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

Comparing libeio/eio.c (file contents):
Revision 1.32 by root, Sat Jun 6 18:06:55 2009 UTC vs.
Revision 1.126 by root, Fri Dec 28 07:33:41 2012 UTC

1/* 1/*
2 * libeio implementation 2 * libeio implementation
3 * 3 *
4 * Copyright (c) 2007,2008,2009 Marc Alexander Lehmann <libeio@schmorp.de> 4 * Copyright (c) 2007,2008,2009,2010,2011,2012 Marc Alexander Lehmann <libeio@schmorp.de>
5 * All rights reserved. 5 * All rights reserved.
6 * 6 *
7 * Redistribution and use in source and binary forms, with or without modifica- 7 * Redistribution and use in source and binary forms, with or without modifica-
8 * tion, are permitted provided that the following conditions are met: 8 * tion, are permitted provided that the following conditions are met:
9 * 9 *
10 * 1. Redistributions of source code must retain the above copyright notice, 10 * 1. Redistributions of source code must retain the above copyright notice,
11 * this list of conditions and the following disclaimer. 11 * this list of conditions and the following disclaimer.
12 * 12 *
13 * 2. Redistributions in binary form must reproduce the above copyright 13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the 14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution. 15 * documentation and/or other materials provided with the distribution.
16 * 16 *
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR IMPLIED 17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR IMPLIED
18 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MER- 18 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MER-
19 * CHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO 19 * CHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO
20 * EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPE- 20 * EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPE-
21 * CIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, 21 * CIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
35 * and other provisions required by the GPL. If you do not delete the 35 * and other provisions required by the GPL. If you do not delete the
36 * provisions above, a recipient may use your version of this file under 36 * provisions above, a recipient may use your version of this file under
37 * either the BSD or the GPL. 37 * either the BSD or the GPL.
38 */ 38 */
39 39
40#ifndef _WIN32
41# include "config.h"
42#endif
43
40#include "eio.h" 44#include "eio.h"
45#include "ecb.h"
41 46
42#ifdef EIO_STACKSIZE 47#ifdef EIO_STACKSIZE
43# define XTHREAD_STACKSIZE EIO_STACKSIZE 48# define X_STACKSIZE EIO_STACKSIZE
44#endif 49#endif
45#include "xthread.h" 50#include "xthread.h"
46 51
47#include <errno.h> 52#include <errno.h>
48#include <stddef.h> 53#include <stddef.h>
53#include <sys/stat.h> 58#include <sys/stat.h>
54#include <limits.h> 59#include <limits.h>
55#include <fcntl.h> 60#include <fcntl.h>
56#include <assert.h> 61#include <assert.h>
57 62
63/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
64/* intptr_t only comes from stdint.h, says idiot openbsd coder */
65#if HAVE_STDINT_H
66# include <stdint.h>
67#endif
68
69#ifndef ECANCELED
70# define ECANCELED EDOM
71#endif
72#ifndef ELOOP
73# define ELOOP EDOM
74#endif
75
76#if !defined(ENOTSOCK) && defined(WSAENOTSOCK)
77# define ENOTSOCK WSAENOTSOCK
78#endif
79
80static void eio_destroy (eio_req *req);
81
58#ifndef EIO_FINISH 82#ifndef EIO_FINISH
59# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0 83# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
60#endif 84#endif
61 85
62#ifndef EIO_DESTROY 86#ifndef EIO_DESTROY
65 89
66#ifndef EIO_FEED 90#ifndef EIO_FEED
67# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0) 91# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)
68#endif 92#endif
69 93
94#ifndef EIO_FD_TO_WIN32_HANDLE
95# define EIO_FD_TO_WIN32_HANDLE(fd) _get_osfhandle (fd)
96#endif
97#ifndef EIO_WIN32_HANDLE_TO_FD
98# define EIO_WIN32_HANDLE_TO_FD(handle) _open_osfhandle (handle, 0)
99#endif
100
101#define EIO_ERRNO(errval,retval) ((errno = errval), retval)
102
103#define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1)
104
70#ifdef _WIN32 105#ifdef _WIN32
71 106
72 /*doh*/ 107 #undef PAGESIZE
108 #define PAGESIZE 4096 /* GetSystemInfo? */
109
110 /* TODO: look at how perl does stat (non-sloppy), unlink (ro-files), utime, link */
111
112 #ifdef EIO_STRUCT_STATI64
113 /* look at perl's non-sloppy stat */
114 #define stat(path,buf) _stati64 (path,buf)
115 #define fstat(fd,buf) _fstati64 (fd,buf)
116 #endif
117 #define lstat(path,buf) stat (path,buf)
118 #define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
119 #define mkdir(path,mode) _mkdir (path)
120 #define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
121
122 #define chmod(path,mode) _chmod (path, mode)
123 #define dup(fd) _dup (fd)
124 #define dup2(fd1,fd2) _dup2 (fd1, fd2)
125
126 #define fchmod(fd,mode) EIO_ENOSYS ()
127 #define chown(path,uid,gid) EIO_ENOSYS ()
128 #define fchown(fd,uid,gid) EIO_ENOSYS ()
129 #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
130 #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
131 #define mknod(path,mode,dev) EIO_ENOSYS ()
132 #define sync() EIO_ENOSYS ()
133 #define readlink(path,buf,s) EIO_ENOSYS ()
134 #define statvfs(path,buf) EIO_ENOSYS ()
135 #define fstatvfs(fd,buf) EIO_ENOSYS ()
136
137 #define pread(fd,buf,count,offset) eio__pread (fd, buf, count, offset)
138 #define pwrite(fd,buf,count,offset) eio__pwrite (fd, buf, count, offset)
139
140 #if __GNUC__
141 typedef long long eio_off_t; /* signed for compatibility to msvc */
142 #else
143 typedef __int64 eio_off_t; /* unsigned not supported by msvc */
144 #endif
145
146 static eio_ssize_t
147 eio__pread (int fd, void *buf, eio_ssize_t count, eio_off_t offset)
148 {
149 OVERLAPPED o = { 0 };
150 DWORD got;
151
152 o.Offset = offset;
153 o.OffsetHigh = offset >> 32;
154
155 return ReadFile ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd), buf, count, &got, &o)
156 ? got : -1;
157 }
158
159 static eio_ssize_t
160 eio__pwrite (int fd, void *buf, eio_ssize_t count, eio_off_t offset)
161 {
162 OVERLAPPED o = { 0 };
163 DWORD got;
164
165 o.Offset = offset;
166 o.OffsetHigh = offset >> 32;
167
168 return WriteFile ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd), buf, count, &got, &o)
169 ? got : -1;
170 }
171
172 /* rename() uses MoveFile, which fails to overwrite */
173 #define rename(old,neu) eio__rename (old, neu)
174
175 static int
176 eio__rename (const char *old, const char *neu)
177 {
178 if (MoveFileEx (old, neu, MOVEFILE_REPLACE_EXISTING))
179 return 0;
180
181 /* should steal _dosmaperr */
182 switch (GetLastError ())
183 {
184 case ERROR_FILE_NOT_FOUND:
185 case ERROR_PATH_NOT_FOUND:
186 case ERROR_INVALID_DRIVE:
187 case ERROR_NO_MORE_FILES:
188 case ERROR_BAD_NETPATH:
189 case ERROR_BAD_NET_NAME:
190 case ERROR_BAD_PATHNAME:
191 case ERROR_FILENAME_EXCED_RANGE:
192 errno = ENOENT;
193 break;
194
195 default:
196 errno = EACCES;
197 break;
198 }
199
200 return -1;
201 }
202
203 /* we could even stat and see if it exists */
204 static int
205 symlink (const char *old, const char *neu)
206 {
207 #if WINVER >= 0x0600
208 if (CreateSymbolicLink (neu, old, 1))
209 return 0;
210
211 if (CreateSymbolicLink (neu, old, 0))
212 return 0;
213 #endif
214
215 return EIO_ERRNO (ENOENT, -1);
216 }
217
218 /* POSIX API only */
219 #define CreateHardLink(neu,old,flags) 0
220 #define CreateSymbolicLink(neu,old,flags) 0
221
222 struct statvfs
223 {
224 int dummy;
225 };
226
227 #define DT_DIR EIO_DT_DIR
228 #define DT_REG EIO_DT_REG
229 #define D_NAME(entp) entp.cFileName
230 #define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
231
73#else 232#else
74 233
75# include "config.h"
76# include <sys/time.h> 234 #include <sys/time.h>
77# include <sys/select.h> 235 #include <sys/select.h>
236 #include <sys/statvfs.h>
78# include <unistd.h> 237 #include <unistd.h>
238 #include <signal.h>
239 #include <dirent.h>
240
241 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
242 #include <sys/mman.h>
243 #endif
244
245 #define D_NAME(entp) entp->d_name
246
247 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
248 #if __FreeBSD__ || __NetBSD__ || __OpenBSD__
249 #define _DIRENT_HAVE_D_TYPE /* sigh */
250 #define D_INO(de) (de)->d_fileno
251 #define D_NAMLEN(de) (de)->d_namlen
252 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
253 #define D_INO(de) (de)->d_ino
254 #endif
255
256 #ifdef _D_EXACT_NAMLEN
257 #undef D_NAMLEN
258 #define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
259 #endif
260
261 #ifdef _DIRENT_HAVE_D_TYPE
262 #define D_TYPE(de) (de)->d_type
263 #endif
264
265 #ifndef EIO_STRUCT_DIRENT
266 #define EIO_STRUCT_DIRENT struct dirent
267 #endif
268
269#endif
270
271#if HAVE_UTIMES
79# include <utime.h> 272# include <utime.h>
80# include <signal.h>
81# include <dirent.h>
82
83/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
84# if defined(__FreeBSD__) || defined(__NetBSD__) || defined(__OpenBSD__)
85# define D_INO(de) (de)->d_fileno
86# define _DIRENT_HAVE_D_TYPE /* sigh */
87# elif defined(__linux) || defined(d_ino) || _XOPEN_SOURCE >= 600
88# define D_INO(de) (de)->d_ino
89# endif 273#endif
90 274
91# ifdef _DIRENT_HAVE_D_TYPE 275#if HAVE_SYS_SYSCALL_H
92# define D_TYPE(de) (de)->d_type 276# include <sys/syscall.h>
93# endif 277#endif
94 278
95# ifndef EIO_STRUCT_DIRENT 279#if HAVE_SYS_PRCTL_H
96# define EIO_STRUCT_DIRENT struct dirent 280# include <sys/prctl.h>
97# endif
98
99#endif 281#endif
100 282
101#if HAVE_SENDFILE 283#if HAVE_SENDFILE
102# if __linux 284# if __linux
103# include <sys/sendfile.h> 285# include <sys/sendfile.h>
104# elif __freebsd 286# elif __FreeBSD__ || defined __APPLE__
105# include <sys/socket.h> 287# include <sys/socket.h>
106# include <sys/uio.h> 288# include <sys/uio.h>
107# elif __hpux 289# elif __hpux
108# include <sys/socket.h> 290# include <sys/socket.h>
109# elif __solaris /* not yet */ 291# elif __solaris
110# include <sys/sendfile.h> 292# include <sys/sendfile.h>
111# else 293# else
112# error sendfile support requested but not available 294# error sendfile support requested but not available
113# endif 295# endif
114#endif 296#endif
115 297
116#ifndef D_TYPE 298#ifndef D_TYPE
117# define D_TYPE(de) 0 299# define D_TYPE(de) 0
118#endif 300#endif
119
120#ifndef D_INO 301#ifndef D_INO
121# define D_INO(de) 0 302# define D_INO(de) 0
122#endif 303#endif
123 304#ifndef D_NAMLEN
124/* number of seconds after which an idle threads exit */ 305# define D_NAMLEN(entp) strlen (D_NAME (entp))
125#define IDLE_TIMEOUT 10 306#endif
126 307
127/* used for struct dirent, AIX doesn't provide it */ 308/* used for struct dirent, AIX doesn't provide it */
128#ifndef NAME_MAX 309#ifndef NAME_MAX
129# define NAME_MAX 4096 310# define NAME_MAX 4096
130#endif 311#endif
131 312
313/* used for readlink etc. */
314#ifndef PATH_MAX
315# define PATH_MAX 4096
316#endif
317
132/* buffer size for various temporary buffers */ 318/* buffer size for various temporary buffers */
133#define EIO_BUFSIZE 65536 319#define EIO_BUFSIZE 65536
134 320
135#define dBUF \ 321#define dBUF \
136 char *eio_buf; \
137 ETP_WORKER_LOCK (self); \
138 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 322 char *eio_buf = malloc (EIO_BUFSIZE); \
139 ETP_WORKER_UNLOCK (self); \
140 errno = ENOMEM; \ 323 errno = ENOMEM; \
141 if (!eio_buf) \ 324 if (!eio_buf) \
142 return -1; 325 return -1
326
327#define FUBd \
328 free (eio_buf)
143 329
144#define EIO_TICKS ((1000000 + 1023) >> 10) 330#define EIO_TICKS ((1000000 + 1023) >> 10)
145 331
146/*****************************************************************************/ 332/*****************************************************************************/
147 333
148#if __GNUC__ >= 3 334struct tmpbuf
149# define expect(expr,value) __builtin_expect ((expr),(value)) 335{
