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

Comparing libeio/eio.c (file contents):
Revision 1.76 by root, Tue Jul 5 09:24:12 2011 UTC vs.
Revision 1.126 by root, Fri Dec 28 07:33:41 2012 UTC

1/* 1/*
2 * libeio implementation 2 * libeio implementation
3 * 3 *
4 * Copyright (c) 2007,2008,2009,2010,2011 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,
… …
43 43
44#include "eio.h" 44#include "eio.h"
45#include "ecb.h" 45#include "ecb.h"
46 46
47#ifdef EIO_STACKSIZE 47#ifdef EIO_STACKSIZE
48# define XTHREAD_STACKSIZE EIO_STACKSIZE 48# define X_STACKSIZE EIO_STACKSIZE
49#endif 49#endif
50#include "xthread.h" 50#include "xthread.h"
51 51
52#include <errno.h> 52#include <errno.h>
53#include <stddef.h> 53#include <stddef.h>
54#include <stdlib.h> 54#include <stdlib.h>
55#include <string.h> 55#include <string.h>
56#include <errno.h> 56#include <errno.h>
57#include <sys/types.h> 57#include <sys/types.h>
58#include <sys/stat.h> 58#include <sys/stat.h>
59#include <sys/statvfs.h>
60#include <limits.h> 59#include <limits.h>
61#include <fcntl.h> 60#include <fcntl.h>
62#include <assert.h> 61#include <assert.h>
63 62
64/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */ 63/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
65/* intptr_t only comes form stdint.h, says idiot openbsd coder */ 64/* intptr_t only comes from stdint.h, says idiot openbsd coder */
66#if HAVE_STDINT_H 65#if HAVE_STDINT_H
67# include <stdint.h> 66# include <stdint.h>
68#endif 67#endif
69 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
70#ifndef EIO_FINISH 82#ifndef EIO_FINISH
71# 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
72#endif 84#endif
73 85
74#ifndef EIO_DESTROY 86#ifndef EIO_DESTROY
… …
77 89
78#ifndef EIO_FEED 90#ifndef EIO_FEED
79# 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)
80#endif 92#endif
81 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
82#ifdef _WIN32 105#ifdef _WIN32
83 106
84 /*doh*/ 107 #undef PAGESIZE
108 #define PAGESIZE 4096 /* GetSystemInfo? */
109
110 /* TODO: look at how perl does stat (non-sloppy), unlink (ro-files), utime, link */
111
112 #ifdef EIO_STRUCT_STATI64
113 /* look at perl's non-sloppy stat */
114 #define stat(path,buf) _stati64 (path,buf)
115 #define fstat(fd,buf) _fstati64 (fd,buf)
116 #endif
117 #define lstat(path,buf) stat (path,buf)
118 #define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
119 #define mkdir(path,mode) _mkdir (path)
120 #define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
121
122 #define chmod(path,mode) _chmod (path, mode)
123 #define dup(fd) _dup (fd)
124 #define dup2(fd1,fd2) _dup2 (fd1, fd2)
125
126 #define fchmod(fd,mode) EIO_ENOSYS ()
127 #define chown(path,uid,gid) EIO_ENOSYS ()
128 #define fchown(fd,uid,gid) EIO_ENOSYS ()
129 #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
130 #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
131 #define mknod(path,mode,dev) EIO_ENOSYS ()
132 #define sync() EIO_ENOSYS ()
133 #define readlink(path,buf,s) EIO_ENOSYS ()
134 #define statvfs(path,buf) EIO_ENOSYS ()
135 #define fstatvfs(fd,buf) EIO_ENOSYS ()
136
137 #define pread(fd,buf,count,offset) eio__pread (fd, buf, count, offset)
138 #define pwrite(fd,buf,count,offset) eio__pwrite (fd, buf, count, offset)
139
140 #if __GNUC__
141 typedef long long eio_off_t; /* signed for compatibility to msvc */
142 #else
143 typedef __int64 eio_off_t; /* unsigned not supported by msvc */
144 #endif
145
146 static eio_ssize_t
147 eio__pread (int fd, void *buf, eio_ssize_t count, eio_off_t offset)
148 {
149 OVERLAPPED o = { 0 };
150 DWORD got;
151
152 o.Offset = offset;
153 o.OffsetHigh = offset >> 32;
154
155 return ReadFile ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd), buf, count, &got, &o)
156 ? got : -1;
157 }
158
159 static eio_ssize_t
160 eio__pwrite (int fd, void *buf, eio_ssize_t count, eio_off_t offset)
161 {
162 OVERLAPPED o = { 0 };
163 DWORD got;
164
165 o.Offset = offset;
166 o.OffsetHigh = offset >> 32;
167
168 return WriteFile ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd), buf, count, &got, &o)
169 ? got : -1;
170 }
171
172 /* rename() uses MoveFile, which fails to overwrite */
173 #define rename(old,neu) eio__rename (old, neu)
174
175 static int
176 eio__rename (const char *old, const char *neu)
177 {
178 if (MoveFileEx (old, neu, MOVEFILE_REPLACE_EXISTING))
179 return 0;
180
181 /* should steal _dosmaperr */
182 switch (GetLastError ())
183 {
184 case ERROR_FILE_NOT_FOUND:
185 case ERROR_PATH_NOT_FOUND:
186 case ERROR_INVALID_DRIVE:
187 case ERROR_NO_MORE_FILES:
188 case ERROR_BAD_NETPATH:
189 case ERROR_BAD_NET_NAME:
190 case ERROR_BAD_PATHNAME:
191 case ERROR_FILENAME_EXCED_RANGE:
192 errno = ENOENT;
193 break;
194
195 default:
196 errno = EACCES;
197 break;
198 }
199
200 return -1;
201 }
202
203 /* we could even stat and see if it exists */
204 static int
205 symlink (const char *old, const char *neu)
206 {
207 #if WINVER >= 0x0600
208 if (CreateSymbolicLink (neu, old, 1))
209 return 0;
210
211 if (CreateSymbolicLink (neu, old, 0))
212 return 0;
213 #endif
214
215 return EIO_ERRNO (ENOENT, -1);
216 }
217
218 /* POSIX API only */
219 #define CreateHardLink(neu,old,flags) 0
220 #define CreateSymbolicLink(neu,old,flags) 0
221
222 struct statvfs
223 {
224 int dummy;
225 };
226
227 #define DT_DIR EIO_DT_DIR
228 #define DT_REG EIO_DT_REG
229 #define D_NAME(entp) entp.cFileName
230 #define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
231
85#else 232#else
86 233
87# include <sys/time.h> 234 #include <sys/time.h>
88# include <sys/select.h> 235 #include <sys/select.h>
236 #include <sys/statvfs.h>
89# 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
90# include <utime.h> 272# include <utime.h>
91# include <signal.h>
92# include <dirent.h>
93
94#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
95# include <sys/mman.h>
96#endif
97
98/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
99# if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
100# define _DIRENT_HAVE_D_TYPE /* sigh */
101# define D_INO(de) (de)->d_fileno
102# define D_NAMLEN(de) (de)->d_namlen
103# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
104# define D_INO(de) (de)->d_ino
105# endif 273#endif
106 274
107#ifdef _D_EXACT_NAMLEN 275#if HAVE_SYS_SYSCALL_H
108# undef D_NAMLEN 276# include <sys/syscall.h>
109# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
110#endif
111
112# ifdef _DIRENT_HAVE_D_TYPE
113# define D_TYPE(de) (de)->d_type
114# endif 277#endif
115 278
116# ifndef EIO_STRUCT_DIRENT 279#if HAVE_SYS_PRCTL_H
117# define EIO_STRUCT_DIRENT struct dirent 280# include <sys/prctl.h>
118# endif
119
120#endif 281#endif
121 282
122#if HAVE_SENDFILE 283#if HAVE_SENDFILE
123# if __linux 284# if __linux
124# include <sys/sendfile.h> 285# include <sys/sendfile.h>
… …
139#endif 300#endif
140#ifndef D_INO 301#ifndef D_INO
141# define D_INO(de) 0 302# define D_INO(de) 0
142#endif 303#endif
143#ifndef D_NAMLEN 304#ifndef D_NAMLEN
144# define D_NAMLEN(de) strlen ((de)->d_name) 305# define D_NAMLEN(entp) strlen (D_NAME (entp))
145#endif 306#endif
146 307
147/* used for struct dirent, AIX doesn't provide it */ 308/* used for struct dirent, AIX doesn't provide it */
148#ifndef NAME_MAX 309#ifndef NAME_MAX
149# define NAME_MAX 4096 310# define NAME_MAX 4096
… …
155#endif 316#endif
156 317
157/* buffer size for various temporary buffers */ 318/* buffer size for various temporary buffers */
158#define EIO_BUFSIZE 65536 319#define EIO_BUFSIZE 65536
159 320
160#define dBUF \ 321#define dBUF \
161 char *eio_buf; \
162 ETP_WORKER_LOCK (self); \
163 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 322 char *eio_buf = malloc (EIO_BUFSIZE); \
164 ETP_WORKER_UNLOCK (self); \
165 errno = ENOMEM; \ 323 errno = ENOMEM; \
166 if (!eio_buf) \ 324 if (!eio_buf) \
167 return -1; 325 return -1
326
327#define FUBd \
328 free (eio_buf)
168 329
169#define EIO_TICKS ((1000000 + 1023) >> 10) 330#define EIO_TICKS ((1000000 + 1023) >> 10)
331
332/*****************************************************************************/
333
334struct tmpbuf
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
360#else
361 #define HAVE_AT 0
362 static const char *wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path);
363#endif
364
365struct eio_pwd
366{
367#if HAVE_AT
368 int fd;
369#endif
370 int len;
371 char str[1]; /* actually, a 0-terminated canonical path */
372};
373
374/*****************************************************************************/
170 375
