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

Comparing libeio/eio.c (file contents):
Revision 1.30 by root, Wed Jun 3 12:24:49 2009 UTC vs.
Revision 1.125 by root, Thu Dec 6 12:21:41 2012 UTC

1/* 1/*
2 * libeio implementation 2 * libeio implementation
3 * 3 *
4 * Copyright (c) 2007,2008 Marc Alexander Lehmann <libeio@schmorp.de> 4 * Copyright (c) 2007,2008,2009,2010,2011,2012 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)
73 231
74#else 232#else
75 233
76# include "config.h"
77# include <sys/time.h> 234 #include <sys/time.h>
78# include <sys/select.h> 235 #include <sys/select.h>
236 #include <sys/statvfs.h>
79# 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
80# include <utime.h> 272# include <utime.h>
81# include <signal.h>
82# include <dirent.h>
83
84# ifndef EIO_STRUCT_DIRENT
85# define EIO_STRUCT_DIRENT struct dirent
86# endif 273#endif
87 274
275#if HAVE_SYS_SYSCALL_H
276# include <sys/syscall.h>
277#endif
278
279#if HAVE_SYS_PRCTL_H
280# include <sys/prctl.h>
88#endif 281#endif
89 282
90#if HAVE_SENDFILE 283#if HAVE_SENDFILE
91# if __linux 284# if __linux
92# include <sys/sendfile.h> 285# include <sys/sendfile.h>
93# elif __freebsd 286# elif __FreeBSD__ || defined __APPLE__
94# include <sys/socket.h> 287# include <sys/socket.h>
95# include <sys/uio.h> 288# include <sys/uio.h>
96# elif __hpux 289# elif __hpux
97# include <sys/socket.h> 290# include <sys/socket.h>
98# elif __solaris /* not yet */ 291# elif __solaris
99# include <sys/sendfile.h> 292# include <sys/sendfile.h>
100# else 293# else
101# error sendfile support requested but not available 294# error sendfile support requested but not available
102# endif 295# endif
103#endif 296#endif
104 297
105/* number of seconds after which an idle threads exit */ 298#ifndef D_TYPE
106#define IDLE_TIMEOUT 10 299# define D_TYPE(de) 0
300#endif
301#ifndef D_INO
302# define D_INO(de) 0
303#endif
304#ifndef D_NAMLEN
305# define D_NAMLEN(entp) strlen (D_NAME (entp))
306#endif
107 307
108/* used for struct dirent, AIX doesn't provide it */ 308/* used for struct dirent, AIX doesn't provide it */
109#ifndef NAME_MAX 309#ifndef NAME_MAX
110# define NAME_MAX 4096 310# define NAME_MAX 4096
111#endif 311#endif
112 312
313/* used for readlink etc. */
314#ifndef PATH_MAX
315# define PATH_MAX 4096
316#endif
317
113/* buffer size for various temporary buffers */ 318/* buffer size for various temporary buffers */
114#define EIO_BUFSIZE 65536 319#define EIO_BUFSIZE 65536
115 320
116#define dBUF \ 321#define dBUF \
117 char *eio_buf; \
118 ETP_WORKER_LOCK (self); \
119 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 322 char *eio_buf = malloc (EIO_BUFSIZE); \
120 ETP_WORKER_UNLOCK (self); \
121 errno = ENOMEM; \ 323 errno = ENOMEM; \
122 if (!eio_buf) \ 324 if (!eio_buf) \
123 return -1; 325 return -1
326
327#define FUBd \
328 free (eio_buf)
124 329
125#define EIO_TICKS ((1000000 + 1023) >> 10) 330#define EIO_TICKS ((1000000 + 1023) >> 10)
126 331
127/*****************************************************************************/ 332/*****************************************************************************/
128 333
129#if __GNUC__ >= 3 334struct tmpbuf
130# 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
131#else 360#else
132# 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);
133#endif 363#endif
134 364
135#define expect_false(expr) expect ((expr) != 0, 0) 365struct eio_pwd
136#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};
137 373
138/*****************************************************************************/ 374/*****************************************************************************/
139 375
140#define ETP_PRI_MIN EIO_PRI_MIN 376#define ETP_PRI_MIN EIO_PRI_MIN
141#define ETP_PRI_MAX EIO_PRI_MAX 377#define ETP_PRI_MAX EIO_PRI_MAX
… …
147static int eio_finish (eio_req *req); 383static int eio_finish (eio_req *req);
148#define ETP_FINISH(req) eio_finish (req) 384#define ETP_FINISH(req) eio_finish (req)
149static void eio_execute (struct etp_worker *self, eio_req *req); 385static void eio_execute (struct etp_worker *self, eio_req *req);
150#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) 386#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
151 387
152#define ETP_WORKER_CLEAR(req) \
153 if (wrk->dbuf) \
154 { \
155 free (wrk->dbuf); \
156 wrk->dbuf = 0; \
157 } \
158 \
159 if (wrk->dirp) \
160 { \
161 closedir (wrk->dirp); \
162 wrk->dirp = 0; \
163 }
164#define ETP_WORKER_COMMON \
165 void *dbuf; \
166 DIR *dirp;
167
168/*****************************************************************************/ 388/*****************************************************************************/
169 389
170#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) 390#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
171 391
172/* calculcate time difference in ~1/EIO_TICKS of a second */ 392/* calculate time difference in ~1/EIO_TICKS of a second */
393ecb_inline int
173static int tvdiff (struct timeval *tv1, struct timeval *tv2) 394tvdiff (struct timeval *tv1, struct timeval *tv2)
174{ 395{
175 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 396 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
176 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 397 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
177} 398}
178 399
… …
182static void (*done_poll_cb) (void); 403static void (*done_poll_cb) (void);
183 404
184static unsigned int max_poll_time; /* reslock */ 405static unsigned int max_poll_time; /* reslock */
185static unsigned int max_poll_reqs; /* reslock */ 406static unsigned int max_poll_reqs; /* reslock */
186 407
187static volatile unsigned int nreqs; /* reqlock */ 408static unsigned int nreqs; /* reqlock */
188static volatile unsigned int nready; /* reqlock */ 409static unsigned int nready; /* reqlock */
189static volatile unsigned int npending; /* reqlock */ 410static unsigned int npending; /* reqlock */
190static 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 */
191 413
192static mutex_t wrklock = X_MUTEX_INIT; 414static xmutex_t wrklock;
193static mutex_t reslock = X_MUTEX_INIT; 415static xmutex_t reslock;
194static mutex_t reqlock = X_MUTEX_INIT; 416static xmutex_t reqlock;
195static cond_t reqwait = X_COND_INIT; 417static xcond_t reqwait;
196
197#if !HAVE_PREADWRITE
198/*
199 * make our pread/pwrite emulation safe against themselves, but not against
200 * normal read/write by using a mutex. slows down execution a lot,
201 * but that's your problem, not mine.
