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

Comparing libeio/eio.c (file contents):
Revision 1.53 by root, Sat Jan 9 10:03:09 2010 UTC vs.
Revision 1.128 by root, Mon Feb 18 03:16:30 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>
49#include <stdlib.h> 54#include <stdlib.h>
50#include <string.h> 55#include <string.h>
51#include <errno.h> 56#include <errno.h>
52#include <sys/types.h> 57#include <sys/types.h>
53#include <sys/stat.h> 58#include <sys/stat.h>
54#include <sys/statvfs.h>
55#include <limits.h> 59#include <limits.h>
56#include <fcntl.h> 60#include <fcntl.h>
57#include <assert.h> 61#include <assert.h>
58 62
63/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
64/* intptr_t only comes from stdint.h, says idiot openbsd coder */
65#if HAVE_STDINT_H
66# include <stdint.h>
67#endif
68
69#ifndef ECANCELED
70# define ECANCELED EDOM
71#endif
72#ifndef ELOOP
73# define ELOOP EDOM
74#endif
75
76#if !defined(ENOTSOCK) && defined(WSAENOTSOCK)
77# define ENOTSOCK WSAENOTSOCK
78#endif
79
80static void eio_destroy (eio_req *req);
81
59#ifndef EIO_FINISH 82#ifndef EIO_FINISH
60# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0 83# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
61#endif 84#endif
62 85
63#ifndef EIO_DESTROY 86#ifndef EIO_DESTROY
… …
66 89
67#ifndef EIO_FEED 90#ifndef EIO_FEED
68# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0) 91# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)
69#endif 92#endif
70 93
94#ifndef EIO_FD_TO_WIN32_HANDLE
95# define EIO_FD_TO_WIN32_HANDLE(fd) _get_osfhandle (fd)
96#endif
97#ifndef EIO_WIN32_HANDLE_TO_FD
98# define EIO_WIN32_HANDLE_TO_FD(handle) _open_osfhandle (handle, 0)
99#endif
100
101#define EIO_ERRNO(errval,retval) ((errno = errval), retval)
102
103#define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1)
104
71#ifdef _WIN32 105#ifdef _WIN32
72 106
73 /*doh*/ 107 #undef PAGESIZE
108 #define PAGESIZE 4096 /* GetSystemInfo? */
109
110 /* TODO: look at how perl does stat (non-sloppy), unlink (ro-files), utime, link */
111
112 #ifdef EIO_STRUCT_STATI64
113 /* look at perl's non-sloppy stat */
114 #define stat(path,buf) _stati64 (path,buf)
115 #define fstat(fd,buf) _fstati64 (fd,buf)
116 #endif
117 #define lstat(path,buf) stat (path,buf)
118 #define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
119 #define mkdir(path,mode) _mkdir (path)
120 #define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
121
122 #define chmod(path,mode) _chmod (path, mode)
123 #define dup(fd) _dup (fd)
124 #define dup2(fd1,fd2) _dup2 (fd1, fd2)
125
126 #define fchmod(fd,mode) EIO_ENOSYS ()
127 #define chown(path,uid,gid) EIO_ENOSYS ()
128 #define fchown(fd,uid,gid) EIO_ENOSYS ()
129 #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
130 #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
131 #define mknod(path,mode,dev) EIO_ENOSYS ()
132 #define sync() EIO_ENOSYS ()
133 #define readlink(path,buf,s) EIO_ENOSYS ()
134 #define statvfs(path,buf) EIO_ENOSYS ()
135 #define fstatvfs(fd,buf) EIO_ENOSYS ()
136
137 #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
74#else 232#else
75 233
76# include "config.h"
77# include <sys/time.h> 234 #include <sys/time.h>
78# include <sys/select.h> 235 #include <sys/select.h>
79# include <sys/mman.h>
80# include <unistd.h> 236 #include <unistd.h>
237 #include <signal.h>
238 #include <dirent.h>
239
240 #ifdef 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
81# include <utime.h> 280# include <utime.h>
82# include <signal.h>
83# include <dirent.h>
84
85/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
86# if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
87# define _DIRENT_HAVE_D_TYPE /* sigh */
88# define D_INO(de) (de)->d_fileno
89# define D_NAMLEN(de) (de)->d_namlen
90# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
91# define D_INO(de) (de)->d_ino
92# endif 281#endif
93 282
94#ifdef _D_EXACT_NAMLEN 283#if HAVE_SYS_SYSCALL_H
95# undef D_NAMLEN 284# include <sys/syscall.h>
96# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
97#endif
98
99# ifdef _DIRENT_HAVE_D_TYPE
100# define D_TYPE(de) (de)->d_type
101# endif 285#endif
102 286
103# ifndef EIO_STRUCT_DIRENT 287#if HAVE_SYS_PRCTL_H
104# define EIO_STRUCT_DIRENT struct dirent 288# include <sys/prctl.h>
105# endif
106
107#endif 289#endif
108 290
109#if HAVE_SENDFILE 291#if HAVE_SENDFILE
110# if __linux 292# if __linux
111# include <sys/sendfile.h> 293# include <sys/sendfile.h>
… …
126#endif 308#endif
127#ifndef D_INO 309#ifndef D_INO
128# define D_INO(de) 0 310# define D_INO(de) 0
129#endif 311#endif
130#ifndef D_NAMLEN 312#ifndef D_NAMLEN
131# define D_NAMLEN(de) strlen ((de)->d_name) 313# define D_NAMLEN(entp) strlen (D_NAME (entp))
132#endif 314#endif
133
134/* number of seconds after which an idle threads exit */
135#define IDLE_TIMEOUT 10
136 315
137/* used for struct dirent, AIX doesn't provide it */ 316/* used for struct dirent, AIX doesn't provide it */
138#ifndef NAME_MAX 317#ifndef NAME_MAX
139# define NAME_MAX 4096 318# define NAME_MAX 4096
140#endif 319#endif
141 320
321/* used for readlink etc. */
322#ifndef PATH_MAX
323# define PATH_MAX 4096
324#endif
325
142/* buffer size for various temporary buffers */ 326/* buffer size for various temporary buffers */
143#define EIO_BUFSIZE 65536 327#define EIO_BUFSIZE 65536
144 328
145#define dBUF \ 329#define dBUF \
146 char *eio_buf; \
147 ETP_WORKER_LOCK (self); \
148 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 330 char *eio_buf = malloc (EIO_BUFSIZE); \
149 ETP_WORKER_UNLOCK (self); \
150 errno = ENOMEM; \ 331 errno = ENOMEM; \
151 if (!eio_buf) \ 332 if (!eio_buf) \
152 return -1; 333 return -1
334
335#define FUBd \
336 free (eio_buf)
153 337
154#define EIO_TICKS ((1000000 + 1023) >> 10) 338#define EIO_TICKS ((1000000 + 1023) >> 10)
155 339
156/*****************************************************************************/ 340/*****************************************************************************/
157 341
158#if __GNUC__ >= 3 342struct tmpbuf
159# 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
160#else 368#else
161# 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);
162#endif 371#endif
163 372
164#define expect_false(expr) expect ((expr) != 0, 0) 373struct eio_pwd
165#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};
166 381
167/*****************************************************************************/ 382/*****************************************************************************/
168 383
169#define ETP_PRI_MIN EIO_PRI_MIN 384#define ETP_PRI_MIN EIO_PRI_MIN
170#define ETP_PRI_MAX EIO_PRI_MAX 385#define ETP_PRI_MAX EIO_PRI_MAX
… …
176static int eio_finish (eio_req *req); 391static int eio_finish (eio_req *req);
177#define ETP_FINISH(req) eio_finish (req) 392#define ETP_FINISH(req) eio_finish (req)
178static void eio_execute (struct etp_worker *self, eio_req *req); 393static void eio_execute (struct etp_worker *self, eio_req *req);
179#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) 394#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
180 395
181#define ETP_WORKER_CLEAR(req) \
182 if (wrk->dbuf) \
183 { \
184 free (wrk->dbuf); \
185 wrk->dbuf = 0; \
186 } \
187 \
188 if (wrk->dirp) \
189 { \
190 closedir (wrk->dirp); \
191 wrk->dirp = 0; \
192 }
193
194#define ETP_WORKER_COMMON \
195 void *dbuf; \
196 DIR *dirp;
197
198/*****************************************************************************/ 396/*****************************************************************************/
199 397
200#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) 398#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
201 399
202/* calculate time difference in ~1/EIO_TICKS of a second */ 400/* calculate time difference in ~1/EIO_TICKS of a second */
401ecb_inline int
203static int tvdiff (struct timeval *tv1, struct timeval *tv2) 402tvdiff (struct timeval *tv1, struct timeval *tv2)
204{ 403{
205 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 404 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
206 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 405 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
207} 406}
208 407
… …
212static void (*done_poll_cb) (void); 411static void (*done_poll_cb) (void);
213 412
214static unsigned int max_poll_time; /* reslock */ 413static unsigned int max_poll_time; /* reslock */
215static unsigned int max_poll_reqs; /* reslock */ 414static unsigned int max_poll_reqs; /* reslock */
216 415
217static volatile unsigned int nreqs; /* reqlock */ 416static unsigned int nreqs; /* reqlock */
218static volatile unsigned int nready; /* reqlock */ 417static unsigned int nready; /* reqlock */
219static volatile unsigned int npending; /* reqlock */ 418static unsigned int npending; /* reqlock */
220static 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 */
221 421
222static mutex_t wrklock = X_MUTEX_INIT; 422static xmutex_t wrklock;
223static mutex_t reslock = X_MUTEX_INIT; 423static xmutex_t reslock;
224static mutex_t reqlock = X_MUTEX_INIT; 424static xmutex_t reqlock;
225static cond_t reqwait = X_COND_INIT; 425static xcond_t reqwait;
226
227#if !HAVE_PREADWRITE
228/*
229 * make our pread/pwrite emulation safe against themselves, but not against
230 * normal read/write by using a mutex. slows down execution a lot,
231 * but that's your problem, not mine.
