
1495
1496 errno = ENAMETOOLONG;
1497 if (res + 1 + len + 1 >= tmp1)
1498 return -1;
1499
1500 /* copy one component */
1501 *res = '/';
1502 memcpy (res + 1, beg, len);
1503
1504 /* zero-terminate, for readlink */
1505 res [len + 1] = 0;
1506
1507 /* now check if it's a symlink */
1508 linklen = readlink (tmpbuf->ptr, tmp1, PATH_MAX);
1509
1510 if (linklen < 0)
1511 {
1512 if (errno != EINVAL)
1513 return -1;
1514
1515 /* it's a normal directory. hopefully */
1516 res += len + 1;
1517 }
1518 else
1519 {
1520 /* yay, it was a symlink - build new path in tmp2 */
1521 int rellen = strlen (rel);
1522
1523 errno = ENAMETOOLONG;
1524 if (linklen + 1 + rellen >= PATH_MAX)
1525 return -1;
1526
1527 errno = ELOOP;
1528 if (!--symlinks)
1529 return -1;
1530
1531 if (*tmp1 == '/')
1532 res = tmpbuf->ptr; /* symlink resolves to an absolute path */
1533
1534 /* we need to be careful, as rel might point into tmp2 already */
1535 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1536 tmp2 [linklen] = '/';
1537 memcpy (tmp2, tmp1, linklen);
1538
1539 rel = tmp2;
1540 }
1541 }
1542
1543 /* special case for the lone root path */
1544 if (res == tmpbuf->ptr)
1545 *res++ = '/';
1546
1547 return res - (char *)tmpbuf->ptr;
1018} 1548}
1019 1549
1020static signed char 1550static signed char
1021eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) 1551eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1022{ 1552{
1023 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */ 1553 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1024 : a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0; 1554 : a->inode < b->inode ? -1
1555 : a->inode > b->inode ? 1
1556 : 0;
1025} 1557}
1026 1558
1027#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0 1559#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1028 1560
1029#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */ 1561#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1030#define EIO_SORT_FAST 60 /* when to only use insertion sort */ 1562#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1031 1563
1032static void 1564static void
1033eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1565eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1034{ 1566{
1035 unsigned char bits [9 + sizeof (ino_t) * 8]; 1567 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1036 unsigned char *bit = bits; 1568 unsigned char *bit = bits;
1037 1569
1038 assert (CHAR_BIT == 8); 1570 assert (CHAR_BIT == 8);
1039 assert (sizeof (eio_dirent) * 8 < 256); 1571 assert (sizeof (eio_dirent) * 8 < 256);
1040 assert (offsetof (eio_dirent, inode)); /* we use 0 as sentinel */ 1572 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1041 assert (offsetof (eio_dirent, score)); /* we use 0 as sentinel */ 1573 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1042 1574
1043 if (size <= EIO_SORT_FAST) 1575 if (size <= EIO_SORT_FAST)
1044 return; 1576 return;
1045 1577
1046 /* first prepare an array of bits to test in our radix sort */ 1578 /* first prepare an array of bits to test in our radix sort */
1047 /* try to take endianness into account, as well as differences in ino_t sizes */ 1579 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1048 /* inode_bits must contain all inodes ORed together */ 1580 /* inode_bits must contain all inodes ORed together */
1049 /* which is used to skip bits that are 0 everywhere, which is very common */ 1581 /* which is used to skip bits that are 0 everywhere, which is very common */
1050 { 1582 {
1051 ino_t endianness; 1583 eio_ino_t endianness;
1052 int i, j; 1584 int i, j;
1053 1585
1054 /* we store the byte offset of byte n into byte n of "endianness" */ 1586 /* we store the byte offset of byte n into byte n of "endianness" */
1055 for (i = 0; i < sizeof (ino_t); ++i) 1587 for (i = 0; i < sizeof (eio_ino_t); ++i)
1056 ((unsigned char *)&endianness)[i] = i; 1588 ((unsigned char *)&endianness)[i] = i;
1057 1589
1058 *bit++ = 0; 1590 *bit++ = 0;
1059 1591
1060 for (i = 0; i < sizeof (ino_t); ++i) 1592 for (i = 0; i < sizeof (eio_ino_t); ++i)
1061 { 1593 {
1062 /* shifting off the byte offsets out of "endianness" */ 1594 /* shifting off the byte offsets out of "endianness" */
1063 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8; 1595 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1064 endianness >>= 8; 1596 endianness >>= 8;
1065 1597
1066 for (j = 0; j < 8; ++j) 1598 for (j = 0; j < 8; ++j)
1067 if (inode_bits & (((ino_t)1) << (i * 8 + j))) 1599 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1068 *bit++ = offs + j; 1600 *bit++ = offs + j;
1069 } 1601 }
1070 1602
1071 for (j = 0; j < 8; ++j) 1603 for (j = 0; j < 8; ++j)
1072 if (score_bits & (1 << j)) 1604 if (score_bits & (1 << j))
1073 *bit++ = offsetof (eio_dirent, score) * 8 + j; 1605 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1074 } 1606 }
1075 1607
1076 /* now actually do the sorting (a variant of MSD radix sort) */ 1608 /* now actually do the sorting (a variant of MSD radix sort) */
1077 { 1609 {
1078 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base; 1610 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1079 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end; 1611 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1080 unsigned char *bit_stk [9 + sizeof (ino_t) * 8]; 1612 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1081 int stk_idx = 0; 1613 int stk_idx = 0;