336 void *ptr;
337 int len;
338};
339
340static void *
341tmpbuf_get (struct tmpbuf *buf, int len)
342{
343 if (buf->len < len)
344 {
345 free (buf->ptr);
346 buf->ptr = malloc (buf->len = len);
347 }
348
349 return buf->ptr;
350}
351
352struct tmpbuf;
353
354#if _POSIX_VERSION >= 200809L
355 #define HAVE_AT 1
356 #define WD2FD(wd) ((wd) ? (wd)->fd : AT_FDCWD)
357 #ifndef O_SEARCH
358 #define O_SEARCH O_RDONLY
359 #endif
150#else 360#else
151# define expect(expr,value) (expr) 361 #define HAVE_AT 0
362 static const char *wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path);
152#endif 363#endif
153 364
154#define expect_false(expr) expect ((expr) != 0, 0) 365struct eio_pwd
155#define expect_true(expr) expect ((expr) != 0, 1) 366{
367#if HAVE_AT
368 int fd;
369#endif
370 int len;
371 char str[1]; /* actually, a 0-terminated canonical path */
372};
156 373
157/*****************************************************************************/ 374/*****************************************************************************/
158 375
159#define ETP_PRI_MIN EIO_PRI_MIN 376#define ETP_PRI_MIN EIO_PRI_MIN
160#define ETP_PRI_MAX EIO_PRI_MAX 377#define ETP_PRI_MAX EIO_PRI_MAX
166static int eio_finish (eio_req *req); 383static int eio_finish (eio_req *req);
167#define ETP_FINISH(req) eio_finish (req) 384#define ETP_FINISH(req) eio_finish (req)
168static void eio_execute (struct etp_worker *self, eio_req *req); 385static void eio_execute (struct etp_worker *self, eio_req *req);
169#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) 386#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
170 387
171#define ETP_WORKER_CLEAR(req) \
172 if (wrk->dbuf) \
173 { \
174 free (wrk->dbuf); \
175 wrk->dbuf = 0; \
176 } \
177 \
178 if (wrk->dirp) \
179 { \
180 closedir (wrk->dirp); \
181 wrk->dirp = 0; \
182 }
183
184#define ETP_WORKER_COMMON \
185 void *dbuf; \
186 DIR *dirp;
187
188/*****************************************************************************/ 388/*****************************************************************************/
189 389
190#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) 390#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
191 391
192/* calculcate time difference in ~1/EIO_TICKS of a second */ 392/* calculate time difference in ~1/EIO_TICKS of a second */
393ecb_inline int
193static int tvdiff (struct timeval *tv1, struct timeval *tv2) 394tvdiff (struct timeval *tv1, struct timeval *tv2)
194{ 395{
195 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 396 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
196 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 397 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
197} 398}
198 399
202static void (*done_poll_cb) (void); 403static void (*done_poll_cb) (void);
203 404
204static unsigned int max_poll_time; /* reslock */ 405static unsigned int max_poll_time; /* reslock */
205static unsigned int max_poll_reqs; /* reslock */ 406static unsigned int max_poll_reqs; /* reslock */
206 407
207static volatile unsigned int nreqs; /* reqlock */ 408static unsigned int nreqs; /* reqlock */
208static volatile unsigned int nready; /* reqlock */ 409static unsigned int nready; /* reqlock */
209static volatile unsigned int npending; /* reqlock */ 410static unsigned int npending; /* reqlock */
210static volatile unsigned int max_idle = 4; 411static unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
412static unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
211 413
212static mutex_t wrklock = X_MUTEX_INIT; 414static xmutex_t wrklock;
213static mutex_t reslock = X_MUTEX_INIT; 415static xmutex_t reslock;
214static mutex_t reqlock = X_MUTEX_INIT; 416static xmutex_t reqlock;
215static cond_t reqwait = X_COND_INIT; 417static xcond_t reqwait;
216
217#if !HAVE_PREADWRITE
218/*
219 * make our pread/pwrite emulation safe against themselves, but not against
220 * normal read/write by using a mutex. slows down execution a lot,
221 * but that's your problem, not mine.
222 */
223static mutex_t preadwritelock = X_MUTEX_INIT;
224#endif
225 418
226typedef struct etp_worker 419typedef struct etp_worker
227{ 420{
421 struct tmpbuf tmpbuf;
422
228 /* locked by wrklock */ 423 /* locked by wrklock */
229 struct etp_worker *prev, *next; 424 struct etp_worker *prev, *next;
230 425
231 thread_t tid; 426 xthread_t tid;
232 427
233 /* locked by reslock, reqlock or wrklock */ 428#ifdef ETP_WORKER_COMMON
234 ETP_REQ *req; /* currently processed request */
235
236 ETP_WORKER_COMMON 429 ETP_WORKER_COMMON
430#endif
237} etp_worker; 431} etp_worker;
238 432
239static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */ 433static etp_worker wrk_first; /* NOT etp */
240 434
241#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock) 435#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
242#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock) 436#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
243 437
244/* worker threads management */ 438/* worker threads management */
245 439
440static void
246static void etp_worker_clear (etp_worker *wrk) 441etp_worker_clear (etp_worker *wrk)
247{ 442{
248 ETP_WORKER_CLEAR (wrk);
249} 443}
250 444
445static void ecb_cold
251static void etp_worker_free (etp_worker *wrk) 446etp_worker_free (etp_worker *wrk)
252{ 447{
448 free (wrk->tmpbuf.ptr);
449
253 wrk->next->prev = wrk->prev; 450 wrk->next->prev = wrk->prev;
254 wrk->prev->next = wrk->next; 451 wrk->prev->next = wrk->next;
255 452
256 free (wrk); 453 free (wrk);
257} 454}
258 455
259static unsigned int etp_nreqs (void) 456static unsigned int
457etp_nreqs (void)
260{ 458{
261 int retval; 459 int retval;
262 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 460 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
263 retval = nreqs; 461 retval = nreqs;
264 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 462 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
265 return retval; 463 return retval;
266} 464}
267 465
268static unsigned int etp_nready (void) 466static unsigned int
467etp_nready (void)
269{ 468{
270 unsigned int retval; 469 unsigned int retval;
271 470
272 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 471 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
273 retval = nready; 472 retval = nready;
274 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 473 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
275 474
276 return retval; 475 return retval;
277} 476}
278 477
279static unsigned int etp_npending (void) 478static unsigned int
479etp_npending (void)
280{ 480{
281 unsigned int retval; 481 unsigned int retval;
282 482
283 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 483 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
284 retval = npending; 484 retval = npending;
285 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 485 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
286 486
287 return retval; 487 return retval;
288} 488}
289 489
290static unsigned int etp_nthreads (void) 490static unsigned int
491etp_nthreads (void)
291{ 492{
292 unsigned int retval; 493 unsigned int retval;
293 494
294 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 495 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
295 retval = started; 496 retval = started;
309} etp_reqq; 510} etp_reqq;
310 511
311static etp_reqq req_queue; 512static etp_reqq req_queue;
312static etp_reqq res_queue; 513static etp_reqq res_queue;
313 514
515static void ecb_noinline ecb_cold
516reqq_init (etp_reqq *q)
517{
518 int pri;
519
520 for (pri = 0; pri < ETP_NUM_PRI; ++pri)
521 q->qs[pri] = q->qe[pri] = 0;
522
523 q->size = 0;
524}
525
526static int ecb_noinline
314static int reqq_push (etp_reqq *q, ETP_REQ *req) 527reqq_push (etp_reqq *q, ETP_REQ *req)
315{ 528{
316 int pri = req->pri; 529 int pri = req->pri;
317 req->next = 0; 530 req->next = 0;
318 531
319 if (q->qe[pri]) 532 if (q->qe[pri])
325 q->qe[pri] = q->qs[pri] = req; 538 q->qe[pri] = q->qs[pri] = req;
326 539
327 return q->size++; 540 return q->size++;
328} 541}
329 542
543static ETP_REQ * ecb_noinline
330static ETP_REQ *reqq_shift (etp_reqq *q) 544reqq_shift (etp_reqq *q)
331{ 545{
332 int pri; 546 int pri;
333 547
334 if (!q->size) 548 if (!q->size)
335 return 0; 549 return 0;
350 } 564 }
351 565
352 abort (); 566 abort ();
353} 567}
354 568
355static void etp_atfork_prepare (void) 569static int ecb_cold
570etp_init (void (*want_poll)(void), void (*done_poll)(void))
356{ 571{
357 X_LOCK (wrklock); 572 X_MUTEX_CREATE (wrklock);
358 X_LOCK (reqlock); 573 X_MUTEX_CREATE (reslock);
359 X_LOCK (reslock); 574 X_MUTEX_CREATE (reqlock);
360#if !HAVE_PREADWRITE 575 X_COND_CREATE (reqwait);
361 X_LOCK (preadwritelock);
362#endif
363}
364 576
365static void etp_atfork_parent (void) 577 reqq_init (&req_queue);
366{ 578 reqq_init (&res_queue);
367#if !HAVE_PREADWRITE
368 X_UNLOCK (preadwritelock);
369#endif
370 X_UNLOCK (reslock);
371 X_UNLOCK (reqlock);
372 X_UNLOCK (wrklock);
373}
374 579
375static void etp_atfork_child (void) 580 wrk_first.next =
376{ 581 wrk_first.prev = &wrk_first;
377 ETP_REQ *prv;
378
379 while ((prv = reqq_shift (&req_queue)))
380 ETP_DESTROY (prv);
381
382 while ((prv = reqq_shift (&res_queue)))
383 ETP_DESTROY (prv);
384
385 while (wrk_first.next != &wrk_first)
386 {
387 etp_worker *wrk = wrk_first.next;
388
389 if (wrk->req)
390 ETP_DESTROY (wrk->req);
391
392 etp_worker_clear (wrk);
393 etp_worker_free (wrk);
394 }
395 582
396 started = 0; 583 started = 0;
397 idle = 0; 584 idle = 0;
398 nreqs = 0; 585 nreqs = 0;
399 nready = 0; 586 nready = 0;
400 npending = 0; 587 npending = 0;
401 588
402 etp_atfork_parent ();
403}
404
405static void
406etp_once_init (void)
407{
408 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
409}
410
411static int
412etp_init (void (*want_poll)(void), void (*done_poll)(void))
413{
414 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
415
416 pthread_once (&doinit, etp_once_init);
417
418 want_poll_cb = want_poll; 589 want_poll_cb = want_poll;
419 done_poll_cb = done_poll; 590 done_poll_cb = done_poll;
420 591
421 return 0; 592 return 0;
422} 593}
423 594
424X_THREAD_PROC (etp_proc); 595X_THREAD_PROC (etp_proc);
425 596
597static void ecb_cold
426static void etp_start_thread (void) 598etp_start_thread (void)
427{ 599{
428 etp_worker *wrk = calloc (1, sizeof (etp_worker)); 600 etp_worker *wrk = calloc (1, sizeof (etp_worker));
429 601
430 /*TODO*/ 602 /*TODO*/
431 assert (("unable to allocate worker thread data", wrk)); 603 assert (("unable to allocate worker thread data", wrk));
432 604
433 X_LOCK (wrklock); 605 X_LOCK (wrklock);
434 606
435 if (thread_create (&wrk->tid, etp_proc, (void *)wrk)) 607 if (xthread_create (&wrk->tid, etp_proc, (void *)wrk))
436 { 608 {
437 wrk->prev = &wrk_first; 609 wrk->prev = &wrk_first;
438 wrk->next = wrk_first.next; 610 wrk->next = wrk_first.next;
439 wrk_first.next->prev = wrk; 611 wrk_first.next->prev = wrk;
440 wrk_first.next = wrk; 612 wrk_first.next = wrk;
444 free (wrk); 616 free (wrk);
445 617
446 X_UNLOCK (wrklock); 618 X_UNLOCK (wrklock);
447} 619}
448 620
621static void
449static void etp_maybe_start_thread (void) 622etp_maybe_start_thread (void)
450{ 623{
451 if (expect_true (etp_nthreads () >= wanted)) 624 if (ecb_expect_true (etp_nthreads () >= wanted))
452 return; 625 return;
453 626
454 /* todo: maybe use idle here, but might be less exact */ 627 /* todo: maybe use idle here, but might be less exact */
455 if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ())) 628 if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
456 return; 629 return;
457 630
458 etp_start_thread (); 631 etp_start_thread ();
459} 632}
460 633
634static void ecb_cold
461static void etp_end_thread (void) 635etp_end_thread (void)
462{ 636{
463 eio_req *req = calloc (1, sizeof (eio_req)); 637 eio_req *req = calloc (1, sizeof (eio_req)); /* will be freed by worker */