171#define ETP_PRI_MIN EIO_PRI_MIN 376#define ETP_PRI_MIN EIO_PRI_MIN
172#define ETP_PRI_MAX EIO_PRI_MAX 377#define ETP_PRI_MAX EIO_PRI_MAX
173 378
174struct etp_worker; 379struct etp_worker;
… …
178static int eio_finish (eio_req *req); 383static int eio_finish (eio_req *req);
179#define ETP_FINISH(req) eio_finish (req) 384#define ETP_FINISH(req) eio_finish (req)
180static void eio_execute (struct etp_worker *self, eio_req *req); 385static void eio_execute (struct etp_worker *self, eio_req *req);
181#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) 386#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
182 387
183#define ETP_WORKER_CLEAR(req) \
184 if (wrk->dbuf) \
185 { \
186 free (wrk->dbuf); \
187 wrk->dbuf = 0; \
188 } \
189 \
190 if (wrk->dirp) \
191 { \
192 closedir (wrk->dirp); \
193 wrk->dirp = 0; \
194 }
195
196#define ETP_WORKER_COMMON \
197 void *dbuf; \
198 DIR *dirp;
199
200/*****************************************************************************/ 388/*****************************************************************************/
201 389
202#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) 390#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
203 391
204/* calculate time difference in ~1/EIO_TICKS of a second */ 392/* calculate time difference in ~1/EIO_TICKS of a second */
… …
215static void (*done_poll_cb) (void); 403static void (*done_poll_cb) (void);
216 404
217static unsigned int max_poll_time; /* reslock */ 405static unsigned int max_poll_time; /* reslock */
218static unsigned int max_poll_reqs; /* reslock */ 406static unsigned int max_poll_reqs; /* reslock */
219 407
220static volatile unsigned int nreqs; /* reqlock */ 408static unsigned int nreqs; /* reqlock */
221static volatile unsigned int nready; /* reqlock */ 409static unsigned int nready; /* reqlock */
222static volatile unsigned int npending; /* reqlock */ 410static unsigned int npending; /* reqlock */
223static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */ 411static unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
224static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */ 412static unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
225 413
226static xmutex_t wrklock; 414static xmutex_t wrklock;
227static xmutex_t reslock; 415static xmutex_t reslock;
228static xmutex_t reqlock; 416static xmutex_t reqlock;
229static xcond_t reqwait; 417static xcond_t reqwait;
230 418
231#if !HAVE_PREADWRITE
232/*
233 * make our pread/pwrite emulation safe against themselves, but not against
234 * normal read/write by using a mutex. slows down execution a lot,
235 * but that's your problem, not mine.
236 */
237static xmutex_t preadwritelock = X_MUTEX_INIT;
238#endif
239
240typedef struct etp_worker 419typedef struct etp_worker
241{ 420{
421 struct tmpbuf tmpbuf;
422
242 /* locked by wrklock */ 423 /* locked by wrklock */
243 struct etp_worker *prev, *next; 424 struct etp_worker *prev, *next;
244 425
245 xthread_t tid; 426 xthread_t tid;
246 427
247 /* locked by reslock, reqlock or wrklock */ 428#ifdef ETP_WORKER_COMMON
248 ETP_REQ *req; /* currently processed request */
249
250 ETP_WORKER_COMMON 429 ETP_WORKER_COMMON
430#endif
251} etp_worker; 431} etp_worker;
252 432
253static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */ 433static etp_worker wrk_first; /* NOT etp */
254 434
255#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock) 435#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
256#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock) 436#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
257 437
258/* worker threads management */ 438/* worker threads management */
259 439
260static void ecb_cold 440static void
261etp_worker_clear (etp_worker *wrk) 441etp_worker_clear (etp_worker *wrk)
262{ 442{
263 ETP_WORKER_CLEAR (wrk);
264} 443}
265 444
266static void ecb_cold 445static void ecb_cold
267etp_worker_free (etp_worker *wrk) 446etp_worker_free (etp_worker *wrk)
268{ 447{
448 free (wrk->tmpbuf.ptr);
449
269 wrk->next->prev = wrk->prev; 450 wrk->next->prev = wrk->prev;
270 wrk->prev->next = wrk->next; 451 wrk->prev->next = wrk->next;
271 452
272 free (wrk); 453 free (wrk);
273} 454}
… …
329} etp_reqq; 510} etp_reqq;
330 511
331static etp_reqq req_queue; 512static etp_reqq req_queue;
332static etp_reqq res_queue; 513static etp_reqq res_queue;
333 514
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
334static int ecb_noinline 526static int ecb_noinline
335reqq_push (etp_reqq *q, ETP_REQ *req) 527reqq_push (etp_reqq *q, ETP_REQ *req)
336{ 528{
337 int pri = req->pri; 529 int pri = req->pri;
338 req->next = 0; 530 req->next = 0;
… …
372 } 564 }
373 565
374 abort (); 566 abort ();
375} 567}
376 568
377static void ecb_cold 569static int ecb_cold
378etp_thread_init (void) 570etp_init (void (*want_poll)(void), void (*done_poll)(void))
379{ 571{
380 X_MUTEX_CREATE (wrklock); 572 X_MUTEX_CREATE (wrklock);
381 X_MUTEX_CREATE (reslock); 573 X_MUTEX_CREATE (reslock);
382 X_MUTEX_CREATE (reqlock); 574 X_MUTEX_CREATE (reqlock);
383 X_COND_CREATE (reqwait); 575 X_COND_CREATE (reqwait);
384}
385 576
386static void ecb_cold 577 reqq_init (&req_queue);
387etp_atfork_prepare (void) 578 reqq_init (&res_queue);
388{
389 X_LOCK (wrklock);
390 X_LOCK (reqlock);
391 X_LOCK (reslock);
392#if !HAVE_PREADWRITE
393 X_LOCK (preadwritelock);
394#endif
395}
396 579
397static void ecb_cold 580 wrk_first.next =
398etp_atfork_parent (void) 581 wrk_first.prev = &wrk_first;
399{
400#if !HAVE_PREADWRITE
401 X_UNLOCK (preadwritelock);
402#endif
403 X_UNLOCK (reslock);
404 X_UNLOCK (reqlock);
405 X_UNLOCK (wrklock);
406}
407
408static void ecb_cold
409etp_atfork_child (void)
410{
411 ETP_REQ *prv;
412
413 while ((prv = reqq_shift (&req_queue)))
414 ETP_DESTROY (prv);
415
416 while ((prv = reqq_shift (&res_queue)))
417 ETP_DESTROY (prv);
418
419 while (wrk_first.next != &wrk_first)
420 {
421 etp_worker *wrk = wrk_first.next;
422
423 if (wrk->req)
424 ETP_DESTROY (wrk->req);
425
426 etp_worker_clear (wrk);
427 etp_worker_free (wrk);
428 }
429 582
430 started = 0; 583 started = 0;
431 idle = 0; 584 idle = 0;
432 nreqs = 0; 585 nreqs = 0;
433 nready = 0; 586 nready = 0;
434 npending = 0; 587 npending = 0;
435 588
436 etp_thread_init ();
437}
438
439static void ecb_cold
440etp_once_init (void)
441{
442 etp_thread_init ();
443 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
444}
445
446static int ecb_cold
447etp_init (void (*want_poll)(void), void (*done_poll)(void))
448{
449 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
450
451 pthread_once (&doinit, etp_once_init);
452
453 want_poll_cb = want_poll; 589 want_poll_cb = want_poll;
454 done_poll_cb = done_poll; 590 done_poll_cb = done_poll;
455 591
456 return 0; 592 return 0;
457} 593}
… …
466 /*TODO*/ 602 /*TODO*/
467 assert (("unable to allocate worker thread data", wrk)); 603 assert (("unable to allocate worker thread data", wrk));
468 604
469 X_LOCK (wrklock); 605 X_LOCK (wrklock);
470 606
471 if (thread_create (&wrk->tid, etp_proc, (void *)wrk)) 607 if (xthread_create (&wrk->tid, etp_proc, (void *)wrk))
472 { 608 {
473 wrk->prev = &wrk_first; 609 wrk->prev = &wrk_first;
474 wrk->next = wrk_first.next; 610 wrk->next = wrk_first.next;
475 wrk_first.next->prev = wrk; 611 wrk_first.next->prev = wrk;
476 wrk_first.next = wrk; 612 wrk_first.next = wrk;
… …
496} 632}
497 633
498static void ecb_cold 634static void ecb_cold
499etp_end_thread (void) 635etp_end_thread (void)
500{ 636{
501 eio_req *req = calloc (1, sizeof (eio_req)); 637 eio_req *req = calloc (1, sizeof (eio_req)); /* will be freed by worker */
502 638
503 req->type = -1; 639 req->type = -1;
504 req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 640 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
505 641
506 X_LOCK (reqlock); 642 X_LOCK (reqlock);
… …
583} 719}
584 720
585static void 721static void
586etp_cancel (ETP_REQ *req) 722etp_cancel (ETP_REQ *req)
587{ 723{
588 X_LOCK (wrklock); 724 req->cancelled = 1;
589 req->flags |= EIO_FLAG_CANCELLED;
590 X_UNLOCK (wrklock);
591 725
592 eio_grp_cancel (req); 726 eio_grp_cancel (req);
593} 727}
594 728
595static void 729static void
… …
711 return eio_finish (grp); 845 return eio_finish (grp);
712 else 846 else
713 return 0; 847 return 0;
714} 848}
715 849
716void 850static void
717eio_destroy (eio_req *req) 851eio_destroy (eio_req *req)
718{ 852{
719 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); 853 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