1082 1614
1083 base_stk [stk_idx] = dents; 1615 base_stk [stk_idx] = dents;
1084 end_stk [stk_idx] = dents + size; 1616 end_stk [stk_idx] = dents + size;
1085 bit_stk [stk_idx] = bit - 1; 1617 bit_stk [stk_idx] = bit - 1;
… …
1164 } 1696 }
1165 } 1697 }
1166} 1698}
1167 1699
1168static void 1700static void
1169eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1701eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1170{ 1702{
1171 if (size <= 1) 1703 if (size <= 1)
1172 return; /* our insertion sort relies on size > 0 */ 1704 return; /* our insertion sort relies on size > 0 */
1173 1705
1174 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */ 1706 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
… …
1182 1714
1183/* read a full directory */ 1715/* read a full directory */
1184static void 1716static void
1185eio__scandir (eio_req *req, etp_worker *self) 1717eio__scandir (eio_req *req, etp_worker *self)
1186{ 1718{
1187 DIR *dirp;
1188 EIO_STRUCT_DIRENT *entp;
1189 char *name, *names; 1719 char *name, *names;
1190 int namesalloc = 4096; 1720 int namesalloc = 4096 - sizeof (void *) * 4;
1191 int namesoffs = 0; 1721 int namesoffs = 0;
1192 int flags = req->int1; 1722 int flags = req->int1;
1193 eio_dirent *dents = 0; 1723 eio_dirent *dents = 0;
1194 int dentalloc = 128; 1724 int dentalloc = 128;
1195 int dentoffs = 0; 1725 int dentoffs = 0;
1196 ino_t inode_bits = 0; 1726 eio_ino_t inode_bits = 0;
1727#ifdef _WIN32
1728 HANDLE dirp;
1729 WIN32_FIND_DATA entp;
1730#else
1731 DIR *dirp;
1732 EIO_STRUCT_DIRENT *entp;
1733#endif
1197 1734
1198 req->result = -1; 1735 req->result = -1;
1199 1736
1200 if (!(flags & EIO_READDIR_DENTS)) 1737 if (!(flags & EIO_READDIR_DENTS))
1201 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); 1738 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1202 1739
1203 X_LOCK (wrklock); 1740#ifdef _WIN32
1204 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 1741 {
1742 int len = strlen ((const char *)req->ptr1);
1743 char *path = malloc (MAX_PATH);
1744 const char *fmt;
1745 const char *reqpath = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1746
1747 if (!len)
1748 fmt = "./*";
1749 else if (reqpath[len - 1] == '/' || reqpath[len - 1] == '\\')
1750 fmt = "%s*";
1751 else
1752 fmt = "%s/*";
1753
1754 _snprintf (path, MAX_PATH, fmt, reqpath);
1755 dirp = FindFirstFile (path, &entp);
1756 free (path);
1757
1758 if (dirp == INVALID_HANDLE_VALUE)
1759 {
1760 /* should steal _dosmaperr */
1761 switch (GetLastError ())
1762 {
1763 case ERROR_FILE_NOT_FOUND:
1764 req->result = 0;
1765 break;
1766
1767 case ERROR_INVALID_NAME:
1768 case ERROR_PATH_NOT_FOUND:
1769 case ERROR_NO_MORE_FILES:
1770 errno = ENOENT;
1771 break;
1772
1773 case ERROR_NOT_ENOUGH_MEMORY:
1774 errno = ENOMEM;
1775 break;
1776
1777 default:
1778 errno = EINVAL;
1779 break;
1780 }
1781
1782 return;
1783 }
1784 }
1785#else
1786 #if HAVE_AT
1787 if (req->wd)
1788 {
1789 int fd = openat (WD2FD (req->wd), req->ptr1, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1790
1791 if (fd < 0)
1792 return;
1793
1794 dirp = fdopendir (fd);
1795
1796 if (!dirp)
1797 close (fd);
1798 }
1799 else
1205 self->dirp = dirp = opendir (req->ptr1); 1800 dirp = opendir (req->ptr1);
1801 #else
1802 dirp = opendir (wd_expand (&self->tmpbuf, req->wd, req->ptr1));
1803 #endif
1804
1805 if (!dirp)
1806 return;
1807#endif
1808
1809 if (req->flags & EIO_FLAG_PTR1_FREE)
1810 free (req->ptr1);
1811
1206 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1812 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1207 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; 1813 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1208 req->ptr2 = names = malloc (namesalloc); 1814 req->ptr2 = names = malloc (namesalloc);
1209 X_UNLOCK (wrklock);
1210 1815
1211 if (dirp && names && (!flags || dents)) 1816 if (!names || (flags && !dents))
1817 return;
1818
1212 for (;;) 1819 for (;;)
1213 { 1820 {
1821 int done;
1822
1823#ifdef _WIN32
1824 done = !dirp;
1825#else
1214 errno = 0; 1826 errno = 0;
1215 entp = readdir (dirp); 1827 entp = readdir (dirp);
1828 done = !entp;
1829#endif
1216 1830
1217 if (!entp) 1831 if (done)
1218 { 1832 {
1833#ifndef _WIN32
1834 int old_errno = errno;
1835 closedir (dirp);
1836 errno = old_errno;
1837
1219 if (errno) 1838 if (errno)
1220 break; 1839 break;
1840#endif
1221 1841
1222 /* sort etc. */ 1842 /* sort etc. */
1223 req->int1 = flags; 1843 req->int1 = flags;
1224 req->result = dentoffs; 1844 req->result = dentoffs;
1225 1845
1226 if (flags & EIO_READDIR_STAT_ORDER) 1846 if (flags & EIO_READDIR_STAT_ORDER)
1227 eio_dent_sort (dents, dentoffs, 0, inode_bits); /* sort by inode exclusively */ 1847 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1228 else if (flags & EIO_READDIR_DIRS_FIRST) 1848 else if (flags & EIO_READDIR_DIRS_FIRST)
1229 if (flags & EIO_READDIR_FOUND_UNKNOWN) 1849 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1230 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */ 1850 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1231 else 1851 else
1852 {
1853 /* in this case, all is known, and we just put dirs first and sort them */
1854 eio_dirent *oth = dents + dentoffs;