464 638
465 req->type = -1; 639 req->type = -1;
466 req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 640 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
467 641
468 X_LOCK (reqlock); 642 X_LOCK (reqlock);
473 X_LOCK (wrklock); 647 X_LOCK (wrklock);
474 --started; 648 --started;
475 X_UNLOCK (wrklock); 649 X_UNLOCK (wrklock);
476} 650}
477 651
478static int etp_poll (void) 652static int
653etp_poll (void)
479{ 654{
480 unsigned int maxreqs; 655 unsigned int maxreqs;
481 unsigned int maxtime; 656 unsigned int maxtime;
482 struct timeval tv_start, tv_now; 657 struct timeval tv_start, tv_now;
483 658
513 688
514 X_LOCK (reqlock); 689 X_LOCK (reqlock);
515 --nreqs; 690 --nreqs;
516 X_UNLOCK (reqlock); 691 X_UNLOCK (reqlock);
517 692
518 if (expect_false (req->type == EIO_GROUP && req->size)) 693 if (ecb_expect_false (req->type == EIO_GROUP && req->size))
519 { 694 {
520 req->int1 = 1; /* mark request as delayed */ 695 req->int1 = 1; /* mark request as delayed */
521 continue; 696 continue;
522 } 697 }
523 else 698 else
524 { 699 {
525 int res = ETP_FINISH (req); 700 int res = ETP_FINISH (req);
526 if (expect_false (res)) 701 if (ecb_expect_false (res))
527 return res; 702 return res;
528 } 703 }
529 704
530 if (expect_false (maxreqs && !--maxreqs)) 705 if (ecb_expect_false (maxreqs && !--maxreqs))
531 break; 706 break;
532 707
533 if (maxtime) 708 if (maxtime)
534 { 709 {
535 gettimeofday (&tv_now, 0); 710 gettimeofday (&tv_now, 0);
541 716
542 errno = EAGAIN; 717 errno = EAGAIN;
543 return -1; 718 return -1;
544} 719}
545 720
721static void
546static void etp_cancel (ETP_REQ *req) 722etp_cancel (ETP_REQ *req)
547{ 723{
548 X_LOCK (wrklock); 724 req->cancelled = 1;
549 req->flags |= EIO_FLAG_CANCELLED;
550 X_UNLOCK (wrklock);
551 725
552 eio_grp_cancel (req); 726 eio_grp_cancel (req);
553} 727}
554 728
729static void
555static void etp_submit (ETP_REQ *req) 730etp_submit (ETP_REQ *req)
556{ 731{
557 req->pri -= ETP_PRI_MIN; 732 req->pri -= ETP_PRI_MIN;
558 733
559 if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN; 734 if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
560 if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 735 if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
561 736
562 if (expect_false (req->type == EIO_GROUP)) 737 if (ecb_expect_false (req->type == EIO_GROUP))
563 { 738 {
564 /* I hope this is worth it :/ */ 739 /* I hope this is worth it :/ */
565 X_LOCK (reqlock); 740 X_LOCK (reqlock);
566 ++nreqs; 741 ++nreqs;
567 X_UNLOCK (reqlock); 742 X_UNLOCK (reqlock);
586 761
587 etp_maybe_start_thread (); 762 etp_maybe_start_thread ();
588 } 763 }
589} 764}
590 765
766static void ecb_cold
591static void etp_set_max_poll_time (double nseconds) 767etp_set_max_poll_time (double nseconds)
592{ 768{
593 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 769 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
594 max_poll_time = nseconds; 770 max_poll_time = nseconds * EIO_TICKS;
595 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 771 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
596} 772}
597 773
774static void ecb_cold
598static void etp_set_max_poll_reqs (unsigned int maxreqs) 775etp_set_max_poll_reqs (unsigned int maxreqs)
599{ 776{
600 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 777 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
601 max_poll_reqs = maxreqs; 778 max_poll_reqs = maxreqs;
602 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 779 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
603} 780}
604 781
782static void ecb_cold
605static void etp_set_max_idle (unsigned int nthreads) 783etp_set_max_idle (unsigned int nthreads)
606{ 784{
607 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 785 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
608 max_idle = nthreads <= 0 ? 1 : nthreads; 786 max_idle = nthreads;
609 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 787 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
610} 788}
611 789
790static void ecb_cold
791etp_set_idle_timeout (unsigned int seconds)
792{
793 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
794 idle_timeout = seconds;
795 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
796}
797
798static void ecb_cold
612static void etp_set_min_parallel (unsigned int nthreads) 799etp_set_min_parallel (unsigned int nthreads)
613{ 800{
614 if (wanted < nthreads) 801 if (wanted < nthreads)
615 wanted = nthreads; 802 wanted = nthreads;
616} 803}
617 804
805static void ecb_cold
618static void etp_set_max_parallel (unsigned int nthreads) 806etp_set_max_parallel (unsigned int nthreads)
619{ 807{
620 if (wanted > nthreads) 808 if (wanted > nthreads)
621 wanted = nthreads; 809 wanted = nthreads;
622 810
623 while (started > wanted) 811 while (started > wanted)
624 etp_end_thread (); 812 etp_end_thread ();
625} 813}
626 814
627/*****************************************************************************/ 815/*****************************************************************************/
628 816
817static void
629static void grp_try_feed (eio_req *grp) 818grp_try_feed (eio_req *grp)
630{ 819{
631 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 820 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
632 { 821 {
633 grp->flags &= ~EIO_FLAG_GROUPADD; 822 grp->flags &= ~EIO_FLAG_GROUPADD;
634 823
641 break; 830 break;
642 } 831 }
643 } 832 }
644} 833}
645 834
835static int
646static int grp_dec (eio_req *grp) 836grp_dec (eio_req *grp)
647{ 837{
648 --grp->size; 838 --grp->size;
649 839
650 /* call feeder, if applicable */ 840 /* call feeder, if applicable */
651 grp_try_feed (grp); 841 grp_try_feed (grp);
655 return eio_finish (grp); 845 return eio_finish (grp);
656 else 846 else
657 return 0; 847 return 0;
658} 848}
659 849
850static void
660void eio_destroy (eio_req *req) 851eio_destroy (eio_req *req)
661{ 852{
662 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); 853 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
663 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); 854 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
664 855
665 EIO_DESTROY (req); 856 EIO_DESTROY (req);
666} 857}
667 858
859static int
668static int eio_finish (eio_req *req) 860eio_finish (eio_req *req)
669{ 861{
670 int res = EIO_FINISH (req); 862 int res = EIO_FINISH (req);
671 863
672 if (req->grp) 864 if (req->grp)
673 { 865 {
681 if (grp->grp_first == req) 873 if (grp->grp_first == req)
682 grp->grp_first = req->grp_next; 874 grp->grp_first = req->grp_next;
683 875
684 res2 = grp_dec (grp); 876 res2 = grp_dec (grp);
685 877
686 if (!res && res2) 878 if (!res)
687 res = res2; 879 res = res2;
688 } 880 }
689 881
690 eio_destroy (req); 882 eio_destroy (req);
691 883
692 return res; 884 return res;
693} 885}
694 886
887void
695void eio_grp_cancel (eio_req *grp) 888eio_grp_cancel (eio_req *grp)
696{ 889{
697 for (grp = grp->grp_first; grp; grp = grp->grp_next) 890 for (grp = grp->grp_first; grp; grp = grp->grp_next)
698 eio_cancel (grp); 891 eio_cancel (grp);
699} 892}
700 893
894void
701void eio_cancel (eio_req *req) 895eio_cancel (eio_req *req)
702{ 896{
703 etp_cancel (req); 897 etp_cancel (req);
704} 898}
705 899
900void
706void eio_submit (eio_req *req) 901eio_submit (eio_req *req)
707{ 902{
708 etp_submit (req); 903 etp_submit (req);
709} 904}
710 905
711unsigned int eio_nreqs (void) 906unsigned int
907eio_nreqs (void)
712{ 908{
713 return etp_nreqs (); 909 return etp_nreqs ();
714} 910}
715 911
716unsigned int eio_nready (void) 912unsigned int
913eio_nready (void)
717{ 914{
718 return etp_nready (); 915 return etp_nready ();
719} 916}
720 917
721unsigned int eio_npending (void) 918unsigned int
919eio_npending (void)
722{ 920{
723 return etp_npending (); 921 return etp_npending ();
724} 922}
725 923
726unsigned int eio_nthreads (void) 924unsigned int ecb_cold
925eio_nthreads (void)
727{ 926{
728 return etp_nthreads (); 927 return etp_nthreads ();
729} 928}
730 929
930void ecb_cold
731void eio_set_max_poll_time (double nseconds) 931eio_set_max_poll_time (double nseconds)
732{ 932{
733 etp_set_max_poll_time (nseconds); 933 etp_set_max_poll_time (nseconds);
734} 934}
735 935
936void ecb_cold
736void eio_set_max_poll_reqs (unsigned int maxreqs) 937eio_set_max_poll_reqs (unsigned int maxreqs)
737{ 938{
738 etp_set_max_poll_reqs (maxreqs); 939 etp_set_max_poll_reqs (maxreqs);
739} 940}
740 941
942void ecb_cold
741void eio_set_max_idle (unsigned int nthreads) 943eio_set_max_idle (unsigned int nthreads)
742{ 944{
743 etp_set_max_idle (nthreads); 945 etp_set_max_idle (nthreads);
744} 946}
745 947
948void ecb_cold
949eio_set_idle_timeout (unsigned int seconds)
950{
951 etp_set_idle_timeout (seconds);
952}
953
954void ecb_cold
746void eio_set_min_parallel (unsigned int nthreads) 955eio_set_min_parallel (unsigned int nthreads)
747{ 956{
748 etp_set_min_parallel (nthreads); 957 etp_set_min_parallel (nthreads);
749} 958}
750 959
960void ecb_cold
751void eio_set_max_parallel (unsigned int nthreads) 961eio_set_max_parallel (unsigned int nthreads)
752{ 962{
753 etp_set_max_parallel (nthreads); 963 etp_set_max_parallel (nthreads);
754} 964}
755 965
756int eio_poll (void) 966int eio_poll (void)
759} 969}
760 970
761/*****************************************************************************/ 971/*****************************************************************************/
762/* work around various missing functions */ 972/* work around various missing functions */
763 973
764#if !HAVE_PREADWRITE
765# undef pread
766# undef pwrite
767# define pread eio__pread
768# define pwrite eio__pwrite
769
770static ssize_t
771eio__pread (int fd, void *buf, size_t count, off_t offset)
772{
773 ssize_t res;
774 off_t ooffset;
775
776 X_LOCK (preadwritelock);
777 ooffset = lseek (fd, 0, SEEK_CUR);
778 lseek (fd, offset, SEEK_SET);
779 res = read (fd, buf, count);
780 lseek (fd, ooffset, SEEK_SET);
781 X_UNLOCK (preadwritelock);
782
783 return res;
784}
785
786static ssize_t
787eio__pwrite (int fd, void *buf, size_t count, off_t offset)
788{
789 ssize_t res;
790 off_t ooffset;
791
792 X_LOCK (preadwritelock);
793 ooffset = lseek (fd, 0, SEEK_CUR);
794 lseek (fd, offset, SEEK_SET);
795 res = write (fd, buf, count);
796 lseek (fd, ooffset, SEEK_SET);
797 X_UNLOCK (preadwritelock);
798
799 return res;
800}
801#endif
802
803#ifndef HAVE_FUTIMES 974#ifndef HAVE_UTIMES
804 975
805# undef utimes 976# undef utimes
806# undef futimes
807# define utimes(path,times) eio__utimes (path, times) 977# define utimes(path,times) eio__utimes (path, times)
808# define futimes(fd,times) eio__futimes (fd, times)
809 978
810static int 979static int
811eio__utimes (const char *filename, const struct timeval times[2]) 980eio__utimes (const char *filename, const struct timeval times[2])
812{ 981{
813 if (times) 982 if (times)
821 } 990 }
822 else 991 else
823 return utime (filename, 0); 992 return utime (filename, 0);
824} 993}
825 994
995#endif
996
997#ifndef HAVE_FUTIMES
998
999# undef futimes
1000# define futimes(fd,times) eio__futimes (fd, times)
1001
1002static int
826static int eio__futimes (int fd, const struct timeval tv[2]) 1003eio__futimes (int fd, const struct timeval tv[2])
827{ 1004{
828 errno = ENOSYS; 1005 errno = ENOSYS;
829 return -1; 1006 return -1;
830} 1007}
831 1008
834#if !HAVE_FDATASYNC 1011#if !HAVE_FDATASYNC
835# undef fdatasync 1012# undef fdatasync
836# define fdatasync(fd) fsync (fd) 1013# define fdatasync(fd) fsync (fd)
837#endif 1014#endif
838 1015
1016static int
1017eio__syncfs (int fd)
1018{
1019 int res;
1020
1021#if HAVE_SYS_SYNCFS
1022 res = (int)syscall (__NR_syncfs, (int)(fd));
1023#else
1024 res = EIO_ENOSYS ();
1025#endif
1026
1027 if (res < 0 && errno == ENOSYS && fd >= 0)
1028 sync ();
1029
1030 return res;
1031}
1032
839/* sync_file_range always needs emulation */ 1033/* sync_file_range always needs emulation */