202 */
203static mutex_t preadwritelock = X_MUTEX_INIT;
204#endif
205 418
206typedef struct etp_worker 419typedef struct etp_worker
207{ 420{
421 struct tmpbuf tmpbuf;
422
208 /* locked by wrklock */ 423 /* locked by wrklock */
209 struct etp_worker *prev, *next; 424 struct etp_worker *prev, *next;
210 425
211 thread_t tid; 426 xthread_t tid;
212 427
213 /* locked by reslock, reqlock or wrklock */ 428#ifdef ETP_WORKER_COMMON
214 ETP_REQ *req; /* currently processed request */
215
216 ETP_WORKER_COMMON 429 ETP_WORKER_COMMON
430#endif
217} etp_worker; 431} etp_worker;
218 432
219static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */ 433static etp_worker wrk_first; /* NOT etp */
220 434
221#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock) 435#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
222#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock) 436#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
223 437
224/* worker threads management */ 438/* worker threads management */
225 439
440static void
226static void etp_worker_clear (etp_worker *wrk) 441etp_worker_clear (etp_worker *wrk)
227{ 442{
228 ETP_WORKER_CLEAR (wrk);
229} 443}
230 444
445static void ecb_cold
231static void etp_worker_free (etp_worker *wrk) 446etp_worker_free (etp_worker *wrk)
232{ 447{
448 free (wrk->tmpbuf.ptr);
449
233 wrk->next->prev = wrk->prev; 450 wrk->next->prev = wrk->prev;
234 wrk->prev->next = wrk->next; 451 wrk->prev->next = wrk->next;
235 452
236 free (wrk); 453 free (wrk);
237} 454}
238 455
239static unsigned int etp_nreqs (void) 456static unsigned int
457etp_nreqs (void)
240{ 458{
241 int retval; 459 int retval;
242 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 460 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
243 retval = nreqs; 461 retval = nreqs;
244 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 462 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
245 return retval; 463 return retval;
246} 464}
247 465
248static unsigned int etp_nready (void) 466static unsigned int
467etp_nready (void)
249{ 468{
250 unsigned int retval; 469 unsigned int retval;
251 470
252 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 471 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
253 retval = nready; 472 retval = nready;
254 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 473 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
255 474
256 return retval; 475 return retval;
257} 476}
258 477
259static unsigned int etp_npending (void) 478static unsigned int
479etp_npending (void)
260{ 480{
261 unsigned int retval; 481 unsigned int retval;
262 482
263 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 483 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
264 retval = npending; 484 retval = npending;
265 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 485 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
266 486
267 return retval; 487 return retval;
268} 488}
269 489
270static unsigned int etp_nthreads (void) 490static unsigned int
491etp_nthreads (void)
271{ 492{
272 unsigned int retval; 493 unsigned int retval;
273 494
274 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 495 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
275 retval = started; 496 retval = started;
… …
289} etp_reqq; 510} etp_reqq;
290 511
291static etp_reqq req_queue; 512static etp_reqq req_queue;
292static etp_reqq res_queue; 513static etp_reqq res_queue;
293 514
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
294static int reqq_push (etp_reqq *q, ETP_REQ *req) 527reqq_push (etp_reqq *q, ETP_REQ *req)
295{ 528{
296 int pri = req->pri; 529 int pri = req->pri;
297 req->next = 0; 530 req->next = 0;
298 531
299 if (q->qe[pri]) 532 if (q->qe[pri])
… …
305 q->qe[pri] = q->qs[pri] = req; 538 q->qe[pri] = q->qs[pri] = req;
306 539
307 return q->size++; 540 return q->size++;
308} 541}
309 542
543static ETP_REQ * ecb_noinline
310static ETP_REQ *reqq_shift (etp_reqq *q) 544reqq_shift (etp_reqq *q)
311{ 545{
312 int pri; 546 int pri;
313 547
314 if (!q->size) 548 if (!q->size)
315 return 0; 549 return 0;
… …
330 } 564 }
331 565
332 abort (); 566 abort ();
333} 567}
334 568
335static void etp_atfork_prepare (void) 569static int ecb_cold
570etp_init (void (*want_poll)(void), void (*done_poll)(void))
336{ 571{
337 X_LOCK (wrklock); 572 X_MUTEX_CREATE (wrklock);
338 X_LOCK (reqlock); 573 X_MUTEX_CREATE (reslock);
339 X_LOCK (reslock); 574 X_MUTEX_CREATE (reqlock);
340#if !HAVE_PREADWRITE 575 X_COND_CREATE (reqwait);
341 X_LOCK (preadwritelock);
342#endif
343}
344 576
345static void etp_atfork_parent (void) 577 reqq_init (&req_queue);
346{ 578 reqq_init (&res_queue);
347#if !HAVE_PREADWRITE
348 X_UNLOCK (preadwritelock);
349#endif
350 X_UNLOCK (reslock);
351 X_UNLOCK (reqlock);
352 X_UNLOCK (wrklock);
353}
354 579
355static void etp_atfork_child (void) 580 wrk_first.next =
356{ 581 wrk_first.prev = &wrk_first;
357 ETP_REQ *prv;
358
359 while ((prv = reqq_shift (&req_queue)))
360 ETP_DESTROY (prv);
361
362 while ((prv = reqq_shift (&res_queue)))
363 ETP_DESTROY (prv);
364
365 while (wrk_first.next != &wrk_first)
366 {
367 etp_worker *wrk = wrk_first.next;
368
369 if (wrk->req)
370 ETP_DESTROY (wrk->req);
371
372 etp_worker_clear (wrk);
373 etp_worker_free (wrk);
374 }
375 582
376 started = 0; 583 started = 0;
377 idle = 0; 584 idle = 0;
378 nreqs = 0; 585 nreqs = 0;
379 nready = 0; 586 nready = 0;
380 npending = 0; 587 npending = 0;
381 588
382 etp_atfork_parent ();
383}
384
385static void
386etp_once_init (void)
387{
388 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
389}
390
391static int
392etp_init (void (*want_poll)(void), void (*done_poll)(void))
393{
394 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
395
396 pthread_once (&doinit, etp_once_init);
397
398 want_poll_cb = want_poll; 589 want_poll_cb = want_poll;
399 done_poll_cb = done_poll; 590 done_poll_cb = done_poll;
400 591
401 return 0; 592 return 0;
402} 593}
403 594
404X_THREAD_PROC (etp_proc); 595X_THREAD_PROC (etp_proc);
405 596
597static void ecb_cold
406static void etp_start_thread (void) 598etp_start_thread (void)
407{ 599{
408 etp_worker *wrk = calloc (1, sizeof (etp_worker)); 600 etp_worker *wrk = calloc (1, sizeof (etp_worker));
409 601
410 /*TODO*/ 602 /*TODO*/
411 assert (("unable to allocate worker thread data", wrk)); 603 assert (("unable to allocate worker thread data", wrk));
412 604
413 X_LOCK (wrklock); 605 X_LOCK (wrklock);
414 606
415 if (thread_create (&wrk->tid, etp_proc, (void *)wrk)) 607 if (xthread_create (&wrk->tid, etp_proc, (void *)wrk))
416 { 608 {
417 wrk->prev = &wrk_first; 609 wrk->prev = &wrk_first;
418 wrk->next = wrk_first.next; 610 wrk->next = wrk_first.next;
419 wrk_first.next->prev = wrk; 611 wrk_first.next->prev = wrk;
420 wrk_first.next = wrk; 612 wrk_first.next = wrk;
… …
424 free (wrk); 616 free (wrk);
425 617
426 X_UNLOCK (wrklock); 618 X_UNLOCK (wrklock);
427} 619}
428 620
621static void
429static void etp_maybe_start_thread (void) 622etp_maybe_start_thread (void)
430{ 623{
431 if (expect_true (etp_nthreads () >= wanted)) 624 if (ecb_expect_true (etp_nthreads () >= wanted))
432 return; 625 return;
433 626
434 /* todo: maybe use idle here, but might be less exact */ 627 /* todo: maybe use idle here, but might be less exact */
435 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 ()))
436 return; 629 return;
437 630
438 etp_start_thread (); 631 etp_start_thread ();
439} 632}
440 633
634static void ecb_cold
441static void etp_end_thread (void) 635etp_end_thread (void)
442{ 636{
443 eio_req *req = calloc (1, sizeof (eio_req)); 637 eio_req *req = calloc (1, sizeof (eio_req)); /* will be freed by worker */
444 638
445 req->type = -1; 639 req->type = -1;
446 req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 640 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
447 641
448 X_LOCK (reqlock); 642 X_LOCK (reqlock);
… …
453 X_LOCK (wrklock); 647 X_LOCK (wrklock);
454 --started; 648 --started;
455 X_UNLOCK (wrklock); 649 X_UNLOCK (wrklock);
456} 650}
457 651
458static int etp_poll (void) 652static int
653etp_poll (void)
459{ 654{
460 unsigned int maxreqs; 655 unsigned int maxreqs;
461 unsigned int maxtime; 656 unsigned int maxtime;
462 struct timeval tv_start, tv_now; 657 struct timeval tv_start, tv_now;
463 658
… …
493 688
494 X_LOCK (reqlock); 689 X_LOCK (reqlock);
495 --nreqs; 690 --nreqs;
496 X_UNLOCK (reqlock); 691 X_UNLOCK (reqlock);
497 692
498 if (expect_false (req->type == EIO_GROUP && req->size)) 693 if (ecb_expect_false (req->type == EIO_GROUP && req->size))
499 { 694 {
500 req->int1 = 1; /* mark request as delayed */ 695 req->int1 = 1; /* mark request as delayed */
501 continue; 696 continue;
502 } 697 }
503 else 698 else
504 { 699 {
505 int res = ETP_FINISH (req); 700 int res = ETP_FINISH (req);
506 if (expect_false (res)) 701 if (ecb_expect_false (res))
507 return res; 702 return res;
508 } 703 }
509 704
510 if (expect_false (maxreqs && !--maxreqs)) 705 if (ecb_expect_false (maxreqs && !--maxreqs))
511 break; 706 break;
512 707
513 if (maxtime) 708 if (maxtime)
514 { 709 {
515 gettimeofday (&tv_now, 0); 710 gettimeofday (&tv_now, 0);
… …
521 716
522 errno = EAGAIN; 717 errno = EAGAIN;
523 return -1; 718 return -1;
524} 719}
525 720
721static void
526static void etp_cancel (ETP_REQ *req) 722etp_cancel (ETP_REQ *req)
527{ 723{
528 X_LOCK (wrklock); 724 req->cancelled = 1;
529 req->flags |= EIO_FLAG_CANCELLED;
530 X_UNLOCK (wrklock);
531 725
532 eio_grp_cancel (req); 726 eio_grp_cancel (req);
533} 727}
534 728
729static void
535static void etp_submit (ETP_REQ *req) 730etp_submit (ETP_REQ *req)
536{ 731{
537 req->pri -= ETP_PRI_MIN; 732 req->pri -= ETP_PRI_MIN;
538 733
539 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;
540 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;
541 736
542 if (expect_false (req->type == EIO_GROUP)) 737 if (ecb_expect_false (req->type == EIO_GROUP))
543 { 738 {
544 /* I hope this is worth it :/ */ 739 /* I hope this is worth it :/ */
545 X_LOCK (reqlock); 740 X_LOCK (reqlock);
546 ++nreqs; 741 ++nreqs;
547 X_UNLOCK (reqlock); 742 X_UNLOCK (reqlock);
… …
566 761
567 etp_maybe_start_thread (); 762 etp_maybe_start_thread ();
568 } 763 }
569} 764}
570 765
766static void ecb_cold
571static void etp_set_max_poll_time (double nseconds) 767etp_set_max_poll_time (double nseconds)
572{ 768{
573 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 769 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
574 max_poll_time = nseconds; 770 max_poll_time = nseconds * EIO_TICKS;
575 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 771 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