1855 eio_dirent *dir = dents;
1856
1857 /* now partition dirs to the front, and non-dirs to the back */
1858 /* by walking from both sides and swapping if necessary */
1859 while (oth > dir)
1860 {
1861 if (dir->type == EIO_DT_DIR)
1862 ++dir;
1863 else if ((--oth)->type == EIO_DT_DIR)
1864 {
1865 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1866
1867 ++dir;
1868 }
1869 }
1870
1871 /* now sort the dirs only (dirs all have the same score) */
1872 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1873 }
1874
1875 break;
1876 }
1877
1878 /* now add the entry to our list(s) */
1879 name = D_NAME (entp);
1880
1881 /* skip . and .. entries */
1882 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1883 {
1884 int len = D_NAMLEN (entp) + 1;
1885
1886 while (ecb_expect_false (namesoffs + len > namesalloc))
1887 {
1888 namesalloc *= 2;
1889 req->ptr2 = names = realloc (names, namesalloc);
1890
1891 if (!names)
1892 break;
1893 }
1894
1895 memcpy (names + namesoffs, name, len);
1896
1897 if (dents)
1898 {
1899 struct eio_dirent *ent;
1900
1901 if (ecb_expect_false (dentoffs == dentalloc))
1232 { 1902 {
1233 /* in this case, all is known, and we just put dirs first and sort them */ 1903 dentalloc *= 2;
1904 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1905
1906 if (!dents)
1907 break;
1908 }
1909
1234 eio_dirent *oth = dents + dentoffs; 1910 ent = dents + dentoffs;
1235 eio_dirent *dir = dents;
1236 1911
1237 /* now partition dirs to the front, and non-dirs to the back */ 1912 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1238 /* by walking from both sides and swapping if necessary */ 1913 ent->namelen = len - 1;
1239 /* also clear score, so it doesn't influence sorting */ 1914 ent->inode = D_INO (entp);
1240 while (oth > dir) 1915
1916 inode_bits |= ent->inode;
1917
1918 switch (D_TYPE (entp))
1919 {
1920 default:
1921 ent->type = EIO_DT_UNKNOWN;
1922 flags |= EIO_READDIR_FOUND_UNKNOWN;
1923 break;
1924
1925 #ifdef DT_FIFO
1926 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1927 #endif
1928 #ifdef DT_CHR
1929 case DT_CHR: ent->type = EIO_DT_CHR; break;
1930 #endif
1931 #ifdef DT_MPC
1932 case DT_MPC: ent->type = EIO_DT_MPC; break;
1933 #endif
1934 #ifdef DT_DIR
1935 case DT_DIR: ent->type = EIO_DT_DIR; break;
1936 #endif
1937 #ifdef DT_NAM
1938 case DT_NAM: ent->type = EIO_DT_NAM; break;
1939 #endif
1940 #ifdef DT_BLK
1941 case DT_BLK: ent->type = EIO_DT_BLK; break;
1942 #endif
1943 #ifdef DT_MPB
1944 case DT_MPB: ent->type = EIO_DT_MPB; break;
1945 #endif
1946 #ifdef DT_REG
1947 case DT_REG: ent->type = EIO_DT_REG; break;
1948 #endif
1949 #ifdef DT_NWK
1950 case DT_NWK: ent->type = EIO_DT_NWK; break;
1951 #endif
1952 #ifdef DT_CMP
1953 case DT_CMP: ent->type = EIO_DT_CMP; break;
1954 #endif
1955 #ifdef DT_LNK
1956 case DT_LNK: ent->type = EIO_DT_LNK; break;
1957 #endif
1958 #ifdef DT_SOCK
1959 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1960 #endif
1961 #ifdef DT_DOOR
1962 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1963 #endif
1964 #ifdef DT_WHT
1965 case DT_WHT: ent->type = EIO_DT_WHT; break;
1966 #endif
1967 }
1968
1969 ent->score = 7;
1970
1971 if (flags & EIO_READDIR_DIRS_FIRST)
1972 {
1973 if (ent->type == EIO_DT_UNKNOWN)
1241 { 1974 {
1242 if (dir->type == EIO_DT_DIR) 1975 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1243 ++dir; 1976 ent->score = 1;
1244 else if ((--oth)->type == EIO_DT_DIR) 1977 else if (!strchr (name, '.')) /* absence of dots indicate likely dirs */
1245 { 1978 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1246 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1247
1248 ++dir;
1249 }
1250 } 1979 }
1251 1980 else if (ent->type == EIO_DT_DIR)
1252 /* now sort the dirs only */ 1981 ent->score = 0;
1253 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1254 } 1982 }
1255
1256 break;
1257 }
1258
1259 /* now add the entry to our list(s) */
1260 name = entp->d_name;
1261
1262 /* skip . and .. entries */
1263 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1264 {
1265 int len = D_NAMLEN (entp) + 1;
1266
1267 while (expect_false (namesoffs + len > namesalloc))
1268 {
1269 namesalloc *= 2;
1270 X_LOCK (wrklock);
1271 req->ptr2 = names = realloc (names, namesalloc);
1272 X_UNLOCK (wrklock);
1273
1274 if (!names)
1275 break;
1276 } 1983 }
1277 1984
1278 memcpy (names + namesoffs, name, len);
1279
1280 if (dents)
1281 {
1282 struct eio_dirent *ent;
1283
1284 if (expect_false (dentoffs == dentalloc))
1285 {
1286 dentalloc *= 2;
1287 X_LOCK (wrklock);
1288 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1289 X_UNLOCK (wrklock);
1290
1291 if (!dents)
1292 break;
1293 }
1294
1295 ent = dents + dentoffs;
1296
1297 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1298 ent->namelen = len - 1;
1299 ent->inode = D_INO (entp);
1300
1301 inode_bits |= ent->inode;
1302
1303 switch (D_TYPE (entp))
1304 {
1305 default:
1306 ent->type = EIO_DT_UNKNOWN;
1307 flags |= EIO_READDIR_FOUND_UNKNOWN;
1308 break;
1309
1310 #ifdef DT_FIFO
1311 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1312 #endif
1313 #ifdef DT_CHR
1314 case DT_CHR: ent->type = EIO_DT_CHR; break;
1315 #endif
1316 #ifdef DT_MPC
1317 case DT_MPC: ent->type = EIO_DT_MPC; break;
1318 #endif
1319 #ifdef DT_DIR
1320 case DT_DIR: ent->type = EIO_DT_DIR; break;
1321 #endif
1322 #ifdef DT_NAM