232 */
233static mutex_t preadwritelock = X_MUTEX_INIT;
234#endif
235 426
236typedef struct etp_worker 427typedef struct etp_worker
237{ 428{
429 struct tmpbuf tmpbuf;
430
238 /* locked by wrklock */ 431 /* locked by wrklock */
239 struct etp_worker *prev, *next; 432 struct etp_worker *prev, *next;
240 433
241 thread_t tid; 434 xthread_t tid;
242 435
243 /* locked by reslock, reqlock or wrklock */ 436#ifdef ETP_WORKER_COMMON
244 ETP_REQ *req; /* currently processed request */
245
246 ETP_WORKER_COMMON 437 ETP_WORKER_COMMON
438#endif
247} etp_worker; 439} etp_worker;
248 440
249static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */ 441static etp_worker wrk_first; /* NOT etp */
250 442
251#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock) 443#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
252#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock) 444#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
253 445
254/* worker threads management */ 446/* worker threads management */
255 447
448static void
256static void etp_worker_clear (etp_worker *wrk) 449etp_worker_clear (etp_worker *wrk)
257{ 450{
258 ETP_WORKER_CLEAR (wrk);
259} 451}
260 452
453static void ecb_cold
261static void etp_worker_free (etp_worker *wrk) 454etp_worker_free (etp_worker *wrk)
262{ 455{
456 free (wrk->tmpbuf.ptr);
457
263 wrk->next->prev = wrk->prev; 458 wrk->next->prev = wrk->prev;
264 wrk->prev->next = wrk->next; 459 wrk->prev->next = wrk->next;
265 460
266 free (wrk); 461 free (wrk);
267} 462}
268 463
269static unsigned int etp_nreqs (void) 464static unsigned int
465etp_nreqs (void)
270{ 466{
271 int retval; 467 int retval;
272 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 468 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
273 retval = nreqs; 469 retval = nreqs;
274 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 470 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
275 return retval; 471 return retval;
276} 472}
277 473
278static unsigned int etp_nready (void) 474static unsigned int
475etp_nready (void)
279{ 476{
280 unsigned int retval; 477 unsigned int retval;
281 478
282 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 479 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
283 retval = nready; 480 retval = nready;
284 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 481 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
285 482
286 return retval; 483 return retval;
287} 484}
288 485
289static unsigned int etp_npending (void) 486static unsigned int
487etp_npending (void)
290{ 488{
291 unsigned int retval; 489 unsigned int retval;
292 490
293 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 491 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
294 retval = npending; 492 retval = npending;
295 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 493 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
296 494
297 return retval; 495 return retval;
298} 496}
299 497
300static unsigned int etp_nthreads (void) 498static unsigned int
499etp_nthreads (void)
301{ 500{
302 unsigned int retval; 501 unsigned int retval;
303 502
304 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 503 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
305 retval = started; 504 retval = started;
… …
319} etp_reqq; 518} etp_reqq;
320 519
321static etp_reqq req_queue; 520static etp_reqq req_queue;
322static etp_reqq res_queue; 521static etp_reqq res_queue;
323 522
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
324static int reqq_push (etp_reqq *q, ETP_REQ *req) 535reqq_push (etp_reqq *q, ETP_REQ *req)
325{ 536{
326 int pri = req->pri; 537 int pri = req->pri;
327 req->next = 0; 538 req->next = 0;
328 539
329 if (q->qe[pri]) 540 if (q->qe[pri])
… …
335 q->qe[pri] = q->qs[pri] = req; 546 q->qe[pri] = q->qs[pri] = req;
336 547
337 return q->size++; 548 return q->size++;
338} 549}
339 550
551static ETP_REQ * ecb_noinline
340static ETP_REQ *reqq_shift (etp_reqq *q) 552reqq_shift (etp_reqq *q)
341{ 553{
342 int pri; 554 int pri;
343 555
344 if (!q->size) 556 if (!q->size)
345 return 0; 557 return 0;
… …
360 } 572 }
361 573
362 abort (); 574 abort ();
363} 575}
364 576
365static void etp_atfork_prepare (void) 577static int ecb_cold
578etp_init (void (*want_poll)(void), void (*done_poll)(void))
366{ 579{
367 X_LOCK (wrklock); 580 X_MUTEX_CREATE (wrklock);
368 X_LOCK (reqlock); 581 X_MUTEX_CREATE (reslock);
369 X_LOCK (reslock); 582 X_MUTEX_CREATE (reqlock);
370#if !HAVE_PREADWRITE 583 X_COND_CREATE (reqwait);
371 X_LOCK (preadwritelock);
372#endif
373}
374 584
375static void etp_atfork_parent (void) 585 reqq_init (&req_queue);
376{ 586 reqq_init (&res_queue);
377#if !HAVE_PREADWRITE
378 X_UNLOCK (preadwritelock);
379#endif
380 X_UNLOCK (reslock);
381 X_UNLOCK (reqlock);
382 X_UNLOCK (wrklock);
383}
384 587
385static void etp_atfork_child (void) 588 wrk_first.next =
386{ 589 wrk_first.prev = &wrk_first;
387 ETP_REQ *prv;
388
389 while ((prv = reqq_shift (&req_queue)))
390 ETP_DESTROY (prv);
391
392 while ((prv = reqq_shift (&res_queue)))
393 ETP_DESTROY (prv);
394
395 while (wrk_first.next != &wrk_first)
396 {
397 etp_worker *wrk = wrk_first.next;
398
399 if (wrk->req)
400 ETP_DESTROY (wrk->req);
401
402 etp_worker_clear (wrk);
403 etp_worker_free (wrk);
404 }
405 590
406 started = 0; 591 started = 0;
407 idle = 0; 592 idle = 0;
408 nreqs = 0; 593 nreqs = 0;
409 nready = 0; 594 nready = 0;
410 npending = 0; 595 npending = 0;
411 596
412 etp_atfork_parent ();
413}
414
415static void
416etp_once_init (void)
417{
418 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
419}
420
421static int
422etp_init (void (*want_poll)(void), void (*done_poll)(void))
423{
424 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
425
426 pthread_once (&doinit, etp_once_init);
427
428 want_poll_cb = want_poll; 597 want_poll_cb = want_poll;
429 done_poll_cb = done_poll; 598 done_poll_cb = done_poll;
430 599
431 return 0; 600 return 0;
432} 601}
433 602
434X_THREAD_PROC (etp_proc); 603X_THREAD_PROC (etp_proc);
435 604
605static void ecb_cold
436static void etp_start_thread (void) 606etp_start_thread (void)
437{ 607{
438 etp_worker *wrk = calloc (1, sizeof (etp_worker)); 608 etp_worker *wrk = calloc (1, sizeof (etp_worker));
439 609
440 /*TODO*/ 610 /*TODO*/
441 assert (("unable to allocate worker thread data", wrk)); 611 assert (("unable to allocate worker thread data", wrk));
442 612
443 X_LOCK (wrklock); 613 X_LOCK (wrklock);
444 614
445 if (thread_create (&wrk->tid, etp_proc, (void *)wrk)) 615 if (xthread_create (&wrk->tid, etp_proc, (void *)wrk))
446 { 616 {
447 wrk->prev = &wrk_first; 617 wrk->prev = &wrk_first;
448 wrk->next = wrk_first.next; 618 wrk->next = wrk_first.next;
449 wrk_first.next->prev = wrk; 619 wrk_first.next->prev = wrk;
450 wrk_first.next = wrk; 620 wrk_first.next = wrk;
… …
454 free (wrk); 624 free (wrk);
455 625
456 X_UNLOCK (wrklock); 626 X_UNLOCK (wrklock);
457} 627}
458 628
629static void
459static void etp_maybe_start_thread (void) 630etp_maybe_start_thread (void)
460{ 631{