840int 1034static int
841eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags) 1035eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
842{ 1036{
843#if HAVE_SYNC_FILE_RANGE 1037#if HAVE_SYNC_FILE_RANGE
844 int res; 1038 int res;
845 1039
858 if (!res || errno != ENOSYS) 1052 if (!res || errno != ENOSYS)
859 return res; 1053 return res;
860#endif 1054#endif
861 1055
862 /* even though we could play tricks with the flags, it's better to always 1056 /* even though we could play tricks with the flags, it's better to always
863 * call fdatasync, as thta matches the expectation of it's users best */ 1057 * call fdatasync, as that matches the expectation of its users best */
864 return fdatasync (fd); 1058 return fdatasync (fd);
1059}
1060
1061static int
1062eio__fallocate (int fd, int mode, off_t offset, size_t len)
1063{
1064#if HAVE_LINUX_FALLOCATE
1065 return fallocate (fd, mode, offset, len);
1066#else
1067 return EIO_ENOSYS ();
1068#endif
865} 1069}
866 1070
867#if !HAVE_READAHEAD 1071#if !HAVE_READAHEAD
868# undef readahead 1072# undef readahead
869# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 1073# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
870 1074
871static ssize_t 1075static eio_ssize_t
872eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) 1076eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
873{ 1077{
874 size_t todo = count; 1078 size_t todo = count;
875 dBUF; 1079 dBUF;
876 1080
881 pread (fd, eio_buf, len, offset); 1085 pread (fd, eio_buf, len, offset);
882 offset += len; 1086 offset += len;
883 todo -= len; 1087 todo -= len;
884 } 1088 }
885 1089
886 errno = 0; 1090 FUBd;
1091
1092 /* linux's readahead basically only fails for EBADF or EINVAL (not mmappable) */
1093 /* but not for e.g. EIO or eof, so we also never fail */
887 return count; 1094 return 0;
888} 1095}
889 1096
890#endif 1097#endif
891 1098
892/* sendfile always needs emulation */ 1099/* sendfile always needs emulation */
893static ssize_t 1100static eio_ssize_t
894eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) 1101eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
895{ 1102{
1103 eio_ssize_t written = 0;
896 ssize_t res; 1104 eio_ssize_t res;
897 1105
898 if (!count) 1106 if (!count)
899 return 0; 1107 return 0;
900 1108
1109 for (;;)
1110 {
1111#ifdef __APPLE__
1112# undef HAVE_SENDFILE /* broken, as everything on os x */
1113#endif
901#if HAVE_SENDFILE 1114#if HAVE_SENDFILE
902# if __linux 1115# if __linux
1116 off_t soffset = offset;
903 res = sendfile (ofd, ifd, &offset, count); 1117 res = sendfile (ofd, ifd, &soffset, count);
904 1118
905# elif __freebsd 1119# elif __FreeBSD__
906 /* 1120 /*
907 * Of course, the freebsd sendfile is a dire hack with no thoughts 1121 * Of course, the freebsd sendfile is a dire hack with no thoughts
908 * wasted on making it similar to other I/O functions. 1122 * wasted on making it similar to other I/O functions.
909 */ 1123 */
910 {
911 off_t sbytes; 1124 off_t sbytes;
912 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 1125 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
913 1126
914 if (res < 0 && sbytes) 1127 #if 0 /* according to the manpage, this is correct, but broken behaviour */
915 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 1128 /* freebsd' sendfile will return 0 on success */
1129 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
1130 /* not on e.g. EIO or EPIPE - sounds broken */
1131 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
916 res = sbytes; 1132 res = sbytes;
917 } 1133 #endif
1134
1135 /* according to source inspection, this is correct, and useful behaviour */
1136 if (sbytes)
1137 res = sbytes;
1138
1139# elif defined __APPLE__
1140 off_t sbytes = count;
1141 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
1142
1143 /* according to the manpage, sbytes is always valid */
1144 if (sbytes)
1145 res = sbytes;
918 1146
919# elif __hpux 1147# elif __hpux
920 res = sendfile (ofd, ifd, offset, count, 0, 0); 1148 res = sendfile (ofd, ifd, offset, count, 0, 0);
921 1149
922# elif __solaris 1150# elif __solaris
923 {
924 struct sendfilevec vec; 1151 struct sendfilevec vec;
925 size_t sbytes; 1152 size_t sbytes;
926 1153
927 vec.sfv_fd = ifd; 1154 vec.sfv_fd = ifd;
928 vec.sfv_flag = 0; 1155 vec.sfv_flag = 0;
929 vec.sfv_off = offset; 1156 vec.sfv_off = offset;
930 vec.sfv_len = count; 1157 vec.sfv_len = count;
931 1158
932 res = sendfilev (ofd, &vec, 1, &sbytes); 1159 res = sendfilev (ofd, &vec, 1, &sbytes);
933 1160
934 if (res < 0 && sbytes) 1161 if (res < 0 && sbytes)
935 res = sbytes; 1162 res = sbytes;
936 }
937 1163
938# endif 1164# endif
1165
1166#elif defined (_WIN32) && 0
1167 /* does not work, just for documentation of what would need to be done */
1168 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
1169 /* libeio guarantees that the file offset does not change, and windows */
1170 /* has no way to get an independent handle to the same file description */
1171 HANDLE h = TO_SOCKET (ifd);
1172 SetFilePointer (h, offset, 0, FILE_BEGIN);
1173 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1174
939#else 1175#else
940 res = -1; 1176 res = EIO_ENOSYS ();
941 errno = ENOSYS;
942#endif 1177#endif
943 1178
1179 /* we assume sendfile can copy at least 128mb in one go */
1180 if (res <= 128 * 1024 * 1024)
1181 {
1182 if (res > 0)
1183 written += res;
1184
1185 if (written)
1186 return written;
1187
1188 break;
1189 }
1190 else
1191 {
1192 /* if we requested more, then probably the kernel was lazy */
1193 written += res;
1194 offset += res;
1195 count -= res;
1196
1197 if (!count)
1198 return written;
1199 }
1200 }
1201
944 if (res < 0 1202 if (res < 0
945 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1203 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1204 /* BSDs */
1205#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1206 || errno == ENOTSUP
1207#endif
1208#ifdef EOPNOTSUPP /* windows */
1209 || errno == EOPNOTSUPP /* BSDs */
1210#endif
946#if __solaris 1211#if __solaris
947 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1212 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
948#endif 1213#endif
949 ) 1214 )
950 ) 1215 )
954 1219
955 res = 0; 1220 res = 0;
956 1221
957 while (count) 1222 while (count)
958 { 1223 {
959 ssize_t cnt; 1224 eio_ssize_t cnt;
960 1225
961 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 1226 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
962 1227
963 if (cnt <= 0) 1228 if (cnt <= 0)
964 { 1229 {
976 1241
977 offset += cnt; 1242 offset += cnt;
978 res += cnt; 1243 res += cnt;
979 count -= cnt; 1244 count -= cnt;
980 } 1245 }
1246
1247 FUBd;
981 } 1248 }
982 1249
983 return res; 1250 return res;
984} 1251}
985 1252
1253#ifdef PAGESIZE
1254# define eio_pagesize() PAGESIZE
1255#else
1256static intptr_t
1257eio_pagesize (void)
1258{
1259 static intptr_t page;
1260
1261 if (!page)
1262 page = sysconf (_SC_PAGESIZE);
1263
1264 return page;
1265}
1266#endif
1267
1268static void
1269eio_page_align (void **addr, size_t *length)
1270{
1271 intptr_t mask = eio_pagesize () - 1;
1272
1273 /* round down addr */
1274 intptr_t adj = mask & (intptr_t)*addr;
1275
1276 *addr = (void *)((intptr_t)*addr - adj);
1277 *length += adj;
1278
1279 /* round up length */
1280 *length = (*length + mask) & ~mask;
1281}
1282
1283#if !_POSIX_MEMLOCK
1284# define eio__mlockall(a) EIO_ENOSYS ()
1285#else
1286
986static int 1287static int
1288eio__mlockall (int flags)
1289{
1290 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1291 extern int mallopt (int, int);
1292 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1293 #endif
1294
1295 if (EIO_MCL_CURRENT != MCL_CURRENT
1296 || EIO_MCL_FUTURE != MCL_FUTURE)
1297 {
1298 flags = 0
1299 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1300 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1301 }
1302
1303 return mlockall (flags);
1304}
1305#endif
1306
1307#if !_POSIX_MEMLOCK_RANGE
1308# define eio__mlock(a,b) EIO_ENOSYS ()
1309#else
1310
1311static int
1312eio__mlock (void *addr, size_t length)
1313{
1314 eio_page_align (&addr, &length);
1315
1316 return mlock (addr, length);
1317}
1318
1319#endif
1320
1321#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1322# define eio__msync(a,b,c) EIO_ENOSYS ()
1323#else
1324
1325static int
1326eio__msync (void *mem, size_t len, int flags)
1327{
1328 eio_page_align (&mem, &len);
1329
1330 if (EIO_MS_ASYNC != MS_SYNC
1331 || EIO_MS_INVALIDATE != MS_INVALIDATE
1332 || EIO_MS_SYNC != MS_SYNC)
1333 {
1334 flags = 0
1335 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1336 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1337 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1338 }
1339
1340 return msync (mem, len, flags);
1341}
1342
1343#endif
1344
1345static int
1346eio__mtouch (eio_req *req)
1347{
1348 void *mem = req->ptr2;
1349 size_t len = req->size;
1350 int flags = req->int1;
1351
1352 eio_page_align (&mem, &len);
1353
1354 {
1355 intptr_t addr = (intptr_t)mem;
1356 intptr_t end = addr + len;
1357 intptr_t page = eio_pagesize ();
1358
1359 if (addr < end)
1360 if (flags & EIO_MT_MODIFY) /* modify */
1361 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1362 else
1363 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1364 }
1365
1366 return 0;
1367}
1368
1369/*****************************************************************************/
1370/* requests implemented outside eio_execute, because they are so large */
1371
1372static void
1373eio__lseek (eio_req *req)
1374{
1375 /* this usually gets optimised away completely, or your compiler sucks, */
1376 /* or the whence constants really are not 0, 1, 2 */
1377 int whence = req->int2 == EIO_SEEK_SET ? SEEK_SET
1378 : req->int2 == EIO_SEEK_CUR ? SEEK_CUR
1379 : req->int2 == EIO_SEEK_END ? SEEK_END
1380 : req->int2;
1381
1382 req->offs = lseek (req->int1, req->offs, whence);
1383 req->result = req->offs == (off_t)-1 ? -1 : 0;
1384}
1385
1386/* result will always end up in tmpbuf, there is always space for adding a 0-byte */
1387static int
1388eio__realpath (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
1389{
1390 const char *rel = path;
1391 char *res;
1392 char *tmp1, *tmp2;
1393#if SYMLOOP_MAX > 32
1394 int symlinks = SYMLOOP_MAX;
1395#else
1396 int symlinks = 32;
1397#endif
1398
1399 errno = EINVAL;
1400 if (!rel)
1401 return -1;
1402
1403 errno = ENOENT;
1404 if (!*rel)
1405 return -1;
1406
1407 res = tmpbuf_get (tmpbuf, PATH_MAX * 3);
1408 tmp1 = res + PATH_MAX;
1409 tmp2 = tmp1 + PATH_MAX;
1410
1411#if 0 /* disabled, the musl way to do things is just too racy */
1412#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1413 /* on linux we may be able to ask the kernel */
1414 {
1415 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1416
1417 if (fd >= 0)
1418 {
1419 sprintf (tmp1, "/proc/self/fd/%d", fd);
1420 req->result = readlink (tmp1, res, PATH_MAX);
1421 /* here we should probably stat the open file and the disk file, to make sure they still match */
1422 close (fd);
1423
1424 if (req->result > 0)
1425 goto done;
1426 }
1427 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1428 return -1;
1429 }
1430#endif
1431#endif
1432
1433 if (*rel != '/')
1434 {
1435 int len;
1436
1437 errno = ENOENT;
1438 if (wd == EIO_INVALID_WD)
1439 return -1;
1440
1441 if (wd == EIO_CWD)
1442 {
1443 if (!getcwd (res, PATH_MAX))
1444 return -1;
1445
1446 len = strlen (res);
1447 }
1448 else
1449 memcpy (res, wd->str, len = wd->len);
1450
1451 if (res [1]) /* only use if not / */
1452 res += len;
1453 }
1454
1455 while (*rel)
1456 {
1457 eio_ssize_t len, linklen;
1458 const char *beg = rel;
1459
1460 while (*rel && *rel != '/')
1461 ++rel;
1462
1463 len = rel - beg;
1464
1465 if (!len) /* skip slashes */
1466 {
1467 ++rel;
1468 continue;
1469 }
1470
1471 if (beg [0] == '.')