576} 772}
577 773
774static void ecb_cold
578static void etp_set_max_poll_reqs (unsigned int maxreqs) 775etp_set_max_poll_reqs (unsigned int maxreqs)
579{ 776{
580 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 777 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
581 max_poll_reqs = maxreqs; 778 max_poll_reqs = maxreqs;
582 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 779 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
583} 780}
584 781
782static void ecb_cold
585static void etp_set_max_idle (unsigned int nthreads) 783etp_set_max_idle (unsigned int nthreads)
586{ 784{
587 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 785 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
588 max_idle = nthreads <= 0 ? 1 : nthreads; 786 max_idle = nthreads;
589 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 787 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
590} 788}
591 789
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
592static void etp_set_min_parallel (unsigned int nthreads) 799etp_set_min_parallel (unsigned int nthreads)
593{ 800{
594 if (wanted < nthreads) 801 if (wanted < nthreads)
595 wanted = nthreads; 802 wanted = nthreads;
596} 803}
597 804
805static void ecb_cold
598static void etp_set_max_parallel (unsigned int nthreads) 806etp_set_max_parallel (unsigned int nthreads)
599{ 807{
600 if (wanted > nthreads) 808 if (wanted > nthreads)
601 wanted = nthreads; 809 wanted = nthreads;
602 810
603 while (started > wanted) 811 while (started > wanted)
604 etp_end_thread (); 812 etp_end_thread ();
605} 813}
606 814
607/*****************************************************************************/ 815/*****************************************************************************/
608 816
817static void
609static void grp_try_feed (eio_req *grp) 818grp_try_feed (eio_req *grp)
610{ 819{
611 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 820 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
612 { 821 {
613 grp->flags &= ~EIO_FLAG_GROUPADD; 822 grp->flags &= ~EIO_FLAG_GROUPADD;
614 823
… …
621 break; 830 break;
622 } 831 }
623 } 832 }
624} 833}
625 834
835static int
626static int grp_dec (eio_req *grp) 836grp_dec (eio_req *grp)
627{ 837{
628 --grp->size; 838 --grp->size;
629 839
630 /* call feeder, if applicable */ 840 /* call feeder, if applicable */
631 grp_try_feed (grp); 841 grp_try_feed (grp);
… …
635 return eio_finish (grp); 845 return eio_finish (grp);
636 else 846 else
637 return 0; 847 return 0;
638} 848}
639 849
850static void
640void eio_destroy (eio_req *req) 851eio_destroy (eio_req *req)
641{ 852{
642 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); 853 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
643 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); 854 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
644 855
645 EIO_DESTROY (req); 856 EIO_DESTROY (req);
646} 857}
647 858
859static int
648static int eio_finish (eio_req *req) 860eio_finish (eio_req *req)
649{ 861{
650 int res = EIO_FINISH (req); 862 int res = EIO_FINISH (req);
651 863
652 if (req->grp) 864 if (req->grp)
653 { 865 {
… …
661 if (grp->grp_first == req) 873 if (grp->grp_first == req)
662 grp->grp_first = req->grp_next; 874 grp->grp_first = req->grp_next;
663 875
664 res2 = grp_dec (grp); 876 res2 = grp_dec (grp);
665 877
666 if (!res && res2) 878 if (!res)
667 res = res2; 879 res = res2;
668 } 880 }
669 881
670 eio_destroy (req); 882 eio_destroy (req);
671 883
672 return res; 884 return res;
673} 885}
674 886
887void
675void eio_grp_cancel (eio_req *grp) 888eio_grp_cancel (eio_req *grp)
676{ 889{
677 for (grp = grp->grp_first; grp; grp = grp->grp_next) 890 for (grp = grp->grp_first; grp; grp = grp->grp_next)
678 eio_cancel (grp); 891 eio_cancel (grp);
679} 892}
680 893
894void
681void eio_cancel (eio_req *req) 895eio_cancel (eio_req *req)
682{ 896{
683 etp_cancel (req); 897 etp_cancel (req);
684} 898}
685 899
900void
686void eio_submit (eio_req *req) 901eio_submit (eio_req *req)
687{ 902{
688 etp_submit (req); 903 etp_submit (req);
689} 904}
690 905
691unsigned int eio_nreqs (void) 906unsigned int
907eio_nreqs (void)
692{ 908{
693 return etp_nreqs (); 909 return etp_nreqs ();
694} 910}
695 911
696unsigned int eio_nready (void) 912unsigned int
913eio_nready (void)
697{ 914{
698 return etp_nready (); 915 return etp_nready ();
699} 916}
700 917
701unsigned int eio_npending (void) 918unsigned int
919eio_npending (void)
702{ 920{
703 return etp_npending (); 921 return etp_npending ();
704} 922}
705 923
706unsigned int eio_nthreads (void) 924unsigned int ecb_cold
925eio_nthreads (void)
707{ 926{
708 return etp_nthreads (); 927 return etp_nthreads ();
709} 928}
710 929
930void ecb_cold
711void eio_set_max_poll_time (double nseconds) 931eio_set_max_poll_time (double nseconds)
712{ 932{
713 etp_set_max_poll_time (nseconds); 933 etp_set_max_poll_time (nseconds);
714} 934}
715 935
936void ecb_cold
716void eio_set_max_poll_reqs (unsigned int maxreqs) 937eio_set_max_poll_reqs (unsigned int maxreqs)
717{ 938{
718 etp_set_max_poll_reqs (maxreqs); 939 etp_set_max_poll_reqs (maxreqs);
719} 940}
720 941
942void ecb_cold
721void eio_set_max_idle (unsigned int nthreads) 943eio_set_max_idle (unsigned int nthreads)
722{ 944{
723 etp_set_max_idle (nthreads); 945 etp_set_max_idle (nthreads);
724} 946}
725 947
948void ecb_cold
949eio_set_idle_timeout (unsigned int seconds)
950{
951 etp_set_idle_timeout (seconds);
952}
953
954void ecb_cold
726void eio_set_min_parallel (unsigned int nthreads) 955eio_set_min_parallel (unsigned int nthreads)
727{ 956{
728 etp_set_min_parallel (nthreads); 957 etp_set_min_parallel (nthreads);
729} 958}
730 959
960void ecb_cold
731void eio_set_max_parallel (unsigned int nthreads) 961eio_set_max_parallel (unsigned int nthreads)
732{ 962{
733 etp_set_max_parallel (nthreads); 963 etp_set_max_parallel (nthreads);
734} 964}
735 965
736int eio_poll (void) 966int eio_poll (void)
… …
739} 969}
740 970
741/*****************************************************************************/ 971/*****************************************************************************/
742/* work around various missing functions */ 972/* work around various missing functions */
743 973
744#if !HAVE_PREADWRITE
745# undef pread
746# undef pwrite
747# define pread eio__pread
748# define pwrite eio__pwrite
749
750static ssize_t
751eio__pread (int fd, void *buf, size_t count, off_t offset)
752{
753 ssize_t res;
754 off_t ooffset;
755
756 X_LOCK (preadwritelock);
757 ooffset = lseek (fd, 0, SEEK_CUR);
758 lseek (fd, offset, SEEK_SET);
759 res = read (fd, buf, count);
760 lseek (fd, ooffset, SEEK_SET);
761 X_UNLOCK (preadwritelock);
762
763 return res;
764}
765
766static ssize_t
767eio__pwrite (int fd, void *buf, size_t count, off_t offset)
768{
769 ssize_t res;
770 off_t ooffset;
771
772 X_LOCK (preadwritelock);
773 ooffset = lseek (fd, 0, SEEK_CUR);
774 lseek (fd, offset, SEEK_SET);
775 res = write (fd, buf, count);
776 lseek (fd, ooffset, SEEK_SET);
777 X_UNLOCK (preadwritelock);
778
779 return res;
780}
781#endif
782
783#ifndef HAVE_FUTIMES 974#ifndef HAVE_UTIMES
784 975
785# undef utimes 976# undef utimes
786# undef futimes
787# define utimes(path,times) eio__utimes (path, times) 977# define utimes(path,times) eio__utimes (path, times)
788# define futimes(fd,times) eio__futimes (fd, times)
789 978
790static int 979static int
791eio__utimes (const char *filename, const struct timeval times[2]) 980eio__utimes (const char *filename, const struct timeval times[2])
792{ 981{
793 if (times) 982 if (times)
… …
801 } 990 }
802 else 991 else
803 return utime (filename, 0); 992 return utime (filename, 0);
804} 993}
805 994
995#endif
996
997#ifndef HAVE_FUTIMES
998
999# undef futimes
1000# define futimes(fd,times) eio__futimes (fd, times)
1001
1002static int
806static int eio__futimes (int fd, const struct timeval tv[2]) 1003eio__futimes (int fd, const struct timeval tv[2])
807{ 1004{
808 errno = ENOSYS; 1005 errno = ENOSYS;
809 return -1; 1006 return -1;
810} 1007}
811 1008
… …
814#if !HAVE_FDATASYNC 1011#if !HAVE_FDATASYNC
815# undef fdatasync 1012# undef fdatasync
816# define fdatasync(fd) fsync (fd) 1013# define fdatasync(fd) fsync (fd)
817#endif 1014#endif
818 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
819/* sync_file_range always needs emulation */ 1033/* sync_file_range always needs emulation */
820int 1034static int
821eio__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)
822{ 1036{
823#if HAVE_SYNC_FILE_RANGE 1037#if HAVE_SYNC_FILE_RANGE
824 int res; 1038 int res;
825 1039
… …
838 if (!res || errno != ENOSYS) 1052 if (!res || errno != ENOSYS)
839 return res; 1053 return res;
840#endif 1054#endif
841 1055
842 /* 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
843 * call fdatasync, as thta matches the expectation of it's users best */ 1057 * call fdatasync, as that matches the expectation of its users best */
844 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
845} 1069}
846 1070
847#if !HAVE_READAHEAD 1071#if !HAVE_READAHEAD
848# undef readahead 1072# undef readahead
849# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 1073# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
850 1074
851static ssize_t 1075static eio_ssize_t
852eio__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)
853{ 1077{
854 size_t todo = count; 1078 size_t todo = count;
855 dBUF; 1079 dBUF;
856 1080
… …
861 pread (fd, eio_buf, len, offset); 1085 pread (fd, eio_buf, len, offset);
862 offset += len; 1086 offset += len;
863 todo -= len; 1087 todo -= len;
864 } 1088 }
865 1089
866 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 */
867 return count; 1094 return 0;
868} 1095}
869 1096
870#endif 1097#endif
871 1098
872/* sendfile always needs emulation */ 1099/* sendfile always needs emulation */
873static ssize_t 1100static eio_ssize_t
874eio__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)
875{ 1102{
1103 eio_ssize_t written = 0;
876 ssize_t res; 1104 eio_ssize_t res;
877 1105
878 if (!count) 1106 if (!count)
879 return 0; 1107 return 0;
880 1108
1109 for (;;)
1110 {
1111#ifdef __APPLE__
1112# undef HAVE_SENDFILE /* broken, as everything on os x */
1113#endif
881#if HAVE_SENDFILE 1114#if HAVE_SENDFILE
882# if __linux 1115# if __linux
1116 off_t soffset = offset;
883 res = sendfile (ofd, ifd, &offset, count); 1117 res = sendfile (ofd, ifd, &soffset, count);
884 1118
885# elif __freebsd 1119# elif __FreeBSD__
886 /* 1120 /*
887 * 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
888 * wasted on making it similar to other I/O functions. 1122 * wasted on making it similar to other I/O functions.