461 if (expect_true (etp_nthreads () >= wanted)) 632 if (ecb_expect_true (etp_nthreads () >= wanted))
462 return; 633 return;
463 634
464 /* todo: maybe use idle here, but might be less exact */ 635 /* todo: maybe use idle here, but might be less exact */
465 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 ()))
466 return; 637 return;
467 638
468 etp_start_thread (); 639 etp_start_thread ();
469} 640}
470 641
642static void ecb_cold
471static void etp_end_thread (void) 643etp_end_thread (void)
472{ 644{
473 eio_req *req = calloc (1, sizeof (eio_req)); 645 eio_req *req = calloc (1, sizeof (eio_req)); /* will be freed by worker */
474 646
475 req->type = -1; 647 req->type = -1;
476 req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 648 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
477 649
478 X_LOCK (reqlock); 650 X_LOCK (reqlock);
… …
483 X_LOCK (wrklock); 655 X_LOCK (wrklock);
484 --started; 656 --started;
485 X_UNLOCK (wrklock); 657 X_UNLOCK (wrklock);
486} 658}
487 659
488static int etp_poll (void) 660static int
661etp_poll (void)
489{ 662{
490 unsigned int maxreqs; 663 unsigned int maxreqs;
491 unsigned int maxtime; 664 unsigned int maxtime;
492 struct timeval tv_start, tv_now; 665 struct timeval tv_start, tv_now;
493 666
… …
523 696
524 X_LOCK (reqlock); 697 X_LOCK (reqlock);
525 --nreqs; 698 --nreqs;
526 X_UNLOCK (reqlock); 699 X_UNLOCK (reqlock);
527 700
528 if (expect_false (req->type == EIO_GROUP && req->size)) 701 if (ecb_expect_false (req->type == EIO_GROUP && req->size))
529 { 702 {
530 req->int1 = 1; /* mark request as delayed */ 703 req->int1 = 1; /* mark request as delayed */
531 continue; 704 continue;
532 } 705 }
533 else 706 else
534 { 707 {
535 int res = ETP_FINISH (req); 708 int res = ETP_FINISH (req);
536 if (expect_false (res)) 709 if (ecb_expect_false (res))
537 return res; 710 return res;
538 } 711 }
539 712
540 if (expect_false (maxreqs && !--maxreqs)) 713 if (ecb_expect_false (maxreqs && !--maxreqs))
541 break; 714 break;
542 715
543 if (maxtime) 716 if (maxtime)
544 { 717 {
545 gettimeofday (&tv_now, 0); 718 gettimeofday (&tv_now, 0);
… …
551 724
552 errno = EAGAIN; 725 errno = EAGAIN;
553 return -1; 726 return -1;
554} 727}
555 728
729static void
556static void etp_cancel (ETP_REQ *req) 730etp_cancel (ETP_REQ *req)
557{ 731{
558 X_LOCK (wrklock); 732 req->cancelled = 1;
559 req->flags |= EIO_FLAG_CANCELLED;
560 X_UNLOCK (wrklock);
561 733
562 eio_grp_cancel (req); 734 eio_grp_cancel (req);
563} 735}
564 736
737static void
565static void etp_submit (ETP_REQ *req) 738etp_submit (ETP_REQ *req)
566{ 739{
567 req->pri -= ETP_PRI_MIN; 740 req->pri -= ETP_PRI_MIN;
568 741
569 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;
570 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;
571 744
572 if (expect_false (req->type == EIO_GROUP)) 745 if (ecb_expect_false (req->type == EIO_GROUP))
573 { 746 {
574 /* I hope this is worth it :/ */ 747 /* I hope this is worth it :/ */
575 X_LOCK (reqlock); 748 X_LOCK (reqlock);
576 ++nreqs; 749 ++nreqs;
577 X_UNLOCK (reqlock); 750 X_UNLOCK (reqlock);
… …
596 769
597 etp_maybe_start_thread (); 770 etp_maybe_start_thread ();
598 } 771 }
599} 772}
600 773
774static void ecb_cold
601static void etp_set_max_poll_time (double nseconds) 775etp_set_max_poll_time (double nseconds)
602{ 776{
603 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 777 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
604 max_poll_time = nseconds * EIO_TICKS; 778 max_poll_time = nseconds * EIO_TICKS;
605 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 779 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
606} 780}
607 781
782static void ecb_cold
608static void etp_set_max_poll_reqs (unsigned int maxreqs) 783etp_set_max_poll_reqs (unsigned int maxreqs)
609{ 784{
610 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 785 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
611 max_poll_reqs = maxreqs; 786 max_poll_reqs = maxreqs;
612 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 787 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
613} 788}
614 789
790static void ecb_cold
615static void etp_set_max_idle (unsigned int nthreads) 791etp_set_max_idle (unsigned int nthreads)
616{ 792{
617 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 793 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
618 max_idle = nthreads <= 0 ? 1 : nthreads; 794 max_idle = nthreads;
619 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 795 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
620} 796}
621 797
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
622static void etp_set_min_parallel (unsigned int nthreads) 807etp_set_min_parallel (unsigned int nthreads)
623{ 808{
624 if (wanted < nthreads) 809 if (wanted < nthreads)
625 wanted = nthreads; 810 wanted = nthreads;
626} 811}
627 812
813static void ecb_cold
628static void etp_set_max_parallel (unsigned int nthreads) 814etp_set_max_parallel (unsigned int nthreads)
629{ 815{
630 if (wanted > nthreads) 816 if (wanted > nthreads)
631 wanted = nthreads; 817 wanted = nthreads;
632 818
633 while (started > wanted) 819 while (started > wanted)
634 etp_end_thread (); 820 etp_end_thread ();
635} 821}
636 822
637/*****************************************************************************/ 823/*****************************************************************************/
638 824
825static void
639static void grp_try_feed (eio_req *grp) 826grp_try_feed (eio_req *grp)
640{ 827{
641 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 828 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
642 { 829 {
643 grp->flags &= ~EIO_FLAG_GROUPADD; 830 grp->flags &= ~EIO_FLAG_GROUPADD;
644 831
… …
651 break; 838 break;
652 } 839 }
653 } 840 }
654} 841}
655 842
843static int
656static int grp_dec (eio_req *grp) 844grp_dec (eio_req *grp)
657{ 845{
658 --grp->size; 846 --grp->size;
659 847
660 /* call feeder, if applicable */ 848 /* call feeder, if applicable */
661 grp_try_feed (grp); 849 grp_try_feed (grp);
… …
665 return eio_finish (grp); 853 return eio_finish (grp);
666 else 854 else
667 return 0; 855 return 0;
668} 856}
669 857
858static void
670void eio_destroy (eio_req *req) 859eio_destroy (eio_req *req)
671{ 860{
672 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); 861 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
673 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); 862 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
674 863
675 EIO_DESTROY (req); 864 EIO_DESTROY (req);
676} 865}
677 866
867static int
678static int eio_finish (eio_req *req) 868eio_finish (eio_req *req)
679{ 869{
680 int res = EIO_FINISH (req); 870 int res = EIO_FINISH (req);
681 871
682 if (req->grp) 872 if (req->grp)
683 { 873 {
… …
691 if (grp->grp_first == req) 881 if (grp->grp_first == req)
692 grp->grp_first = req->grp_next; 882 grp->grp_first = req->grp_next;
693 883
694 res2 = grp_dec (grp); 884 res2 = grp_dec (grp);
695 885
696 if (!res && res2) 886 if (!res)
697 res = res2; 887 res = res2;
698 } 888 }
699 889
700 eio_destroy (req); 890 eio_destroy (req);
701 891