720 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); 854 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
721 855
… …
739 if (grp->grp_first == req) 873 if (grp->grp_first == req)
740 grp->grp_first = req->grp_next; 874 grp->grp_first = req->grp_next;
741 875
742 res2 = grp_dec (grp); 876 res2 = grp_dec (grp);
743 877
744 if (!res && res2) 878 if (!res)
745 res = res2; 879 res = res2;
746 } 880 }
747 881
748 eio_destroy (req); 882 eio_destroy (req);
749 883
… …
834 return etp_poll (); 968 return etp_poll ();
835} 969}
836 970
837/*****************************************************************************/ 971/*****************************************************************************/
838/* work around various missing functions */ 972/* work around various missing functions */
839
840#if _POSIX_VERSION < 200809L
841# define realpath(path,resolved_path) (errno = ENOSYS, 0)
842#endif
843
844#if !HAVE_PREADWRITE
845# undef pread
846# undef pwrite
847# define pread eio__pread
848# define pwrite eio__pwrite
849
850static ssize_t
851eio__pread (int fd, void *buf, size_t count, off_t offset)
852{
853 ssize_t res;
854 off_t ooffset;
855
856 X_LOCK (preadwritelock);
857 ooffset = lseek (fd, 0, SEEK_CUR);
858 lseek (fd, offset, SEEK_SET);
859 res = read (fd, buf, count);
860 lseek (fd, ooffset, SEEK_SET);
861 X_UNLOCK (preadwritelock);
862
863 return res;
864}
865
866static ssize_t
867eio__pwrite (int fd, void *buf, size_t count, off_t offset)
868{
869 ssize_t res;
870 off_t ooffset;
871
872 X_LOCK (preadwritelock);
873 ooffset = lseek (fd, 0, SEEK_CUR);
874 lseek (fd, offset, SEEK_SET);
875 res = write (fd, buf, count);
876 lseek (fd, ooffset, SEEK_SET);
877 X_UNLOCK (preadwritelock);
878
879 return res;
880}
881#endif
882 973
883#ifndef HAVE_UTIMES 974#ifndef HAVE_UTIMES
884 975
885# undef utimes 976# undef utimes
886# define utimes(path,times) eio__utimes (path, times) 977# define utimes(path,times) eio__utimes (path, times)
… …
919 1010
920#if !HAVE_FDATASYNC 1011#if !HAVE_FDATASYNC
921# undef fdatasync 1012# undef fdatasync
922# define fdatasync(fd) fsync (fd) 1013# define fdatasync(fd) fsync (fd)
923#endif 1014#endif
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}
924 1032
925/* sync_file_range always needs emulation */ 1033/* sync_file_range always needs emulation */
926static int 1034static int
927eio__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)
928{ 1036{
… …
948 /* 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
949 * call fdatasync, as that matches the expectation of its users best */ 1057 * call fdatasync, as that matches the expectation of its users best */
950 return fdatasync (fd); 1058 return fdatasync (fd);
951} 1059}
952 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
1069}
1070
953#if !HAVE_READAHEAD 1071#if !HAVE_READAHEAD
954# undef readahead 1072# undef readahead
955# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 1073# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
956 1074
957static ssize_t 1075static eio_ssize_t
958eio__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)
959{ 1077{
960 size_t todo = count; 1078 size_t todo = count;
961 dBUF; 1079 dBUF;
962 1080
… …
967 pread (fd, eio_buf, len, offset); 1085 pread (fd, eio_buf, len, offset);
968 offset += len; 1086 offset += len;
969 todo -= len; 1087 todo -= len;
970 } 1088 }
971 1089
972 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 */
973 return count; 1094 return 0;
974} 1095}
975 1096
976#endif 1097#endif
977 1098
978/* sendfile always needs emulation */ 1099/* sendfile always needs emulation */
979static ssize_t 1100static eio_ssize_t
980eio__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)
981{ 1102{
982 ssize_t written = 0; 1103 eio_ssize_t written = 0;
983 ssize_t res; 1104 eio_ssize_t res;
984 1105
985 if (!count) 1106 if (!count)
986 return 0; 1107 return 0;
987 1108
988 for (;;) 1109 for (;;)
989 { 1110 {
1111#ifdef __APPLE__
1112# undef HAVE_SENDFILE /* broken, as everything on os x */
1113#endif
990#if HAVE_SENDFILE 1114#if HAVE_SENDFILE
991# if __linux 1115# if __linux
992 off_t soffset = offset; 1116 off_t soffset = offset;
993 res = sendfile (ofd, ifd, &soffset, count); 1117 res = sendfile (ofd, ifd, &soffset, count);
994 1118
… …
1010 1134
1011 /* according to source inspection, this is correct, and useful behaviour */ 1135 /* according to source inspection, this is correct, and useful behaviour */
1012 if (sbytes) 1136 if (sbytes)
1013 res = sbytes; 1137 res = sbytes;
1014 1138
1015# elif defined (__APPLE__) && 0 /* broken, as everything on os x */ 1139# elif defined __APPLE__
1016 off_t sbytes = count; 1140 off_t sbytes = count;
1017 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0); 1141 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
1018 1142
1019 /* according to the manpage, sbytes is always valid */ 1143 /* according to the manpage, sbytes is always valid */
1020 if (sbytes) 1144 if (sbytes)
… …
1037 if (res < 0 && sbytes) 1161 if (res < 0 && sbytes)
1038 res = sbytes; 1162 res = sbytes;
1039 1163
1040# endif 1164# endif
1041 1165
1042#elif defined (_WIN32) 1166#elif defined (_WIN32) && 0
1043 /* does not work, just for documentation of what would need to be done */ 1167 /* does not work, just for documentation of what would need to be done */
1044 /* actually, cannot be done like this, as TransmitFile changes the file offset, */ 1168 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
1045 /* libeio guarantees that the file offset does not change, and windows */ 1169 /* libeio guarantees that the file offset does not change, and windows */
1046 /* has no way to get an independent handle to the same file description */ 1170 /* has no way to get an independent handle to the same file description */
1047 HANDLE h = TO_SOCKET (ifd); 1171 HANDLE h = TO_SOCKET (ifd);
1048 SetFilePointer (h, offset, 0, FILE_BEGIN); 1172 SetFilePointer (h, offset, 0, FILE_BEGIN);
1049 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0); 1173 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1050 1174
1051#else 1175#else
1052 res = -1; 1176 res = EIO_ENOSYS ();
1053 errno = ENOSYS;
1054#endif 1177#endif
1055 1178
1056 /* we assume sendfile can copy at least 128mb in one go */ 1179 /* we assume sendfile can copy at least 128mb in one go */
1057 if (res <= 128 * 1024 * 1024) 1180 if (res <= 128 * 1024 * 1024)
1058 { 1181 {
… …
1080 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1203 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1081 /* BSDs */ 1204 /* BSDs */
1082#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */ 1205#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1083 || errno == ENOTSUP 1206 || errno == ENOTSUP
1084#endif 1207#endif
1208#ifdef EOPNOTSUPP /* windows */
1085 || errno == EOPNOTSUPP /* BSDs */ 1209 || errno == EOPNOTSUPP /* BSDs */
1210#endif
1086#if __solaris 1211#if __solaris
1087 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1212 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
1088#endif 1213#endif
1089 ) 1214 )
1090 ) 1215 )
… …
1094 1219
1095 res = 0; 1220 res = 0;
1096 1221
1097 while (count) 1222 while (count)
1098 { 1223 {
1099 ssize_t cnt; 1224 eio_ssize_t cnt;
1100 1225
1101 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);
1102 1227
1103 if (cnt <= 0) 1228 if (cnt <= 0)
1104 { 1229 {
… …
1116 1241
1117 offset += cnt; 1242 offset += cnt;
1118 res += cnt; 1243 res += cnt;
1119 count -= cnt; 1244 count -= cnt;
1120 } 1245 }
1246
1247 FUBd;
1121 } 1248 }
1122 1249
1123 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 /* here we should probably stat the open file and the disk file, to make sure they still match */
1422 close (fd);
1423
1424 if (req->result > 0)
1425 goto done;
1426 }
1427 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1428 return -1;
1429 }
1430#endif
1431#endif
1432
1433 if (*rel != '/')
1434 {
1435 int len;
1436
1437 errno = ENOENT;
1438 if (wd == EIO_INVALID_WD)
1439 return -1;
1440
1441 if (wd == EIO_CWD)
1442 {
1443 if (!getcwd (res, PATH_MAX))
1444 return -1;
1445
1446 len = strlen (res);
1447 }
1448 else
1449 memcpy (res, wd->str, len = wd->len);
1450
1451 if (res [1]) /* only use if not / */
1452 res += len;
1453 }
1454
1455 while (*rel)
1456 {
1457 eio_ssize_t len, linklen;
1458 const char *beg = rel;
1459
1460 while (*rel && *rel != '/')
1461 ++rel;
1462
1463 len = rel - beg;
1464
1465 if (!len) /* skip slashes */
1466 {
1467 ++rel;
1468 continue;
1469 }
1470
1471 if (beg [0] == '.')