1323 case DT_NAM: ent->type = EIO_DT_NAM; break;
1324 #endif
1325 #ifdef DT_BLK
1326 case DT_BLK: ent->type = EIO_DT_BLK; break;
1327 #endif
1328 #ifdef DT_MPB
1329 case DT_MPB: ent->type = EIO_DT_MPB; break;
1330 #endif
1331 #ifdef DT_REG
1332 case DT_REG: ent->type = EIO_DT_REG; break;
1333 #endif
1334 #ifdef DT_NWK
1335 case DT_NWK: ent->type = EIO_DT_NWK; break;
1336 #endif
1337 #ifdef DT_CMP
1338 case DT_CMP: ent->type = EIO_DT_CMP; break;
1339 #endif
1340 #ifdef DT_LNK
1341 case DT_LNK: ent->type = EIO_DT_LNK; break;
1342 #endif
1343 #ifdef DT_SOCK
1344 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1345 #endif
1346 #ifdef DT_DOOR
1347 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1348 #endif
1349 #ifdef DT_WHT
1350 case DT_WHT: ent->type = EIO_DT_WHT; break;
1351 #endif
1352 }
1353
1354 ent->score = 7;
1355
1356 if (flags & EIO_READDIR_DIRS_FIRST)
1357 {
1358 if (ent->type == EIO_DT_UNKNOWN)
1359 {
1360 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1361 ent->score = 1;
1362 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1363 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1364 }
1365 else if (ent->type == EIO_DT_DIR)
1366 ent->score = 0;
1367 }
1368 }
1369
1370 namesoffs += len; 1985 namesoffs += len;
1371 ++dentoffs; 1986 ++dentoffs;
1372 } 1987 }
1373 1988
1374 if (EIO_CANCELLED (req)) 1989 if (EIO_CANCELLED (req))
1375 { 1990 {
1376 errno = ECANCELED; 1991 errno = ECANCELED;
1377 break; 1992 break;
1378 } 1993 }
1994
1995#ifdef _WIN32
1996 if (!FindNextFile (dirp, &entp))
1997 {
1998 FindClose (dirp);
1999 dirp = 0;
1379 } 2000 }
2001#endif
2002 }
1380} 2003}
1381 2004
1382#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) 2005/*****************************************************************************/
1383# define eio__msync(a,b,c) ((errno = ENOSYS), -1) 2006/* working directory stuff */
1384#else 2007/* various deficiencies in the posix 2008 api force us to */
2008/* keep the absolute path in string form at all times */
2009/* fuck yeah. */
1385 2010
1386int 2011#if !HAVE_AT
1387eio__msync (void *mem, size_t len, int flags) 2012
2013/* a bit like realpath, but usually faster because it doesn'T have to return */
2014/* an absolute or canonical path */
2015static const char *
2016wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
1388{ 2017{
1389 if (EIO_MS_ASYNC != MS_SYNC 2018 if (!wd || *path == '/')
1390 || EIO_MS_INVALIDATE != MS_INVALIDATE 2019 return path;
1391 || EIO_MS_SYNC != MS_SYNC) 2020
2021 if (path [0] == '.' && !path [1])
2022 return wd->str;
2023
2024 {
2025 int l1 = wd->len;
2026 int l2 = strlen (path);
2027
2028 char *res = tmpbuf_get (tmpbuf, l1 + l2 + 2);
2029
2030 memcpy (res, wd->str, l1);
2031 res [l1] = '/';
2032 memcpy (res + l1 + 1, path, l2 + 1);
2033
2034 return res;
2035 }
2036}
2037
2038#endif
2039
2040static eio_wd
2041eio__wd_open_sync (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
2042{
2043 int fd;
2044 eio_wd res;
2045 int len = eio__realpath (tmpbuf, wd, path);
2046
2047 if (len < 0)
2048 return EIO_INVALID_WD;
2049
2050#if HAVE_AT
2051 fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
2052
2053 if (fd < 0)
2054 return EIO_INVALID_WD;
2055#endif
2056
2057 res = malloc (sizeof (*res) + len); /* one extra 0-byte */
2058
2059#if HAVE_AT
2060 res->fd = fd;
2061#endif
2062
2063 res->len = len;
2064 memcpy (res->str, tmpbuf->ptr, len);
2065 res->str [len] = 0;
2066
2067 return res;
2068}
2069
2070eio_wd
2071eio_wd_open_sync (eio_wd wd, const char *path)
2072{
2073 struct tmpbuf tmpbuf = { 0 };
2074 wd = eio__wd_open_sync (&tmpbuf, wd, path);
2075 free (tmpbuf.ptr);
2076
2077 return wd;
2078}
2079
2080void
2081eio_wd_close_sync (eio_wd wd)
2082{
2083 if (wd != EIO_INVALID_WD && wd != EIO_CWD)
1392 { 2084 {
1393 flags = 0 2085 #if HAVE_AT
1394 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0) 2086 close (wd->fd);
1395 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0) 2087 #endif
1396 | (flags & EIO_MS_SYNC ? MS_SYNC : 0); 2088 free (wd);
1397 } 2089 }
1398
1399 return msync (mem, len, flags);
1400} 2090}
1401 2091
1402#endif 2092#if HAVE_AT
1403 2093
1404int 2094/* they forgot these */
1405eio__mtouch (void *mem, size_t len, int flags)
1406{
1407 intptr_t addr = (intptr_t)mem;
1408 intptr_t end = addr + len;
1409#ifdef PAGESIZE
1410 const intptr_t page = PAGESIZE;
1411#else
1412 static intptr_t page;
1413 2095
1414 if (!page) 2096static int
1415 page = sysconf (_SC_PAGESIZE); 2097eio__truncateat (int dirfd, const char *path, off_t length)
1416#endif 2098{
2099 int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
2100 int res;
1417 2101
1418 /* round down to start of page, although this is probably useless */ 2102 if (fd < 0)
1419 addr &= ~(page - 1); /* assume page size is always a power of two */ 2103 return fd;
1420 2104
1421 if (addr < end) 2105 res = ftruncate (fd, length);
1422 if (flags & EIO_MT_MODIFY) /* modify */ 2106 close (fd);
1423 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
1424 else
1425 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
1426
1427 return 0; 2107 return res;
1428} 2108}
2109
2110static int
2111eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
2112{
2113 int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
2114 int res;
2115
2116 if (fd < 0)
2117 return fd;
2118
2119 res = fstatvfs (fd, buf);