702 return res; 892 return res;
703} 893}
704 894
895void
705void eio_grp_cancel (eio_req *grp) 896eio_grp_cancel (eio_req *grp)
706{ 897{
707 for (grp = grp->grp_first; grp; grp = grp->grp_next) 898 for (grp = grp->grp_first; grp; grp = grp->grp_next)
708 eio_cancel (grp); 899 eio_cancel (grp);
709} 900}
710 901
902void
711void eio_cancel (eio_req *req) 903eio_cancel (eio_req *req)
712{ 904{
713 etp_cancel (req); 905 etp_cancel (req);
714} 906}
715 907
908void
716void eio_submit (eio_req *req) 909eio_submit (eio_req *req)
717{ 910{
718 etp_submit (req); 911 etp_submit (req);
719} 912}
720 913
721unsigned int eio_nreqs (void) 914unsigned int
915eio_nreqs (void)
722{ 916{
723 return etp_nreqs (); 917 return etp_nreqs ();
724} 918}
725 919
726unsigned int eio_nready (void) 920unsigned int
921eio_nready (void)
727{ 922{
728 return etp_nready (); 923 return etp_nready ();
729} 924}
730 925
731unsigned int eio_npending (void) 926unsigned int
927eio_npending (void)
732{ 928{
733 return etp_npending (); 929 return etp_npending ();
734} 930}
735 931
736unsigned int eio_nthreads (void) 932unsigned int ecb_cold
933eio_nthreads (void)
737{ 934{
738 return etp_nthreads (); 935 return etp_nthreads ();
739} 936}
740 937
938void ecb_cold
741void eio_set_max_poll_time (double nseconds) 939eio_set_max_poll_time (double nseconds)
742{ 940{
743 etp_set_max_poll_time (nseconds); 941 etp_set_max_poll_time (nseconds);
744} 942}
745 943
944void ecb_cold
746void eio_set_max_poll_reqs (unsigned int maxreqs) 945eio_set_max_poll_reqs (unsigned int maxreqs)
747{ 946{
748 etp_set_max_poll_reqs (maxreqs); 947 etp_set_max_poll_reqs (maxreqs);
749} 948}
750 949
950void ecb_cold
751void eio_set_max_idle (unsigned int nthreads) 951eio_set_max_idle (unsigned int nthreads)
752{ 952{
753 etp_set_max_idle (nthreads); 953 etp_set_max_idle (nthreads);
754} 954}
755 955
956void ecb_cold
957eio_set_idle_timeout (unsigned int seconds)
958{
959 etp_set_idle_timeout (seconds);
960}
961
962void ecb_cold
756void eio_set_min_parallel (unsigned int nthreads) 963eio_set_min_parallel (unsigned int nthreads)
757{ 964{
758 etp_set_min_parallel (nthreads); 965 etp_set_min_parallel (nthreads);
759} 966}
760 967
968void ecb_cold
761void eio_set_max_parallel (unsigned int nthreads) 969eio_set_max_parallel (unsigned int nthreads)
762{ 970{
763 etp_set_max_parallel (nthreads); 971 etp_set_max_parallel (nthreads);
764} 972}
765 973
766int eio_poll (void) 974int eio_poll (void)
… …
769} 977}
770 978
771/*****************************************************************************/ 979/*****************************************************************************/
772/* work around various missing functions */ 980/* work around various missing functions */
773 981
774#if !HAVE_PREADWRITE
775# undef pread
776# undef pwrite
777# define pread eio__pread
778# define pwrite eio__pwrite
779
780static ssize_t
781eio__pread (int fd, void *buf, size_t count, off_t offset)
782{
783 ssize_t res;
784 off_t ooffset;
785
786 X_LOCK (preadwritelock);
787 ooffset = lseek (fd, 0, SEEK_CUR);
788 lseek (fd, offset, SEEK_SET);
789 res = read (fd, buf, count);
790 lseek (fd, ooffset, SEEK_SET);
791 X_UNLOCK (preadwritelock);
792
793 return res;
794}
795
796static ssize_t
797eio__pwrite (int fd, void *buf, size_t count, off_t offset)
798{
799 ssize_t res;
800 off_t ooffset;
801
802 X_LOCK (preadwritelock);
803 ooffset = lseek (fd, 0, SEEK_CUR);
804 lseek (fd, offset, SEEK_SET);
805 res = write (fd, buf, count);
806 lseek (fd, ooffset, SEEK_SET);
807 X_UNLOCK (preadwritelock);
808
809 return res;
810}
811#endif
812
813#ifndef HAVE_FUTIMES 982#ifndef HAVE_UTIMES
814 983
815# undef utimes 984# undef utimes
816# undef futimes
817# define utimes(path,times) eio__utimes (path, times) 985# define utimes(path,times) eio__utimes (path, times)
818# define futimes(fd,times) eio__futimes (fd, times)
819 986
820static int 987static int
821eio__utimes (const char *filename, const struct timeval times[2]) 988eio__utimes (const char *filename, const struct timeval times[2])
822{ 989{
823 if (times) 990 if (times)
… …
831 } 998 }
832 else 999 else
833 return utime (filename, 0); 1000 return utime (filename, 0);
834} 1001}
835 1002
1003#endif
1004
1005#ifndef HAVE_FUTIMES
1006
1007# undef futimes
1008# define futimes(fd,times) eio__futimes (fd, times)
1009
1010static int
836static int eio__futimes (int fd, const struct timeval tv[2]) 1011eio__futimes (int fd, const struct timeval tv[2])
837{ 1012{
838 errno = ENOSYS; 1013 errno = ENOSYS;
839 return -1; 1014 return -1;
840} 1015}
841 1016
… …
844#if !HAVE_FDATASYNC 1019#if !HAVE_FDATASYNC
845# undef fdatasync 1020# undef fdatasync
846# define fdatasync(fd) fsync (fd) 1021# define fdatasync(fd) fsync (fd)
847#endif 1022#endif
848 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
849/* sync_file_range always needs emulation */ 1041/* sync_file_range always needs emulation */
850int 1042static int
851eio__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)
852{ 1044{
853#if HAVE_SYNC_FILE_RANGE 1045#if HAVE_SYNC_FILE_RANGE
854 int res; 1046 int res;
855 1047
… …
872 /* 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
873 * call fdatasync, as that matches the expectation of its users best */ 1065 * call fdatasync, as that matches the expectation of its users best */
874 return fdatasync (fd); 1066 return fdatasync (fd);
875} 1067}
876 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
877#if !HAVE_READAHEAD 1079#if !HAVE_READAHEAD
878# undef readahead 1080# undef readahead
879# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 1081# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
880 1082
881static ssize_t 1083static eio_ssize_t
882eio__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)
883{ 1085{
884 size_t todo = count; 1086 size_t todo = count;
885 dBUF; 1087 dBUF;
886 1088
… …
891 pread (fd, eio_buf, len, offset); 1093 pread (fd, eio_buf, len, offset);
892 offset += len; 1094 offset += len;
893 todo -= len; 1095 todo -= len;
894 } 1096 }
895 1097
896 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 */
897 return count; 1102 return 0;
898} 1103}
899 1104
900#endif 1105#endif
901 1106
902/* sendfile always needs emulation */ 1107/* sendfile always needs emulation */
903static ssize_t 1108static eio_ssize_t
904eio__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)
905{ 1110{
1111 eio_ssize_t written = 0;
906 ssize_t res; 1112 eio_ssize_t res;
907 1113
908 if (!count) 1114 if (!count)
909 return 0; 1115 return 0;
910 1116
1117 for (;;)
1118 {
1119#ifdef __APPLE__
1120# undef HAVE_SENDFILE /* broken, as everything on os x */
1121#endif
911#if HAVE_SENDFILE 1122#if HAVE_SENDFILE
912# if __linux 1123# if __linux
1124 off_t soffset = offset;
913 res = sendfile (ofd, ifd, &offset, count); 1125 res = sendfile (ofd, ifd, &soffset, count);
914 1126
915# elif __FreeBSD__ 1127# elif __FreeBSD__
916 /* 1128 /*
917 * 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
918 * wasted on making it similar to other I/O functions. 1130 * wasted on making it similar to other I/O functions.