1472 {
1473 if (len == 1)
1474 continue; /* . - nop */
1475
1476 if (beg [1] == '.' && len == 2)
1477 {
1478 /* .. - back up one component, if possible */
1479
1480 while (res != tmpbuf->ptr)
1481 if (*--res == '/')
1482 break;
1483
1484 continue;
1485 }
1486 }
1487
1488 errno = ENAMETOOLONG;
1489 if (res + 1 + len + 1 >= tmp1)
1490 return -1;
1491
1492 /* copy one component */
1493 *res = '/';
1494 memcpy (res + 1, beg, len);
1495
1496 /* zero-terminate, for readlink */
1497 res [len + 1] = 0;
1498
1499 /* now check if it's a symlink */
1500 linklen = readlink (tmpbuf->ptr, tmp1, PATH_MAX);
1501
1502 if (linklen < 0)
1503 {
1504 if (errno != EINVAL)
1505 return -1;
1506
1507 /* it's a normal directory. hopefully */
1508 res += len + 1;
1509 }
1510 else
1511 {
1512 /* yay, it was a symlink - build new path in tmp2 */
1513 int rellen = strlen (rel);
1514
1515 errno = ENAMETOOLONG;
1516 if (linklen + 1 + rellen >= PATH_MAX)
1517 return -1;
1518
1519 errno = ELOOP;
1520 if (!--symlinks)
1521 return -1;
1522
1523 if (*tmp1 == '/')
1524 res = tmpbuf->ptr; /* symlink resolves to an absolute path */
1525
1526 /* we need to be careful, as rel might point into tmp2 already */
1527 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1528 tmp2 [linklen] = '/';
1529 memcpy (tmp2, tmp1, linklen);
1530
1531 rel = tmp2;
1532 }
1533 }
1534
1535 /* special case for the lone root path */
1536 if (res == tmpbuf->ptr)
1537 *res++ = '/';
1538
1539 return res - (char *)tmpbuf->ptr;
1540}
1541
1542static signed char
987eio_dent_cmp (const void *a_, const void *b_) 1543eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
988{ 1544{
989 const eio_dirent *a = (const eio_dirent *)a_; 1545 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
990 const eio_dirent *b = (const eio_dirent *)b_; 1546 : a->inode < b->inode ? -1
1547 : a->inode > b->inode ? 1
1548 : 0;
1549}
991 1550
992 return (int)b->score - (int)a->score ? (int)b->score - (int)a->score 1551#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
993 : a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0; /* int might be < ino_t */ 1552
1553#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1554#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1555
1556static void
1557eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1558{
1559 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1560 unsigned char *bit = bits;
1561
1562 assert (CHAR_BIT == 8);
1563 assert (sizeof (eio_dirent) * 8 < 256);
1564 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1565 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1566
1567 if (size <= EIO_SORT_FAST)
1568 return;
1569
1570 /* first prepare an array of bits to test in our radix sort */
1571 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1572 /* inode_bits must contain all inodes ORed together */
1573 /* which is used to skip bits that are 0 everywhere, which is very common */
1574 {
1575 eio_ino_t endianness;
1576 int i, j;
1577
1578 /* we store the byte offset of byte n into byte n of "endianness" */
1579 for (i = 0; i < sizeof (eio_ino_t); ++i)
1580 ((unsigned char *)&endianness)[i] = i;
1581
1582 *bit++ = 0;
1583
1584 for (i = 0; i < sizeof (eio_ino_t); ++i)
1585 {
1586 /* shifting off the byte offsets out of "endianness" */
1587 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1588 endianness >>= 8;
1589
1590 for (j = 0; j < 8; ++j)
1591 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1592 *bit++ = offs + j;
1593 }
1594
1595 for (j = 0; j < 8; ++j)
1596 if (score_bits & (1 << j))
1597 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1598 }
1599
1600 /* now actually do the sorting (a variant of MSD radix sort) */
1601 {
1602 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1603 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1604 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1605 int stk_idx = 0;
1606
1607 base_stk [stk_idx] = dents;
1608 end_stk [stk_idx] = dents + size;
1609 bit_stk [stk_idx] = bit - 1;
1610
1611 do
1612 {
1613 base = base_stk [stk_idx];
1614 end = end_stk [stk_idx];
1615 bit = bit_stk [stk_idx];
1616
1617 for (;;)
1618 {
1619 unsigned char O = *bit >> 3;
1620 unsigned char M = 1 << (*bit & 7);
1621
1622 eio_dirent *a = base;
1623 eio_dirent *b = end;
1624
1625 if (b - a < EIO_SORT_CUTOFF)
1626 break;
1627
1628 /* now bit-partition the array on the bit */
1629 /* this ugly asymmetric loop seems to perform much better than typical */
1630 /* partition algos found in the literature */
1631 do
1632 if (!(((unsigned char *)a)[O] & M))
1633 ++a;
1634 else if (!(((unsigned char *)--b)[O] & M))
1635 {
1636 eio_dirent tmp = *a; *a = *b; *b = tmp;
1637 ++a;
1638 }
1639 while (b > a);
1640
1641 /* next bit, or stop, if no bits left in this path */
1642 if (!*--bit)
1643 break;
1644
1645 base_stk [stk_idx] = a;
1646 end_stk [stk_idx] = end;
1647 bit_stk [stk_idx] = bit;
1648 ++stk_idx;
1649
1650 end = a;
1651 }
1652 }
1653 while (stk_idx--);
1654 }
1655}
1656
1657static void
1658eio_dent_insertion_sort (eio_dirent *dents, int size)
1659{
1660 /* first move the smallest element to the front, to act as a sentinel */
1661 {
1662 int i;
1663 eio_dirent *min = dents;
1664
1665 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1666 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1667 if (EIO_DENT_CMP (dents [i], <, *min))
1668 min = &dents [i];
1669
1670 /* swap elements 0 and j (minimum) */
1671 {
1672 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1673 }
1674 }
1675
1676 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1677 {
1678 eio_dirent *i, *j;
1679
1680 for (i = dents + 1; i < dents + size; ++i)
1681 {
1682 eio_dirent value = *i;
1683
1684 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1685 j [1] = j [0];
1686
1687 j [1] = value;
1688 }
1689 }
1690}
1691
1692static void
1693eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1694{
1695 if (size <= 1)
1696 return; /* our insertion sort relies on size > 0 */
1697
1698 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1699 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1700 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1701
1702 /* use an insertion sort at the end, or for small arrays, */
1703 /* as insertion sort is more efficient for small partitions */
1704 eio_dent_insertion_sort (dents, size);
994} 1705}
995 1706
996/* read a full directory */ 1707/* read a full directory */
997static void 1708static void
998eio__scandir (eio_req *req, etp_worker *self) 1709eio__scandir (eio_req *req, etp_worker *self)
999{ 1710{
1000 DIR *dirp;
1001 EIO_STRUCT_DIRENT *entp;
1002 unsigned char *name, *names; 1711 char *name, *names;
1003 int namesalloc = 4096; 1712 int namesalloc = 4096 - sizeof (void *) * 4;
1004 int namesoffs = 0; 1713 int namesoffs = 0;
1005 int flags = req->int1; 1714 int flags = req->int1;
1006 eio_dirent *dents = 0; 1715 eio_dirent *dents = 0;
1007 int dentalloc = 128; 1716 int dentalloc = 128;
1008 int dentoffs = 0; 1717 int dentoffs = 0;
1718 eio_ino_t inode_bits = 0;
1719#ifdef _WIN32
1720 HANDLE dirp;
1721 WIN32_FIND_DATA entp;
1722#else
1723 DIR *dirp;
1724 EIO_STRUCT_DIRENT *entp;
1725#endif
1009 1726
1010 req->result = -1; 1727 req->result = -1;
1011 1728
1012 if (!(flags & EIO_READDIR_DENTS)) 1729 if (!(flags & EIO_READDIR_DENTS))
1013 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); 1730 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1014 1731
1015 X_LOCK (wrklock); 1732#ifdef _WIN32
1016 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 1733 {
1734 int len = strlen ((const char *)req->ptr1);
1735 char *path = malloc (MAX_PATH);
1736 const char *fmt;
1737 const char *reqpath = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1738
1739 if (!len)
1740 fmt = "./*";
1741 else if (reqpath[len - 1] == '/' || reqpath[len - 1] == '\\')
1742 fmt = "%s*";
1743 else
1744 fmt = "%s/*";
1745
1746 _snprintf (path, MAX_PATH, fmt, reqpath);
1747 dirp = FindFirstFile (path, &entp);
1748 free (path);
1749
1750 if (dirp == INVALID_HANDLE_VALUE)
1751 {
1752 /* should steal _dosmaperr */
1753 switch (GetLastError ())
1754 {
1755 case ERROR_FILE_NOT_FOUND:
1756 req->result = 0;
1757 break;
1758
1759 case ERROR_INVALID_NAME:
1760 case ERROR_PATH_NOT_FOUND:
1761 case ERROR_NO_MORE_FILES:
1762 errno = ENOENT;
1763 break;
1764
1765 case ERROR_NOT_ENOUGH_MEMORY:
1766 errno = ENOMEM;
1767 break;
1768
1769 default:
1770 errno = EINVAL;
1771 break;
1772 }
1773
1774 return;
1775 }
1776 }
1777#else
1778 #if HAVE_AT
1779 if (req->wd)
1780 {
1781 int fd = openat (WD2FD (req->wd), req->ptr1, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1782
1783 if (fd < 0)
1784 return;
1785
1786 dirp = fdopendir (fd);
1787
1788 if (!dirp)
1789 close (fd);
1790 }
1791 else
1017 self->dirp = dirp = opendir (req->ptr1); 1792 dirp = opendir (req->ptr1);
1793 #else
1794 dirp = opendir (wd_expand (&self->tmpbuf, req->wd, req->ptr1));
1795 #endif
1796
1797 if (!dirp)
1798 return;
1799#endif
1800
1801 if (req->flags & EIO_FLAG_PTR1_FREE)
1802 free (req->ptr1);
1803
1018 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1804 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1019 req->ptr1 = names = malloc (namesalloc);
1020 req->ptr2 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; 1805 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1021 X_UNLOCK (wrklock); 1806 req->ptr2 = names = malloc (namesalloc);
1022 1807
1023 if (dirp && names && (!flags || dents)) 1808 if (!names || (flags && !dents))
1809 return;
1810
1024 for (;;) 1811 for (;;)
1025 { 1812 {
1813 int done;
1814
1815#ifdef _WIN32
1816 done = !dirp;
1817#else
1026 errno = 0; 1818 errno = 0;
1027 entp = readdir (dirp); 1819 entp = readdir (dirp);
1820 done = !entp;
1821#endif
1028 1822
1029 if (!entp) 1823 if (done)
1030 { 1824 {
1825#ifndef _WIN32
1826 int old_errno = errno;
1827 closedir (dirp);
1828 errno = old_errno;
1829
1031 if (errno) 1830 if (errno)
1032 break; 1831 break;
1832#endif
1033 1833
1034 /* sort etc. */ 1834 /* sort etc. */
1035 req->int1 = flags; 1835 req->int1 = flags;
1036 req->result = dentoffs; 1836 req->result = dentoffs;
1037 1837
1038 if (dents)
1039 {
1040 eio_dirent *ent = dents + dentoffs;
1041
1042 while (ent > dents)
1043 (--ent)->name = names + (size_t)ent->name;
1044 }
1045
1046 if (flags & EIO_READDIR_STAT_ORDER 1838 if (flags & EIO_READDIR_STAT_ORDER)
1047 || !(~flags & (EIO_READDIR_DIRS_FIRST | EIO_READDIR_FOUND_UNKNOWN))) 1839 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1048 {