2120 close (fd);
2121 return res;
2122
2123}
2124
2125#endif
1429 2126
1430/*****************************************************************************/ 2127/*****************************************************************************/
1431 2128
1432#define ALLOC(len) \ 2129#define ALLOC(len) \
1433 if (!req->ptr2) \ 2130 if (!req->ptr2) \
… …
1442 req->result = -1; \ 2139 req->result = -1; \
1443 break; \ 2140 break; \
1444 } \ 2141 } \
1445 } 2142 }
1446 2143
2144static void ecb_noinline ecb_cold
2145etp_proc_init (void)
2146{
2147#if HAVE_PRCTL_SET_NAME
2148 /* provide a more sensible "thread name" */
2149 char name[16 + 1];
2150 const int namelen = sizeof (name) - 1;
2151 int len;
2152
2153 prctl (PR_GET_NAME, (unsigned long)name, 0, 0, 0);
2154 name [namelen] = 0;
2155 len = strlen (name);
2156 strcpy (name + (len <= namelen - 4 ? len : namelen - 4), "/eio");
2157 prctl (PR_SET_NAME, (unsigned long)name, 0, 0, 0);
2158#endif
2159}
2160
1447X_THREAD_PROC (etp_proc) 2161X_THREAD_PROC (etp_proc)
1448{ 2162{
1449 ETP_REQ *req; 2163 ETP_REQ *req;
1450 struct timespec ts; 2164 struct timespec ts;
1451 etp_worker *self = (etp_worker *)thr_arg; 2165 etp_worker *self = (etp_worker *)thr_arg;
1452 2166
2167 etp_proc_init ();
2168
1453 /* try to distribute timeouts somewhat randomly */ 2169 /* try to distribute timeouts somewhat evenly */
1454 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 2170 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1455 2171
1456 for (;;) 2172 for (;;)
1457 { 2173 {
2174 ts.tv_sec = 0;
2175
1458 X_LOCK (reqlock); 2176 X_LOCK (reqlock);
1459 2177
1460 for (;;) 2178 for (;;)
1461 { 2179 {
1462 self->req = req = reqq_shift (&req_queue); 2180 req = reqq_shift (&req_queue);
1463 2181
1464 if (req) 2182 if (req)
1465 break; 2183 break;
1466 2184
2185 if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
2186 {
2187 X_UNLOCK (reqlock);
2188 X_LOCK (wrklock);
2189 --started;
2190 X_UNLOCK (wrklock);
2191 goto quit;
2192 }
2193
1467 ++idle; 2194 ++idle;
1468 2195
1469 ts.tv_sec = time (0) + IDLE_TIMEOUT; 2196 if (idle <= max_idle)
1470 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 2197 /* we are allowed to idle, so do so without any timeout */
2198 X_COND_WAIT (reqwait, reqlock);
2199 else
1471 { 2200 {
1472 if (idle > max_idle) 2201 /* initialise timeout once */
1473 { 2202 if (!ts.tv_sec)
1474 --idle; 2203 ts.tv_sec = time (0) + idle_timeout;
1475 X_UNLOCK (reqlock);
1476 X_LOCK (wrklock);
1477 --started;
1478 X_UNLOCK (wrklock);
1479 goto quit;
1480 }
1481 2204
1482 /* we are allowed to idle, so do so without any timeout */
1483 X_COND_WAIT (reqwait, reqlock); 2205 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
2206 ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1484 } 2207 }
1485 2208
1486 --idle; 2209 --idle;
1487 } 2210 }
1488 2211
… …
1491 X_UNLOCK (reqlock); 2214 X_UNLOCK (reqlock);
1492 2215
1493 if (req->type < 0) 2216 if (req->type < 0)
1494 goto quit; 2217 goto quit;
1495 2218
1496 if (!EIO_CANCELLED (req))
1497 ETP_EXECUTE (self, req); 2219 ETP_EXECUTE (self, req);
1498 2220
1499 X_LOCK (reslock); 2221 X_LOCK (reslock);
1500 2222
1501 ++npending; 2223 ++npending;
1502 2224
1503 if (!reqq_push (&res_queue, req) && want_poll_cb) 2225 if (!reqq_push (&res_queue, req) && want_poll_cb)
1504 want_poll_cb (); 2226 want_poll_cb ();
1505 2227
1506 self->req = 0;
1507 etp_worker_clear (self); 2228 etp_worker_clear (self);
1508 2229
1509 X_UNLOCK (reslock); 2230 X_UNLOCK (reslock);
1510 } 2231 }
1511 2232
1512quit: 2233quit:
2234 free (req);
2235
1513 X_LOCK (wrklock); 2236 X_LOCK (wrklock);
1514 etp_worker_free (self); 2237 etp_worker_free (self);
1515 X_UNLOCK (wrklock); 2238 X_UNLOCK (wrklock);
1516 2239
1517 return 0; 2240 return 0;
1518} 2241}
1519 2242
1520/*****************************************************************************/ 2243/*****************************************************************************/
1521 2244
2245int ecb_cold
1522int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 2246eio_init (void (*want_poll)(void), void (*done_poll)(void))
1523{ 2247{
1524 return etp_init (want_poll, done_poll); 2248 return etp_init (want_poll, done_poll);
1525} 2249}
1526 2250
2251ecb_inline void
1527static void eio_api_destroy (eio_req *req) 2252eio_api_destroy (eio_req *req)
1528{ 2253{
1529 free (req); 2254 free (req);
1530} 2255}
1531 2256
1532#define REQ(rtype) \ 2257#define REQ(rtype) \
1533 eio_req *req; \ 2258 eio_req *req; \
1534 \ 2259 \
1535 req = (eio_req *)calloc (1, sizeof *req); \ 2260 req = (eio_req *)calloc (1, sizeof *req); \
1536 if (!req) \ 2261 if (!req) \
1537 return 0; \ 2262 return 0; \
… …
1551 { \ 2276 { \
1552 eio_api_destroy (req); \ 2277 eio_api_destroy (req); \
1553 return 0; \ 2278 return 0; \
1554 } 2279 }
1555 2280
2281#define SINGLEDOT(ptr) (0[(char *)(ptr)] == '.' && !1[(char *)(ptr)])
2282
2283static void
1556static void eio_execute (etp_worker *self, eio_req *req) 2284eio_execute (etp_worker *self, eio_req *req)
1557{ 2285{
1558 errno = 0; 2286#if HAVE_AT
2287 int dirfd;
2288#else
2289 const char *path;
2290#endif
2291
2292 if (ecb_expect_false (EIO_CANCELLED (req)))
2293 {
2294 req->result = -1;
2295 req->errorno = ECANCELED;
2296 return;
2297 }
2298
2299 if (ecb_expect_false (req->wd == EIO_INVALID_WD))
2300 {