919 */ 1131 */
920 {
921 off_t sbytes; 1132 off_t sbytes;
922 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 1133 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
923 1134
924 #if 0 /* according to the manpage, this is correct, but broken behaviour */ 1135 #if 0 /* according to the manpage, this is correct, but broken behaviour */
925 /* freebsd' sendfile will return 0 on success */ 1136 /* freebsd' sendfile will return 0 on success */
926 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */ 1137 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
927 /* not on e.g. EIO or EPIPE - sounds broken */ 1138 /* not on e.g. EIO or EPIPE - sounds broken */
928 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0) 1139 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
929 res = sbytes; 1140 res = sbytes;
930 #endif 1141 #endif
931 1142
932 /* according to source inspection, this is correct, and useful behaviour */ 1143 /* according to source inspection, this is correct, and useful behaviour */
933 if (sbytes) 1144 if (sbytes)
934 res = sbytes; 1145 res = sbytes;
935 }
936 1146
937# elif defined (__APPLE__) 1147# elif defined __APPLE__
938
939 {
940 off_t sbytes = count; 1148 off_t sbytes = count;
941 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0); 1149 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
942 1150
943 /* according to the manpage, sbytes is always valid */ 1151 /* according to the manpage, sbytes is always valid */
944 if (sbytes) 1152 if (sbytes)
945 res = sbytes; 1153 res = sbytes;
946 }
947 1154
948# elif __hpux 1155# elif __hpux
949 res = sendfile (ofd, ifd, offset, count, 0, 0); 1156 res = sendfile (ofd, ifd, offset, count, 0, 0);
950 1157
951# elif __solaris 1158# elif __solaris
952 {
953 struct sendfilevec vec; 1159 struct sendfilevec vec;
954 size_t sbytes; 1160 size_t sbytes;
955 1161
956 vec.sfv_fd = ifd; 1162 vec.sfv_fd = ifd;
957 vec.sfv_flag = 0; 1163 vec.sfv_flag = 0;
958 vec.sfv_off = offset; 1164 vec.sfv_off = offset;
959 vec.sfv_len = count; 1165 vec.sfv_len = count;
960 1166
961 res = sendfilev (ofd, &vec, 1, &sbytes); 1167 res = sendfilev (ofd, &vec, 1, &sbytes);
962 1168
963 if (res < 0 && sbytes) 1169 if (res < 0 && sbytes)
964 res = sbytes; 1170 res = sbytes;
965 }
966 1171
967# endif 1172# endif
968 1173
969#elif defined (_WIN32) 1174#elif defined (_WIN32) && 0
970
971 /* 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 */
972 { 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 */
973 HANDLE h = TO_SOCKET (ifd); 1179 HANDLE h = TO_SOCKET (ifd);
974 SetFilePointer (h, offset, 0, FILE_BEGIN); 1180 SetFilePointer (h, offset, 0, FILE_BEGIN);
975 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0); 1181 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
976 }
977 1182
978#else 1183#else
979 res = -1; 1184 res = EIO_ENOSYS ();
980 errno = ENOSYS;
981#endif 1185#endif
982 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
983 if (res < 0 1210 if (res < 0
984 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1211 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
985 /* BSDs */ 1212 /* BSDs */
986#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... */
987 || errno == ENOTSUP 1214 || errno == ENOTSUP
988#endif 1215#endif
1216#ifdef EOPNOTSUPP /* windows */
989 || errno == EOPNOTSUPP /* BSDs */ 1217 || errno == EOPNOTSUPP /* BSDs */
1218#endif
990#if __solaris 1219#if __solaris
991 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1220 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
992#endif 1221#endif
993 ) 1222 )
994 ) 1223 )
… …
998 1227
999 res = 0; 1228 res = 0;
1000 1229
1001 while (count) 1230 while (count)
1002 { 1231 {
1003 ssize_t cnt; 1232 eio_ssize_t cnt;
1004 1233
1005 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);
1006 1235
1007 if (cnt <= 0) 1236 if (cnt <= 0)
1008 { 1237 {
… …
1020 1249
1021 offset += cnt; 1250 offset += cnt;
1022 res += cnt; 1251 res += cnt;
1023 count -= cnt; 1252 count -= cnt;
1024 } 1253 }
1254
1255 FUBd;
1025 } 1256 }
1026 1257
1027 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;
1028} 1548}
1029 1549
1030static signed char 1550static signed char
1031eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) 1551eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1032{ 1552{
1033 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 */
1034 : 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;
1035} 1557}
1036 1558
1037#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
1038 1560
1039#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 */
1040#define EIO_SORT_FAST 60 /* when to only use insertion sort */ 1562#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1041 1563
1042static void 1564static void
1043eio_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)
1044{ 1566{
1045 unsigned char bits [9 + sizeof (ino_t) * 8]; 1567 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1046 unsigned char *bit = bits; 1568 unsigned char *bit = bits;
1047 1569
1048 assert (CHAR_BIT == 8); 1570 assert (CHAR_BIT == 8);
1049 assert (sizeof (eio_dirent) * 8 < 256); 1571 assert (sizeof (eio_dirent) * 8 < 256);
1050 assert (offsetof (eio_dirent, inode)); /* we use 0 as sentinel */ 1572 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1051 assert (offsetof (eio_dirent, score)); /* we use 0 as sentinel */ 1573 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1052 1574
1053 if (size <= EIO_SORT_FAST) 1575 if (size <= EIO_SORT_FAST)
1054 return; 1576 return;
1055 1577
1056 /* 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 */
1057 /* 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 */
1058 /* inode_bits must contain all inodes ORed together */ 1580 /* inode_bits must contain all inodes ORed together */
1059 /* 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 */
1060 { 1582 {
1061 ino_t endianness; 1583 eio_ino_t endianness;
1062 int i, j; 1584 int i, j;
1063 1585
1064 /* 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" */
1065 for (i = 0; i < sizeof (ino_t); ++i) 1587 for (i = 0; i < sizeof (eio_ino_t); ++i)
1066 ((unsigned char *)&endianness)[i] = i; 1588 ((unsigned char *)&endianness)[i] = i;
1067 1589
1068 *bit++ = 0; 1590 *bit++ = 0;
1069 1591
1070 for (i = 0; i < sizeof (ino_t); ++i) 1592 for (i = 0; i < sizeof (eio_ino_t); ++i)
1071 { 1593 {
1072 /* shifting off the byte offsets out of "endianness" */ 1594 /* shifting off the byte offsets out of "endianness" */
1073 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8; 1595 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1074 endianness >>= 8; 1596 endianness >>= 8;
1075 1597
1076 for (j = 0; j < 8; ++j) 1598 for (j = 0; j < 8; ++j)
1077 if (inode_bits & (((ino_t)1) << (i * 8 + j))) 1599 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1078 *bit++ = offs + j; 1600 *bit++ = offs + j;
1079 } 1601 }
1080 1602
1081 for (j = 0; j < 8; ++j) 1603 for (j = 0; j < 8; ++j)
1082 if (score_bits & (1 << j)) 1604 if (score_bits & (1 << j))
1083 *bit++ = offsetof (eio_dirent, score) * 8 + j; 1605 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1084 } 1606 }
1085 1607
1086 /* now actually do the sorting (a variant of MSD radix sort) */ 1608 /* now actually do the sorting (a variant of MSD radix sort) */
1087 { 1609 {
1088 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base; 1610 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1089 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end; 1611 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1090 unsigned char *bit_stk [9 + sizeof (ino_t) * 8]; 1612 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1091 int stk_idx = 0; 1613 int stk_idx = 0;
1092 1614
1093 base_stk [stk_idx] = dents; 1615 base_stk [stk_idx] = dents;
1094 end_stk [stk_idx] = dents + size; 1616 end_stk [stk_idx] = dents + size;
1095 bit_stk [stk_idx] = bit - 1; 1617 bit_stk [stk_idx] = bit - 1;
… …
1174 } 1696 }
1175 } 1697 }
1176} 1698}
1177 1699
1178static void 1700static void
1179eio_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)
1180{ 1702{
1181 if (size <= 1) 1703 if (size <= 1)
1182 return; /* our insertion sort relies on size > 0 */ 1704 return; /* our insertion sort relies on size > 0 */
1183 1705
1184 /* 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 */
… …
1192 1714
1193/* read a full directory */ 1715/* read a full directory */
1194static void 1716static void
1195eio__scandir (eio_req *req, etp_worker *self) 1717eio__scandir (eio_req *req, etp_worker *self)
1196{ 1718{
1197 DIR *dirp;
1198 EIO_STRUCT_DIRENT *entp;
1199 char *name, *names; 1719 char *name, *names;
1200 int namesalloc = 4096; 1720 int namesalloc = 4096 - sizeof (void *) * 4;
1201 int namesoffs = 0; 1721 int namesoffs = 0;
1202 int flags = req->int1; 1722 int flags = req->int1;
1203 eio_dirent *dents = 0; 1723 eio_dirent *dents = 0;