889 */ 1123 */
890 {
891 off_t sbytes; 1124 off_t sbytes;
892 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 1125 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
893 1126
894 if (res < 0 && sbytes) 1127 #if 0 /* according to the manpage, this is correct, but broken behaviour */
895 /* 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)
896 res = sbytes; 1132 res = sbytes;
897 } 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;
898 1146
899# elif __hpux 1147# elif __hpux
900 res = sendfile (ofd, ifd, offset, count, 0, 0); 1148 res = sendfile (ofd, ifd, offset, count, 0, 0);
901 1149
902# elif __solaris 1150# elif __solaris
903 {
904 struct sendfilevec vec; 1151 struct sendfilevec vec;
905 size_t sbytes; 1152 size_t sbytes;
906 1153
907 vec.sfv_fd = ifd; 1154 vec.sfv_fd = ifd;
908 vec.sfv_flag = 0; 1155 vec.sfv_flag = 0;
909 vec.sfv_off = offset; 1156 vec.sfv_off = offset;
910 vec.sfv_len = count; 1157 vec.sfv_len = count;
911 1158
912 res = sendfilev (ofd, &vec, 1, &sbytes); 1159 res = sendfilev (ofd, &vec, 1, &sbytes);
913 1160
914 if (res < 0 && sbytes) 1161 if (res < 0 && sbytes)
915 res = sbytes; 1162 res = sbytes;
916 }
917 1163
918# 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
919#else 1175#else
920 res = -1; 1176 res = EIO_ENOSYS ();
921 errno = ENOSYS;
922#endif 1177#endif
923 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
924 if (res < 0 1202 if (res < 0
925 && (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
926#if __solaris 1211#if __solaris
927 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1212 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
928#endif 1213#endif
929 ) 1214 )
930 ) 1215 )
… …
934 1219
935 res = 0; 1220 res = 0;
936 1221
937 while (count) 1222 while (count)
938 { 1223 {
939 ssize_t cnt; 1224 eio_ssize_t cnt;
940 1225
941 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);
942 1227
943 if (cnt <= 0) 1228 if (cnt <= 0)
944 { 1229 {
… …
956 1241
957 offset += cnt; 1242 offset += cnt;
958 res += cnt; 1243 res += cnt;
959 count -= cnt; 1244 count -= cnt;
960 } 1245 }
1246
1247 FUBd;
961 } 1248 }
962 1249
963 return res; 1250 return res;
1251}
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
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 close (fd);
1422
1423 /* here we should probably stat the open file and the disk file, to make sure they still match */
1424
1425 if (req->result > 0)
1426 goto done;
1427 }
1428 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1429 return;
1430 }
1431#endif
1432#endif
1433
1434 if (*rel != '/')
1435 {
1436 int len;
1437
1438 errno = ENOENT;
1439 if (wd == EIO_INVALID_WD)
1440 return -1;
1441
1442 if (wd == EIO_CWD)
1443 {
1444 if (!getcwd (res, PATH_MAX))
1445 return -1;
1446
1447 len = strlen (res);
1448 }
1449 else
1450 memcpy (res, wd->str, len = wd->len);
1451
1452 if (res [1]) /* only use if not / */
1453 res += len;
1454 }
1455
1456 while (*rel)
1457 {
1458 eio_ssize_t len, linklen;
1459 const char *beg = rel;
1460
1461 while (*rel && *rel != '/')
1462 ++rel;
1463
1464 len = rel - beg;
1465
1466 if (!len) /* skip slashes */
1467 {
1468 ++rel;
1469 continue;
1470 }
1471
1472 if (beg [0] == '.')
1473 {
1474 if (len == 1)
1475 continue; /* . - nop */
1476
1477 if (beg [1] == '.' && len == 2)
1478 {
1479 /* .. - back up one component, if possible */
1480
1481 while (res != tmpbuf->ptr)
1482 if (*--res == '/')
1483 break;
1484
1485 continue;
1486 }
1487 }
1488
1489 errno = ENAMETOOLONG;
1490 if (res + 1 + len + 1 >= tmp1)
1491 return -1;
1492
1493 /* copy one component */
1494 *res = '/';
1495 memcpy (res + 1, beg, len);
1496
1497 /* zero-terminate, for readlink */
1498 res [len + 1] = 0;
1499
1500 /* now check if it's a symlink */
1501 linklen = readlink (tmpbuf->ptr, tmp1, PATH_MAX);
1502
1503 if (linklen < 0)
1504 {
1505 if (errno != EINVAL)
1506 return -1;
1507
1508 /* it's a normal directory. hopefully */
1509 res += len + 1;
1510 }
1511 else
1512 {
1513 /* yay, it was a symlink - build new path in tmp2 */
1514 int rellen = strlen (rel);
1515
1516 errno = ENAMETOOLONG;
1517 if (linklen + 1 + rellen >= PATH_MAX)
1518 return -1;
1519
1520 errno = ELOOP;
1521 if (!--symlinks)
1522 return -1;
1523
1524 if (*tmp1 == '/')
1525 res = tmpbuf->ptr; /* symlink resolves to an absolute path */
1526
1527 /* we need to be careful, as rel might point into tmp2 already */
1528 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1529 tmp2 [linklen] = '/';
1530 memcpy (tmp2, tmp1, linklen);
1531
1532 rel = tmp2;
1533 }
1534 }
1535
1536 /* special case for the lone root path */
1537 if (res == tmpbuf->ptr)
1538 *res++ = '/';
1539
1540 return res - (char *)tmpbuf->ptr;
1541}
1542
1543static signed char
1544eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1545{
1546 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1547 : a->inode < b->inode ? -1
1548 : a->inode > b->inode ? 1
1549 : 0;
1550}
1551
1552#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1553
1554#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1555#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1556
1557static void
1558eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1559{
1560 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1561 unsigned char *bit = bits;
1562
1563 assert (CHAR_BIT == 8);
1564 assert (sizeof (eio_dirent) * 8 < 256);
1565 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1566 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1567
1568 if (size <= EIO_SORT_FAST)
1569 return;
1570
1571 /* first prepare an array of bits to test in our radix sort */
1572 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1573 /* inode_bits must contain all inodes ORed together */
1574 /* which is used to skip bits that are 0 everywhere, which is very common */
1575 {
1576 eio_ino_t endianness;
1577 int i, j;
1578
1579 /* we store the byte offset of byte n into byte n of "endianness" */
1580 for (i = 0; i < sizeof (eio_ino_t); ++i)
1581 ((unsigned char *)&endianness)[i] = i;
1582
1583 *bit++ = 0;
1584
1585 for (i = 0; i < sizeof (eio_ino_t); ++i)
1586 {
1587 /* shifting off the byte offsets out of "endianness" */
1588 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1589 endianness >>= 8;
1590
1591 for (j = 0; j < 8; ++j)
1592 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1593 *bit++ = offs + j;
1594 }
1595
1596 for (j = 0; j < 8; ++j)
1597 if (score_bits & (1 << j))
1598 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1599 }
1600
1601 /* now actually do the sorting (a variant of MSD radix sort) */
1602 {
1603 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1604 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1605 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1606 int stk_idx = 0;
1607
1608 base_stk [stk_idx] = dents;
1609 end_stk [stk_idx] = dents + size;
1610 bit_stk [stk_idx] = bit - 1;
1611
1612 do
1613 {
1614 base = base_stk [stk_idx];
1615 end = end_stk [stk_idx];
1616 bit = bit_stk [stk_idx];
1617
1618 for (;;)
1619 {
1620 unsigned char O = *bit >> 3;
1621 unsigned char M = 1 << (*bit & 7);
1622
1623 eio_dirent *a = base;
1624 eio_dirent *b = end;
1625
1626 if (b - a < EIO_SORT_CUTOFF)
1627 break;
1628
1629 /* now bit-partition the array on the bit */
1630 /* this ugly asymmetric loop seems to perform much better than typical */
1631 /* partition algos found in the literature */
1632 do
1633 if (!(((unsigned char *)a)[O] & M))
1634 ++a;
1635 else if (!(((unsigned char *)--b)[O] & M))
1636 {
1637 eio_dirent tmp = *a; *a = *b; *b = tmp;
1638 ++a;
1639 }
1640 while (b > a);
1641
1642 /* next bit, or stop, if no bits left in this path */
1643 if (!*--bit)
1644 break;
1645
1646 base_stk [stk_idx] = a;
1647 end_stk [stk_idx] = end;
1648 bit_stk [stk_idx] = bit;
1649 ++stk_idx;
1650
1651 end = a;
1652 }
1653 }
1654 while (stk_idx--);
1655 }
1656}
1657
1658static void
1659eio_dent_insertion_sort (eio_dirent *dents, int size)
1660{
1661 /* first move the smallest element to the front, to act as a sentinel */
1662 {
1663 int i;
1664 eio_dirent *min = dents;
1665