2301 req->result = -1;
2302 req->errorno = ENOENT;
2303 return;
2304 }
2305
2306 if (req->type >= EIO_OPEN)
2307 {
2308 #if HAVE_AT
2309 dirfd = WD2FD (req->wd);
2310 #else
2311 path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
2312 #endif
2313 }
1559 2314
1560 switch (req->type) 2315 switch (req->type)
1561 { 2316 {
2317 case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
2318 req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
2319 break;
2320 case EIO_WD_CLOSE: req->result = 0;
2321 eio_wd_close_sync (req->wd); break;
2322
2323 case EIO_SEEK: eio__lseek (req); break;
1562 case EIO_READ: ALLOC (req->size); 2324 case EIO_READ: ALLOC (req->size);
1563 req->result = req->offs >= 0 2325 req->result = req->offs >= 0
1564 ? pread (req->int1, req->ptr2, req->size, req->offs) 2326 ? pread (req->int1, req->ptr2, req->size, req->offs)
1565 : read (req->int1, req->ptr2, req->size); break; 2327 : read (req->int1, req->ptr2, req->size); break;
1566 case EIO_WRITE: req->result = req->offs >= 0 2328 case EIO_WRITE: req->result = req->offs >= 0
1567 ? pwrite (req->int1, req->ptr2, req->size, req->offs) 2329 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1568 : write (req->int1, req->ptr2, req->size); break; 2330 : write (req->int1, req->ptr2, req->size); break;
1569 2331
1570 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break; 2332 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1571 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break; 2333 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
2334
2335#if HAVE_AT
1572 2336
1573 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2337 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1574 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2338 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break;
1575 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2339 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1576 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2340 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, AT_SYMLINK_NOFOLLOW); break;
1577 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1578 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1579
1580 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; 2341 case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break;
1581 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1582 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break; 2342 case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break;
1583 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1584 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break; 2343 case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break;
1585 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1586
1587 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break; 2344 case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break;
1588 case EIO_CLOSE: req->result = close (req->int1); break; 2345
1589 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1590 case EIO_UNLINK: req->result = unlink (req->ptr1); break; 2346 case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
1591 case EIO_RMDIR: req->result = rmdir (req->ptr1); break; 2347 case EIO_RMDIR: /* complications arise because "." cannot be removed, so we might have to expand */
2348 req->result = req->wd && SINGLEDOT (req->ptr1)
2349 ? rmdir (req->wd->str)
2350 : unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break;
1592 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; 2351 case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); break;
1593 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; 2352 case EIO_RENAME: /* complications arise because "." cannot be renamed, so we might have to expand */
2353 req->result = req->wd && SINGLEDOT (req->ptr1)
2354 ? rename (req->wd->str, req->ptr2)
2355 : renameat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2); break;
1594 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; 2356 case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2, 0); break;
1595 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; 2357 case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
1596 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break; 2358 case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1597
1598 case EIO_READLINK: ALLOC (NAME_MAX); 2359 case EIO_READLINK: ALLOC (PATH_MAX);
1599 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; 2360 req->result = readlinkat (dirfd, req->ptr1, req->ptr2, PATH_MAX); break;
1600 2361 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1601 case EIO_SYNC: req->result = 0; sync (); break; 2362 req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1602 case EIO_FSYNC: req->result = fsync (req->int1); break;
1603 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1604 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1605 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;
1606 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1607
1608 case EIO_READDIR: eio__scandir (req, self); break;
1609
1610 case EIO_BUSY: 2363 case EIO_UTIME:
1611#ifdef _WIN32 2364 case EIO_FUTIME:
1612 Sleep (req->nv1 * 1000.); 2365 {
2366 struct timespec ts[2];
2367 struct timespec *times;
2368
2369 if (req->nv1 != -1. || req->nv2 != -1.)