1049 /* pray your qsort doesn't use quicksort */
1050 qsort (dents, dentoffs, sizeof (*dents), eio_dent_cmp); /* score depends of DIRS_FIRST */
1051 }
1052 else if (flags & EIO_READDIR_DIRS_FIRST) 1840 else if (flags & EIO_READDIR_DIRS_FIRST)
1841 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1842 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1843 else
1053 { 1844 {
1054 /* in this case, all is known, and we just put dirs first and sort them */ 1845 /* in this case, all is known, and we just put dirs first and sort them */
1055 eio_dirent *ent = dents + dentoffs; 1846 eio_dirent *oth = dents + dentoffs;
1056 eio_dirent *dir = dents; 1847 eio_dirent *dir = dents;
1057 1848
1849 /* now partition dirs to the front, and non-dirs to the back */
1850 /* by walking from both sides and swapping if necessary */
1058 while (ent > dir) 1851 while (oth > dir)
1059 { 1852 {
1060 if (dir->type == DT_DIR) 1853 if (dir->type == EIO_DT_DIR)
1061 ++dir; 1854 ++dir;
1062 else 1855 else if ((--oth)->type == EIO_DT_DIR)
1063 { 1856 {
1064 --ent;
1065
1066 if (ent->type == DT_DIR)
1067 {
1068 eio_dirent tmp = *dir; 1857 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1069 *dir = *ent;
1070 *ent = tmp;
1071 1858
1072 ++dir; 1859 ++dir;
1073 }
1074 } 1860 }
1075 } 1861 }
1076 1862
1077 /* now sort the dirs only */ 1863 /* now sort the dirs only (dirs all have the same score) */
1078 qsort (dents, dir - dents, sizeof (*dents), eio_dent_cmp); 1864 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1079 } 1865 }
1080 1866
1081 /* only provide the names array unless DENTS is specified */
1082 if (!(flags & EIO_READDIR_DENTS))
1083 {
1084 X_LOCK (wrklock);
1085 assert (!dents);
1086 req->ptr1 = 0;
1087 req->ptr2 = names;
1088 X_UNLOCK (wrklock);
1089 }
1090
1091 break; 1867 break;
1092 } 1868 }
1093 1869
1094 /* now add the entry to our list(s) */ 1870 /* now add the entry to our list(s) */
1095 name = entp->d_name; 1871 name = D_NAME (entp);
1096 1872
1097 /* skip . and .. entries */ 1873 /* skip . and .. entries */
1098 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1874 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1099 { 1875 {
1100 int len = strlen (name) + 1; 1876 int len = D_NAMLEN (entp) + 1;
1101 1877
1102 while (expect_false (namesoffs + len > namesalloc)) 1878 while (ecb_expect_false (namesoffs + len > namesalloc))
1103 { 1879 {
1104 namesalloc *= 2; 1880 namesalloc *= 2;
1105 X_LOCK (wrklock);
1106 req->ptr1 = names = realloc (names, namesalloc); 1881 req->ptr2 = names = realloc (names, namesalloc);
1107 X_UNLOCK (wrklock);
1108 1882
1109 if (!names) 1883 if (!names)
1110 break; 1884 break;
1111 } 1885 }
1112 1886
1113 memcpy (names + namesoffs, name, len); 1887 memcpy (names + namesoffs, name, len);
1114 1888
1115 if (dents) 1889 if (dents)
1116 { 1890 {
1117 struct eio_dirent *ent; 1891 struct eio_dirent *ent;
1118 1892
1119 if (expect_false (dentoffs == dentalloc)) 1893 if (ecb_expect_false (dentoffs == dentalloc))
1120 { 1894 {
1121 dentalloc *= 2; 1895 dentalloc *= 2;
1122 X_LOCK (wrklock);
1123 req->ptr2 = dents = realloc (dents, dentalloc * sizeof (eio_dirent)); 1896 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1124 X_UNLOCK (wrklock);
1125 1897
1126 if (!dents) 1898 if (!dents)
1127 break; 1899 break;
1128 } 1900 }
1129 1901
1130 ent = dents + dentoffs; 1902 ent = dents + dentoffs;
1131 1903
1132 ent->name = (char *)(size_t)namesoffs; /* rather dirtily we store the offset in the pointer */ 1904 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1133 ent->namelen = len - 1; 1905 ent->namelen = len - 1;
1134 ent->inode = D_INO (entp); 1906 ent->inode = D_INO (entp);
1135 1907
1908 inode_bits |= ent->inode;
1909
1136 switch (D_TYPE (entp)) 1910 switch (D_TYPE (entp))
1137 { 1911 {
1138 default: 1912 default:
1139 ent->type = EIO_DT_UNKNOWN; 1913 ent->type = EIO_DT_UNKNOWN;
1140 flags |= EIO_READDIR_FOUND_UNKNOWN; 1914 flags |= EIO_READDIR_FOUND_UNKNOWN;
1141 break; 1915 break;
1142 1916
1143 #ifdef DT_FIFO 1917 #ifdef DT_FIFO
1144 case DT_FIFO: ent->type = EIO_DT_FIFO; break; 1918 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1145 #endif 1919 #endif
1146 #ifdef DT_CHR 1920 #ifdef DT_CHR
1147 case DT_CHR: ent->type = EIO_DT_CHR; break; 1921 case DT_CHR: ent->type = EIO_DT_CHR; break;
1148 #endif 1922 #endif
1923 #ifdef DT_MPC
1924 case DT_MPC: ent->type = EIO_DT_MPC; break;
1925 #endif
1149 #ifdef DT_DIR 1926 #ifdef DT_DIR
1150 case DT_DIR: ent->type = EIO_DT_DIR; break; 1927 case DT_DIR: ent->type = EIO_DT_DIR; break;
1151 #endif 1928 #endif
1929 #ifdef DT_NAM
1930 case DT_NAM: ent->type = EIO_DT_NAM; break;
1931 #endif
1152 #ifdef DT_BLK 1932 #ifdef DT_BLK
1153 case DT_BLK: ent->type = EIO_DT_BLK; break; 1933 case DT_BLK: ent->type = EIO_DT_BLK; break;
1154 #endif 1934 #endif
1935 #ifdef DT_MPB
1936 case DT_MPB: ent->type = EIO_DT_MPB; break;
1937 #endif
1155 #ifdef DT_REG 1938 #ifdef DT_REG
1156 case DT_REG: ent->type = EIO_DT_REG; break; 1939 case DT_REG: ent->type = EIO_DT_REG; break;
1157 #endif 1940 #endif
1941 #ifdef DT_NWK
1942 case DT_NWK: ent->type = EIO_DT_NWK; break;
1943 #endif
1944 #ifdef DT_CMP
1945 case DT_CMP: ent->type = EIO_DT_CMP; break;
1946 #endif
1158 #ifdef DT_LNK 1947 #ifdef DT_LNK
1159 case DT_LNK: ent->type = EIO_DT_LNK; break; 1948 case DT_LNK: ent->type = EIO_DT_LNK; break;
1160 #endif 1949 #endif
1161 #ifdef DT_SOCK 1950 #ifdef DT_SOCK
1162 case DT_SOCK: ent->type = EIO_DT_SOCK; break; 1951 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1163 #endif 1952 #endif
1953 #ifdef DT_DOOR
1954 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1955 #endif
1164 #ifdef DT_WHT 1956 #ifdef DT_WHT
1165 case DT_WHT: ent->type = EIO_DT_WHT; break; 1957 case DT_WHT: ent->type = EIO_DT_WHT; break;
1166 #endif 1958 #endif
1167 } 1959 }
1168 1960
1169 ent->score = 0; 1961 ent->score = 7;
1170 1962
1171 if (flags & EIO_READDIR_DIRS_FIRST) 1963 if (flags & EIO_READDIR_DIRS_FIRST)
1172 { 1964 {
1173 if (ent->type == EIO_DT_UNKNOWN) 1965 if (ent->type == EIO_DT_UNKNOWN)
1174 { 1966 {
1175 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */ 1967 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1176 ent->score = 98; 1968 ent->score = 1;
1177 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */ 1969 else if (!strchr (name, '.')) /* absence of dots indicate likely dirs */
1178 ent->score = len <= 4 ? 5 : len <= 7 ? 4 : 1; /* shorter == more likely dir, but avoid too many classes */ 1970 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1179 } 1971 }
1180 else if (ent->type == DT_DIR) 1972 else if (ent->type == EIO_DT_DIR)
1181 ent->score = 100; 1973 ent->score = 0;
1182 } 1974 }
1183 } 1975 }
1184 1976
1185 namesoffs += len; 1977 namesoffs += len;
1186 ++dentoffs; 1978 ++dentoffs;
1187 } 1979 }
1980
1981 if (EIO_CANCELLED (req))
1982 {
1983 errno = ECANCELED;
1984 break;
1188 } 1985 }
1189 else
1190 req->result = -1;
1191}
1192 1986
1193#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) 1987#ifdef _WIN32
1194# undef msync 1988 if (!FindNextFile (dirp, &entp))
1195# define msync(a,b,c) ((errno = ENOSYS), -1) 1989 {
1990 FindClose (dirp);
1991 dirp = 0;
1992 }
1196#endif 1993#endif
1994 }
1995}
1197 1996
1198int 1997/*****************************************************************************/
1199eio__mtouch (void *mem, size_t len, int flags) 1998/* working directory stuff */
1200{ 1999/* various deficiencies in the posix 2008 api force us to */
1201 intptr_t addr = (intptr_t)mem; 2000/* keep the absolute path in string form at all times */
1202 intptr_t end = addr + len; 2001/* fuck yeah. */
1203#ifdef PAGESIZE
1204 const intptr_t page = PAGESIZE;
1205#else
1206 static intptr_t page;
1207 2002
1208 if (!page) 2003#if !HAVE_AT
1209 page = sysconf (_SC_PAGESIZE);
1210#endif
1211 2004
1212 addr &= ~(page - 1); /* assume page size is always a power of two */ 2005/* a bit like realpath, but usually faster because it doesn'T have to return */
2006/* an absolute or canonical path */
2007static const char *
2008wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
2009{
2010 if (!wd || *path == '/')
2011 return path;
1213 2012
1214 if (addr < end) 2013 if (path [0] == '.' && !path [1])
1215 if (flags) /* modify */ 2014 return wd->str;
1216 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
1217 else
1218 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
1219 2015
2016 {
2017 int l1 = wd->len;
2018 int l2 = strlen (path);
2019
2020 char *res = tmpbuf_get (tmpbuf, l1 + l2 + 2);
2021
2022 memcpy (res, wd->str, l1);
2023 res [l1] = '/';
2024 memcpy (res + l1 + 1, path, l2 + 1);
2025
2026 return res;
2027 }
2028}
2029
2030#endif
2031
2032static eio_wd
2033eio__wd_open_sync (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
2034{
2035 int fd;
2036 eio_wd res;
2037 int len = eio__realpath (tmpbuf, wd, path);
2038
2039 if (len < 0)
2040 return EIO_INVALID_WD;
2041
2042#if HAVE_AT
2043 fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
2044
2045 if (fd < 0)
2046 return EIO_INVALID_WD;
2047#endif
2048
2049 res = malloc (sizeof (*res) + len); /* one extra 0-byte */
2050
2051#if HAVE_AT
2052 res->fd = fd;
2053#endif
2054
2055 res->len = len;
2056 memcpy (res->str, tmpbuf->ptr, len);
2057 res->str [len] = 0;
2058
2059 return res;
2060}
2061
2062eio_wd
2063eio_wd_open_sync (eio_wd wd, const char *path)
2064{
2065 struct tmpbuf tmpbuf = { 0 };
2066 wd = eio__wd_open_sync (&tmpbuf, wd, path);
2067 free (tmpbuf.ptr);
2068
1220 return 0; 2069 return wd;
1221} 2070}
2071
2072void
2073eio_wd_close_sync (eio_wd wd)
2074{
2075 if (wd != EIO_INVALID_WD && wd != EIO_CWD)
2076 {
2077 #if HAVE_AT
2078 close (wd->fd);
2079 #endif
2080 free (wd);
2081 }
2082}
2083
2084#if HAVE_AT
2085
2086/* they forgot these */
2087
2088static int
2089eio__truncateat (int dirfd, const char *path, off_t length)
2090{
2091 int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
2092 int res;
2093
2094 if (fd < 0)
2095 return fd;
2096
2097 res = ftruncate (fd, length);
2098 close (fd);
2099 return res;
2100}
2101
2102static int
2103eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
2104{
2105 int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
2106 int res;
2107
2108 if (fd < 0)
2109 return fd;
2110
2111 res = fstatvfs (fd, buf);
2112 close (fd);
2113 return res;
2114
2115}
2116
2117#endif
1222 2118
1223/*****************************************************************************/ 2119/*****************************************************************************/