1666 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1667 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1668 if (EIO_DENT_CMP (dents [i], <, *min))
1669 min = &dents [i];
1670
1671 /* swap elements 0 and j (minimum) */
1672 {
1673 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1674 }
1675 }
1676
1677 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1678 {
1679 eio_dirent *i, *j;
1680
1681 for (i = dents + 1; i < dents + size; ++i)
1682 {
1683 eio_dirent value = *i;
1684
1685 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1686 j [1] = j [0];
1687
1688 j [1] = value;
1689 }
1690 }
1691}
1692
1693static void
1694eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1695{
1696 if (size <= 1)
1697 return; /* our insertion sort relies on size > 0 */
1698
1699 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1700 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1701 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1702
1703 /* use an insertion sort at the end, or for small arrays, */
1704 /* as insertion sort is more efficient for small partitions */
1705 eio_dent_insertion_sort (dents, size);
964} 1706}
965 1707
966/* read a full directory */ 1708/* read a full directory */
967static void 1709static void
968eio__scandir (eio_req *req, etp_worker *self) 1710eio__scandir (eio_req *req, etp_worker *self)
969{ 1711{
1712 char *name, *names;
1713 int namesalloc = 4096 - sizeof (void *) * 4;
1714 int namesoffs = 0;
1715 int flags = req->int1;
1716 eio_dirent *dents = 0;
1717 int dentalloc = 128;
1718 int dentoffs = 0;
1719 eio_ino_t inode_bits = 0;
1720#ifdef _WIN32
1721 HANDLE dirp;
1722 WIN32_FIND_DATA entp;
1723#else
970 DIR *dirp; 1724 DIR *dirp;
971 EIO_STRUCT_DIRENT *entp; 1725 EIO_STRUCT_DIRENT *entp;
972 char *name, *names; 1726#endif
973 int memlen = 4096;
974 int memofs = 0;
975 int res = 0;
976 1727
977 X_LOCK (wrklock); 1728 req->result = -1;
978 /* the corresponding closedir is in ETP_WORKER_CLEAR */
979 self->dirp = dirp = opendir (req->ptr1);
980 req->flags |= EIO_FLAG_PTR2_FREE;
981 req->ptr2 = names = malloc (memlen);
982 X_UNLOCK (wrklock);
983 1729
984 if (dirp && names) 1730 if (!(flags & EIO_READDIR_DENTS))
985 for (;;) 1731 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1732
1733#ifdef _WIN32
1734 {
1735 int len = strlen ((const char *)req->ptr1);
1736 char *path = malloc (MAX_PATH);
1737 const char *fmt;
1738 const char *reqpath = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1739
1740 if (!len)
1741 fmt = "./*";
1742 else if (reqpath[len - 1] == '/' || reqpath[len - 1] == '\\')
1743 fmt = "%s*";
1744 else
1745 fmt = "%s/*";
1746
1747 _snprintf (path, MAX_PATH, fmt, reqpath);
1748 dirp = FindFirstFile (path, &entp);
1749 free (path);
1750
1751 if (dirp == INVALID_HANDLE_VALUE)
1752 {
1753 /* should steal _dosmaperr */
1754 switch (GetLastError ())
1755 {
1756 case ERROR_FILE_NOT_FOUND:
1757 req->result = 0;
1758 break;
1759
1760 case ERROR_INVALID_NAME:
1761 case ERROR_PATH_NOT_FOUND:
1762 case ERROR_NO_MORE_FILES:
1763 errno = ENOENT;
1764 break;
1765
1766 case ERROR_NOT_ENOUGH_MEMORY:
1767 errno = ENOMEM;
1768 break;
1769
1770 default:
1771 errno = EINVAL;
1772 break;
1773 }
1774
1775 return;
1776 }
1777 }
1778#else
1779 #if HAVE_AT
1780 if (req->wd)
986 { 1781 {
1782 int fd = openat (WD2FD (req->wd), req->ptr1, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1783
1784 if (fd < 0)
1785 return;
1786
1787 dirp = fdopendir (fd);
1788
1789 if (!dirp)
1790 close (fd);
1791 }
1792 else
1793 dirp = opendir (req->ptr1);
1794 #else
1795 dirp = opendir (wd_expand (&self->tmpbuf, req->wd, req->ptr1));
1796 #endif
1797
1798 if (!dirp)
1799 return;
1800#endif
1801
1802 if (req->flags & EIO_FLAG_PTR1_FREE)
1803 free (req->ptr1);
1804
1805 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1806 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1807 req->ptr2 = names = malloc (namesalloc);
1808
1809 if (!names || (flags && !dents))
1810 return;
1811
1812 for (;;)
1813 {
1814 int done;
1815
1816#ifdef _WIN32
1817 done = !dirp;
1818#else
987 errno = 0; 1819 errno = 0;
988 entp = readdir (dirp); 1820 entp = readdir (dirp);
1821 done = !entp;
1822#endif
989 1823
1824 if (done)
1825 {
1826#ifndef _WIN32
1827 int old_errno = errno;
1828 closedir (dirp);
1829 errno = old_errno;
1830
990 if (!entp) 1831 if (errno)
1832 break;
1833#endif
1834
1835 /* sort etc. */
1836 req->int1 = flags;
1837 req->result = dentoffs;
1838
1839 if (flags & EIO_READDIR_STAT_ORDER)
1840 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1841 else if (flags & EIO_READDIR_DIRS_FIRST)
1842 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1843 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1844 else
1845 {
1846 /* in this case, all is known, and we just put dirs first and sort them */
1847 eio_dirent *oth = dents + dentoffs;
1848 eio_dirent *dir = dents;
1849
1850 /* now partition dirs to the front, and non-dirs to the back */
1851 /* by walking from both sides and swapping if necessary */
1852 while (oth > dir)
1853 {
1854 if (dir->type == EIO_DT_DIR)
1855 ++dir;
1856 else if ((--oth)->type == EIO_DT_DIR)
1857 {
1858 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1859
1860 ++dir;
1861 }
1862 }
1863
1864 /* now sort the dirs only (dirs all have the same score) */
1865 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1866 }
1867
991 break; 1868 break;
1869 }
992 1870
993 name = entp->d_name; 1871 /* now add the entry to our list(s) */
1872 name = D_NAME (entp);
994 1873
1874 /* skip . and .. entries */
995 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1875 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
996 { 1876 {
997 int len = strlen (name) + 1; 1877 int len = D_NAMLEN (entp) + 1;
998 1878
999 res++; 1879 while (ecb_expect_false (namesoffs + len > namesalloc))
1000
1001 while (memofs + len > memlen)
1002 { 1880 {
1003 memlen *= 2; 1881 namesalloc *= 2;
1004 X_LOCK (wrklock);
1005 req->ptr2 = names = realloc (names, memlen); 1882 req->ptr2 = names = realloc (names, namesalloc);
1006 X_UNLOCK (wrklock);
1007 1883
1008 if (!names) 1884 if (!names)
1885 break;
1886 }
1887
1888 memcpy (names + namesoffs, name, len);
1889
1890 if (dents)
1891 {
1892 struct eio_dirent *ent;
1893
1894 if (ecb_expect_false (dentoffs == dentalloc))
1895 {
1896 dentalloc *= 2;
1897 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1898
1899 if (!dents)
1009 break; 1900 break;
1010 } 1901 }
1011 1902
1012 memcpy (names + memofs, name, len); 1903 ent = dents + dentoffs;
1013 memofs += len; 1904
1905 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1906 ent->namelen = len - 1;
1907 ent->inode = D_INO (entp);
1908
1909 inode_bits |= ent->inode;
1910
1911 switch (D_TYPE (entp))
1912 {
1913 default:
1914 ent->type = EIO_DT_UNKNOWN;
1915 flags |= EIO_READDIR_FOUND_UNKNOWN;
1916 break;
1917
1918 #ifdef DT_FIFO
1919 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1920 #endif
1921 #ifdef DT_CHR
1922 case DT_CHR: ent->type = EIO_DT_CHR; break;
1923 #endif
1924 #ifdef DT_MPC
1925 case DT_MPC: ent->type = EIO_DT_MPC; break;
1926 #endif
1927 #ifdef DT_DIR
1928 case DT_DIR: ent->type = EIO_DT_DIR; break;
1929 #endif
1930 #ifdef DT_NAM
1931 case DT_NAM: ent->type = EIO_DT_NAM; break;
1932 #endif
1933 #ifdef DT_BLK
1934 case DT_BLK: ent->type = EIO_DT_BLK; break;
1935 #endif
1936 #ifdef DT_MPB
1937 case DT_MPB: ent->type = EIO_DT_MPB; break;
1938 #endif
1939 #ifdef DT_REG
1940 case DT_REG: ent->type = EIO_DT_REG; break;
1941 #endif
1942 #ifdef DT_NWK
1943 case DT_NWK: ent->type = EIO_DT_NWK; break;
1944 #endif
1945 #ifdef DT_CMP
1946 case DT_CMP: ent->type = EIO_DT_CMP; break;
1947 #endif
1948 #ifdef DT_LNK
1949 case DT_LNK: ent->type = EIO_DT_LNK; break;
1950 #endif
1951 #ifdef DT_SOCK
1952 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1953 #endif
1954 #ifdef DT_DOOR
1955 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1956 #endif
1957 #ifdef DT_WHT
1958 case DT_WHT: ent->type = EIO_DT_WHT; break;
1959 #endif
1960 }
1961
1962 ent->score = 7;
1963
1964 if (flags & EIO_READDIR_DIRS_FIRST)
1965 {
1966 if (ent->type == EIO_DT_UNKNOWN)
1967 {