2370 {
2371 ts[0].tv_sec = req->nv1;
2372 ts[0].tv_nsec = (req->nv1 - ts[0].tv_sec) * 1e9;
2373 ts[1].tv_sec = req->nv2;
2374 ts[1].tv_nsec = (req->nv2 - ts[1].tv_sec) * 1e9;
2375
2376 times = ts;
2377 }
2378 else
2379 times = 0;
2380
2381 req->result = req->type == EIO_FUTIME
2382 ? futimens (req->int1, times)
2383 : utimensat (dirfd, req->ptr1, times, 0);
2384 }
2385 break;
2386
1613#else 2387#else
1614 {
1615 struct timeval tv;
1616 2388
1617 tv.tv_sec = req->nv1; 2389 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1618 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; 2390 req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
2391 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2392 req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
2393 case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
2394 case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
2395 case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
2396 case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
1619 2397
1620 req->result = select (0, 0, 0, 0, &tv); 2398 case EIO_UNLINK: req->result = unlink (path ); break;
1621 } 2399 case EIO_RMDIR: req->result = rmdir (path ); break;
1622#endif 2400 case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
1623 break; 2401 case EIO_RENAME: req->result = rename (path , req->ptr2); break;
2402 case EIO_LINK: req->result = link (path , req->ptr2); break;
2403 case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
2404 case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
2405 case EIO_READLINK: ALLOC (PATH_MAX);
2406 req->result = readlink (path, req->ptr2, PATH_MAX); break;
2407 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2408 req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
1624 2409
1625 case EIO_UTIME: 2410 case EIO_UTIME:
1626 case EIO_FUTIME: 2411 case EIO_FUTIME:
1627 { 2412 {
1628 struct timeval tv[2]; 2413 struct timeval tv[2];
1629 struct timeval *times; 2414 struct timeval *times;
1630 2415
1631 if (req->nv1 != -1. || req->nv2 != -1.) 2416 if (req->nv1 != -1. || req->nv2 != -1.)
1632 { 2417 {
1633 tv[0].tv_sec = req->nv1; 2418 tv[0].tv_sec = req->nv1;
1634 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.; 2419 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
1635 tv[1].tv_sec = req->nv2; 2420 tv[1].tv_sec = req->nv2;
1636 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.; 2421 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
1637 2422
1638 times = tv; 2423 times = tv;
1639 } 2424 }
1640 else 2425 else
1641 times = 0; 2426 times = 0;
1642
1643 2427
1644 req->result = req->type == EIO_FUTIME 2428 req->result = req->type == EIO_FUTIME
1645 ? futimes (req->int1, times) 2429 ? futimes (req->int1, times)
1646 : utimes (req->ptr1, times); 2430 : utimes (req->ptr1, times);
1647 } 2431 }
1648 break; 2432 break;
1649 2433
2434#endif
2435
2436 case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
2437 {
2438 ALLOC (req->result);
2439 memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
2440 }
2441 break;
2442
2443 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2444 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
2445
2446 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2447 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2448
2449 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
2450 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
2451 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
2452
2453 case EIO_CLOSE: req->result = close (req->int1); break;
2454 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
2455 case EIO_SYNC: req->result = 0; sync (); break;
2456 case EIO_FSYNC: req->result = fsync (req->int1); break;
2457 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
2458 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
2459 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
2460 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
2461 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
2462 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
2463 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
2464 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
2465
2466 case EIO_READDIR: eio__scandir (req, self); break;
2467
2468 case EIO_BUSY:
2469#ifdef _WIN32
2470 Sleep (req->nv1 * 1e3);
2471#else
2472 {
2473 struct timeval tv;
2474
2475 tv.tv_sec = req->nv1;
2476 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
2477
2478 req->result = select (0, 0, 0, 0, &tv);
2479 }
2480#endif
2481 break;
2482
1650 case EIO_GROUP: 2483 case EIO_GROUP:
1651 abort (); /* handled in eio_request */ 2484 abort (); /* handled in eio_request */
1652 2485
1653 case EIO_NOP: 2486 case EIO_NOP:
1654 req->result = 0; 2487 req->result = 0;
1655 break; 2488 break;
1656 2489
1657 case EIO_CUSTOM: 2490 case EIO_CUSTOM:
1658 ((void (*)(eio_req *))req->feed) (req); 2491 req->feed (req);
1659 break; 2492 break;
1660 2493
1661 default: 2494 default:
1662 req->result = -1; 2495 req->result = EIO_ENOSYS ();
1663 break; 2496 break;
1664 } 2497 }
1665 2498
1666 req->errorno = errno; 2499 req->errorno = errno;
1667} 2500}
1668 2501
1669#ifndef EIO_NO_WRAPPERS 2502#ifndef EIO_NO_WRAPPERS
1670 2503
2504eio_req *eio_wd_open (const char *path, int pri, eio_cb cb, void *data)
2505{
2506 REQ (EIO_WD_OPEN); PATH; SEND;
2507}
2508
2509eio_req *eio_wd_close (eio_wd wd, int pri, eio_cb cb, void *data)
2510{
2511 REQ (EIO_WD_CLOSE); req->wd = wd; SEND;
2512}
2513
1671eio_req *eio_nop (int pri, eio_cb cb, void *data) 2514eio_req *eio_nop (int pri, eio_cb cb, void *data)
1672{ 2515{
1673 REQ (EIO_NOP); SEND; 2516 REQ (EIO_NOP); SEND;
1674} 2517}
1675 2518
… …
1691eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2534eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1692{ 2535{
1693 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2536 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1694} 2537}
1695 2538
2539eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2540{
2541 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2542}
2543
2544eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
2545{
2546 REQ (EIO_SYNCFS); req->int1 = fd; SEND;
2547}
2548
2549eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2550{
2551 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2552}
2553
1696eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2554eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1697{ 2555{
1698 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2556 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1699} 2557}
1700 2558
1701eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) 2559eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
1702{ 2560{
1703 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; 2561 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
1704} 2562}
1705 2563
1706eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) 2564eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
1707{ 2565{
1708 REQ (EIO_FDATASYNC); req->int1 = fd; SEND; 2566 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2567}
2568
2569eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2570{
2571 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
1709} 2572}
1710 2573
1711eio_req *eio_close (int fd, int pri, eio_cb cb, void *data) 2574eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