1472 {
1473 if (len == 1)
1474 continue; /* . - nop */
1475
1476 if (beg [1] == '.' && len == 2)
1477 {
1478 /* .. - back up one component, if possible */
1479
1480 while (res != tmpbuf->ptr)
1481 if (*--res == '/')
1482 break;
1483
1484 continue;
1485 }
1486 }
1487
1488 errno = ENAMETOOLONG;
1489 if (res + 1 + len + 1 >= tmp1)
1490 return -1;
1491
1492 /* copy one component */
1493 *res = '/';
1494 memcpy (res + 1, beg, len);
1495
1496 /* zero-terminate, for readlink */
1497 res [len + 1] = 0;
1498
1499 /* now check if it's a symlink */
1500 linklen = readlink (tmpbuf->ptr, tmp1, PATH_MAX);
1501
1502 if (linklen < 0)
1503 {
1504 if (errno != EINVAL)
1505 return -1;
1506
1507 /* it's a normal directory. hopefully */
1508 res += len + 1;
1509 }
1510 else
1511 {
1512 /* yay, it was a symlink - build new path in tmp2 */
1513 int rellen = strlen (rel);
1514
1515 errno = ENAMETOOLONG;
1516 if (linklen + 1 + rellen >= PATH_MAX)
1517 return -1;
1518
1519 errno = ELOOP;
1520 if (!--symlinks)
1521 return -1;
1522
1523 if (*tmp1 == '/')
1524 res = tmpbuf->ptr; /* symlink resolves to an absolute path */
1525
1526 /* we need to be careful, as rel might point into tmp2 already */
1527 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1528 tmp2 [linklen] = '/';
1529 memcpy (tmp2, tmp1, linklen);
1530
1531 rel = tmp2;
1532 }
1533 }
1534
1535 /* special case for the lone root path */
1536 if (res == tmpbuf->ptr)
1537 *res++ = '/';
1538
1539 return res - (char *)tmpbuf->ptr;
1124} 1540}
1125 1541
1126static signed char 1542static signed char
1127eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) 1543eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1128{ 1544{
… …
1136 1552
1137#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */ 1553#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1138#define EIO_SORT_FAST 60 /* when to only use insertion sort */ 1554#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1139 1555
1140static void 1556static void
1141eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1557eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1142{ 1558{
1143 unsigned char bits [9 + sizeof (ino_t) * 8]; 1559 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1144 unsigned char *bit = bits; 1560 unsigned char *bit = bits;
1145 1561
1146 assert (CHAR_BIT == 8); 1562 assert (CHAR_BIT == 8);
1147 assert (sizeof (eio_dirent) * 8 < 256); 1563 assert (sizeof (eio_dirent) * 8 < 256);
1148 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */ 1564 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
… …
1150 1566
1151 if (size <= EIO_SORT_FAST) 1567 if (size <= EIO_SORT_FAST)
1152 return; 1568 return;
1153 1569
1154 /* first prepare an array of bits to test in our radix sort */ 1570 /* first prepare an array of bits to test in our radix sort */
1155 /* try to take endianness into account, as well as differences in ino_t sizes */ 1571 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1156 /* inode_bits must contain all inodes ORed together */ 1572 /* inode_bits must contain all inodes ORed together */
1157 /* which is used to skip bits that are 0 everywhere, which is very common */ 1573 /* which is used to skip bits that are 0 everywhere, which is very common */
1158 { 1574 {
1159 ino_t endianness; 1575 eio_ino_t endianness;
1160 int i, j; 1576 int i, j;
1161 1577
1162 /* we store the byte offset of byte n into byte n of "endianness" */ 1578 /* we store the byte offset of byte n into byte n of "endianness" */
1163 for (i = 0; i < sizeof (ino_t); ++i) 1579 for (i = 0; i < sizeof (eio_ino_t); ++i)
1164 ((unsigned char *)&endianness)[i] = i; 1580 ((unsigned char *)&endianness)[i] = i;
1165 1581
1166 *bit++ = 0; 1582 *bit++ = 0;
1167 1583
1168 for (i = 0; i < sizeof (ino_t); ++i) 1584 for (i = 0; i < sizeof (eio_ino_t); ++i)
1169 { 1585 {
1170 /* shifting off the byte offsets out of "endianness" */ 1586 /* shifting off the byte offsets out of "endianness" */
1171 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8; 1587 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1172 endianness >>= 8; 1588 endianness >>= 8;
1173 1589
1174 for (j = 0; j < 8; ++j) 1590 for (j = 0; j < 8; ++j)
1175 if (inode_bits & (((ino_t)1) << (i * 8 + j))) 1591 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1176 *bit++ = offs + j; 1592 *bit++ = offs + j;
1177 } 1593 }
1178 1594
1179 for (j = 0; j < 8; ++j) 1595 for (j = 0; j < 8; ++j)
1180 if (score_bits & (1 << j)) 1596 if (score_bits & (1 << j))
1181 *bit++ = offsetof (eio_dirent, score) * 8 + j; 1597 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1182 } 1598 }
1183 1599
1184 /* now actually do the sorting (a variant of MSD radix sort) */ 1600 /* now actually do the sorting (a variant of MSD radix sort) */
1185 { 1601 {
1186 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base; 1602 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1187 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end; 1603 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1188 unsigned char *bit_stk [9 + sizeof (ino_t) * 8]; 1604 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1189 int stk_idx = 0; 1605 int stk_idx = 0;
1190 1606
1191 base_stk [stk_idx] = dents; 1607 base_stk [stk_idx] = dents;
1192 end_stk [stk_idx] = dents + size; 1608 end_stk [stk_idx] = dents + size;
1193 bit_stk [stk_idx] = bit - 1; 1609 bit_stk [stk_idx] = bit - 1;
… …
1272 } 1688 }
1273 } 1689 }
1274} 1690}
1275 1691
1276static void 1692static void
1277eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1693eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1278{ 1694{
1279 if (size <= 1) 1695 if (size <= 1)
1280 return; /* our insertion sort relies on size > 0 */ 1696 return; /* our insertion sort relies on size > 0 */
1281 1697
1282 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */ 1698 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
… …
1290 1706
1291/* read a full directory */ 1707/* read a full directory */
1292static void 1708static void
1293eio__scandir (eio_req *req, etp_worker *self) 1709eio__scandir (eio_req *req, etp_worker *self)
1294{ 1710{
1295 DIR *dirp;
1296 EIO_STRUCT_DIRENT *entp;
1297 char *name, *names; 1711 char *name, *names;
1298 int namesalloc = 4096; 1712 int namesalloc = 4096 - sizeof (void *) * 4;
1299 int namesoffs = 0; 1713 int namesoffs = 0;
1300 int flags = req->int1; 1714 int flags = req->int1;
1301 eio_dirent *dents = 0; 1715 eio_dirent *dents = 0;
1302 int dentalloc = 128; 1716 int dentalloc = 128;
1303 int dentoffs = 0; 1717 int dentoffs = 0;
1304 ino_t inode_bits = 0; 1718 eio_ino_t inode_bits = 0;
1719#ifdef _WIN32
1720 HANDLE dirp;
1721 WIN32_FIND_DATA entp;
1722#else
1723 DIR *dirp;
1724 EIO_STRUCT_DIRENT *entp;
1725#endif
1305 1726
1306 req->result = -1; 1727 req->result = -1;
1307 1728
1308 if (!(flags & EIO_READDIR_DENTS)) 1729 if (!(flags & EIO_READDIR_DENTS))
1309 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); 1730 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1310 1731
1311 X_LOCK (wrklock); 1732#ifdef _WIN32
1312 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 1733 {
1734 int len = strlen ((const char *)req->ptr1);
1735 char *path = malloc (MAX_PATH);
1736 const char *fmt;
1737 const char *reqpath = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1738
1739 if (!len)
1740 fmt = "./*";
1741 else if (reqpath[len - 1] == '/' || reqpath[len - 1] == '\\')
1742 fmt = "%s*";
1743 else
1744 fmt = "%s/*";
1745
1746 _snprintf (path, MAX_PATH, fmt, reqpath);
1747 dirp = FindFirstFile (path, &entp);
1748 free (path);
1749
1750 if (dirp == INVALID_HANDLE_VALUE)
1751 {
1752 /* should steal _dosmaperr */
1753 switch (GetLastError ())
1754 {
1755 case ERROR_FILE_NOT_FOUND:
1756 req->result = 0;
1757 break;
1758
1759 case ERROR_INVALID_NAME:
1760 case ERROR_PATH_NOT_FOUND:
1761 case ERROR_NO_MORE_FILES:
1762 errno = ENOENT;
1763 break;
1764
1765 case ERROR_NOT_ENOUGH_MEMORY:
1766 errno = ENOMEM;
1767 break;
1768
1769 default:
1770 errno = EINVAL;
1771 break;
1772 }
1773
1774 return;
1775 }
1776 }
1777#else
1778 #if HAVE_AT
1779 if (req->wd)
1780 {
1781 int fd = openat (WD2FD (req->wd), req->ptr1, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1782
1783 if (fd < 0)
1784 return;
1785
1786 dirp = fdopendir (fd);
1787
1788 if (!dirp)
1789 close (fd);
1790 }
1791 else
1313 self->dirp = dirp = opendir (req->ptr1); 1792 dirp = opendir (req->ptr1);
1793 #else
1794 dirp = opendir (wd_expand (&self->tmpbuf, req->wd, req->ptr1));
1795 #endif
1796
1797 if (!dirp)
1798 return;
1799#endif
1314 1800
1315 if (req->flags & EIO_FLAG_PTR1_FREE) 1801 if (req->flags & EIO_FLAG_PTR1_FREE)
1316 free (req->ptr1); 1802 free (req->ptr1);
1317 1803
1318 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1804 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1319 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; 1805 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1320 req->ptr2 = names = malloc (namesalloc); 1806 req->ptr2 = names = malloc (namesalloc);
1321 X_UNLOCK (wrklock);
1322 1807
1323 if (dirp && names && (!flags || dents)) 1808 if (!names || (flags && !dents))
1809 return;
1810
1324 for (;;) 1811 for (;;)
1325 { 1812 {
1813 int done;
1814
1815#ifdef _WIN32
1816 done = !dirp;
1817#else
1326 errno = 0; 1818 errno = 0;
1327 entp = readdir (dirp); 1819 entp = readdir (dirp);
1820 done = !entp;
1821#endif
1328 1822
1329 if (!entp) 1823 if (done)
1330 { 1824 {
1825#ifndef _WIN32
1826 int old_errno = errno;
1827 closedir (dirp);
1828 errno = old_errno;
1829
1331 if (errno) 1830 if (errno)
1332 break; 1831 break;
1832#endif
1333 1833
1334 /* sort etc. */ 1834 /* sort etc. */
1335 req->int1 = flags; 1835 req->int1 = flags;
1336 req->result = dentoffs; 1836 req->result = dentoffs;
1337 1837
1338 if (flags & EIO_READDIR_STAT_ORDER) 1838 if (flags & EIO_READDIR_STAT_ORDER)
1339 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits); 1839 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1340 else if (flags & EIO_READDIR_DIRS_FIRST) 1840 else if (flags & EIO_READDIR_DIRS_FIRST)
1341 if (flags & EIO_READDIR_FOUND_UNKNOWN) 1841 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1342 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */ 1842 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1343 else 1843 else
1844 {
1845 /* in this case, all is known, and we just put dirs first and sort them */
1846 eio_dirent *oth = dents + dentoffs;
1847 eio_dirent *dir = dents;
1848
1849 /* now partition dirs to the front, and non-dirs to the back */
1850 /* by walking from both sides and swapping if necessary */
1851 while (oth > dir)
1852 {
1853 if (dir->type == EIO_DT_DIR)
1854 ++dir;
1855 else if ((--oth)->type == EIO_DT_DIR)
1856 {
1857 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1858
1859 ++dir;
1860 }
1861 }
1862
1863 /* now sort the dirs only (dirs all have the same score) */
1864 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1865 }
1866
1867 break;
1868 }
1869
1870 /* now add the entry to our list(s) */
1871 name = D_NAME (entp);
1872
1873 /* skip . and .. entries */
1874 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1875 {
1876 int len = D_NAMLEN (entp) + 1;
1877
1878 while (ecb_expect_false (namesoffs + len > namesalloc))
1879 {
1880 namesalloc *= 2;
1881 req->ptr2 = names = realloc (names, namesalloc);
1882
1883 if (!names)
1884 break;
1885 }
1886
1887 memcpy (names + namesoffs, name, len);
1888
1889 if (dents)
1890 {
1891 struct eio_dirent *ent;
1892
1893 if (ecb_expect_false (dentoffs == dentalloc))
1344 { 1894 {
1345 /* in this case, all is known, and we just put dirs first and sort them */ 1895 dentalloc *= 2;
1896 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1897
1898 if (!dents)
1899 break;
1900 }
1901
1346 eio_dirent *oth = dents + dentoffs; 1902 ent = dents + dentoffs;
1347 eio_dirent *dir = dents;
1348 1903
1349 /* now partition dirs to the front, and non-dirs to the back */ 1904 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1350 /* by walking from both sides and swapping if necessary */ 1905 ent->namelen = len - 1;
1351 while (oth > dir) 1906 ent->inode = D_INO (entp);
1907
1908 inode_bits |= ent->inode;
1909
1910 switch (D_TYPE (entp))
1911 {
1912 default:
1913 ent->type = EIO_DT_UNKNOWN;
1914 flags |= EIO_READDIR_FOUND_UNKNOWN;
1915 break;
1916
1917 #ifdef DT_FIFO
1918 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1919 #endif
1920 #ifdef DT_CHR
1921 case DT_CHR: ent->type = EIO_DT_CHR; break;
1922 #endif
1923 #ifdef DT_MPC
1924 case DT_MPC: ent->type = EIO_DT_MPC; break;
1925 #endif
1926 #ifdef DT_DIR
1927 case DT_DIR: ent->type = EIO_DT_DIR; break;
1928 #endif
1929 #ifdef DT_NAM
1930 case DT_NAM: ent->type = EIO_DT_NAM; break;
1931 #endif
1932 #ifdef DT_BLK
1933 case DT_BLK: ent->type = EIO_DT_BLK; break;
1934 #endif
1935 #ifdef DT_MPB
1936 case DT_MPB: ent->type = EIO_DT_MPB; break;
1937 #endif
1938 #ifdef DT_REG
1939 case DT_REG: ent->type = EIO_DT_REG; break;
1940 #endif
1941 #ifdef DT_NWK
1942 case DT_NWK: ent->type = EIO_DT_NWK; break;
1943 #endif
1944 #ifdef DT_CMP
1945 case DT_CMP: ent->type = EIO_DT_CMP; break;
1946 #endif
1947 #ifdef DT_LNK
1948 case DT_LNK: ent->type = EIO_DT_LNK; break;
1949 #endif
1950 #ifdef DT_SOCK
1951 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1952 #endif
1953 #ifdef DT_DOOR
1954 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1955 #endif
1956 #ifdef DT_WHT
1957 case DT_WHT: ent->type = EIO_DT_WHT; break;
1958 #endif
1959 }
1960
1961 ent->score = 7;
1962
1963 if (flags & EIO_READDIR_DIRS_FIRST)
1964 {
1965 if (ent->type == EIO_DT_UNKNOWN)
1352 { 1966 {
1353 if (dir->type == EIO_DT_DIR) 1967 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1354 ++dir; 1968 ent->score = 1;
1355 else if ((--oth)->type == EIO_DT_DIR) 1969 else if (!strchr (name, '.')) /* absence of dots indicate likely dirs */
1356 { 1970 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1357 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1358
1359 ++dir;
1360 }
1361 } 1971 }
1362 1972 else if (ent->type == EIO_DT_DIR)
1363 /* now sort the dirs only (dirs all have the same score) */ 1973 ent->score = 0;
1364 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1365 } 1974 }
1366
1367 break;
1368 }
1369
1370 /* now add the entry to our list(s) */
1371 name = entp->d_name;
1372
1373 /* skip . and .. entries */
1374 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1375 {
1376 int len = D_NAMLEN (entp) + 1;
1377
1378 while (ecb_expect_false (namesoffs + len > namesalloc))
1379 {
1380 namesalloc *= 2;
1381 X_LOCK (wrklock);
1382 req->ptr2 = names = realloc (names, namesalloc);
1383 X_UNLOCK (wrklock);
1384
1385 if (!names)
1386 break;
1387 } 1975 }
1388 1976
1389 memcpy (names + namesoffs, name, len);
1390
1391 if (dents)
1392 {
1393 struct eio_dirent *ent;
1394
1395 if (ecb_expect_false (dentoffs == dentalloc))
1396 {
1397 dentalloc *= 2;
1398 X_LOCK (wrklock);
1399 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1400 X_UNLOCK (wrklock);
1401
1402 if (!dents)
1403 break;
1404 }
1405
1406 ent = dents + dentoffs;
1407
1408 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1409 ent->namelen = len - 1;
1410 ent->inode = D_INO (entp);
1411
1412 inode_bits |= ent->inode;
1413
1414 switch (D_TYPE (entp))
1415 {
1416 default:
1417 ent->type = EIO_DT_UNKNOWN;
1418 flags |= EIO_READDIR_FOUND_UNKNOWN;
1419 break;
1420
1421 #ifdef DT_FIFO
1422 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1423 #endif
1424 #ifdef DT_CHR
1425 case DT_CHR: ent->type = EIO_DT_CHR; break;
1426 #endif
1427 #ifdef DT_MPC
1428 case DT_MPC: ent->type = EIO_DT_MPC; break;
1429 #endif
1430 #ifdef DT_DIR
1431 case DT_DIR: ent->type = EIO_DT_DIR; break;
1432 #endif
1433 #ifdef DT_NAM
1434 case DT_NAM: ent->type = EIO_DT_NAM; break;
1435 #endif
1436 #ifdef DT_BLK
1437 case DT_BLK: ent->type = EIO_DT_BLK; break;
1438 #endif
1439 #ifdef DT_MPB
1440 case DT_MPB: ent->type = EIO_DT_MPB; break;
1441 #endif
1442 #ifdef DT_REG
1443 case DT_REG: ent->type = EIO_DT_REG; break;
1444 #endif
1445 #ifdef DT_NWK
1446 case DT_NWK: ent->type = EIO_DT_NWK; break;
1447 #endif
1448 #ifdef DT_CMP
1449 case DT_CMP: ent->type = EIO_DT_CMP; break;
1450 #endif
1451 #ifdef DT_LNK
1452 case DT_LNK: ent->type = EIO_DT_LNK; break;
1453 #endif
1454 #ifdef DT_SOCK
1455 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1456 #endif
1457 #ifdef DT_DOOR
1458 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1459 #endif
1460 #ifdef DT_WHT
1461 case DT_WHT: ent->type = EIO_DT_WHT; break;
1462 #endif
1463 }
1464
1465 ent->score = 7;
1466
1467 if (flags & EIO_READDIR_DIRS_FIRST)
1468 {
1469 if (ent->type == EIO_DT_UNKNOWN)
1470 {
1471 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1472 ent->score = 1;
1473 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1474 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1475 }
1476 else if (ent->type == EIO_DT_DIR)
1477 ent->score = 0;
1478 }
1479 }
1480
1481 namesoffs += len; 1977 namesoffs += len;
1482 ++dentoffs; 1978 ++dentoffs;
1483 } 1979 }
1484 1980
1485 if (EIO_CANCELLED (req)) 1981 if (EIO_CANCELLED (req))
1486 { 1982 {
1487 errno = ECANCELED; 1983 errno = ECANCELED;
1488 break; 1984 break;
1489 } 1985 }
1986
1987#ifdef _WIN32
1988 if (!FindNextFile (dirp, &entp))
1989 {
1990 FindClose (dirp);
1991 dirp = 0;
1490 } 1992 }
1993#endif
1994 }
1491} 1995}
1492 1996
1493#ifdef PAGESIZE 1997/*****************************************************************************/
1494# define eio_pagesize() PAGESIZE 1998/* working directory stuff */
1495#else 1999/* various deficiencies in the posix 2008 api force us to */
1496static intptr_t 2000/* keep the absolute path in string form at all times */
1497eio_pagesize (void) 2001/* fuck yeah. */
1498{
1499 static intptr_t page;
1500 2002
1501 if (!page) 2003#if !HAVE_AT
1502 page = sysconf (_SC_PAGESIZE);
1503 2004
2005/* a bit like realpath, but usually faster because it doesn'T have to return */
2006/* an absolute or canonical path */
2007static const char *
2008wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
2009{
2010 if (!wd || *path == '/')
2011 return path;
2012
2013 if (path [0] == '.' && !path [1])
2014 return wd->str;
2015
2016 {
2017 int l1 = wd->len;
2018 int l2 = strlen (path);
2019
2020 char *res = tmpbuf_get (tmpbuf, l1 + l2 + 2);
2021
2022 memcpy (res, wd->str, l1);
2023 res [l1] = '/';
2024 memcpy (res + l1 + 1, path, l2 + 1);
2025
2026 return res;
2027 }
2028}
2029
2030#endif
2031
2032static eio_wd
2033eio__wd_open_sync (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
2034{
2035 int fd;
2036 eio_wd res;
2037 int len = eio__realpath (tmpbuf, wd, path);
2038
2039 if (len < 0)
2040 return EIO_INVALID_WD;
2041
2042#if HAVE_AT
2043 fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
2044
2045 if (fd < 0)
2046 return EIO_INVALID_WD;
2047#endif
2048
2049 res = malloc (sizeof (*res) + len); /* one extra 0-byte */
2050
2051#if HAVE_AT
2052 res->fd = fd;
2053#endif
2054
2055 res->len = len;
2056 memcpy (res->str, tmpbuf->ptr, len);
2057 res->str [len] = 0;
2058
1504 return page; 2059 return res;
1505} 2060}
1506#endif
1507 2061
1508static void 2062eio_wd
1509eio_page_align (void **addr, size_t *length) 2063eio_wd_open_sync (eio_wd wd, const char *path)
1510{ 2064{
1511 intptr_t mask = eio_pagesize () - 1; 2065 struct tmpbuf tmpbuf = { 0 };
2066 wd = eio__wd_open_sync (&tmpbuf, wd, path);
2067 free (tmpbuf.ptr);
1512 2068
1513 /* round down addr */ 2069 return wd;
1514 intptr_t adj = mask & (intptr_t)*addr;
1515
1516 *addr = (void *)((intptr_t)*addr - adj);
1517 *length += adj;
1518
1519 /* round up length */
1520 *length = (*length + mask) & ~mask;
1521} 2070}
1522 2071
1523#if !_POSIX_MEMLOCK 2072void
1524# define eio__mlockall(a) ((errno = ENOSYS), -1) 2073eio_wd_close_sync (eio_wd wd)
1525#else 2074{
2075 if (wd != EIO_INVALID_WD && wd != EIO_CWD)
2076 {
2077 #if HAVE_AT
2078 close (wd->fd);
2079 #endif
2080 free (wd);
2081 }
2082}
2083
2084#if HAVE_AT
2085
2086/* they forgot these */
1526 2087
1527static int 2088static int
1528eio__mlockall (int flags) 2089eio__truncateat (int dirfd, const char *path, off_t length)
1529{ 2090{
1530 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7 2091 int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
1531 extern int mallopt (int, int); 2092 int res;
1532 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1533 #endif
1534 2093
1535 if (EIO_MCL_CURRENT != MCL_CURRENT 2094 if (fd < 0)
1536 || EIO_MCL_FUTURE != MCL_FUTURE) 2095 return fd;
1537 {
1538 flags = 0
1539 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1540 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1541 }
1542 2096
1543 return mlockall (flags); 2097 res = ftruncate (fd, length);
2098 close (fd);
2099 return res;
1544} 2100}
1545#endif
1546
1547#if !_POSIX_MEMLOCK_RANGE
1548# define eio__mlock(a,b) ((errno = ENOSYS), -1)
1549#else
1550 2101
1551static int 2102static int
1552eio__mlock (void *addr, size_t length) 2103eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
1553{ 2104{
1554 eio_page_align (&addr, &length); 2105 int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
2106 int res;
1555 2107
1556 return mlock (addr, length); 2108 if (fd < 0)
1557} 2109 return fd;
1558 2110
1559#endif 2111 res = fstatvfs (fd, buf);
1560 2112 close (fd);
1561#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1562# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
1563#else
1564
1565static int
1566eio__msync (void *mem, size_t len, int flags)
1567{
1568 eio_page_align (&mem, &len);
1569
1570 if (EIO_MS_ASYNC != MS_SYNC
1571 || EIO_MS_INVALIDATE != MS_INVALIDATE
1572 || EIO_MS_SYNC != MS_SYNC)
1573 {
1574 flags = 0
1575 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1576 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1577 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1578 }
1579
1580 return msync (mem, len, flags);
1581}
1582
1583#endif
1584
1585static int
1586eio__mtouch (eio_req *req)
1587{
1588 void *mem = req->ptr2;
1589 size_t len = req->size;
1590 int flags = req->int1;
1591
1592 eio_page_align (&mem, &len);
1593
1594 {
1595 intptr_t addr = (intptr_t)mem;
1596 intptr_t end = addr + len;
1597 intptr_t page = eio_pagesize ();
1598
1599 if (addr < end)
1600 if (flags & EIO_MT_MODIFY) /* modify */
1601 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1602 else
1603 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1604 }
1605
1606 return 0; 2113 return res;
2114
1607} 2115}
2116
2117#endif
1608 2118
1609/*****************************************************************************/ 2119/*****************************************************************************/
1610 2120
1611#define ALLOC(len) \ 2121#define ALLOC(len) \
1612 if (!req->ptr2) \ 2122 if (!req->ptr2) \
… …
1621 req->result = -1; \ 2131 req->result = -1; \
1622 break; \ 2132 break; \
1623 } \ 2133 } \
1624 } 2134 }
1625 2135
2136static void ecb_noinline ecb_cold
2137etp_proc_init (void)
2138{
2139#if HAVE_PRCTL_SET_NAME
2140 /* provide a more sensible "thread name" */
2141 char name[16 + 1];
2142 const int namelen = sizeof (name) - 1;
2143 int len;
2144
2145 prctl (PR_GET_NAME, (unsigned long)name, 0, 0, 0);
2146 name [namelen] = 0;
2147 len = strlen (name);
2148 strcpy (name + (len <= namelen - 4 ? len : namelen - 4), "/eio");
2149 prctl (PR_SET_NAME, (unsigned long)name, 0, 0, 0);
2150#endif
2151}
2152
1626X_THREAD_PROC (etp_proc) 2153X_THREAD_PROC (etp_proc)
1627{ 2154{
1628 ETP_REQ *req; 2155 ETP_REQ *req;
1629 struct timespec ts; 2156 struct timespec ts;
1630 etp_worker *self = (etp_worker *)thr_arg; 2157 etp_worker *self = (etp_worker *)thr_arg;
1631 2158
2159 etp_proc_init ();
2160
1632 /* try to distribute timeouts somewhat randomly */ 2161 /* try to distribute timeouts somewhat evenly */
1633 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 2162 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1634 2163
1635 for (;;) 2164 for (;;)
1636 { 2165 {
2166 ts.tv_sec = 0;
2167
1637 X_LOCK (reqlock); 2168 X_LOCK (reqlock);
1638 2169
1639 for (;;) 2170 for (;;)
1640 { 2171 {
1641 self->req = req = reqq_shift (&req_queue); 2172 req = reqq_shift (&req_queue);
1642 2173
1643 if (req) 2174 if (req)
1644 break; 2175 break;
1645 2176
2177 if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
2178 {
2179 X_UNLOCK (reqlock);
2180 X_LOCK (wrklock);
2181 --started;
2182 X_UNLOCK (wrklock);
2183 goto quit;
2184 }
2185
1646 ++idle; 2186 ++idle;
1647 2187
1648 ts.tv_sec = time (0) + idle_timeout; 2188 if (idle <= max_idle)
1649 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 2189 /* we are allowed to idle, so do so without any timeout */
2190 X_COND_WAIT (reqwait, reqlock);
2191 else
1650 { 2192 {
1651 if (idle > max_idle) 2193 /* initialise timeout once */
1652 { 2194 if (!ts.tv_sec)
1653 --idle; 2195 ts.tv_sec = time (0) + idle_timeout;
1654 X_UNLOCK (reqlock);
1655 X_LOCK (wrklock);
1656 --started;
1657 X_UNLOCK (wrklock);
1658 goto quit;
1659 }
1660 2196
1661 /* we are allowed to idle, so do so without any timeout */
1662 X_COND_WAIT (reqwait, reqlock); 2197 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
2198 ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1663 } 2199 }
1664 2200
1665 --idle; 2201 --idle;
1666 } 2202 }
1667 2203
… …
1670 X_UNLOCK (reqlock); 2206 X_UNLOCK (reqlock);
1671 2207
1672 if (req->type < 0) 2208 if (req->type < 0)
1673 goto quit; 2209 goto quit;
1674 2210
1675 if (!EIO_CANCELLED (req))
1676 ETP_EXECUTE (self, req); 2211 ETP_EXECUTE (self, req);
1677 2212
1678 X_LOCK (reslock); 2213 X_LOCK (reslock);
1679 2214
1680 ++npending; 2215 ++npending;
1681 2216
1682 if (!reqq_push (&res_queue, req) && want_poll_cb) 2217 if (!reqq_push (&res_queue, req) && want_poll_cb)
1683 want_poll_cb (); 2218 want_poll_cb ();
1684 2219
1685 self->req = 0;
1686 etp_worker_clear (self); 2220 etp_worker_clear (self);
1687 2221
1688 X_UNLOCK (reslock); 2222 X_UNLOCK (reslock);
1689 } 2223 }
1690 2224
1691quit: 2225quit:
2226 free (req);
2227
1692 X_LOCK (wrklock); 2228 X_LOCK (wrklock);
1693 etp_worker_free (self); 2229 etp_worker_free (self);
1694 X_UNLOCK (wrklock); 2230 X_UNLOCK (wrklock);
1695 2231
1696 return 0; 2232 return 0;
… …
1708eio_api_destroy (eio_req *req) 2244eio_api_destroy (eio_req *req)
1709{ 2245{
1710 free (req); 2246 free (req);
1711} 2247}
1712 2248
1713#define REQ(rtype) \ 2249#define REQ(rtype) \
1714 eio_req *req; \ 2250 eio_req *req; \
1715 \ 2251 \
1716 req = (eio_req *)calloc (1, sizeof *req); \ 2252 req = (eio_req *)calloc (1, sizeof *req); \
1717 if (!req) \ 2253 if (!req) \
1718 return 0; \ 2254 return 0; \
… …
1732 { \ 2268 { \
1733 eio_api_destroy (req); \ 2269 eio_api_destroy (req); \
1734 return 0; \ 2270 return 0; \
1735 } 2271 }
1736 2272
2273#define SINGLEDOT(ptr) (0[(char *)(ptr)] == '.' && !1[(char *)(ptr)])
2274
1737static void 2275static void
1738eio_execute (etp_worker *self, eio_req *req) 2276eio_execute (etp_worker *self, eio_req *req)
1739{ 2277{
2278#if HAVE_AT
2279 int dirfd;
2280#else
2281 const char *path;
2282#endif
2283
2284 if (ecb_expect_false (EIO_CANCELLED (req)))
2285 {
2286 req->result = -1;
2287 req->errorno = ECANCELED;
2288 return;
2289 }
2290
2291 if (ecb_expect_false (req->wd == EIO_INVALID_WD))
2292 {
2293 req->result = -1;
2294 req->errorno = ENOENT;
2295 return;
2296 }
2297
2298 if (req->type >= EIO_OPEN)
2299 {
2300 #if HAVE_AT
2301 dirfd = WD2FD (req->wd);
2302 #else
2303 path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
2304 #endif
2305 }
2306
1740 switch (req->type) 2307 switch (req->type)
1741 { 2308 {
2309 case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
2310 req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
2311 break;
2312 case EIO_WD_CLOSE: req->result = 0;
2313 eio_wd_close_sync (req->wd); break;
2314
2315 case EIO_SEEK: eio__lseek (req); break;
1742 case EIO_READ: ALLOC (req->size); 2316 case EIO_READ: ALLOC (req->size);
1743 req->result = req->offs >= 0 2317 req->result = req->offs >= 0
1744 ? pread (req->int1, req->ptr2, req->size, req->offs) 2318 ? pread (req->int1, req->ptr2, req->size, req->offs)
1745 : read (req->int1, req->ptr2, req->size); break; 2319 : read (req->int1, req->ptr2, req->size); break;
1746 case EIO_WRITE: req->result = req->offs >= 0 2320 case EIO_WRITE: req->result = req->offs >= 0
1747 ? pwrite (req->int1, req->ptr2, req->size, req->offs) 2321 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1748 : write (req->int1, req->ptr2, req->size); break; 2322 : write (req->int1, req->ptr2, req->size); break;
1749 2323
1750 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break; 2324 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1751 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break; 2325 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
2326
2327#if HAVE_AT
1752 2328
1753 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2329 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1754 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2330 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break;
1755 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2331 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2332 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, AT_SYMLINK_NOFOLLOW); break;
2333 case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break;
2334 case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break;
2335 case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break;
2336 case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break;
2337
2338 case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
2339 case EIO_RMDIR: /* complications arise because "." cannot be removed, so we might have to expand */
2340 req->result = req->wd && SINGLEDOT (req->ptr1)
2341 ? rmdir (req->wd->str)
2342 : unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break;
2343 case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); break;
2344 case EIO_RENAME: /* complications arise because "." cannot be renamed, so we might have to expand */
2345 req->result = req->wd && SINGLEDOT (req->ptr1)
2346 ? rename (req->wd->str, req->ptr2)
2347 : renameat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2); break;
2348 case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2, 0); break;
2349 case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
2350 case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
2351 case EIO_READLINK: ALLOC (PATH_MAX);
2352 req->result = readlinkat (dirfd, req->ptr1, req->ptr2, PATH_MAX); break;
2353 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2354 req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2355 case EIO_UTIME:
2356 case EIO_FUTIME:
2357 {
2358 struct timespec ts[2];
2359 struct timespec *times;
2360
2361 if (req->nv1 != -1. || req->nv2 != -1.)