1968 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1969 ent->score = 1;
1970 else if (!strchr (name, '.')) /* absence of dots indicate likely dirs */
1971 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1972 }
1973 else if (ent->type == EIO_DT_DIR)
1974 ent->score = 0;
1975 }
1014 } 1976 }
1977
1978 namesoffs += len;
1979 ++dentoffs;
1015 } 1980 }
1016 1981
1017 if (errno) 1982 if (EIO_CANCELLED (req))
1018 res = -1; 1983 {
1984 errno = ECANCELED;
1985 break;
1986 }
1987
1988#ifdef _WIN32
1989 if (!FindNextFile (dirp, &entp))
1990 {
1991 FindClose (dirp);
1992 dirp = 0;
1993 }
1994#endif
1995 }
1996}
1997
1998/*****************************************************************************/
1999/* working directory stuff */
2000/* various deficiencies in the posix 2008 api force us to */
2001/* keep the absolute path in string form at all times */
2002/* fuck yeah. */
2003
2004#if !HAVE_AT
2005
2006/* a bit like realpath, but usually faster because it doesn'T have to return */
2007/* an absolute or canonical path */
2008static const char *
2009wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
2010{
2011 if (!wd || *path == '/')
2012 return path;
2013
2014 if (path [0] == '.' && !path [1])
2015 return wd->str;
2016
1019 2017 {
1020 req->result = res; 2018 int l1 = wd->len;
1021} 2019 int l2 = strlen (path);
1022 2020
1023#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) 2021 char *res = tmpbuf_get (tmpbuf, l1 + l2 + 2);
1024# undef msync
1025# define msync(a,b,c) ((errno = ENOSYS), -1)
1026#endif
1027 2022
1028int 2023 memcpy (res, wd->str, l1);
1029eio__mtouch (void *mem, size_t len, int flags) 2024 res [l1] = '/';
1030{ 2025 memcpy (res + l1 + 1, path, l2 + 1);
1031 intptr_t addr = (intptr_t)mem;
1032 intptr_t end = addr + len;
1033#ifdef PAGESIZE
1034 const intptr_t page = PAGESIZE;
1035#else
1036 static intptr_t page;
1037 2026
1038 if (!page) 2027 return res;
1039 page = sysconf (_SC_PAGESIZE); 2028 }
1040#endif 2029}
1041 2030
1042 addr &= ~(page - 1); /* assume page size is always a power of two */ 2031#endif
1043 2032
1044 if (addr < end) 2033static eio_wd
1045 if (flags) /* modify */ 2034eio__wd_open_sync (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
1046 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len); 2035{
1047 else 2036 int fd;
1048 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len); 2037 eio_wd res;
2038 int len = eio__realpath (tmpbuf, wd, path);
1049 2039
2040 if (len < 0)
2041 return EIO_INVALID_WD;
2042
2043#if HAVE_AT
2044 fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
2045
2046 if (fd < 0)
2047 return EIO_INVALID_WD;
2048#endif
2049
2050 res = malloc (sizeof (*res) + len); /* one extra 0-byte */
2051
2052#if HAVE_AT
2053 res->fd = fd;
2054#endif
2055
2056 res->len = len;
2057 memcpy (res->str, tmpbuf->ptr, len);
2058 res->str [len] = 0;
2059
2060 return res;
2061}
2062
2063eio_wd
2064eio_wd_open_sync (eio_wd wd, const char *path)
2065{
2066 struct tmpbuf tmpbuf = { 0 };
2067 wd = eio__wd_open_sync (&tmpbuf, wd, path);
2068 free (tmpbuf.ptr);
2069
1050 return 0; 2070 return wd;
1051} 2071}
2072
2073void
2074eio_wd_close_sync (eio_wd wd)
2075{
2076 if (wd != EIO_INVALID_WD && wd != EIO_CWD)
2077 {
2078 #if HAVE_AT
2079 close (wd->fd);
2080 #endif
2081 free (wd);
2082 }
2083}
2084
2085#if HAVE_AT
2086
2087/* they forgot these */
2088
2089static int
2090eio__truncateat (int dirfd, const char *path, off_t length)
2091{
2092 int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
2093 int res;
2094
2095 if (fd < 0)
2096 return fd;
2097
2098 res = ftruncate (fd, length);
2099 close (fd);
2100 return res;
2101}
2102
2103static int
2104eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
2105{
2106 int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
2107 int res;
2108
2109 if (fd < 0)
2110 return fd;
2111
2112 res = fstatvfs (fd, buf);
2113 close (fd);
2114 return res;
2115
2116}
2117
2118#endif
1052 2119
1053/*****************************************************************************/ 2120/*****************************************************************************/
1054 2121
1055#define ALLOC(len) \ 2122#define ALLOC(len) \
1056 if (!req->ptr2) \ 2123 if (!req->ptr2) \
… …
1065 req->result = -1; \ 2132 req->result = -1; \
1066 break; \ 2133 break; \
1067 } \ 2134 } \
1068 } 2135 }
1069 2136
2137static void ecb_noinline ecb_cold
2138etp_proc_init (void)
2139{
2140#if HAVE_PRCTL_SET_NAME
2141 /* provide a more sensible "thread name" */
2142 char name[16 + 1];
2143 const int namelen = sizeof (name) - 1;
2144 int len;
2145
2146 prctl (PR_GET_NAME, (unsigned long)name, 0, 0, 0);
2147 name [namelen] = 0;
2148 len = strlen (name);
2149 strcpy (name + (len <= namelen - 4 ? len : namelen - 4), "/eio");
2150 prctl (PR_SET_NAME, (unsigned long)name, 0, 0, 0);
2151#endif
2152}
2153
1070X_THREAD_PROC (etp_proc) 2154X_THREAD_PROC (etp_proc)
1071{ 2155{
1072 ETP_REQ *req; 2156 ETP_REQ *req;
1073 struct timespec ts; 2157 struct timespec ts;
1074 etp_worker *self = (etp_worker *)thr_arg; 2158 etp_worker *self = (etp_worker *)thr_arg;
1075 2159
2160 etp_proc_init ();
2161
1076 /* try to distribute timeouts somewhat randomly */ 2162 /* try to distribute timeouts somewhat evenly */
1077 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 2163 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1078 2164
1079 for (;;) 2165 for (;;)
1080 { 2166 {
2167 ts.tv_sec = 0;
2168
1081 X_LOCK (reqlock); 2169 X_LOCK (reqlock);
1082 2170
1083 for (;;) 2171 for (;;)
1084 { 2172 {
1085 self->req = req = reqq_shift (&req_queue); 2173 req = reqq_shift (&req_queue);
1086 2174
1087 if (req) 2175 if (req)
1088 break; 2176 break;
1089 2177
2178 if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
2179 {
2180 X_UNLOCK (reqlock);
2181 X_LOCK (wrklock);
2182 --started;
2183 X_UNLOCK (wrklock);
2184 goto quit;
2185 }
2186
1090 ++idle; 2187 ++idle;
1091 2188
1092 ts.tv_sec = time (0) + IDLE_TIMEOUT; 2189 if (idle <= max_idle)
1093 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 2190 /* we are allowed to idle, so do so without any timeout */
2191 X_COND_WAIT (reqwait, reqlock);
2192 else
1094 { 2193 {
1095 if (idle > max_idle) 2194 /* initialise timeout once */
1096 { 2195 if (!ts.tv_sec)
1097 --idle; 2196 ts.tv_sec = time (0) + idle_timeout;
1098 X_UNLOCK (reqlock);
1099 X_LOCK (wrklock);
1100 --started;
1101 X_UNLOCK (wrklock);
1102 goto quit;
1103 }
1104 2197
1105 /* we are allowed to idle, so do so without any timeout */
1106 X_COND_WAIT (reqwait, reqlock); 2198 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
2199 ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1107 } 2200 }
1108 2201
1109 --idle; 2202 --idle;
1110 } 2203 }
1111 2204
… …
1114 X_UNLOCK (reqlock); 2207 X_UNLOCK (reqlock);
1115 2208
1116 if (req->type < 0) 2209 if (req->type < 0)
1117 goto quit; 2210 goto quit;
1118 2211
1119 if (!EIO_CANCELLED (req))
1120 ETP_EXECUTE (self, req); 2212 ETP_EXECUTE (self, req);
1121 2213
1122 X_LOCK (reslock); 2214 X_LOCK (reslock);
1123 2215
1124 ++npending; 2216 ++npending;
1125 2217
1126 if (!reqq_push (&res_queue, req) && want_poll_cb) 2218 if (!reqq_push (&res_queue, req) && want_poll_cb)
1127 want_poll_cb (); 2219 want_poll_cb ();
1128 2220
1129 self->req = 0;
1130 etp_worker_clear (self); 2221 etp_worker_clear (self);
1131 2222
1132 X_UNLOCK (reslock); 2223 X_UNLOCK (reslock);
1133 } 2224 }
1134 2225
1135quit: 2226quit:
2227 free (req);
2228
1136 X_LOCK (wrklock); 2229 X_LOCK (wrklock);
1137 etp_worker_free (self); 2230 etp_worker_free (self);
1138 X_UNLOCK (wrklock); 2231 X_UNLOCK (wrklock);
1139 2232
1140 return 0; 2233 return 0;
1141} 2234}
1142 2235
1143/*****************************************************************************/ 2236/*****************************************************************************/
1144 2237
2238int ecb_cold
1145int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 2239eio_init (void (*want_poll)(void), void (*done_poll)(void))
1146{ 2240{
1147 return etp_init (want_poll, done_poll); 2241 return etp_init (want_poll, done_poll);
1148} 2242}
1149 2243
2244ecb_inline void