1712{ 2575{
1713 REQ (EIO_CLOSE); req->int1 = fd; SEND; 2576 REQ (EIO_CLOSE); req->int1 = fd; SEND;
… …
1716eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data) 2579eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
1717{ 2580{
1718 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND; 2581 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
1719} 2582}
1720 2583
2584eio_req *eio_seek (int fd, off_t offset, int whence, int pri, eio_cb cb, void *data)
2585{
2586 REQ (EIO_SEEK); req->int1 = fd; req->offs = offset; req->int2 = whence; SEND;
2587}
2588
1721eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data) 2589eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1722{ 2590{
1723 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND; 2591 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1724} 2592}
1725 2593
… …
1731eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) 2599eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1732{ 2600{
1733 REQ (EIO_FSTAT); req->int1 = fd; SEND; 2601 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1734} 2602}
1735 2603
2604eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
2605{
2606 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
2607}
2608
1736eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) 2609eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1737{ 2610{
1738 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; 2611 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1739} 2612}
1740 2613
… …
1746eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) 2619eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1747{ 2620{
1748 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2621 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1749} 2622}
1750 2623
1751eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2624eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1752{ 2625{
1753 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2626 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1754} 2627}
1755 2628
1756eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) 2629eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
… …
1776eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data) 2649eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1777{ 2650{
1778 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2651 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1779} 2652}
1780 2653
1781eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2654eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1782{ 2655{
1783 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2656 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1784} 2657}
1785 2658
1786eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) 2659eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
… …
1802eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data) 2675eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1803{ 2676{
1804 return eio__1path (EIO_READLINK, path, pri, cb, data); 2677 return eio__1path (EIO_READLINK, path, pri, cb, data);
1805} 2678}
1806 2679
2680eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2681{
2682 return eio__1path (EIO_REALPATH, path, pri, cb, data);
2683}
2684
1807eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) 2685eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1808{ 2686{
1809 return eio__1path (EIO_STAT, path, pri, cb, data); 2687 return eio__1path (EIO_STAT, path, pri, cb, data);
1810} 2688}
1811 2689
1812eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) 2690eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1813{ 2691{
1814 return eio__1path (EIO_LSTAT, path, pri, cb, data); 2692 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1815} 2693}
1816 2694
2695eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
2696{
2697 return eio__1path (EIO_STATVFS, path, pri, cb, data);
2698}
2699
1817eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) 2700eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1818{ 2701{
1819 return eio__1path (EIO_UNLINK, path, pri, cb, data); 2702 return eio__1path (EIO_UNLINK, path, pri, cb, data);
1820} 2703}
1821 2704
… …
1829 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND; 2712 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1830} 2713}
1831 2714
1832eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) 2715eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1833{ 2716{
1834 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; 2717 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1835} 2718}
1836 2719
1837static eio_req * 2720static eio_req *
1838eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2721eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1839{ 2722{
… …
1863eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2746eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1864{ 2747{
1865 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2748 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1866} 2749}
1867 2750
1868eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) 2751eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
1869{ 2752{
1870 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND; 2753 REQ (EIO_CUSTOM); req->feed = execute; SEND;
1871} 2754}
1872 2755
1873#endif 2756#endif
1874 2757
1875eio_req *eio_grp (eio_cb cb, void *data) 2758eio_req *eio_grp (eio_cb cb, void *data)
… …
1884#undef SEND 2767#undef SEND
1885 2768
1886/*****************************************************************************/ 2769/*****************************************************************************/
1887/* grp functions */ 2770/* grp functions */
1888 2771
2772void
1889void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2773eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1890{ 2774{
1891 grp->int2 = limit; 2775 grp->int2 = limit;
1892 grp->feed = feed; 2776 grp->feed = feed;
1893 2777
1894 grp_try_feed (grp); 2778 grp_try_feed (grp);
1895} 2779}
1896 2780
2781void
1897void eio_grp_limit (eio_req *grp, int limit) 2782eio_grp_limit (eio_req *grp, int limit)
1898{ 2783{
1899 grp->int2 = limit; 2784 grp->int2 = limit;
1900 2785
1901 grp_try_feed (grp); 2786 grp_try_feed (grp);
1902} 2787}
1903 2788
2789void
1904void eio_grp_add (eio_req *grp, eio_req *req) 2790eio_grp_add (eio_req *grp, eio_req *req)
1905{ 2791{
1906 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2792 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1907 2793
1908 grp->flags |= EIO_FLAG_GROUPADD; 2794 grp->flags |= EIO_FLAG_GROUPADD;
1909 2795
… …
1920} 2806}
1921 2807
1922/*****************************************************************************/ 2808/*****************************************************************************/
1923/* misc garbage */ 2809/* misc garbage */
1924 2810
2811eio_ssize_t
1925ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2812eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1926{ 2813{
1927 etp_worker wrk;
1928 ssize_t ret;
1929
1930 wrk.dbuf = 0;
1931
1932 ret = eio__sendfile (ofd, ifd, offset, count, &wrk); 2814 return eio__sendfile (ofd, ifd, offset, count);
1933
1934 if (wrk.dbuf)
1935 free (wrk.dbuf);
1936
1937 return ret;
1938} 2815}
1939 2816

Diff Legend

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