2362 {
2363 ts[0].tv_sec = req->nv1;
2364 ts[0].tv_nsec = (req->nv1 - ts[0].tv_sec) * 1e9;
2365 ts[1].tv_sec = req->nv2;
2366 ts[1].tv_nsec = (req->nv2 - ts[1].tv_sec) * 1e9;
2367
2368 times = ts;
2369 }
2370 else
2371 times = 0;
2372
2373 req->result = req->type == EIO_FUTIME
2374 ? futimens (req->int1, times)
2375 : utimensat (dirfd, req->ptr1, times, 0);
2376 }
2377 break;
2378
2379#else
2380
2381 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2382 req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
2383 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1756 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2384 req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
2385 case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
2386 case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
2387 case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
2388 case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
2389
2390 case EIO_UNLINK: req->result = unlink (path ); break;
2391 case EIO_RMDIR: req->result = rmdir (path ); break;
2392 case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
2393 case EIO_RENAME: req->result = rename (path , req->ptr2); break;
2394 case EIO_LINK: req->result = link (path , req->ptr2); break;
2395 case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
2396 case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
2397 case EIO_READLINK: ALLOC (PATH_MAX);
2398 req->result = readlink (path, req->ptr2, PATH_MAX); break;
2399 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2400 req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
2401
2402 case EIO_UTIME:
2403 case EIO_FUTIME:
2404 {
2405 struct timeval tv[2];
2406 struct timeval *times;
2407
2408 if (req->nv1 != -1. || req->nv2 != -1.)
2409 {
2410 tv[0].tv_sec = req->nv1;
2411 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
2412 tv[1].tv_sec = req->nv2;
2413 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
2414
2415 times = tv;
2416 }
2417 else
2418 times = 0;
2419
2420 req->result = req->type == EIO_FUTIME
2421 ? futimes (req->int1, times)
2422 : utimes (req->ptr1, times);
2423 }
2424 break;
2425
2426#endif
2427
2428 case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
2429 {
2430 ALLOC (req->result);
2431 memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
2432 }
2433 break;
2434
1757 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2435 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1758 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; 2436 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1759 2437
1760 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1761 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1762 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS)); 2438 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1763 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break; 2439 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1764 2440
1765 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1766 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; 2441 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1767 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1768 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break; 2442 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1769 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1770 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break; 2443 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1771 2444
1772 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
1773 case EIO_CLOSE: req->result = close (req->int1); break; 2445 case EIO_CLOSE: req->result = close (req->int1); break;
1774 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break; 2446 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1775 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
1776 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1777 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1778 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1779 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1780 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1781 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1782
1783 case EIO_REALPATH: req->flags |= EIO_FLAG_PTR2_FREE;
1784 req->ptr2 = realpath (req->ptr1, 0);
1785 req->result = req->ptr2 ? strlen (req->ptr2) : -1;
1786 break;
1787
1788 case EIO_READLINK: ALLOC (PATH_MAX);
1789 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1790
1791 case EIO_SYNC: req->result = 0; sync (); break; 2447 case EIO_SYNC: req->result = 0; sync (); break;
1792 case EIO_FSYNC: req->result = fsync (req->int1); break; 2448 case EIO_FSYNC: req->result = fsync (req->int1); break;
1793 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 2449 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
2450 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
2451 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1794 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break; 2452 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1795 case EIO_MTOUCH: req->result = eio__mtouch (req); break; 2453 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1796 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break; 2454 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1797 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break; 2455 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1798 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break; 2456 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
1799 2457
1800 case EIO_READDIR: eio__scandir (req, self); break; 2458 case EIO_READDIR: eio__scandir (req, self); break;
1801 2459
1802 case EIO_BUSY: 2460 case EIO_BUSY:
1803#ifdef _WIN32 2461#ifdef _WIN32
… …
1812 req->result = select (0, 0, 0, 0, &tv); 2470 req->result = select (0, 0, 0, 0, &tv);
1813 } 2471 }
1814#endif 2472#endif
1815 break; 2473 break;
1816 2474
1817 case EIO_UTIME:
1818 case EIO_FUTIME:
1819 {
1820 struct timeval tv[2];
1821 struct timeval *times;
1822
1823 if (req->nv1 != -1. || req->nv2 != -1.)
1824 {
1825 tv[0].tv_sec = req->nv1;
1826 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
1827 tv[1].tv_sec = req->nv2;
1828 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
1829
1830 times = tv;
1831 }
1832 else
1833 times = 0;
1834
1835 req->result = req->type == EIO_FUTIME
1836 ? futimes (req->int1, times)
1837 : utimes (req->ptr1, times);
1838 }
1839 break;
1840
1841 case EIO_GROUP: 2475 case EIO_GROUP:
1842 abort (); /* handled in eio_request */ 2476 abort (); /* handled in eio_request */
1843 2477
1844 case EIO_NOP: 2478 case EIO_NOP:
1845 req->result = 0; 2479 req->result = 0;
… …
1848 case EIO_CUSTOM: 2482 case EIO_CUSTOM:
1849 req->feed (req); 2483 req->feed (req);
1850 break; 2484 break;
1851 2485
1852 default: 2486 default:
1853 errno = ENOSYS;
1854 req->result = -1; 2487 req->result = EIO_ENOSYS ();
1855 break; 2488 break;
1856 } 2489 }
1857 2490
1858 req->errorno = errno; 2491 req->errorno = errno;
1859} 2492}
1860 2493
1861#ifndef EIO_NO_WRAPPERS 2494#ifndef EIO_NO_WRAPPERS
1862 2495
2496eio_req *eio_wd_open (const char *path, int pri, eio_cb cb, void *data)
2497{
2498 REQ (EIO_WD_OPEN); PATH; SEND;
2499}
2500
2501eio_req *eio_wd_close (eio_wd wd, int pri, eio_cb cb, void *data)
2502{
2503 REQ (EIO_WD_CLOSE); req->wd = wd; SEND;
2504}
2505
1863eio_req *eio_nop (int pri, eio_cb cb, void *data) 2506eio_req *eio_nop (int pri, eio_cb cb, void *data)
1864{ 2507{
1865 REQ (EIO_NOP); SEND; 2508 REQ (EIO_NOP); SEND;
1866} 2509}
1867 2510
… …
1883eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2526eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1884{ 2527{
1885 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2528 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1886} 2529}
1887 2530
2531eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2532{
2533 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2534}
2535
2536eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
2537{
2538 REQ (EIO_SYNCFS); req->int1 = fd; SEND;
2539}
2540
2541eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2542{
2543 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2544}
2545
1888eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2546eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1889{ 2547{
1890 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2548 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1891} 2549}
1892 2550
… …
1898eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data) 2556eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
1899{ 2557{
1900 REQ (EIO_MLOCKALL); req->int1 = flags; SEND; 2558 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
1901} 2559}
1902 2560
1903eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) 2561eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
1904{ 2562{
1905 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; 2563 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
1906}
1907
1908eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1909{
1910 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1911} 2564}
1912 2565
1913eio_req *eio_close (int fd, int pri, eio_cb cb, void *data) 2566eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
1914{ 2567{
1915 REQ (EIO_CLOSE); req->int1 = fd; SEND; 2568 REQ (EIO_CLOSE); req->int1 = fd; SEND;
… …
1918eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data) 2571eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
1919{ 2572{
1920 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND; 2573 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
1921} 2574}
1922 2575
2576eio_req *eio_seek (int fd, off_t offset, int whence, int pri, eio_cb cb, void *data)
2577{
2578 REQ (EIO_SEEK); req->int1 = fd; req->offs = offset; req->int2 = whence; SEND;
2579}
2580
1923eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data) 2581eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1924{ 2582{
1925 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND; 2583 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1926} 2584}
1927 2585
… …
1953eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) 2611eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1954{ 2612{
1955 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2613 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1956} 2614}
1957 2615
1958eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2616eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1959{ 2617{
1960 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2618 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1961} 2619}
1962 2620
1963eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) 2621eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
… …
1983eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data) 2641eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1984{ 2642{
1985 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2643 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1986} 2644}
1987 2645
1988eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2646eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1989{ 2647{
1990 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2648 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1991} 2649}
1992 2650
1993eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) 2651eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
… …
2140} 2798}
2141 2799
2142/*****************************************************************************/ 2800/*****************************************************************************/
2143/* misc garbage */ 2801/* misc garbage */
2144 2802
2145ssize_t 2803eio_ssize_t
2146eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2804eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2147{ 2805{
2148 etp_worker wrk;
2149 ssize_t ret;
2150
2151 wrk.dbuf = 0;
2152
2153 ret = eio__sendfile (ofd, ifd, offset, count, &wrk); 2806 return eio__sendfile (ofd, ifd, offset, count);
2154
2155 if (wrk.dbuf)
2156 free (wrk.dbuf);
2157
2158 return ret;
2159} 2807}
2160 2808

Diff Legend

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