1150static void eio_api_destroy (eio_req *req) 2245eio_api_destroy (eio_req *req)
1151{ 2246{
1152 free (req); 2247 free (req);
1153} 2248}
1154 2249
1155#define REQ(rtype) \ 2250#define REQ(rtype) \
1156 eio_req *req; \ 2251 eio_req *req; \
1157 \ 2252 \
1158 req = (eio_req *)calloc (1, sizeof *req); \ 2253 req = (eio_req *)calloc (1, sizeof *req); \
1159 if (!req) \ 2254 if (!req) \
1160 return 0; \ 2255 return 0; \
… …
1174 { \ 2269 { \
1175 eio_api_destroy (req); \ 2270 eio_api_destroy (req); \
1176 return 0; \ 2271 return 0; \
1177 } 2272 }
1178 2273
2274static void
1179static void eio_execute (etp_worker *self, eio_req *req) 2275eio_execute (etp_worker *self, eio_req *req)
1180{ 2276{
1181 errno = 0; 2277#if HAVE_AT
2278 int dirfd;
2279#else
2280 const char *path;
2281#endif
2282
2283 if (ecb_expect_false (EIO_CANCELLED (req)))
2284 {
2285 req->result = -1;
2286 req->errorno = ECANCELED;
2287 return;
2288 }
2289
2290 if (ecb_expect_false (req->wd == EIO_INVALID_WD))
2291 {
2292 req->result = -1;
2293 req->errorno = ENOENT;
2294 return;
2295 }
2296
2297 if (req->type >= EIO_OPEN)
2298 {
2299 #if HAVE_AT
2300 dirfd = WD2FD (req->wd);
2301 #else
2302 path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
2303 #endif
2304 }
1182 2305
1183 switch (req->type) 2306 switch (req->type)
1184 { 2307 {
2308 case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
2309 req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
2310 break;
2311 case EIO_WD_CLOSE: req->result = 0;
2312 eio_wd_close_sync (req->wd); break;
2313
2314 case EIO_SEEK: eio__lseek (req); break;
1185 case EIO_READ: ALLOC (req->size); 2315 case EIO_READ: ALLOC (req->size);
1186 req->result = req->offs >= 0 2316 req->result = req->offs >= 0
1187 ? pread (req->int1, req->ptr2, req->size, req->offs) 2317 ? pread (req->int1, req->ptr2, req->size, req->offs)
1188 : read (req->int1, req->ptr2, req->size); break; 2318 : read (req->int1, req->ptr2, req->size); break;
1189 case EIO_WRITE: req->result = req->offs >= 0 2319 case EIO_WRITE: req->result = req->offs >= 0
1190 ? pwrite (req->int1, req->ptr2, req->size, req->offs) 2320 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1191 : write (req->int1, req->ptr2, req->size); break; 2321 : write (req->int1, req->ptr2, req->size); break;
1192 2322
1193 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break; 2323 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1194 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break; 2324 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
2325
2326#if HAVE_AT
1195 2327
1196 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2328 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1197 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2329 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break;
1198 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2330 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1199 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2331 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, AT_SYMLINK_NOFOLLOW); break;
1200 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1201 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1202
1203 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; 2332 case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break;
1204 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1205 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break; 2333 case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break;
1206 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1207 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break; 2334 case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break;
1208 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1209
1210 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break; 2335 case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break;
1211 case EIO_CLOSE: req->result = close (req->int1); break; 2336
1212 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1213 case EIO_UNLINK: req->result = unlink (req->ptr1); break; 2337 case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
1214 case EIO_RMDIR: req->result = rmdir (req->ptr1); break; 2338 case EIO_RMDIR: req->result = unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break;
1215 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; 2339 case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); break;
1216 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; 2340 case EIO_RENAME: req->result = renameat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2); break;
1217 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; 2341 case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2, 0); break;
1218 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; 2342 case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
1219 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break; 2343 case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1220
1221 case EIO_READLINK: ALLOC (NAME_MAX); 2344 case EIO_READLINK: ALLOC (PATH_MAX);
1222 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; 2345 req->result = readlinkat (dirfd, req->ptr1, req->ptr2, PATH_MAX); break;
1223 2346 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1224 case EIO_SYNC: req->result = 0; sync (); break; 2347 req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1225 case EIO_FSYNC: req->result = fsync (req->int1); break;
1226 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1227 case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break;
1228 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;
1229 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1230
1231 case EIO_READDIR: eio__scandir (req, self); break;
1232
1233 case EIO_BUSY: 2348 case EIO_UTIME:
1234#ifdef _WIN32 2349 case EIO_FUTIME:
1235 Sleep (req->nv1 * 1000.); 2350 {
2351 struct timespec ts[2];
2352 struct timespec *times;
2353
2354 if (req->nv1 != -1. || req->nv2 != -1.)
2355 {
2356 ts[0].tv_sec = req->nv1;
2357 ts[0].tv_nsec = (req->nv1 - ts[0].tv_sec) * 1e9;
2358 ts[1].tv_sec = req->nv2;
2359 ts[1].tv_nsec = (req->nv2 - ts[1].tv_sec) * 1e9;
2360
2361 times = ts;
2362 }
2363 else
2364 times = 0;
2365
2366 req->result = req->type == EIO_FUTIME
2367 ? futimens (req->int1, times)
2368 : utimensat (dirfd, req->ptr1, times, 0);
2369 }
2370 break;
2371
1236#else 2372#else
1237 {
1238 struct timeval tv;
1239 2373
1240 tv.tv_sec = req->nv1; 2374 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1241 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; 2375 req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
2376 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2377 req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
2378 case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
2379 case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
2380 case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
2381 case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
1242 2382
1243 req->result = select (0, 0, 0, 0, &tv); 2383 case EIO_UNLINK: req->result = unlink (path ); break;
1244 } 2384 case EIO_RMDIR: req->result = rmdir (path ); break;
1245#endif 2385 case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
1246 break; 2386 case EIO_RENAME: req->result = rename (path , req->ptr2); break;
2387 case EIO_LINK: req->result = link (path , req->ptr2); break;
2388 case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
2389 case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
2390 case EIO_READLINK: ALLOC (PATH_MAX);
2391 req->result = readlink (path, req->ptr2, PATH_MAX); break;
2392 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2393 req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
1247 2394
1248 case EIO_UTIME: 2395 case EIO_UTIME:
1249 case EIO_FUTIME: 2396 case EIO_FUTIME:
1250 { 2397 {
1251 struct timeval tv[2]; 2398 struct timeval tv[2];
1252 struct timeval *times; 2399 struct timeval *times;
1253 2400
1254 if (req->nv1 != -1. || req->nv2 != -1.) 2401 if (req->nv1 != -1. || req->nv2 != -1.)