1224 2120
1225#define ALLOC(len) \ 2121#define ALLOC(len) \
1226 if (!req->ptr2) \ 2122 if (!req->ptr2) \
1235 req->result = -1; \ 2131 req->result = -1; \
1236 break; \ 2132 break; \
1237 } \ 2133 } \
1238 } 2134 }
1239 2135
2136static void ecb_noinline ecb_cold
2137etp_proc_init (void)
2138{
2139#if HAVE_PRCTL_SET_NAME
2140 /* provide a more sensible "thread name" */
2141 char name[16 + 1];
2142 const int namelen = sizeof (name) - 1;
2143 int len;
2144
2145 prctl (PR_GET_NAME, (unsigned long)name, 0, 0, 0);
2146 name [namelen] = 0;
2147 len = strlen (name);
2148 strcpy (name + (len <= namelen - 4 ? len : namelen - 4), "/eio");
2149 prctl (PR_SET_NAME, (unsigned long)name, 0, 0, 0);
2150#endif
2151}
2152
1240X_THREAD_PROC (etp_proc) 2153X_THREAD_PROC (etp_proc)
1241{ 2154{
1242 ETP_REQ *req; 2155 ETP_REQ *req;
1243 struct timespec ts; 2156 struct timespec ts;
1244 etp_worker *self = (etp_worker *)thr_arg; 2157 etp_worker *self = (etp_worker *)thr_arg;
1245 2158
2159 etp_proc_init ();
2160
1246 /* try to distribute timeouts somewhat randomly */ 2161 /* try to distribute timeouts somewhat evenly */
1247 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 2162 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1248 2163
1249 for (;;) 2164 for (;;)
1250 { 2165 {
2166 ts.tv_sec = 0;
2167
1251 X_LOCK (reqlock); 2168 X_LOCK (reqlock);
1252 2169
1253 for (;;) 2170 for (;;)
1254 { 2171 {
1255 self->req = req = reqq_shift (&req_queue); 2172 req = reqq_shift (&req_queue);
1256 2173
1257 if (req) 2174 if (req)
1258 break; 2175 break;
1259 2176
2177 if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
2178 {
2179 X_UNLOCK (reqlock);
2180 X_LOCK (wrklock);
2181 --started;
2182 X_UNLOCK (wrklock);
2183 goto quit;
2184 }
2185
1260 ++idle; 2186 ++idle;
1261 2187
1262 ts.tv_sec = time (0) + IDLE_TIMEOUT; 2188 if (idle <= max_idle)
1263 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 2189 /* we are allowed to idle, so do so without any timeout */
2190 X_COND_WAIT (reqwait, reqlock);
2191 else
1264 { 2192 {
1265 if (idle > max_idle) 2193 /* initialise timeout once */
1266 { 2194 if (!ts.tv_sec)
1267 --idle; 2195 ts.tv_sec = time (0) + idle_timeout;
1268 X_UNLOCK (reqlock);
1269 X_LOCK (wrklock);
1270 --started;
1271 X_UNLOCK (wrklock);
1272 goto quit;
1273 }
1274 2196
1275 /* we are allowed to idle, so do so without any timeout */
1276 X_COND_WAIT (reqwait, reqlock); 2197 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
2198 ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1277 } 2199 }
1278 2200
1279 --idle; 2201 --idle;
1280 } 2202 }
1281 2203
1284 X_UNLOCK (reqlock); 2206 X_UNLOCK (reqlock);
1285 2207
1286 if (req->type < 0) 2208 if (req->type < 0)
1287 goto quit; 2209 goto quit;
1288 2210
1289 if (!EIO_CANCELLED (req))
1290 ETP_EXECUTE (self, req); 2211 ETP_EXECUTE (self, req);
1291 2212
1292 X_LOCK (reslock); 2213 X_LOCK (reslock);
1293 2214
1294 ++npending; 2215 ++npending;
1295 2216
1296 if (!reqq_push (&res_queue, req) && want_poll_cb) 2217 if (!reqq_push (&res_queue, req) && want_poll_cb)
1297 want_poll_cb (); 2218 want_poll_cb ();
1298 2219
1299 self->req = 0;
1300 etp_worker_clear (self); 2220 etp_worker_clear (self);
1301 2221
1302 X_UNLOCK (reslock); 2222 X_UNLOCK (reslock);
1303 } 2223 }
1304 2224
1305quit: 2225quit:
2226 free (req);
2227
1306 X_LOCK (wrklock); 2228 X_LOCK (wrklock);
1307 etp_worker_free (self); 2229 etp_worker_free (self);
1308 X_UNLOCK (wrklock); 2230 X_UNLOCK (wrklock);
1309 2231
1310 return 0; 2232 return 0;
1311} 2233}
1312 2234
1313/*****************************************************************************/ 2235/*****************************************************************************/
1314 2236
2237int ecb_cold
1315int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 2238eio_init (void (*want_poll)(void), void (*done_poll)(void))
1316{ 2239{
1317 return etp_init (want_poll, done_poll); 2240 return etp_init (want_poll, done_poll);
1318} 2241}
1319 2242
2243ecb_inline void
1320static void eio_api_destroy (eio_req *req) 2244eio_api_destroy (eio_req *req)
1321{ 2245{
1322 free (req); 2246 free (req);
1323} 2247}
1324 2248
1325#define REQ(rtype) \ 2249#define REQ(rtype) \
1326 eio_req *req; \ 2250 eio_req *req; \
1327 \ 2251 \
1328 req = (eio_req *)calloc (1, sizeof *req); \ 2252 req = (eio_req *)calloc (1, sizeof *req); \
1329 if (!req) \ 2253 if (!req) \
1330 return 0; \ 2254 return 0; \
1344 { \ 2268 { \
1345 eio_api_destroy (req); \ 2269 eio_api_destroy (req); \
1346 return 0; \ 2270 return 0; \
1347 } 2271 }
1348 2272
2273#define SINGLEDOT(ptr) (0[(char *)(ptr)] == '.' && !1[(char *)(ptr)])
2274
2275static void
1349static void eio_execute (etp_worker *self, eio_req *req) 2276eio_execute (etp_worker *self, eio_req *req)
1350{ 2277{
1351 errno = 0; 2278#if HAVE_AT
2279 int dirfd;
2280#else
2281 const char *path;
2282#endif
2283
2284 if (ecb_expect_false (EIO_CANCELLED (req)))
2285 {
2286 req->result = -1;
2287 req->errorno = ECANCELED;
2288 return;
2289 }
2290
2291 if (ecb_expect_false (req->wd == EIO_INVALID_WD))
2292 {
2293 req->result = -1;
2294 req->errorno = ENOENT;
2295 return;
2296 }
2297
2298 if (req->type >= EIO_OPEN)
2299 {
2300 #if HAVE_AT
2301 dirfd = WD2FD (req->wd);
2302 #else
2303 path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
2304 #endif
2305 }
1352 2306
1353 switch (req->type) 2307 switch (req->type)
1354 { 2308 {
2309 case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
2310 req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
2311 break;
2312 case EIO_WD_CLOSE: req->result = 0;
2313 eio_wd_close_sync (req->wd); break;
2314
2315 case EIO_SEEK: eio__lseek (req); break;
1355 case EIO_READ: ALLOC (req->size); 2316 case EIO_READ: ALLOC (req->size);
1356 req->result = req->offs >= 0 2317 req->result = req->offs >= 0
1357 ? pread (req->int1, req->ptr2, req->size, req->offs) 2318 ? pread (req->int1, req->ptr2, req->size, req->offs)
1358 : read (req->int1, req->ptr2, req->size); break; 2319 : read (req->int1, req->ptr2, req->size); break;
1359 case EIO_WRITE: req->result = req->offs >= 0 2320 case EIO_WRITE: req->result = req->offs >= 0
1360 ? pwrite (req->int1, req->ptr2, req->size, req->offs) 2321 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1361 : write (req->int1, req->ptr2, req->size); break; 2322 : write (req->int1, req->ptr2, req->size); break;
1362 2323
1363 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break; 2324 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1364 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break; 2325 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
2326
2327#if HAVE_AT
1365 2328
1366 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2329 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1367 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2330 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break;
1368 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2331 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1369 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2332 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, AT_SYMLINK_NOFOLLOW); break;
1370 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1371 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1372
1373 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; 2333 case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break;
1374 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1375 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break; 2334 case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break;
1376 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1377 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break; 2335 case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break;
1378 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1379
1380 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break; 2336 case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break;
1381 case EIO_CLOSE: req->result = close (req->int1); break; 2337
1382 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1383 case EIO_UNLINK: req->result = unlink (req->ptr1); break; 2338 case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
1384 case EIO_RMDIR: req->result = rmdir (req->ptr1); break; 2339 case EIO_RMDIR: /* complications arise because "." cannot be removed, so we might have to expand */
2340 req->result = req->wd && SINGLEDOT (req->ptr1)
2341 ? rmdir (req->wd->str)
2342 : unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break;
1385 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; 2343 case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); break;
1386 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; 2344 case EIO_RENAME: /* complications arise because "." cannot be renamed, so we might have to expand */
2345 req->result = req->wd && SINGLEDOT (req->ptr1)
2346 ? rename (req->wd->str, req->ptr2)
2347 : renameat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2); break;
1387 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; 2348 case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2, 0); break;
1388 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; 2349 case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
1389 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break; 2350 case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1390
1391 case EIO_READLINK: ALLOC (NAME_MAX); 2351 case EIO_READLINK: ALLOC (PATH_MAX);
1392 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; 2352 req->result = readlinkat (dirfd, req->ptr1, req->ptr2, PATH_MAX); break;
1393 2353 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1394 case EIO_SYNC: req->result = 0; sync (); break; 2354 req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1395 case EIO_FSYNC: req->result = fsync (req->int1); break;
1396 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1397 case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break;
1398 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;
1399 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1400
1401 case EIO_READDIR: eio__scandir (req, self); break;
1402
1403 case EIO_BUSY: 2355 case EIO_UTIME:
1404#ifdef _WIN32 2356 case EIO_FUTIME:
1405 Sleep (req->nv1 * 1000.); 2357 {
2358 struct timespec ts[2];
2359 struct timespec *times;
2360
2361 if (req->nv1 != -1. || req->nv2 != -1.)