1204 int dentalloc = 128; 1724 int dentalloc = 128;
1205 int dentoffs = 0; 1725 int dentoffs = 0;
1206 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
1207 1734
1208 req->result = -1; 1735 req->result = -1;
1209 1736
1210 if (!(flags & EIO_READDIR_DENTS)) 1737 if (!(flags & EIO_READDIR_DENTS))
1211 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); 1738 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1212 1739
1213 X_LOCK (wrklock); 1740#ifdef _WIN32
1214 /* 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
1215 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
1216 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1812 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1217 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; 1813 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1218 req->ptr2 = names = malloc (namesalloc); 1814 req->ptr2 = names = malloc (namesalloc);
1219 X_UNLOCK (wrklock);
1220 1815
1221 if (dirp && names && (!flags || dents)) 1816 if (!names || (flags && !dents))
1817 return;
1818
1222 for (;;) 1819 for (;;)
1223 { 1820 {
1821 int done;
1822
1823#ifdef _WIN32
1824 done = !dirp;
1825#else
1224 errno = 0; 1826 errno = 0;
1225 entp = readdir (dirp); 1827 entp = readdir (dirp);
1828 done = !entp;
1829#endif
1226 1830
1227 if (!entp) 1831 if (done)
1228 { 1832 {
1833#ifndef _WIN32
1834 int old_errno = errno;
1835 closedir (dirp);
1836 errno = old_errno;
1837
1229 if (errno) 1838 if (errno)
1230 break; 1839 break;
1840#endif
1231 1841
1232 /* sort etc. */ 1842 /* sort etc. */
1233 req->int1 = flags; 1843 req->int1 = flags;
1234 req->result = dentoffs; 1844 req->result = dentoffs;
1235 1845
1236 if (flags & EIO_READDIR_STAT_ORDER) 1846 if (flags & EIO_READDIR_STAT_ORDER)
1237 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);
1238 else if (flags & EIO_READDIR_DIRS_FIRST) 1848 else if (flags & EIO_READDIR_DIRS_FIRST)
1239 if (flags & EIO_READDIR_FOUND_UNKNOWN) 1849 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1240 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 */
1241 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))
1242 { 1902 {
1243 /* 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
1244 eio_dirent *oth = dents + dentoffs; 1910 ent = dents + dentoffs;
1245 eio_dirent *dir = dents;
1246 1911
1247 /* 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 */
1248 /* by walking from both sides and swapping if necessary */ 1913 ent->namelen = len - 1;
1249 /* also clear score, so it doesn't influence sorting */ 1914 ent->inode = D_INO (entp);
1250 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)
1251 { 1974 {
1252 if (dir->type == EIO_DT_DIR) 1975 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1253 ++dir; 1976 ent->score = 1;
1254 else if ((--oth)->type == EIO_DT_DIR) 1977 else if (!strchr (name, '.')) /* absence of dots indicate likely dirs */
1255 { 1978 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1256 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1257
1258 ++dir;
1259 }
1260 } 1979 }
1261 1980 else if (ent->type == EIO_DT_DIR)
1262 /* now sort the dirs only */ 1981 ent->score = 0;
1263 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1264 } 1982 }
1265
1266 break;
1267 }
1268
1269 /* now add the entry to our list(s) */
1270 name = entp->d_name;
1271
1272 /* skip . and .. entries */
1273 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1274 {
1275 int len = D_NAMLEN (entp) + 1;
1276
1277 while (expect_false (namesoffs + len > namesalloc))
1278 {
1279 namesalloc *= 2;
1280 X_LOCK (wrklock);
1281 req->ptr2 = names = realloc (names, namesalloc);
1282 X_UNLOCK (wrklock);
1283
1284 if (!names)
1285 break;
1286 } 1983 }
1287 1984
1288 memcpy (names + namesoffs, name, len);
1289
1290 if (dents)
1291 {
1292 struct eio_dirent *ent;
1293
1294 if (expect_false (dentoffs == dentalloc))
1295 {
1296 dentalloc *= 2;
1297 X_LOCK (wrklock);
1298 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1299 X_UNLOCK (wrklock);
1300
1301 if (!dents)
1302 break;
1303 }
1304
1305 ent = dents + dentoffs;
1306
1307 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1308 ent->namelen = len - 1;
1309 ent->inode = D_INO (entp);
1310
1311 inode_bits |= ent->inode;
1312
1313 switch (D_TYPE (entp))
1314 {
1315 default:
1316 ent->type = EIO_DT_UNKNOWN;
1317 flags |= EIO_READDIR_FOUND_UNKNOWN;
1318 break;
1319
1320 #ifdef DT_FIFO
1321 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1322 #endif
1323 #ifdef DT_CHR
1324 case DT_CHR: ent->type = EIO_DT_CHR; break;
1325 #endif
1326 #ifdef DT_MPC
1327 case DT_MPC: ent->type = EIO_DT_MPC; break;
1328 #endif
1329 #ifdef DT_DIR
1330 case DT_DIR: ent->type = EIO_DT_DIR; break;
1331 #endif
1332 #ifdef DT_NAM
1333 case DT_NAM: ent->type = EIO_DT_NAM; break;
1334 #endif
1335 #ifdef DT_BLK
1336 case DT_BLK: ent->type = EIO_DT_BLK; break;
1337 #endif
1338 #ifdef DT_MPB
1339 case DT_MPB: ent->type = EIO_DT_MPB; break;
1340 #endif
1341 #ifdef DT_REG
1342 case DT_REG: ent->type = EIO_DT_REG; break;
1343 #endif
1344 #ifdef DT_NWK
1345 case DT_NWK: ent->type = EIO_DT_NWK; break;
1346 #endif
1347 #ifdef DT_CMP
1348 case DT_CMP: ent->type = EIO_DT_CMP; break;
1349 #endif
1350 #ifdef DT_LNK
1351 case DT_LNK: ent->type = EIO_DT_LNK; break;
1352 #endif
1353 #ifdef DT_SOCK
1354 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1355 #endif
1356 #ifdef DT_DOOR
1357 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1358 #endif
1359 #ifdef DT_WHT
1360 case DT_WHT: ent->type = EIO_DT_WHT; break;
1361 #endif
1362 }
1363
1364 ent->score = 7;
1365
1366 if (flags & EIO_READDIR_DIRS_FIRST)
1367 {
1368 if (ent->type == EIO_DT_UNKNOWN)
1369 {
1370 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1371 ent->score = 1;
1372 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1373 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1374 }
1375 else if (ent->type == EIO_DT_DIR)
1376 ent->score = 0;
1377 }
1378 }
1379
1380 namesoffs += len; 1985 namesoffs += len;
1381 ++dentoffs; 1986 ++dentoffs;
1382 } 1987 }
1383 1988
1384 if (EIO_CANCELLED (req)) 1989 if (EIO_CANCELLED (req))
1385 { 1990 {
1386 errno = ECANCELED; 1991 errno = ECANCELED;
1387 break; 1992 break;
1388 } 1993 }
1994
1995#ifdef _WIN32
1996 if (!FindNextFile (dirp, &entp))
1997 {
1998 FindClose (dirp);
1999 dirp = 0;
1389 } 2000 }
2001#endif
2002 }
1390} 2003}
1391 2004
1392#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) 2005/*****************************************************************************/
1393# define eio__msync(a,b,c) ((errno = ENOSYS), -1) 2006/* working directory stuff */
1394#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. */
1395 2010
1396int 2011#if !HAVE_AT
1397eio__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)
1398{ 2017{
1399 if (EIO_MS_ASYNC != MS_SYNC 2018 if (!wd || *path == '/')
1400 || EIO_MS_INVALIDATE != MS_INVALIDATE 2019 return path;
1401 || 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)
1402 { 2084 {
1403 flags = 0 2085 #if HAVE_AT
1404 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0) 2086 close (wd->fd);
1405 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0) 2087 #endif
1406 | (flags & EIO_MS_SYNC ? MS_SYNC : 0); 2088 free (wd);
1407 } 2089 }
1408
1409 return msync (mem, len, flags);
1410} 2090}
1411 2091
1412#endif 2092#if HAVE_AT
1413 2093
1414int 2094/* they forgot these */
1415eio__mtouch (void *mem, size_t len, int flags)
1416{
1417 intptr_t addr = (intptr_t)mem;
1418 intptr_t end = addr + len;
1419#ifdef PAGESIZE
1420 const intptr_t page = PAGESIZE;
1421#else
1422 static intptr_t page;
1423 2095
1424 if (!page) 2096static int
1425 page = sysconf (_SC_PAGESIZE); 2097eio__truncateat (int dirfd, const char *path, off_t length)
1426#endif 2098{
2099 int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
2100 int res;
1427 2101
1428 /* round down to start of page, although this is probably useless */ 2102 if (fd < 0)
1429 addr &= ~(page - 1); /* assume page size is always a power of two */ 2103 return fd;
1430 2104
1431 if (addr < end) 2105 res = ftruncate (fd, length);
1432 if (flags & EIO_MT_MODIFY) /* modify */ 2106 close (fd);
1433 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
1434 else
1435 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
1436
1437 return 0; 2107 return res;
1438} 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
1439 2126
1440/*****************************************************************************/ 2127/*****************************************************************************/
1441 2128
1442#define ALLOC(len) \ 2129#define ALLOC(len) \
1443 if (!req->ptr2) \ 2130 if (!req->ptr2) \
… …
1452 req->result = -1; \ 2139 req->result = -1; \
1453 break; \ 2140 break; \
1454 } \ 2141 } \
1455 } 2142 }
1456 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
1457X_THREAD_PROC (etp_proc) 2161X_THREAD_PROC (etp_proc)
1458{ 2162{
1459 ETP_REQ *req; 2163 ETP_REQ *req;
1460 struct timespec ts; 2164 struct timespec ts;
1461 etp_worker *self = (etp_worker *)thr_arg; 2165 etp_worker *self = (etp_worker *)thr_arg;
1462 2166
2167 etp_proc_init ();
2168
1463 /* try to distribute timeouts somewhat randomly */ 2169 /* try to distribute timeouts somewhat evenly */
1464 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 2170 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1465 2171
1466 for (;;) 2172 for (;;)
1467 { 2173 {
2174 ts.tv_sec = 0;
2175
1468 X_LOCK (reqlock); 2176 X_LOCK (reqlock);
1469 2177
1470 for (;;) 2178 for (;;)
1471 { 2179 {
1472 self->req = req = reqq_shift (&req_queue); 2180 req = reqq_shift (&req_queue);
1473 2181
1474 if (req) 2182 if (req)
1475 break; 2183 break;
1476 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
1477 ++idle; 2194 ++idle;
1478 2195
1479 ts.tv_sec = time (0) + IDLE_TIMEOUT; 2196 if (idle <= max_idle)
1480 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
1481 { 2200 {
1482 if (idle > max_idle) 2201 /* initialise timeout once */
1483 { 2202 if (!ts.tv_sec)
1484 --idle; 2203 ts.tv_sec = time (0) + idle_timeout;
1485 X_UNLOCK (reqlock);
1486 X_LOCK (wrklock);
1487 --started;
1488 X_UNLOCK (wrklock);
1489 goto quit;
1490 }
1491 2204
1492 /* we are allowed to idle, so do so without any timeout */
1493 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.,.. */
1494 } 2207 }
1495 2208
1496 --idle; 2209 --idle;
1497 } 2210 }
1498 2211
… …
1501 X_UNLOCK (reqlock); 2214 X_UNLOCK (reqlock);
1502 2215
1503 if (req->type < 0) 2216 if (req->type < 0)
1504 goto quit; 2217 goto quit;
1505 2218
1506 if (!EIO_CANCELLED (req))
1507 ETP_EXECUTE (self, req); 2219 ETP_EXECUTE (self, req);
1508 2220
1509 X_LOCK (reslock); 2221 X_LOCK (reslock);
1510 2222
1511 ++npending; 2223 ++npending;
1512 2224
1513 if (!reqq_push (&res_queue, req) && want_poll_cb) 2225 if (!reqq_push (&res_queue, req) && want_poll_cb)
1514 want_poll_cb (); 2226 want_poll_cb ();
1515 2227
1516 self->req = 0;
1517 etp_worker_clear (self); 2228 etp_worker_clear (self);
1518 2229
1519 X_UNLOCK (reslock); 2230 X_UNLOCK (reslock);
1520 } 2231 }
1521 2232
1522quit: 2233quit:
2234 free (req);
2235
1523 X_LOCK (wrklock); 2236 X_LOCK (wrklock);
1524 etp_worker_free (self); 2237 etp_worker_free (self);
1525 X_UNLOCK (wrklock); 2238 X_UNLOCK (wrklock);
1526 2239
1527 return 0; 2240 return 0;
1528} 2241}
1529 2242
1530/*****************************************************************************/ 2243/*****************************************************************************/
1531 2244
2245int ecb_cold
1532int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 2246eio_init (void (*want_poll)(void), void (*done_poll)(void))
1533{ 2247{
1534 return etp_init (want_poll, done_poll); 2248 return etp_init (want_poll, done_poll);
1535} 2249}
1536 2250
2251ecb_inline void
1537static void eio_api_destroy (eio_req *req) 2252eio_api_destroy (eio_req *req)
1538{ 2253{
1539 free (req); 2254 free (req);
1540} 2255}
1541 2256
1542#define REQ(rtype) \ 2257#define REQ(rtype) \
1543 eio_req *req; \ 2258 eio_req *req; \
1544 \ 2259 \
1545 req = (eio_req *)calloc (1, sizeof *req); \ 2260 req = (eio_req *)calloc (1, sizeof *req); \
1546 if (!req) \ 2261 if (!req) \
1547 return 0; \ 2262 return 0; \
… …
1561 { \ 2276 { \
1562 eio_api_destroy (req); \ 2277 eio_api_destroy (req); \
1563 return 0; \ 2278 return 0; \
1564 } 2279 }
1565 2280
2281#define SINGLEDOT(ptr) (0[(char *)(ptr)] == '.' && !1[(char *)(ptr)])
2282
2283static void
1566static void eio_execute (etp_worker *self, eio_req *req) 2284eio_execute (etp_worker *self, eio_req *req)
1567{ 2285{
1568 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 }
1569 2314
1570 switch (req->type) 2315 switch (req->type)
1571 { 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;
1572 case EIO_READ: ALLOC (req->size); 2324 case EIO_READ: ALLOC (req->size);
1573 req->result = req->offs >= 0 2325 req->result = req->offs >= 0
1574 ? pread (req->int1, req->ptr2, req->size, req->offs) 2326 ? pread (req->int1, req->ptr2, req->size, req->offs)
1575 : read (req->int1, req->ptr2, req->size); break; 2327 : read (req->int1, req->ptr2, req->size); break;
1576 case EIO_WRITE: req->result = req->offs >= 0 2328 case EIO_WRITE: req->result = req->offs >= 0
1577 ? pwrite (req->int1, req->ptr2, req->size, req->offs) 2329 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1578 : write (req->int1, req->ptr2, req->size); break; 2330 : write (req->int1, req->ptr2, req->size); break;
1579 2331
1580 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;
1581 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
1582 2336
1583 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2337 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1584 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;
1585 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2339 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1586 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;
1587 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2341 case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break;
1588 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; 2342 case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break;
2343 case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break;
2344 case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break;
1589 2345
2346 case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); 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;
2351 case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); 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;
2356 case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2, 0); break;
2357 case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
2358 case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
2359 case EIO_READLINK: ALLOC (PATH_MAX);
2360 req->result = readlinkat (dirfd, req->ptr1, req->ptr2, PATH_MAX); break;
1590 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS)); 2361 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1591 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break; 2362 req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1592 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1593 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1594
1595 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1596 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1597 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1598 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1599 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1600 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1601
1602 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
1603 case EIO_CLOSE: req->result = close (req->int1); break;
1604 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1605 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
1606 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1607 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1608 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1609 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1610 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1611 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break;
1612
1613 case EIO_READLINK: ALLOC (NAME_MAX);
1614 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
1615
1616 case EIO_SYNC: req->result = 0; sync (); break;
1617 case EIO_FSYNC: req->result = fsync (req->int1); break;
1618 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1619 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1620 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;
1621 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1622
1623 case EIO_READDIR: eio__scandir (req, self); break;
1624
1625 case EIO_BUSY: 2363 case EIO_UTIME:
1626#ifdef _WIN32 2364 case EIO_FUTIME:
1627 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
1628#else 2387#else
1629 {
1630 struct timeval tv;
1631 2388
1632 tv.tv_sec = req->nv1; 2389 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1633 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;
1634 2397
1635 req->result = select (0, 0, 0, 0, &tv); 2398 case EIO_UNLINK: req->result = unlink (path ); break;
1636 } 2399 case EIO_RMDIR: req->result = rmdir (path ); break;
1637#endif 2400 case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
1638 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;
1639 2409
1640 case EIO_UTIME: 2410 case EIO_UTIME:
1641 case EIO_FUTIME: 2411 case EIO_FUTIME:
1642 { 2412 {
1643 struct timeval tv[2]; 2413 struct timeval tv[2];
1644 struct timeval *times; 2414 struct timeval *times;
1645 2415
1646 if (req->nv1 != -1. || req->nv2 != -1.) 2416 if (req->nv1 != -1. || req->nv2 != -1.)
1647 { 2417 {
1648 tv[0].tv_sec = req->nv1; 2418 tv[0].tv_sec = req->nv1;
1649 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.; 2419 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
1650 tv[1].tv_sec = req->nv2; 2420 tv[1].tv_sec = req->nv2;
1651 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.; 2421 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
1652 2422
1653 times = tv; 2423 times = tv;
1654 } 2424 }
1655 else 2425 else
1656 times = 0; 2426 times = 0;
1657
1658 2427
1659 req->result = req->type == EIO_FUTIME 2428 req->result = req->type == EIO_FUTIME
1660 ? futimes (req->int1, times) 2429 ? futimes (req->int1, times)
1661 : utimes (req->ptr1, times); 2430 : utimes (req->ptr1, times);
1662 } 2431 }
1663 break; 2432 break;
1664 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
1665 case EIO_GROUP: 2483 case EIO_GROUP:
1666 abort (); /* handled in eio_request */ 2484 abort (); /* handled in eio_request */
1667 2485
1668 case EIO_NOP: 2486 case EIO_NOP:
1669 req->result = 0; 2487 req->result = 0;
1670 break; 2488 break;
1671 2489
1672 case EIO_CUSTOM: 2490 case EIO_CUSTOM:
1673 ((void (*)(eio_req *))req->feed) (req); 2491 req->feed (req);
1674 break; 2492 break;
1675 2493
1676 default: 2494 default:
1677 req->result = -1; 2495 req->result = EIO_ENOSYS ();
1678 break; 2496 break;
1679 } 2497 }
1680 2498
1681 req->errorno = errno; 2499 req->errorno = errno;
1682} 2500}
1683 2501
1684#ifndef EIO_NO_WRAPPERS 2502#ifndef EIO_NO_WRAPPERS
1685 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
1686eio_req *eio_nop (int pri, eio_cb cb, void *data) 2514eio_req *eio_nop (int pri, eio_cb cb, void *data)
1687{ 2515{
1688 REQ (EIO_NOP); SEND; 2516 REQ (EIO_NOP); SEND;
1689} 2517}
1690 2518
… …
1706eio_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)
1707{ 2535{
1708 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;
1709} 2537}
1710 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
1711eio_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)
1712{ 2555{
1713 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;
1714} 2557}
1715 2558
1716eio_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)
1717{ 2560{
1718 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;
1719} 2562}
1720 2563
1721eio_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)
1722{ 2565{
1723 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;
1724} 2572}
1725 2573
1726eio_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)
1727{ 2575{
1728 REQ (EIO_CLOSE); req->int1 = fd; SEND; 2576 REQ (EIO_CLOSE); req->int1 = fd; SEND;
… …
1731eio_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)
1732{ 2580{
1733 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;
1734} 2582}
1735 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
1736eio_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)
1737{ 2590{
1738 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;
1739} 2592}
1740 2593
… …
1766eio_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)
1767{ 2620{
1768 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2621 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1769} 2622}
1770 2623
1771eio_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)
1772{ 2625{
1773 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;
1774} 2627}
1775 2628
1776eio_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)
… …
1796eio_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)
1797{ 2650{
1798 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2651 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1799} 2652}
1800 2653
1801eio_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)
1802{ 2655{
1803 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;
1804} 2657}
1805 2658
1806eio_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)
… …
1822eio_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)
1823{ 2676{
1824 return eio__1path (EIO_READLINK, path, pri, cb, data); 2677 return eio__1path (EIO_READLINK, path, pri, cb, data);
1825} 2678}
1826 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
1827eio_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)
1828{ 2686{
1829 return eio__1path (EIO_STAT, path, pri, cb, data); 2687 return eio__1path (EIO_STAT, path, pri, cb, data);
1830} 2688}
1831 2689
… …
1854 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND; 2712 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1855} 2713}
1856 2714
1857eio_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)
1858{ 2716{
1859 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;
1860} 2718}
1861 2719
1862static eio_req * 2720static eio_req *
1863eio__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)
1864{ 2722{
… …
1888eio_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)
1889{ 2747{
1890 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2748 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1891} 2749}
1892 2750
1893eio_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)
1894{ 2752{
1895 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND; 2753 REQ (EIO_CUSTOM); req->feed = execute; SEND;
1896} 2754}
1897 2755
1898#endif 2756#endif
1899 2757
1900eio_req *eio_grp (eio_cb cb, void *data) 2758eio_req *eio_grp (eio_cb cb, void *data)
… …
1909#undef SEND 2767#undef SEND
1910 2768
1911/*****************************************************************************/ 2769/*****************************************************************************/
1912/* grp functions */ 2770/* grp functions */
1913 2771
2772void
1914void 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)
1915{ 2774{
1916 grp->int2 = limit; 2775 grp->int2 = limit;
1917 grp->feed = feed; 2776 grp->feed = feed;
1918 2777
1919 grp_try_feed (grp); 2778 grp_try_feed (grp);
1920} 2779}
1921 2780
2781void
1922void eio_grp_limit (eio_req *grp, int limit) 2782eio_grp_limit (eio_req *grp, int limit)
1923{ 2783{
1924 grp->int2 = limit; 2784 grp->int2 = limit;
1925 2785
1926 grp_try_feed (grp); 2786 grp_try_feed (grp);
1927} 2787}
1928 2788
2789void
1929void eio_grp_add (eio_req *grp, eio_req *req) 2790eio_grp_add (eio_req *grp, eio_req *req)
1930{ 2791{
1931 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));
1932 2793
1933 grp->flags |= EIO_FLAG_GROUPADD; 2794 grp->flags |= EIO_FLAG_GROUPADD;
1934 2795
… …
1945} 2806}
1946 2807
1947/*****************************************************************************/ 2808/*****************************************************************************/
1948/* misc garbage */ 2809/* misc garbage */
1949 2810
2811eio_ssize_t
1950ssize_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)
1951{ 2813{
1952 etp_worker wrk;
1953 ssize_t ret;
1954
1955 wrk.dbuf = 0;
1956
1957 ret = eio__sendfile (ofd, ifd, offset, count, &wrk); 2814 return eio__sendfile (ofd, ifd, offset, count);
1958
1959 if (wrk.dbuf)
1960 free (wrk.dbuf);
1961
1962 return ret;
1963} 2815}
1964 2816

Diff Legend

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