1255 { 2402 {
1256 tv[0].tv_sec = req->nv1; 2403 tv[0].tv_sec = req->nv1;
1257 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.; 2404 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
1258 tv[1].tv_sec = req->nv2; 2405 tv[1].tv_sec = req->nv2;
1259 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.; 2406 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
1260 2407
1261 times = tv; 2408 times = tv;
1262 } 2409 }
1263 else 2410 else
1264 times = 0; 2411 times = 0;
1265
1266 2412
1267 req->result = req->type == EIO_FUTIME 2413 req->result = req->type == EIO_FUTIME
1268 ? futimes (req->int1, times) 2414 ? futimes (req->int1, times)
1269 : utimes (req->ptr1, times); 2415 : utimes (req->ptr1, times);
1270 } 2416 }
1271 break; 2417 break;
1272 2418
2419#endif
2420
2421 case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
2422 {
2423 ALLOC (req->result);
2424 memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
2425 }
2426 break;
2427
2428 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2429 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
2430
2431 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2432 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2433
2434 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
2435 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
2436 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
2437
2438 case EIO_CLOSE: req->result = close (req->int1); break;
2439 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
2440 case EIO_SYNC: req->result = 0; sync (); break;
2441 case EIO_FSYNC: req->result = fsync (req->int1); break;
2442 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
2443 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
2444 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
2445 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
2446 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
2447 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
2448 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
2449 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
2450
2451 case EIO_READDIR: eio__scandir (req, self); break;
2452
2453 case EIO_BUSY:
2454#ifdef _WIN32
2455 Sleep (req->nv1 * 1e3);
2456#else
2457 {
2458 struct timeval tv;
2459
2460 tv.tv_sec = req->nv1;
2461 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
2462
2463 req->result = select (0, 0, 0, 0, &tv);
2464 }
2465#endif
2466 break;
2467
1273 case EIO_GROUP: 2468 case EIO_GROUP:
1274 abort (); /* handled in eio_request */ 2469 abort (); /* handled in eio_request */
1275 2470
1276 case EIO_NOP: 2471 case EIO_NOP:
1277 req->result = 0; 2472 req->result = 0;
1278 break; 2473 break;
1279 2474
1280 case EIO_CUSTOM: 2475 case EIO_CUSTOM:
1281 ((void (*)(eio_req *))req->feed) (req); 2476 req->feed (req);
1282 break; 2477 break;
1283 2478
1284 default: 2479 default:
1285 req->result = -1; 2480 req->result = EIO_ENOSYS ();
1286 break; 2481 break;
1287 } 2482 }
1288 2483
1289 req->errorno = errno; 2484 req->errorno = errno;
1290} 2485}
1291 2486
1292#ifndef EIO_NO_WRAPPERS 2487#ifndef EIO_NO_WRAPPERS
1293 2488
2489eio_req *eio_wd_open (const char *path, int pri, eio_cb cb, void *data)
2490{
2491 REQ (EIO_WD_OPEN); PATH; SEND;
2492}
2493
2494eio_req *eio_wd_close (eio_wd wd, int pri, eio_cb cb, void *data)
2495{
2496 REQ (EIO_WD_CLOSE); req->wd = wd; SEND;
2497}
2498
1294eio_req *eio_nop (int pri, eio_cb cb, void *data) 2499eio_req *eio_nop (int pri, eio_cb cb, void *data)
1295{ 2500{
1296 REQ (EIO_NOP); SEND; 2501 REQ (EIO_NOP); SEND;
1297} 2502}
1298 2503
… …
1314eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2519eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1315{ 2520{
1316 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2521 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1317} 2522}
1318 2523
2524eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2525{
2526 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2527}
2528
2529eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
2530{
2531 REQ (EIO_SYNCFS); req->int1 = fd; SEND;
2532}
2533
2534eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2535{
2536 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2537}
2538
1319eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2539eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1320{ 2540{
1321 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2541 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1322} 2542}
1323 2543
1324eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) 2544eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
1325{ 2545{
1326 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; 2546 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
1327} 2547}
1328 2548
1329eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) 2549eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
1330{ 2550{
1331 REQ (EIO_FDATASYNC); req->int1 = fd; SEND; 2551 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2552}
2553
2554eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2555{
2556 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
1332} 2557}
1333 2558
1334eio_req *eio_close (int fd, int pri, eio_cb cb, void *data) 2559eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
1335{ 2560{
1336 REQ (EIO_CLOSE); req->int1 = fd; SEND; 2561 REQ (EIO_CLOSE); req->int1 = fd; SEND;
… …
1339eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data) 2564eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
1340{ 2565{
1341 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND; 2566 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
1342} 2567}
1343 2568
2569eio_req *eio_seek (int fd, off_t offset, int whence, int pri, eio_cb cb, void *data)
2570{
2571 REQ (EIO_SEEK); req->int1 = fd; req->offs = offset; req->int2 = whence; SEND;
2572}
2573
1344eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data) 2574eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1345{ 2575{
1346 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND; 2576 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1347} 2577}
1348 2578
… …
1354eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) 2584eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1355{ 2585{
1356 REQ (EIO_FSTAT); req->int1 = fd; SEND; 2586 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1357} 2587}
1358 2588
2589eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
2590{
2591 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
2592}
2593
1359eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) 2594eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1360{ 2595{
1361 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; 2596 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1362} 2597}
1363 2598
… …
1369eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) 2604eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1370{ 2605{
1371 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2606 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1372} 2607}
1373 2608
1374eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2609eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1375{ 2610{
1376 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2611 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1377} 2612}
1378 2613
1379eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) 2614eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
… …
1399eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data) 2634eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1400{ 2635{
1401 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2636 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1402} 2637}
1403 2638
1404eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2639eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1405{ 2640{
1406 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2641 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1407} 2642}
1408 2643
1409eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) 2644eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
… …
1425eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data) 2660eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1426{ 2661{
1427 return eio__1path (EIO_READLINK, path, pri, cb, data); 2662 return eio__1path (EIO_READLINK, path, pri, cb, data);
1428} 2663}
1429 2664
2665eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2666{
2667 return eio__1path (EIO_REALPATH, path, pri, cb, data);
2668}
2669
1430eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) 2670eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1431{ 2671{
1432 return eio__1path (EIO_STAT, path, pri, cb, data); 2672 return eio__1path (EIO_STAT, path, pri, cb, data);
1433} 2673}
1434 2674
1435eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) 2675eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1436{ 2676{
1437 return eio__1path (EIO_LSTAT, path, pri, cb, data); 2677 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1438} 2678}
1439 2679
2680eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
2681{
2682 return eio__1path (EIO_STATVFS, path, pri, cb, data);
2683}
2684
1440eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) 2685eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1441{ 2686{
1442 return eio__1path (EIO_UNLINK, path, pri, cb, data); 2687 return eio__1path (EIO_UNLINK, path, pri, cb, data);
1443} 2688}
1444 2689
1445eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data) 2690eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1446{ 2691{
1447 return eio__1path (EIO_RMDIR, path, pri, cb, data); 2692 return eio__1path (EIO_RMDIR, path, pri, cb, data);
1448} 2693}
1449 2694
1450eio_req *eio_readdir (const char *path, int pri, eio_cb cb, void *data) 2695eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1451{ 2696{
1452 return eio__1path (EIO_READDIR, path, pri, cb, data); 2697 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1453} 2698}
1454 2699
1455eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) 2700eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1456{ 2701{
1457 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; 2702 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1458} 2703}
1459 2704
1460static eio_req * 2705static eio_req *
1461eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2706eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1462{ 2707{
… …
1486eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2731eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1487{ 2732{
1488 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2733 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1489} 2734}
1490 2735
1491eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) 2736eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
1492{ 2737{
1493 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND; 2738 REQ (EIO_CUSTOM); req->feed = execute; SEND;
1494} 2739}
1495 2740
1496#endif 2741#endif
1497 2742
1498eio_req *eio_grp (eio_cb cb, void *data) 2743eio_req *eio_grp (eio_cb cb, void *data)
… …
1507#undef SEND 2752#undef SEND
1508 2753
1509/*****************************************************************************/ 2754/*****************************************************************************/
1510/* grp functions */ 2755/* grp functions */
1511 2756
2757void
1512void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2758eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1513{ 2759{
1514 grp->int2 = limit; 2760 grp->int2 = limit;
1515 grp->feed = feed; 2761 grp->feed = feed;
1516 2762
1517 grp_try_feed (grp); 2763 grp_try_feed (grp);
1518} 2764}
1519 2765
2766void
1520void eio_grp_limit (eio_req *grp, int limit) 2767eio_grp_limit (eio_req *grp, int limit)
1521{ 2768{
1522 grp->int2 = limit; 2769 grp->int2 = limit;
1523 2770
1524 grp_try_feed (grp); 2771 grp_try_feed (grp);
1525} 2772}
1526 2773
2774void
1527void eio_grp_add (eio_req *grp, eio_req *req) 2775eio_grp_add (eio_req *grp, eio_req *req)
1528{ 2776{
1529 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2777 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1530 2778
1531 grp->flags |= EIO_FLAG_GROUPADD; 2779 grp->flags |= EIO_FLAG_GROUPADD;
1532 2780
… …
1543} 2791}
1544 2792
1545/*****************************************************************************/ 2793/*****************************************************************************/
1546/* misc garbage */ 2794/* misc garbage */
1547 2795
2796eio_ssize_t
1548ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2797eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1549{ 2798{
1550 etp_worker wrk;
1551
1552 wrk.dbuf = 0;
1553
1554 eio__sendfile (ofd, ifd, offset, count, &wrk); 2799 return eio__sendfile (ofd, ifd, offset, count);
1555
1556 if (wrk.dbuf)
1557 free (wrk.dbuf);
1558} 2800}
1559 2801

Diff Legend

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