2362 {
2363 ts[0].tv_sec = req->nv1;
2364 ts[0].tv_nsec = (req->nv1 - ts[0].tv_sec) * 1e9;
2365 ts[1].tv_sec = req->nv2;
2366 ts[1].tv_nsec = (req->nv2 - ts[1].tv_sec) * 1e9;
2367
2368 times = ts;
2369 }
2370 else
2371 times = 0;
2372
2373 req->result = req->type == EIO_FUTIME
2374 ? futimens (req->int1, times)
2375 : utimensat (dirfd, req->ptr1, times, 0);
2376 }
2377 break;
2378
1406#else 2379#else
1407 {
1408 struct timeval tv;
1409 2380
1410 tv.tv_sec = req->nv1; 2381 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1411 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; 2382 req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
2383 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2384 req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
2385 case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
2386 case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
2387 case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
2388 case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
1412 2389
1413 req->result = select (0, 0, 0, 0, &tv); 2390 case EIO_UNLINK: req->result = unlink (path ); break;
1414 } 2391 case EIO_RMDIR: req->result = rmdir (path ); break;
1415#endif 2392 case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
1416 break; 2393 case EIO_RENAME: req->result = rename (path , req->ptr2); break;
2394 case EIO_LINK: req->result = link (path , req->ptr2); break;
2395 case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
2396 case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
2397 case EIO_READLINK: ALLOC (PATH_MAX);
2398 req->result = readlink (path, req->ptr2, PATH_MAX); break;
2399 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2400 req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
1417 2401
1418 case EIO_UTIME: 2402 case EIO_UTIME:
1419 case EIO_FUTIME: 2403 case EIO_FUTIME:
1420 { 2404 {
1421 struct timeval tv[2]; 2405 struct timeval tv[2];
1422 struct timeval *times; 2406 struct timeval *times;
1423 2407
1424 if (req->nv1 != -1. || req->nv2 != -1.) 2408 if (req->nv1 != -1. || req->nv2 != -1.)
1425 { 2409 {
1426 tv[0].tv_sec = req->nv1; 2410 tv[0].tv_sec = req->nv1;
1427 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.; 2411 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
1428 tv[1].tv_sec = req->nv2; 2412 tv[1].tv_sec = req->nv2;
1429 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.; 2413 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
1430 2414
1431 times = tv; 2415 times = tv;
1432 } 2416 }
1433 else 2417 else
1434 times = 0; 2418 times = 0;
1435
1436 2419
1437 req->result = req->type == EIO_FUTIME 2420 req->result = req->type == EIO_FUTIME
1438 ? futimes (req->int1, times) 2421 ? futimes (req->int1, times)
1439 : utimes (req->ptr1, times); 2422 : utimes (req->ptr1, times);
1440 } 2423 }
1441 break; 2424 break;
1442 2425
2426#endif
2427
2428 case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
2429 {
2430 ALLOC (req->result);
2431 memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
2432 }
2433 break;
2434
2435 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2436 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
2437
2438 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2439 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2440
2441 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
2442 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
2443 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
2444
2445 case EIO_CLOSE: req->result = close (req->int1); break;
2446 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
2447 case EIO_SYNC: req->result = 0; sync (); break;
2448 case EIO_FSYNC: req->result = fsync (req->int1); break;
2449 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
2450 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
2451 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
2452 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
2453 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
2454 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
2455 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
2456 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
2457
2458 case EIO_READDIR: eio__scandir (req, self); break;
2459
2460 case EIO_BUSY:
2461#ifdef _WIN32
2462 Sleep (req->nv1 * 1e3);
2463#else
2464 {
2465 struct timeval tv;
2466
2467 tv.tv_sec = req->nv1;
2468 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
2469
2470 req->result = select (0, 0, 0, 0, &tv);
2471 }
2472#endif
2473 break;
2474
1443 case EIO_GROUP: 2475 case EIO_GROUP:
1444 abort (); /* handled in eio_request */ 2476 abort (); /* handled in eio_request */
1445 2477
1446 case EIO_NOP: 2478 case EIO_NOP:
1447 req->result = 0; 2479 req->result = 0;
1448 break; 2480 break;
1449 2481
1450 case EIO_CUSTOM: 2482 case EIO_CUSTOM:
1451 ((void (*)(eio_req *))req->feed) (req); 2483 req->feed (req);
1452 break; 2484 break;
1453 2485
1454 default: 2486 default:
1455 req->result = -1; 2487 req->result = EIO_ENOSYS ();
1456 break; 2488 break;
1457 } 2489 }
1458 2490
1459 req->errorno = errno; 2491 req->errorno = errno;
1460} 2492}
1461 2493
1462#ifndef EIO_NO_WRAPPERS 2494#ifndef EIO_NO_WRAPPERS
1463 2495
2496eio_req *eio_wd_open (const char *path, int pri, eio_cb cb, void *data)
2497{
2498 REQ (EIO_WD_OPEN); PATH; SEND;
2499}
2500
2501eio_req *eio_wd_close (eio_wd wd, int pri, eio_cb cb, void *data)
2502{
2503 REQ (EIO_WD_CLOSE); req->wd = wd; SEND;
2504}
2505
1464eio_req *eio_nop (int pri, eio_cb cb, void *data) 2506eio_req *eio_nop (int pri, eio_cb cb, void *data)
1465{ 2507{
1466 REQ (EIO_NOP); SEND; 2508 REQ (EIO_NOP); SEND;
1467} 2509}
1468 2510
1484eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2526eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1485{ 2527{
1486 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2528 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1487} 2529}
1488 2530
2531eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2532{
2533 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2534}
2535
2536eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
2537{
2538 REQ (EIO_SYNCFS); req->int1 = fd; SEND;
2539}
2540
2541eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2542{
2543 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2544}
2545
1489eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2546eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1490{ 2547{
1491 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2548 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1492} 2549}
1493 2550
1494eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) 2551eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
1495{ 2552{
1496 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; 2553 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
1497} 2554}
1498 2555
1499eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) 2556eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
1500{ 2557{
1501 REQ (EIO_FDATASYNC); req->int1 = fd; SEND; 2558 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2559}
2560
2561eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2562{
2563 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
1502} 2564}
1503 2565
1504eio_req *eio_close (int fd, int pri, eio_cb cb, void *data) 2566eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
1505{ 2567{
1506 REQ (EIO_CLOSE); req->int1 = fd; SEND; 2568 REQ (EIO_CLOSE); req->int1 = fd; SEND;
1509eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data) 2571eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
1510{ 2572{
1511 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND; 2573 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
1512} 2574}
1513 2575
2576eio_req *eio_seek (int fd, off_t offset, int whence, int pri, eio_cb cb, void *data)
2577{
2578 REQ (EIO_SEEK); req->int1 = fd; req->offs = offset; req->int2 = whence; SEND;
2579}
2580
1514eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data) 2581eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1515{ 2582{
1516 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND; 2583 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1517} 2584}
1518 2585
1524eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) 2591eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1525{ 2592{
1526 REQ (EIO_FSTAT); req->int1 = fd; SEND; 2593 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1527} 2594}
1528 2595
2596eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
2597{
2598 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
2599}
2600
1529eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) 2601eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1530{ 2602{
1531 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; 2603 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1532} 2604}
1533 2605
1539eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) 2611eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1540{ 2612{
1541 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2613 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1542} 2614}
1543 2615
1544eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2616eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1545{ 2617{
1546 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2618 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1547} 2619}
1548 2620
1549eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) 2621eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
1569eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data) 2641eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1570{ 2642{
1571 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2643 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1572} 2644}
1573 2645
1574eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2646eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1575{ 2647{
1576 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2648 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1577} 2649}
1578 2650
1579eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) 2651eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
1595eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data) 2667eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1596{ 2668{
1597 return eio__1path (EIO_READLINK, path, pri, cb, data); 2669 return eio__1path (EIO_READLINK, path, pri, cb, data);
1598} 2670}
1599 2671
2672eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2673{
2674 return eio__1path (EIO_REALPATH, path, pri, cb, data);
2675}
2676
1600eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) 2677eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1601{ 2678{
1602 return eio__1path (EIO_STAT, path, pri, cb, data); 2679 return eio__1path (EIO_STAT, path, pri, cb, data);
1603} 2680}
1604 2681
1605eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) 2682eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1606{ 2683{
1607 return eio__1path (EIO_LSTAT, path, pri, cb, data); 2684 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1608} 2685}
1609 2686
2687eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
2688{
2689 return eio__1path (EIO_STATVFS, path, pri, cb, data);
2690}
2691
1610eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) 2692eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1611{ 2693{
1612 return eio__1path (EIO_UNLINK, path, pri, cb, data); 2694 return eio__1path (EIO_UNLINK, path, pri, cb, data);
1613} 2695}
1614 2696
1622 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND; 2704 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1623} 2705}
1624 2706
1625eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) 2707eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1626{ 2708{
1627 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; 2709 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1628} 2710}
1629 2711
1630static eio_req * 2712static eio_req *
1631eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2713eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1632{ 2714{
1656eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2738eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1657{ 2739{
1658 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2740 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1659} 2741}
1660 2742
1661eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) 2743eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
1662{ 2744{
1663 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND; 2745 REQ (EIO_CUSTOM); req->feed = execute; SEND;
1664} 2746}
1665 2747
1666#endif 2748#endif
1667 2749
1668eio_req *eio_grp (eio_cb cb, void *data) 2750eio_req *eio_grp (eio_cb cb, void *data)
1677#undef SEND 2759#undef SEND
1678 2760
1679/*****************************************************************************/ 2761/*****************************************************************************/
1680/* grp functions */ 2762/* grp functions */
1681 2763
2764void
1682void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2765eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1683{ 2766{
1684 grp->int2 = limit; 2767 grp->int2 = limit;
1685 grp->feed = feed; 2768 grp->feed = feed;
1686 2769
1687 grp_try_feed (grp); 2770 grp_try_feed (grp);
1688} 2771}
1689 2772
2773void
1690void eio_grp_limit (eio_req *grp, int limit) 2774eio_grp_limit (eio_req *grp, int limit)
1691{ 2775{
1692 grp->int2 = limit; 2776 grp->int2 = limit;
1693 2777
1694 grp_try_feed (grp); 2778 grp_try_feed (grp);
1695} 2779}
1696 2780
2781void
1697void eio_grp_add (eio_req *grp, eio_req *req) 2782eio_grp_add (eio_req *grp, eio_req *req)
1698{ 2783{
1699 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2784 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1700 2785
1701 grp->flags |= EIO_FLAG_GROUPADD; 2786 grp->flags |= EIO_FLAG_GROUPADD;
1702 2787
1713} 2798}
1714 2799
1715/*****************************************************************************/ 2800/*****************************************************************************/
1716/* misc garbage */ 2801/* misc garbage */
1717 2802
2803eio_ssize_t
1718ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2804eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1719{ 2805{
1720 etp_worker wrk;
1721
1722 wrk.dbuf = 0;
1723
1724 eio__sendfile (ofd, ifd, offset, count, &wrk); 2806 return eio__sendfile (ofd, ifd, offset, count);
1725
1726 if (wrk.dbuf)
1727 free (wrk.dbuf);
1728} 2807}
1729 2808

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines