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

Comparing libeio/eio.c (file contents):
Revision 1.67 by root, Sun Jun 5 20:25:00 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,
… …
40#ifndef _WIN32 40#ifndef _WIN32
41# include "config.h" 41# include "config.h"
42#endif 42#endif
43 43
44#include "eio.h" 44#include "eio.h"
45#include "ecb.h"
45 46
46#ifdef EIO_STACKSIZE 47#ifdef EIO_STACKSIZE
47# define XTHREAD_STACKSIZE EIO_STACKSIZE 48# define X_STACKSIZE EIO_STACKSIZE
48#endif 49#endif
49#include "xthread.h" 50#include "xthread.h"
50 51
51#include <errno.h> 52#include <errno.h>
52#include <stddef.h> 53#include <stddef.h>
53#include <stdlib.h> 54#include <stdlib.h>
54#include <string.h> 55#include <string.h>
55#include <errno.h> 56#include <errno.h>
56#include <sys/types.h> 57#include <sys/types.h>
57#include <sys/stat.h> 58#include <sys/stat.h>
58#include <sys/statvfs.h>
59#include <limits.h> 59#include <limits.h>
60#include <fcntl.h> 60#include <fcntl.h>
61#include <assert.h> 61#include <assert.h>
62 62
63/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */ 63/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
64/* intptr_t only comes form stdint.h, says idiot openbsd coder */ 64/* intptr_t only comes from stdint.h, says idiot openbsd coder */
65#if HAVE_STDINT_H 65#if HAVE_STDINT_H
66# include <stdint.h> 66# include <stdint.h>
67#endif 67#endif
68 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
69#ifndef EIO_FINISH 82#ifndef EIO_FINISH
70# 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
71#endif 84#endif
72 85
73#ifndef EIO_DESTROY 86#ifndef EIO_DESTROY
… …
76 89
77#ifndef EIO_FEED 90#ifndef EIO_FEED
78# 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)
79#endif 92#endif
80 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
81#ifdef _WIN32 105#ifdef _WIN32
82 106
83 /*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
84#else 232#else
85 233
86# include <sys/time.h> 234 #include <sys/time.h>
87# include <sys/select.h> 235 #include <sys/select.h>
236 #include <sys/statvfs.h>
88# 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
89# include <utime.h> 272# include <utime.h>
90# include <signal.h>
91# include <dirent.h>
92
93#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
94# include <sys/mman.h>
95#endif
96
97/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
98# if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
99# define _DIRENT_HAVE_D_TYPE /* sigh */
100# define D_INO(de) (de)->d_fileno
101# define D_NAMLEN(de) (de)->d_namlen
102# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
103# define D_INO(de) (de)->d_ino
104# endif 273#endif
105 274
106#ifdef _D_EXACT_NAMLEN 275#if HAVE_SYS_SYSCALL_H
107# undef D_NAMLEN 276# include <sys/syscall.h>
108# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
109#endif
110
111# ifdef _DIRENT_HAVE_D_TYPE
112# define D_TYPE(de) (de)->d_type
113# endif 277#endif
114 278
115# ifndef EIO_STRUCT_DIRENT 279#if HAVE_SYS_PRCTL_H
116# define EIO_STRUCT_DIRENT struct dirent 280# include <sys/prctl.h>
117# endif
118
119#endif 281#endif
120 282
121#if HAVE_SENDFILE 283#if HAVE_SENDFILE
122# if __linux 284# if __linux
123# include <sys/sendfile.h> 285# include <sys/sendfile.h>
… …
138#endif 300#endif
139#ifndef D_INO 301#ifndef D_INO
140# define D_INO(de) 0 302# define D_INO(de) 0
141#endif 303#endif
142#ifndef D_NAMLEN 304#ifndef D_NAMLEN
143# define D_NAMLEN(de) strlen ((de)->d_name) 305# define D_NAMLEN(entp) strlen (D_NAME (entp))
144#endif 306#endif
145 307
146/* used for struct dirent, AIX doesn't provide it */ 308/* used for struct dirent, AIX doesn't provide it */
147#ifndef NAME_MAX 309#ifndef NAME_MAX
148# define NAME_MAX 4096 310# define NAME_MAX 4096
… …
154#endif 316#endif
155 317
156/* buffer size for various temporary buffers */ 318/* buffer size for various temporary buffers */
157#define EIO_BUFSIZE 65536 319#define EIO_BUFSIZE 65536
158 320
159#define dBUF \ 321#define dBUF \
160 char *eio_buf; \
161 ETP_WORKER_LOCK (self); \
162 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 322 char *eio_buf = malloc (EIO_BUFSIZE); \
163 ETP_WORKER_UNLOCK (self); \
164 errno = ENOMEM; \ 323 errno = ENOMEM; \
165 if (!eio_buf) \ 324 if (!eio_buf) \
166 return -1; 325 return -1
326
327#define FUBd \
328 free (eio_buf)
167 329
168#define EIO_TICKS ((1000000 + 1023) >> 10) 330#define EIO_TICKS ((1000000 + 1023) >> 10)
169 331
170/*****************************************************************************/ 332/*****************************************************************************/
171 333
172#if __GNUC__ >= 3 334struct tmpbuf
173# 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
174#else 360#else
175# 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);
176#endif 363#endif
177 364
178#define expect_false(expr) expect ((expr) != 0, 0) 365struct eio_pwd
179#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};
180 373
181/*****************************************************************************/ 374/*****************************************************************************/
182 375
183#define ETP_PRI_MIN EIO_PRI_MIN 376#define ETP_PRI_MIN EIO_PRI_MIN
184#define ETP_PRI_MAX EIO_PRI_MAX 377#define ETP_PRI_MAX EIO_PRI_MAX
… …
190static int eio_finish (eio_req *req); 383static int eio_finish (eio_req *req);
191#define ETP_FINISH(req) eio_finish (req) 384#define ETP_FINISH(req) eio_finish (req)
192static void eio_execute (struct etp_worker *self, eio_req *req); 385static void eio_execute (struct etp_worker *self, eio_req *req);
193#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) 386#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
194 387
195#define ETP_WORKER_CLEAR(req) \
196 if (wrk->dbuf) \
197 { \
198 free (wrk->dbuf); \
199 wrk->dbuf = 0; \
200 } \
201 \
202 if (wrk->dirp) \
203 { \
204 closedir (wrk->dirp); \
205 wrk->dirp = 0; \
206 }
207
208#define ETP_WORKER_COMMON \
209 void *dbuf; \
210 DIR *dirp;
211
212/*****************************************************************************/ 388/*****************************************************************************/
213 389
214#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) 390#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
215 391
216/* calculate time difference in ~1/EIO_TICKS of a second */ 392/* calculate time difference in ~1/EIO_TICKS of a second */
393ecb_inline int
217static int tvdiff (struct timeval *tv1, struct timeval *tv2) 394tvdiff (struct timeval *tv1, struct timeval *tv2)
218{ 395{
219 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 396 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
220 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 397 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
221} 398}
222 399
… …
226static void (*done_poll_cb) (void); 403static void (*done_poll_cb) (void);
227 404
228static unsigned int max_poll_time; /* reslock */ 405static unsigned int max_poll_time; /* reslock */
229static unsigned int max_poll_reqs; /* reslock */ 406static unsigned int max_poll_reqs; /* reslock */
230 407
231static volatile unsigned int nreqs; /* reqlock */ 408static unsigned int nreqs; /* reqlock */
232static volatile unsigned int nready; /* reqlock */ 409static unsigned int nready; /* reqlock */
233static volatile unsigned int npending; /* reqlock */ 410static unsigned int npending; /* reqlock */
234static 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 */
235static 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 */
236 413
237static xmutex_t wrklock; 414static xmutex_t wrklock;
238static xmutex_t reslock; 415static xmutex_t reslock;
239static xmutex_t reqlock; 416static xmutex_t reqlock;
240static xcond_t reqwait; 417static xcond_t reqwait;
241 418
242#if !HAVE_PREADWRITE
243/*
244 * make our pread/pwrite emulation safe against themselves, but not against
245 * normal read/write by using a mutex. slows down execution a lot,
246 * but that's your problem, not mine.
247 */
248static xmutex_t preadwritelock = X_MUTEX_INIT;
249#endif
250
251typedef struct etp_worker 419typedef struct etp_worker
252{ 420{
421 struct tmpbuf tmpbuf;
422
253 /* locked by wrklock */ 423 /* locked by wrklock */
254 struct etp_worker *prev, *next; 424 struct etp_worker *prev, *next;
255 425
256 xthread_t tid; 426 xthread_t tid;
257 427
258 /* locked by reslock, reqlock or wrklock */ 428#ifdef ETP_WORKER_COMMON
259 ETP_REQ *req; /* currently processed request */
260
261 ETP_WORKER_COMMON 429 ETP_WORKER_COMMON
430#endif
262} etp_worker; 431} etp_worker;
263 432
264static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */ 433static etp_worker wrk_first; /* NOT etp */
265 434
266#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock) 435#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
267#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock) 436#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
268 437
269/* worker threads management */ 438/* worker threads management */
270 439
440static void
271static void etp_worker_clear (etp_worker *wrk) 441etp_worker_clear (etp_worker *wrk)
272{ 442{
273 ETP_WORKER_CLEAR (wrk);
274} 443}
275 444
445static void ecb_cold
276static void etp_worker_free (etp_worker *wrk) 446etp_worker_free (etp_worker *wrk)
277{ 447{
448 free (wrk->tmpbuf.ptr);
449
278 wrk->next->prev = wrk->prev; 450 wrk->next->prev = wrk->prev;
279 wrk->prev->next = wrk->next; 451 wrk->prev->next = wrk->next;
280 452
281 free (wrk); 453 free (wrk);
282} 454}
283 455
284static unsigned int etp_nreqs (void) 456static unsigned int
457etp_nreqs (void)
285{ 458{
286 int retval; 459 int retval;
287 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 460 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
288 retval = nreqs; 461 retval = nreqs;
289 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 462 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
290 return retval; 463 return retval;
291} 464}
292 465
293static unsigned int etp_nready (void) 466static unsigned int
467etp_nready (void)
294{ 468{
295 unsigned int retval; 469 unsigned int retval;
296 470
297 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 471 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
298 retval = nready; 472 retval = nready;
299 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 473 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
300 474
301 return retval; 475 return retval;
302} 476}
303 477
304static unsigned int etp_npending (void) 478static unsigned int
479etp_npending (void)
305{ 480{
306 unsigned int retval; 481 unsigned int retval;
307 482
308 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 483 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
309 retval = npending; 484 retval = npending;
310 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 485 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
311 486
312 return retval; 487 return retval;
313} 488}
314 489
315static unsigned int etp_nthreads (void) 490static unsigned int
491etp_nthreads (void)
316{ 492{
317 unsigned int retval; 493 unsigned int retval;
318 494
319 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 495 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
320 retval = started; 496 retval = started;
… …
334} etp_reqq; 510} etp_reqq;
335 511
336static etp_reqq req_queue; 512static etp_reqq req_queue;
337static etp_reqq res_queue; 513static etp_reqq res_queue;
338 514
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
339static int reqq_push (etp_reqq *q, ETP_REQ *req) 527reqq_push (etp_reqq *q, ETP_REQ *req)
340{ 528{
341 int pri = req->pri; 529 int pri = req->pri;
342 req->next = 0; 530 req->next = 0;
343 531
344 if (q->qe[pri]) 532 if (q->qe[pri])
… …
350 q->qe[pri] = q->qs[pri] = req; 538 q->qe[pri] = q->qs[pri] = req;
351 539
352 return q->size++; 540 return q->size++;
353} 541}
354 542
543static ETP_REQ * ecb_noinline
355static ETP_REQ *reqq_shift (etp_reqq *q) 544reqq_shift (etp_reqq *q)
356{ 545{
357 int pri; 546 int pri;
358 547
359 if (!q->size) 548 if (!q->size)
360 return 0; 549 return 0;
… …
375 } 564 }
376 565
377 abort (); 566 abort ();
378} 567}
379 568
380static void etp_thread_init (void) 569static int ecb_cold
570etp_init (void (*want_poll)(void), void (*done_poll)(void))
381{ 571{
382 X_MUTEX_CREATE (wrklock); 572 X_MUTEX_CREATE (wrklock);
383 X_MUTEX_CREATE (reslock); 573 X_MUTEX_CREATE (reslock);
384 X_MUTEX_CREATE (reqlock); 574 X_MUTEX_CREATE (reqlock);
385 X_COND_CREATE (reqwait); 575 X_COND_CREATE (reqwait);
386}
387 576
388static void etp_atfork_prepare (void) 577 reqq_init (&req_queue);
389{ 578 reqq_init (&res_queue);
390 X_LOCK (wrklock);
391 X_LOCK (reqlock);
392 X_LOCK (reslock);
393#if !HAVE_PREADWRITE
394 X_LOCK (preadwritelock);
395#endif
396}
397 579
398static void etp_atfork_parent (void) 580 wrk_first.next =
399{ 581 wrk_first.prev = &wrk_first;
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 etp_atfork_child (void)
409{
410 ETP_REQ *prv;
411
412 while ((prv = reqq_shift (&req_queue)))
413 ETP_DESTROY (prv);
414
415 while ((prv = reqq_shift (&res_queue)))
416 ETP_DESTROY (prv);
417
418 while (wrk_first.next != &wrk_first)
419 {
420 etp_worker *wrk = wrk_first.next;
421
422 if (wrk->req)
423 ETP_DESTROY (wrk->req);
424
425 etp_worker_clear (wrk);
426 etp_worker_free (wrk);
427 }
428 582
429 started = 0; 583 started = 0;
430 idle = 0; 584 idle = 0;
431 nreqs = 0; 585 nreqs = 0;
432 nready = 0; 586 nready = 0;
433 npending = 0; 587 npending = 0;
434 588
435 etp_thread_init ();
436}
437
438static void
439etp_once_init (void)
440{
441 etp_thread_init ();
442 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
443}
444
445static int
446etp_init (void (*want_poll)(void), void (*done_poll)(void))
447{
448 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
449
450 pthread_once (&doinit, etp_once_init);
451
452 want_poll_cb = want_poll; 589 want_poll_cb = want_poll;
453 done_poll_cb = done_poll; 590 done_poll_cb = done_poll;
454 591
455 return 0; 592 return 0;
456} 593}
457 594
458X_THREAD_PROC (etp_proc); 595X_THREAD_PROC (etp_proc);
459 596
597static void ecb_cold
460static void etp_start_thread (void) 598etp_start_thread (void)
461{ 599{
462 etp_worker *wrk = calloc (1, sizeof (etp_worker)); 600 etp_worker *wrk = calloc (1, sizeof (etp_worker));
463 601
464 /*TODO*/ 602 /*TODO*/
465 assert (("unable to allocate worker thread data", wrk)); 603 assert (("unable to allocate worker thread data", wrk));
466 604
467 X_LOCK (wrklock); 605 X_LOCK (wrklock);
468 606
469 if (thread_create (&wrk->tid, etp_proc, (void *)wrk)) 607 if (xthread_create (&wrk->tid, etp_proc, (void *)wrk))
470 { 608 {
471 wrk->prev = &wrk_first; 609 wrk->prev = &wrk_first;
472 wrk->next = wrk_first.next; 610 wrk->next = wrk_first.next;
473 wrk_first.next->prev = wrk; 611 wrk_first.next->prev = wrk;
474 wrk_first.next = wrk; 612 wrk_first.next = wrk;
… …
478 free (wrk); 616 free (wrk);
479 617
480 X_UNLOCK (wrklock); 618 X_UNLOCK (wrklock);
481} 619}
482 620
621static void
483static void etp_maybe_start_thread (void) 622etp_maybe_start_thread (void)
484{ 623{
485 if (expect_true (etp_nthreads () >= wanted)) 624 if (ecb_expect_true (etp_nthreads () >= wanted))
486 return; 625 return;
487 626
488 /* todo: maybe use idle here, but might be less exact */ 627 /* todo: maybe use idle here, but might be less exact */
489 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 ()))
490 return; 629 return;
491 630
492 etp_start_thread (); 631 etp_start_thread ();
493} 632}
494 633
634static void ecb_cold
495static void etp_end_thread (void) 635etp_end_thread (void)
496{ 636{
497 eio_req *req = calloc (1, sizeof (eio_req)); 637 eio_req *req = calloc (1, sizeof (eio_req)); /* will be freed by worker */
498 638
499 req->type = -1; 639 req->type = -1;
500 req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 640 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
501 641
502 X_LOCK (reqlock); 642 X_LOCK (reqlock);
… …
507 X_LOCK (wrklock); 647 X_LOCK (wrklock);
508 --started; 648 --started;
509 X_UNLOCK (wrklock); 649 X_UNLOCK (wrklock);
510} 650}
511 651
512static int etp_poll (void) 652static int
653etp_poll (void)
513{ 654{
514 unsigned int maxreqs; 655 unsigned int maxreqs;
515 unsigned int maxtime; 656 unsigned int maxtime;
516 struct timeval tv_start, tv_now; 657 struct timeval tv_start, tv_now;
517 658
… …
547 688
548 X_LOCK (reqlock); 689 X_LOCK (reqlock);
549 --nreqs; 690 --nreqs;
550 X_UNLOCK (reqlock); 691 X_UNLOCK (reqlock);
551 692
552 if (expect_false (req->type == EIO_GROUP && req->size)) 693 if (ecb_expect_false (req->type == EIO_GROUP && req->size))
553 { 694 {
554 req->int1 = 1; /* mark request as delayed */ 695 req->int1 = 1; /* mark request as delayed */
555 continue; 696 continue;
556 } 697 }
557 else 698 else
558 { 699 {
559 int res = ETP_FINISH (req); 700 int res = ETP_FINISH (req);
560 if (expect_false (res)) 701 if (ecb_expect_false (res))
561 return res; 702 return res;
562 } 703 }
563 704
564 if (expect_false (maxreqs && !--maxreqs)) 705 if (ecb_expect_false (maxreqs && !--maxreqs))
565 break; 706 break;
566 707
567 if (maxtime) 708 if (maxtime)
568 { 709 {
569 gettimeofday (&tv_now, 0); 710 gettimeofday (&tv_now, 0);
… …
575 716
576 errno = EAGAIN; 717 errno = EAGAIN;
577 return -1; 718 return -1;
578} 719}
579 720
721static void
580static void etp_cancel (ETP_REQ *req) 722etp_cancel (ETP_REQ *req)
581{ 723{
582 X_LOCK (wrklock); 724 req->cancelled = 1;
583 req->flags |= EIO_FLAG_CANCELLED;
584 X_UNLOCK (wrklock);
585 725
586 eio_grp_cancel (req); 726 eio_grp_cancel (req);
587} 727}
588 728
729static void
589static void etp_submit (ETP_REQ *req) 730etp_submit (ETP_REQ *req)
590{ 731{
591 req->pri -= ETP_PRI_MIN; 732 req->pri -= ETP_PRI_MIN;
592 733
593 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;
594 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;
595 736
596 if (expect_false (req->type == EIO_GROUP)) 737 if (ecb_expect_false (req->type == EIO_GROUP))
597 { 738 {
598 /* I hope this is worth it :/ */ 739 /* I hope this is worth it :/ */
599 X_LOCK (reqlock); 740 X_LOCK (reqlock);
600 ++nreqs; 741 ++nreqs;
601 X_UNLOCK (reqlock); 742 X_UNLOCK (reqlock);
… …
620 761
621 etp_maybe_start_thread (); 762 etp_maybe_start_thread ();
622 } 763 }
623} 764}
624 765
766static void ecb_cold
625static void etp_set_max_poll_time (double nseconds) 767etp_set_max_poll_time (double nseconds)
626{ 768{
627 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 769 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
628 max_poll_time = nseconds * EIO_TICKS; 770 max_poll_time = nseconds * EIO_TICKS;
629 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 771 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
630} 772}
631 773
774static void ecb_cold
632static void etp_set_max_poll_reqs (unsigned int maxreqs) 775etp_set_max_poll_reqs (unsigned int maxreqs)
633{ 776{
634 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 777 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
635 max_poll_reqs = maxreqs; 778 max_poll_reqs = maxreqs;
636 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 779 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
637} 780}
638 781
782static void ecb_cold
639static void etp_set_max_idle (unsigned int nthreads) 783etp_set_max_idle (unsigned int nthreads)
640{ 784{
641 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 785 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
642 max_idle = nthreads; 786 max_idle = nthreads;
643 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 787 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
644} 788}
645 789
790static void ecb_cold
646static void etp_set_idle_timeout (unsigned int seconds) 791etp_set_idle_timeout (unsigned int seconds)
647{ 792{
648 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 793 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
649 idle_timeout = seconds; 794 idle_timeout = seconds;
650 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 795 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
651} 796}
652 797
798static void ecb_cold
653static void etp_set_min_parallel (unsigned int nthreads) 799etp_set_min_parallel (unsigned int nthreads)
654{ 800{
655 if (wanted < nthreads) 801 if (wanted < nthreads)
656 wanted = nthreads; 802 wanted = nthreads;
657} 803}
658 804
805static void ecb_cold
659static void etp_set_max_parallel (unsigned int nthreads) 806etp_set_max_parallel (unsigned int nthreads)
660{ 807{
661 if (wanted > nthreads) 808 if (wanted > nthreads)
662 wanted = nthreads; 809 wanted = nthreads;
663 810
664 while (started > wanted) 811 while (started > wanted)
665 etp_end_thread (); 812 etp_end_thread ();
666} 813}
667 814
668/*****************************************************************************/ 815/*****************************************************************************/
669 816
817static void
670static void grp_try_feed (eio_req *grp) 818grp_try_feed (eio_req *grp)
671{ 819{
672 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 820 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
673 { 821 {
674 grp->flags &= ~EIO_FLAG_GROUPADD; 822 grp->flags &= ~EIO_FLAG_GROUPADD;
675 823
… …
682 break; 830 break;
683 } 831 }
684 } 832 }
685} 833}
686 834
835static int
687static int grp_dec (eio_req *grp) 836grp_dec (eio_req *grp)
688{ 837{
689 --grp->size; 838 --grp->size;
690 839
691 /* call feeder, if applicable */ 840 /* call feeder, if applicable */
692 grp_try_feed (grp); 841 grp_try_feed (grp);
… …
696 return eio_finish (grp); 845 return eio_finish (grp);
697 else 846 else
698 return 0; 847 return 0;
699} 848}
700 849
850static void
701void eio_destroy (eio_req *req) 851eio_destroy (eio_req *req)
702{ 852{
703 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); 853 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
704 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); 854 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
705 855
706 EIO_DESTROY (req); 856 EIO_DESTROY (req);
707} 857}
708 858
859static int
709static int eio_finish (eio_req *req) 860eio_finish (eio_req *req)
710{ 861{
711 int res = EIO_FINISH (req); 862 int res = EIO_FINISH (req);
712 863
713 if (req->grp) 864 if (req->grp)
714 { 865 {
… …
722 if (grp->grp_first == req) 873 if (grp->grp_first == req)
723 grp->grp_first = req->grp_next; 874 grp->grp_first = req->grp_next;
724 875
725 res2 = grp_dec (grp); 876 res2 = grp_dec (grp);
726 877
727 if (!res && res2) 878 if (!res)
728 res = res2; 879 res = res2;
729 } 880 }
730 881
731 eio_destroy (req); 882 eio_destroy (req);
732 883
733 return res; 884 return res;
734} 885}
735 886
887void
736void eio_grp_cancel (eio_req *grp) 888eio_grp_cancel (eio_req *grp)
737{ 889{
738 for (grp = grp->grp_first; grp; grp = grp->grp_next) 890 for (grp = grp->grp_first; grp; grp = grp->grp_next)
739 eio_cancel (grp); 891 eio_cancel (grp);
740} 892}
741 893
894void
742void eio_cancel (eio_req *req) 895eio_cancel (eio_req *req)
743{ 896{
744 etp_cancel (req); 897 etp_cancel (req);
745} 898}
746 899
900void
747void eio_submit (eio_req *req) 901eio_submit (eio_req *req)
748{ 902{
749 etp_submit (req); 903 etp_submit (req);
750} 904}
751 905
752unsigned int eio_nreqs (void) 906unsigned int
907eio_nreqs (void)
753{ 908{
754 return etp_nreqs (); 909 return etp_nreqs ();
755} 910}
756 911
757unsigned int eio_nready (void) 912unsigned int
913eio_nready (void)
758{ 914{
759 return etp_nready (); 915 return etp_nready ();
760} 916}
761 917
762unsigned int eio_npending (void) 918unsigned int
919eio_npending (void)
763{ 920{
764 return etp_npending (); 921 return etp_npending ();
765} 922}
766 923
767unsigned int eio_nthreads (void) 924unsigned int ecb_cold
925eio_nthreads (void)
768{ 926{
769 return etp_nthreads (); 927 return etp_nthreads ();
770} 928}
771 929
930void ecb_cold
772void eio_set_max_poll_time (double nseconds) 931eio_set_max_poll_time (double nseconds)
773{ 932{
774 etp_set_max_poll_time (nseconds); 933 etp_set_max_poll_time (nseconds);
775} 934}
776 935
936void ecb_cold
777void eio_set_max_poll_reqs (unsigned int maxreqs) 937eio_set_max_poll_reqs (unsigned int maxreqs)
778{ 938{
779 etp_set_max_poll_reqs (maxreqs); 939 etp_set_max_poll_reqs (maxreqs);
780} 940}
781 941
942void ecb_cold
782void eio_set_max_idle (unsigned int nthreads) 943eio_set_max_idle (unsigned int nthreads)
783{ 944{
784 etp_set_max_idle (nthreads); 945 etp_set_max_idle (nthreads);
785} 946}
786 947
948void ecb_cold
787void eio_set_idle_timeout (unsigned int seconds) 949eio_set_idle_timeout (unsigned int seconds)
788{ 950{
789 etp_set_idle_timeout (seconds); 951 etp_set_idle_timeout (seconds);
790} 952}
791 953
954void ecb_cold
792void eio_set_min_parallel (unsigned int nthreads) 955eio_set_min_parallel (unsigned int nthreads)
793{ 956{
794 etp_set_min_parallel (nthreads); 957 etp_set_min_parallel (nthreads);
795} 958}
796 959
960void ecb_cold
797void eio_set_max_parallel (unsigned int nthreads) 961eio_set_max_parallel (unsigned int nthreads)
798{ 962{
799 etp_set_max_parallel (nthreads); 963 etp_set_max_parallel (nthreads);
800} 964}
801 965
802int eio_poll (void) 966int eio_poll (void)
… …
804 return etp_poll (); 968 return etp_poll ();
805} 969}
806 970
807/*****************************************************************************/ 971/*****************************************************************************/
808/* work around various missing functions */ 972/* work around various missing functions */
809
810#if !HAVE_PREADWRITE
811# undef pread
812# undef pwrite
813# define pread eio__pread
814# define pwrite eio__pwrite
815
816static ssize_t
817eio__pread (int fd, void *buf, size_t count, off_t offset)
818{
819 ssize_t res;
820 off_t ooffset;
821
822 X_LOCK (preadwritelock);
823 ooffset = lseek (fd, 0, SEEK_CUR);
824 lseek (fd, offset, SEEK_SET);
825 res = read (fd, buf, count);
826 lseek (fd, ooffset, SEEK_SET);
827 X_UNLOCK (preadwritelock);
828
829 return res;
830}
831
832static ssize_t
833eio__pwrite (int fd, void *buf, size_t count, off_t offset)
834{
835 ssize_t res;
836 off_t ooffset;
837
838 X_LOCK (preadwritelock);
839 ooffset = lseek (fd, 0, SEEK_CUR);
840 lseek (fd, offset, SEEK_SET);
841 res = write (fd, buf, count);
842 lseek (fd, ooffset, SEEK_SET);
843 X_UNLOCK (preadwritelock);
844
845 return res;
846}
847#endif
848 973
849#ifndef HAVE_UTIMES 974#ifndef HAVE_UTIMES
850 975
851# undef utimes 976# undef utimes
852# define utimes(path,times) eio__utimes (path, times) 977# define utimes(path,times) eio__utimes (path, times)
… …
872#ifndef HAVE_FUTIMES 997#ifndef HAVE_FUTIMES
873 998
874# undef futimes 999# undef futimes
875# define futimes(fd,times) eio__futimes (fd, times) 1000# define futimes(fd,times) eio__futimes (fd, times)
876 1001
1002static int
877static int eio__futimes (int fd, const struct timeval tv[2]) 1003eio__futimes (int fd, const struct timeval tv[2])
878{ 1004{
879 errno = ENOSYS; 1005 errno = ENOSYS;
880 return -1; 1006 return -1;
881} 1007}
882 1008
… …
885#if !HAVE_FDATASYNC 1011#if !HAVE_FDATASYNC
886# undef fdatasync 1012# undef fdatasync
887# define fdatasync(fd) fsync (fd) 1013# define fdatasync(fd) fsync (fd)
888#endif 1014#endif
889 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
890/* sync_file_range always needs emulation */ 1033/* sync_file_range always needs emulation */
891int 1034static int
892eio__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)
893{ 1036{
894#if HAVE_SYNC_FILE_RANGE 1037#if HAVE_SYNC_FILE_RANGE
895 int res; 1038 int res;
896 1039
… …
913 /* 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
914 * call fdatasync, as that matches the expectation of its users best */ 1057 * call fdatasync, as that matches the expectation of its users best */
915 return fdatasync (fd); 1058 return fdatasync (fd);
916} 1059}
917 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
918#if !HAVE_READAHEAD 1071#if !HAVE_READAHEAD
919# undef readahead 1072# undef readahead
920# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 1073# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
921 1074
922static ssize_t 1075static eio_ssize_t
923eio__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)
924{ 1077{
925 size_t todo = count; 1078 size_t todo = count;
926 dBUF; 1079 dBUF;
927 1080
… …
932 pread (fd, eio_buf, len, offset); 1085 pread (fd, eio_buf, len, offset);
933 offset += len; 1086 offset += len;
934 todo -= len; 1087 todo -= len;
935 } 1088 }
936 1089
937 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 */
938 return count; 1094 return 0;
939} 1095}
940 1096
941#endif 1097#endif
942 1098
943/* sendfile always needs emulation */ 1099/* sendfile always needs emulation */
944static ssize_t 1100static eio_ssize_t
945eio__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)
946{ 1102{
1103 eio_ssize_t written = 0;
947 ssize_t res; 1104 eio_ssize_t res;
948 1105
949 if (!count) 1106 if (!count)
950 return 0; 1107 return 0;
951 1108
1109 for (;;)
1110 {
1111#ifdef __APPLE__
1112# undef HAVE_SENDFILE /* broken, as everything on os x */
1113#endif
952#if HAVE_SENDFILE 1114#if HAVE_SENDFILE
953# if __linux 1115# if __linux
1116 off_t soffset = offset;
954 res = sendfile (ofd, ifd, &offset, count); 1117 res = sendfile (ofd, ifd, &soffset, count);
955 1118
956# elif __FreeBSD__ 1119# elif __FreeBSD__
957 /* 1120 /*
958 * 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
959 * wasted on making it similar to other I/O functions. 1122 * wasted on making it similar to other I/O functions.
960 */ 1123 */
961 {
962 off_t sbytes; 1124 off_t sbytes;
963 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 1125 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
964 1126
965 #if 0 /* according to the manpage, this is correct, but broken behaviour */ 1127 #if 0 /* according to the manpage, this is correct, but broken behaviour */
966 /* freebsd' sendfile will return 0 on success */ 1128 /* freebsd' sendfile will return 0 on success */
967 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */ 1129 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
968 /* not on e.g. EIO or EPIPE - sounds broken */ 1130 /* not on e.g. EIO or EPIPE - sounds broken */
969 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0) 1131 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
970 res = sbytes; 1132 res = sbytes;
971 #endif 1133 #endif
972 1134
973 /* according to source inspection, this is correct, and useful behaviour */ 1135 /* according to source inspection, this is correct, and useful behaviour */
974 if (sbytes) 1136 if (sbytes)
975 res = sbytes; 1137 res = sbytes;
976 }
977 1138
978# elif defined (__APPLE__) 1139# elif defined __APPLE__
979
980 {
981 off_t sbytes = count; 1140 off_t sbytes = count;
982 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0); 1141 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
983 1142
984 /* according to the manpage, sbytes is always valid */ 1143 /* according to the manpage, sbytes is always valid */
985 if (sbytes) 1144 if (sbytes)
986 res = sbytes; 1145 res = sbytes;
987 }
988 1146
989# elif __hpux 1147# elif __hpux
990 res = sendfile (ofd, ifd, offset, count, 0, 0); 1148 res = sendfile (ofd, ifd, offset, count, 0, 0);
991 1149
992# elif __solaris 1150# elif __solaris
993 {
994 struct sendfilevec vec; 1151 struct sendfilevec vec;
995 size_t sbytes; 1152 size_t sbytes;
996 1153
997 vec.sfv_fd = ifd; 1154 vec.sfv_fd = ifd;
998 vec.sfv_flag = 0; 1155 vec.sfv_flag = 0;
999 vec.sfv_off = offset; 1156 vec.sfv_off = offset;
1000 vec.sfv_len = count; 1157 vec.sfv_len = count;
1001 1158
1002 res = sendfilev (ofd, &vec, 1, &sbytes); 1159 res = sendfilev (ofd, &vec, 1, &sbytes);
1003 1160
1004 if (res < 0 && sbytes) 1161 if (res < 0 && sbytes)
1005 res = sbytes; 1162 res = sbytes;
1006 }
1007 1163
1008# endif 1164# endif
1009 1165
1010#elif defined (_WIN32) 1166#elif defined (_WIN32) && 0
1011
1012 /* 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 */
1013 { 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 */
1014 HANDLE h = TO_SOCKET (ifd); 1171 HANDLE h = TO_SOCKET (ifd);
1015 SetFilePointer (h, offset, 0, FILE_BEGIN); 1172 SetFilePointer (h, offset, 0, FILE_BEGIN);
1016 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0); 1173 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1017 }
1018 1174
1019#else 1175#else
1020 res = -1; 1176 res = EIO_ENOSYS ();
1021 errno = ENOSYS;
1022#endif 1177#endif
1023 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
1024 if (res < 0 1202 if (res < 0
1025 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1203 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1026 /* BSDs */ 1204 /* BSDs */
1027#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... */
1028 || errno == ENOTSUP 1206 || errno == ENOTSUP
1029#endif 1207#endif
1208#ifdef EOPNOTSUPP /* windows */
1030 || errno == EOPNOTSUPP /* BSDs */ 1209 || errno == EOPNOTSUPP /* BSDs */
1210#endif
1031#if __solaris 1211#if __solaris
1032 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1212 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
1033#endif 1213#endif
1034 ) 1214 )
1035 ) 1215 )
… …
1039 1219
1040 res = 0; 1220 res = 0;
1041 1221
1042 while (count) 1222 while (count)
1043 { 1223 {
1044 ssize_t cnt; 1224 eio_ssize_t cnt;
1045 1225
1046 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);
1047 1227
1048 if (cnt <= 0) 1228 if (cnt <= 0)
1049 { 1229 {
… …
1061 1241
1062 offset += cnt; 1242 offset += cnt;
1063 res += cnt; 1243 res += cnt;
1064 count -= cnt; 1244 count -= cnt;
1065 } 1245 }
1246
1247 FUBd;
1066 } 1248 }
1067 1249
1068 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;
1069} 1540}
1070 1541
1071static signed char 1542static signed char
1072eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) 1543eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1073{ 1544{
… …
1081 1552
1082#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 */
1083#define EIO_SORT_FAST 60 /* when to only use insertion sort */ 1554#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1084 1555
1085static void 1556static void
1086eio_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)
1087{ 1558{
1088 unsigned char bits [9 + sizeof (ino_t) * 8]; 1559 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1089 unsigned char *bit = bits; 1560 unsigned char *bit = bits;
1090 1561
1091 assert (CHAR_BIT == 8); 1562 assert (CHAR_BIT == 8);
1092 assert (sizeof (eio_dirent) * 8 < 256); 1563 assert (sizeof (eio_dirent) * 8 < 256);
1093 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */ 1564 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
… …
1095 1566
1096 if (size <= EIO_SORT_FAST) 1567 if (size <= EIO_SORT_FAST)
1097 return; 1568 return;
1098 1569
1099 /* 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 */
1100 /* 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 */
1101 /* inode_bits must contain all inodes ORed together */ 1572 /* inode_bits must contain all inodes ORed together */
1102 /* 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 */
1103 { 1574 {
1104 ino_t endianness; 1575 eio_ino_t endianness;
1105 int i, j; 1576 int i, j;
1106 1577
1107 /* 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" */
1108 for (i = 0; i < sizeof (ino_t); ++i) 1579 for (i = 0; i < sizeof (eio_ino_t); ++i)
1109 ((unsigned char *)&endianness)[i] = i; 1580 ((unsigned char *)&endianness)[i] = i;
1110 1581
1111 *bit++ = 0; 1582 *bit++ = 0;
1112 1583
1113 for (i = 0; i < sizeof (ino_t); ++i) 1584 for (i = 0; i < sizeof (eio_ino_t); ++i)
1114 { 1585 {
1115 /* shifting off the byte offsets out of "endianness" */ 1586 /* shifting off the byte offsets out of "endianness" */
1116 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8; 1587 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1117 endianness >>= 8; 1588 endianness >>= 8;
1118 1589
1119 for (j = 0; j < 8; ++j) 1590 for (j = 0; j < 8; ++j)
1120 if (inode_bits & (((ino_t)1) << (i * 8 + j))) 1591 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1121 *bit++ = offs + j; 1592 *bit++ = offs + j;
1122 } 1593 }
1123 1594
1124 for (j = 0; j < 8; ++j) 1595 for (j = 0; j < 8; ++j)
1125 if (score_bits & (1 << j)) 1596 if (score_bits & (1 << j))
1126 *bit++ = offsetof (eio_dirent, score) * 8 + j; 1597 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1127 } 1598 }
1128 1599
1129 /* now actually do the sorting (a variant of MSD radix sort) */ 1600 /* now actually do the sorting (a variant of MSD radix sort) */
1130 { 1601 {
1131 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base; 1602 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1132 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end; 1603 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1133 unsigned char *bit_stk [9 + sizeof (ino_t) * 8]; 1604 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1134 int stk_idx = 0; 1605 int stk_idx = 0;
1135 1606
1136 base_stk [stk_idx] = dents; 1607 base_stk [stk_idx] = dents;
1137 end_stk [stk_idx] = dents + size; 1608 end_stk [stk_idx] = dents + size;
1138 bit_stk [stk_idx] = bit - 1; 1609 bit_stk [stk_idx] = bit - 1;
… …
1217 } 1688 }
1218 } 1689 }
1219} 1690}
1220 1691
1221static void 1692static void
1222eio_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)
1223{ 1694{
1224 if (size <= 1) 1695 if (size <= 1)
1225 return; /* our insertion sort relies on size > 0 */ 1696 return; /* our insertion sort relies on size > 0 */
1226 1697
1227 /* 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 */
… …
1235 1706
1236/* read a full directory */ 1707/* read a full directory */
1237static void 1708static void
1238eio__scandir (eio_req *req, etp_worker *self) 1709eio__scandir (eio_req *req, etp_worker *self)
1239{ 1710{
1240 DIR *dirp;
1241 EIO_STRUCT_DIRENT *entp;
1242 char *name, *names; 1711 char *name, *names;
1243 int namesalloc = 4096; 1712 int namesalloc = 4096 - sizeof (void *) * 4;
1244 int namesoffs = 0; 1713 int namesoffs = 0;
1245 int flags = req->int1; 1714 int flags = req->int1;
1246 eio_dirent *dents = 0; 1715 eio_dirent *dents = 0;
1247 int dentalloc = 128; 1716 int dentalloc = 128;
1248 int dentoffs = 0; 1717 int dentoffs = 0;
1249 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
1250 1726
1251 req->result = -1; 1727 req->result = -1;
1252 1728
1253 if (!(flags & EIO_READDIR_DENTS)) 1729 if (!(flags & EIO_READDIR_DENTS))
1254 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); 1730 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1255 1731
1256 X_LOCK (wrklock); 1732#ifdef _WIN32
1257 /* 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
1258 self->dirp = dirp = opendir (req->ptr1); 1792 dirp = opendir (req->ptr1);
1793 #else
1794 dirp = opendir (wd_expand (&self->tmpbuf, req->wd, req->ptr1));
1795 #endif
1796
1797 if (!dirp)
1798 return;
1799#endif
1800
1801 if (req->flags & EIO_FLAG_PTR1_FREE)
1802 free (req->ptr1);
1259 1803
1260 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1804 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1261 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; 1805 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1262 req->ptr2 = names = malloc (namesalloc); 1806 req->ptr2 = names = malloc (namesalloc);
1263 X_UNLOCK (wrklock);
1264 1807
1265 if (dirp && names && (!flags || dents)) 1808 if (!names || (flags && !dents))
1809 return;
1810
1266 for (;;) 1811 for (;;)
1267 { 1812 {
1813 int done;
1814
1815#ifdef _WIN32
1816 done = !dirp;
1817#else
1268 errno = 0; 1818 errno = 0;
1269 entp = readdir (dirp); 1819 entp = readdir (dirp);
1820 done = !entp;
1821#endif
1270 1822
1271 if (!entp) 1823 if (done)
1272 { 1824 {
1825#ifndef _WIN32
1826 int old_errno = errno;
1827 closedir (dirp);
1828 errno = old_errno;
1829
1273 if (errno) 1830 if (errno)
1274 break; 1831 break;
1832#endif
1275 1833
1276 /* sort etc. */ 1834 /* sort etc. */
1277 req->int1 = flags; 1835 req->int1 = flags;
1278 req->result = dentoffs; 1836 req->result = dentoffs;
1279 1837
1280 if (flags & EIO_READDIR_STAT_ORDER) 1838 if (flags & EIO_READDIR_STAT_ORDER)
1281 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);
1282 else if (flags & EIO_READDIR_DIRS_FIRST) 1840 else if (flags & EIO_READDIR_DIRS_FIRST)
1283 if (flags & EIO_READDIR_FOUND_UNKNOWN) 1841 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1284 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 */
1285 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))
1286 { 1894 {
1287 /* 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
1288 eio_dirent *oth = dents + dentoffs; 1902 ent = dents + dentoffs;
1289 eio_dirent *dir = dents;
1290 1903
1291 /* 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 */
1292 /* by walking from both sides and swapping if necessary */ 1905 ent->namelen = len - 1;
1293 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)
1294 { 1966 {
1295 if (dir->type == EIO_DT_DIR) 1967 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1296 ++dir; 1968 ent->score = 1;
1297 else if ((--oth)->type == EIO_DT_DIR) 1969 else if (!strchr (name, '.')) /* absence of dots indicate likely dirs */
1298 { 1970 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1299 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1300
1301 ++dir;
1302 }
1303 } 1971 }
1304 1972 else if (ent->type == EIO_DT_DIR)
1305 /* now sort the dirs only (dirs all have the same score) */ 1973 ent->score = 0;
1306 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1307 } 1974 }
1308
1309 break;
1310 }
1311
1312 /* now add the entry to our list(s) */
1313 name = entp->d_name;
1314
1315 /* skip . and .. entries */
1316 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1317 {
1318 int len = D_NAMLEN (entp) + 1;
1319
1320 while (expect_false (namesoffs + len > namesalloc))
1321 {
1322 namesalloc *= 2;
1323 X_LOCK (wrklock);
1324 req->ptr2 = names = realloc (names, namesalloc);
1325 X_UNLOCK (wrklock);
1326
1327 if (!names)
1328 break;
1329 } 1975 }
1330 1976
1331 memcpy (names + namesoffs, name, len);
1332
1333 if (dents)
1334 {
1335 struct eio_dirent *ent;
1336
1337 if (expect_false (dentoffs == dentalloc))
1338 {
1339 dentalloc *= 2;
1340 X_LOCK (wrklock);
1341 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1342 X_UNLOCK (wrklock);
1343
1344 if (!dents)
1345 break;
1346 }
1347
1348 ent = dents + dentoffs;
1349
1350 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1351 ent->namelen = len - 1;
1352 ent->inode = D_INO (entp);
1353
1354 inode_bits |= ent->inode;
1355
1356 switch (D_TYPE (entp))
1357 {
1358 default:
1359 ent->type = EIO_DT_UNKNOWN;
1360 flags |= EIO_READDIR_FOUND_UNKNOWN;
1361 break;
1362
1363 #ifdef DT_FIFO
1364 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1365 #endif
1366 #ifdef DT_CHR
1367 case DT_CHR: ent->type = EIO_DT_CHR; break;
1368 #endif
1369 #ifdef DT_MPC
1370 case DT_MPC: ent->type = EIO_DT_MPC; break;
1371 #endif
1372 #ifdef DT_DIR
1373 case DT_DIR: ent->type = EIO_DT_DIR; break;
1374 #endif
1375 #ifdef DT_NAM
1376 case DT_NAM: ent->type = EIO_DT_NAM; break;
1377 #endif
1378 #ifdef DT_BLK
1379 case DT_BLK: ent->type = EIO_DT_BLK; break;
1380 #endif
1381 #ifdef DT_MPB
1382 case DT_MPB: ent->type = EIO_DT_MPB; break;
1383 #endif
1384 #ifdef DT_REG
1385 case DT_REG: ent->type = EIO_DT_REG; break;
1386 #endif
1387 #ifdef DT_NWK
1388 case DT_NWK: ent->type = EIO_DT_NWK; break;
1389 #endif
1390 #ifdef DT_CMP
1391 case DT_CMP: ent->type = EIO_DT_CMP; break;
1392 #endif
1393 #ifdef DT_LNK
1394 case DT_LNK: ent->type = EIO_DT_LNK; break;
1395 #endif
1396 #ifdef DT_SOCK
1397 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1398 #endif
1399 #ifdef DT_DOOR
1400 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1401 #endif
1402 #ifdef DT_WHT
1403 case DT_WHT: ent->type = EIO_DT_WHT; break;
1404 #endif
1405 }
1406
1407 ent->score = 7;
1408
1409 if (flags & EIO_READDIR_DIRS_FIRST)
1410 {
1411 if (ent->type == EIO_DT_UNKNOWN)
1412 {
1413 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1414 ent->score = 1;
1415 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1416 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1417 }
1418 else if (ent->type == EIO_DT_DIR)
1419 ent->score = 0;
1420 }
1421 }
1422
1423 namesoffs += len; 1977 namesoffs += len;
1424 ++dentoffs; 1978 ++dentoffs;
1425 } 1979 }
1426 1980
1427 if (EIO_CANCELLED (req)) 1981 if (EIO_CANCELLED (req))
1428 { 1982 {
1429 errno = ECANCELED; 1983 errno = ECANCELED;
1430 break; 1984 break;
1431 } 1985 }
1986
1987#ifdef _WIN32
1988 if (!FindNextFile (dirp, &entp))
1989 {
1990 FindClose (dirp);
1991 dirp = 0;
1432 } 1992 }
1993#endif
1994 }
1433} 1995}
1434 1996
1435#ifdef PAGESIZE 1997/*****************************************************************************/
1436# define eio_pagesize() PAGESIZE 1998/* working directory stuff */
1437#else 1999/* various deficiencies in the posix 2008 api force us to */
1438static intptr_t 2000/* keep the absolute path in string form at all times */
1439eio_pagesize (void) 2001/* fuck yeah. */
1440{
1441 static intptr_t page;
1442 2002
1443 if (!page) 2003#if !HAVE_AT
1444 page = sysconf (_SC_PAGESIZE);
1445 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
1446 return page; 2059 return res;
1447} 2060}
1448#endif
1449 2061
1450static void 2062eio_wd
1451eio_page_align (void **addr, size_t *length) 2063eio_wd_open_sync (eio_wd wd, const char *path)
1452{ 2064{
1453 intptr_t mask = eio_pagesize () - 1; 2065 struct tmpbuf tmpbuf = { 0 };
2066 wd = eio__wd_open_sync (&tmpbuf, wd, path);
2067 free (tmpbuf.ptr);
1454 2068
1455 /* round down addr */ 2069 return wd;
1456 intptr_t adj = mask & (intptr_t)*addr;
1457
1458 *addr = (void *)((intptr_t)*addr - adj);
1459 *length += adj;
1460
1461 /* round up length */
1462 *length = (*length + mask) & ~mask;
1463} 2070}
1464 2071
1465#if !_POSIX_MEMLOCK 2072void
1466# define eio__mlockall(a) ((errno = ENOSYS), -1) 2073eio_wd_close_sync (eio_wd wd)
1467#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 */
1468 2087
1469static int 2088static int
1470eio__mlockall (int flags) 2089eio__truncateat (int dirfd, const char *path, off_t length)
1471{ 2090{
1472 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7 2091 int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
1473 extern int mallopt (int, int); 2092 int res;
1474 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1475 #endif
1476 2093
1477 if (EIO_MCL_CURRENT != MCL_CURRENT 2094 if (fd < 0)
1478 || EIO_MCL_FUTURE != MCL_FUTURE) 2095 return fd;
1479 {
1480 flags = 0
1481 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1482 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1483 }
1484 2096
1485 return mlockall (flags); 2097 res = ftruncate (fd, length);
2098 close (fd);
2099 return res;
1486} 2100}
1487#endif
1488
1489#if !_POSIX_MEMLOCK_RANGE
1490# define eio__mlock(a,b) ((errno = ENOSYS), -1)
1491#else
1492 2101
1493static int 2102static int
1494eio__mlock (void *addr, size_t length) 2103eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
1495{ 2104{
1496 eio_page_align (&addr, &length); 2105 int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
2106 int res;
1497 2107
1498 return mlock (addr, length); 2108 if (fd < 0)
1499} 2109 return fd;
1500 2110
1501#endif 2111 res = fstatvfs (fd, buf);
1502 2112 close (fd);
1503#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1504# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
1505#else
1506
1507int
1508eio__msync (void *mem, size_t len, int flags)
1509{
1510 eio_page_align (&mem, &len);
1511
1512 if (EIO_MS_ASYNC != MS_SYNC
1513 || EIO_MS_INVALIDATE != MS_INVALIDATE
1514 || EIO_MS_SYNC != MS_SYNC)
1515 {
1516 flags = 0
1517 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1518 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1519 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1520 }
1521
1522 return msync (mem, len, flags);
1523}
1524
1525#endif
1526
1527int
1528eio__mtouch (eio_req *req)
1529{
1530 void *mem = req->ptr2;
1531 size_t len = req->size;
1532 int flags = req->int1;
1533
1534 eio_page_align (&mem, &len);
1535
1536 {
1537 intptr_t addr = (intptr_t)mem;
1538 intptr_t end = addr + len;
1539 intptr_t page = eio_pagesize ();
1540
1541 if (addr < end)
1542 if (flags & EIO_MT_MODIFY) /* modify */
1543 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1544 else
1545 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1546 }
1547
1548 return 0; 2113 return res;
2114
1549} 2115}
2116
2117#endif
1550 2118
1551/*****************************************************************************/ 2119/*****************************************************************************/
1552 2120
1553#define ALLOC(len) \ 2121#define ALLOC(len) \
1554 if (!req->ptr2) \ 2122 if (!req->ptr2) \
… …
1563 req->result = -1; \ 2131 req->result = -1; \
1564 break; \ 2132 break; \
1565 } \ 2133 } \
1566 } 2134 }
1567 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
1568X_THREAD_PROC (etp_proc) 2153X_THREAD_PROC (etp_proc)
1569{ 2154{
1570 ETP_REQ *req; 2155 ETP_REQ *req;
1571 struct timespec ts; 2156 struct timespec ts;
1572 etp_worker *self = (etp_worker *)thr_arg; 2157 etp_worker *self = (etp_worker *)thr_arg;
1573 2158
2159 etp_proc_init ();
2160
1574 /* try to distribute timeouts somewhat randomly */ 2161 /* try to distribute timeouts somewhat evenly */
1575 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 2162 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1576 2163
1577 for (;;) 2164 for (;;)
1578 { 2165 {
2166 ts.tv_sec = 0;
2167
1579 X_LOCK (reqlock); 2168 X_LOCK (reqlock);
1580 2169
1581 for (;;) 2170 for (;;)
1582 { 2171 {
1583 self->req = req = reqq_shift (&req_queue); 2172 req = reqq_shift (&req_queue);
1584 2173
1585 if (req) 2174 if (req)
1586 break; 2175 break;
1587 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
1588 ++idle; 2186 ++idle;
1589 2187
1590 ts.tv_sec = time (0) + idle_timeout; 2188 if (idle <= max_idle)
1591 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
1592 { 2192 {
1593 if (idle > max_idle) 2193 /* initialise timeout once */
1594 { 2194 if (!ts.tv_sec)
1595 --idle; 2195 ts.tv_sec = time (0) + idle_timeout;
1596 X_UNLOCK (reqlock);
1597 X_LOCK (wrklock);
1598 --started;
1599 X_UNLOCK (wrklock);
1600 goto quit;
1601 }
1602 2196
1603 /* we are allowed to idle, so do so without any timeout */
1604 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.,.. */
1605 } 2199 }
1606 2200
1607 --idle; 2201 --idle;
1608 } 2202 }
1609 2203
… …
1612 X_UNLOCK (reqlock); 2206 X_UNLOCK (reqlock);
1613 2207
1614 if (req->type < 0) 2208 if (req->type < 0)
1615 goto quit; 2209 goto quit;
1616 2210
1617 if (!EIO_CANCELLED (req))
1618 ETP_EXECUTE (self, req); 2211 ETP_EXECUTE (self, req);
1619 2212
1620 X_LOCK (reslock); 2213 X_LOCK (reslock);
1621 2214
1622 ++npending; 2215 ++npending;
1623 2216
1624 if (!reqq_push (&res_queue, req) && want_poll_cb) 2217 if (!reqq_push (&res_queue, req) && want_poll_cb)
1625 want_poll_cb (); 2218 want_poll_cb ();
1626 2219
1627 self->req = 0;
1628 etp_worker_clear (self); 2220 etp_worker_clear (self);
1629 2221
1630 X_UNLOCK (reslock); 2222 X_UNLOCK (reslock);
1631 } 2223 }
1632 2224
1633quit: 2225quit:
2226 free (req);
2227
1634 X_LOCK (wrklock); 2228 X_LOCK (wrklock);
1635 etp_worker_free (self); 2229 etp_worker_free (self);
1636 X_UNLOCK (wrklock); 2230 X_UNLOCK (wrklock);
1637 2231
1638 return 0; 2232 return 0;
1639} 2233}
1640 2234
1641/*****************************************************************************/ 2235/*****************************************************************************/
1642 2236
2237int ecb_cold
1643int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 2238eio_init (void (*want_poll)(void), void (*done_poll)(void))
1644{ 2239{
1645 return etp_init (want_poll, done_poll); 2240 return etp_init (want_poll, done_poll);
1646} 2241}
1647 2242
2243ecb_inline void
1648static void eio_api_destroy (eio_req *req) 2244eio_api_destroy (eio_req *req)
1649{ 2245{
1650 free (req); 2246 free (req);
1651} 2247}
1652 2248
1653#define REQ(rtype) \ 2249#define REQ(rtype) \
1654 eio_req *req; \ 2250 eio_req *req; \
1655 \ 2251 \
1656 req = (eio_req *)calloc (1, sizeof *req); \ 2252 req = (eio_req *)calloc (1, sizeof *req); \
1657 if (!req) \ 2253 if (!req) \
1658 return 0; \ 2254 return 0; \
… …
1672 { \ 2268 { \
1673 eio_api_destroy (req); \ 2269 eio_api_destroy (req); \
1674 return 0; \ 2270 return 0; \
1675 } 2271 }
1676 2272
2273#define SINGLEDOT(ptr) (0[(char *)(ptr)] == '.' && !1[(char *)(ptr)])
2274
2275static void
1677static void eio_execute (etp_worker *self, eio_req *req) 2276eio_execute (etp_worker *self, eio_req *req)
1678{ 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
1679 switch (req->type) 2307 switch (req->type)
1680 { 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;
1681 case EIO_READ: ALLOC (req->size); 2316 case EIO_READ: ALLOC (req->size);
1682 req->result = req->offs >= 0 2317 req->result = req->offs >= 0
1683 ? pread (req->int1, req->ptr2, req->size, req->offs) 2318 ? pread (req->int1, req->ptr2, req->size, req->offs)
1684 : read (req->int1, req->ptr2, req->size); break; 2319 : read (req->int1, req->ptr2, req->size); break;
1685 case EIO_WRITE: req->result = req->offs >= 0 2320 case EIO_WRITE: req->result = req->offs >= 0
1686 ? pwrite (req->int1, req->ptr2, req->size, req->offs) 2321 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1687 : write (req->int1, req->ptr2, req->size); break; 2322 : write (req->int1, req->ptr2, req->size); break;
1688 2323
1689 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;
1690 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
1691 2328
1692 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2329 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1693 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;
1694 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));
1695 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
1696 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2435 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1697 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; 2436 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1698 2437
1699 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1700 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1701 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS)); 2438 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1702 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break; 2439 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1703 2440
1704 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1705 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;
1706 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1707 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;
1708 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1709 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break; 2443 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1710 2444
1711 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
1712 case EIO_CLOSE: req->result = close (req->int1); break; 2445 case EIO_CLOSE: req->result = close (req->int1); break;
1713 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break; 2446 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1714 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
1715 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1716 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1717 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1718 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1719 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1720 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1721
1722 case EIO_READLINK: ALLOC (PATH_MAX);
1723 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1724
1725 case EIO_SYNC: req->result = 0; sync (); break; 2447 case EIO_SYNC: req->result = 0; sync (); break;
1726 case EIO_FSYNC: req->result = fsync (req->int1); break; 2448 case EIO_FSYNC: req->result = fsync (req->int1); break;
1727 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;
1728 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;
1729 case EIO_MTOUCH: req->result = eio__mtouch (req); break; 2453 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1730 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;
1731 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break; 2455 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1732 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;
1733 2457
1734 case EIO_READDIR: eio__scandir (req, self); break; 2458 case EIO_READDIR: eio__scandir (req, self); break;
1735 2459
1736 case EIO_BUSY: 2460 case EIO_BUSY:
1737#ifdef _WIN32 2461#ifdef _WIN32
… …
1746 req->result = select (0, 0, 0, 0, &tv); 2470 req->result = select (0, 0, 0, 0, &tv);
1747 } 2471 }
1748#endif 2472#endif
1749 break; 2473 break;
1750 2474
1751 case EIO_UTIME:
1752 case EIO_FUTIME:
1753 {
1754 struct timeval tv[2];
1755 struct timeval *times;
1756
1757 if (req->nv1 != -1. || req->nv2 != -1.)
1758 {
1759 tv[0].tv_sec = req->nv1;
1760 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
1761 tv[1].tv_sec = req->nv2;
1762 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
1763
1764 times = tv;
1765 }
1766 else
1767 times = 0;
1768
1769 req->result = req->type == EIO_FUTIME
1770 ? futimes (req->int1, times)
1771 : utimes (req->ptr1, times);
1772 }
1773 break;
1774
1775 case EIO_GROUP: 2475 case EIO_GROUP:
1776 abort (); /* handled in eio_request */ 2476 abort (); /* handled in eio_request */
1777 2477
1778 case EIO_NOP: 2478 case EIO_NOP:
1779 req->result = 0; 2479 req->result = 0;
1780 break; 2480 break;
1781 2481
1782 case EIO_CUSTOM: 2482 case EIO_CUSTOM:
1783 ((void (*)(eio_req *))req->feed) (req); 2483 req->feed (req);
1784 break; 2484 break;
1785 2485
1786 default: 2486 default:
1787 errno = ENOSYS;
1788 req->result = -1; 2487 req->result = EIO_ENOSYS ();
1789 break; 2488 break;
1790 } 2489 }
1791 2490
1792 req->errorno = errno; 2491 req->errorno = errno;
1793} 2492}
1794 2493
1795#ifndef EIO_NO_WRAPPERS 2494#ifndef EIO_NO_WRAPPERS
1796 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
1797eio_req *eio_nop (int pri, eio_cb cb, void *data) 2506eio_req *eio_nop (int pri, eio_cb cb, void *data)
1798{ 2507{
1799 REQ (EIO_NOP); SEND; 2508 REQ (EIO_NOP); SEND;
1800} 2509}
1801 2510
… …
1817eio_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)
1818{ 2527{
1819 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;
1820} 2529}
1821 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
1822eio_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)
1823{ 2547{
1824 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;
1825} 2549}
1826 2550
… …
1832eio_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)
1833{ 2557{
1834 REQ (EIO_MLOCKALL); req->int1 = flags; SEND; 2558 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
1835} 2559}
1836 2560
1837eio_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)
1838{ 2562{
1839 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;
1840}
1841
1842eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1843{
1844 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1845} 2564}
1846 2565
1847eio_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)
1848{ 2567{
1849 REQ (EIO_CLOSE); req->int1 = fd; SEND; 2568 REQ (EIO_CLOSE); req->int1 = fd; SEND;
… …
1852eio_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)
1853{ 2572{
1854 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;
1855} 2574}
1856 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
1857eio_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)
1858{ 2582{
1859 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;
1860} 2584}
1861 2585
… …
1887eio_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)
1888{ 2612{
1889 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2613 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1890} 2614}
1891 2615
1892eio_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)
1893{ 2617{
1894 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;
1895} 2619}
1896 2620
1897eio_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)
… …
1917eio_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)
1918{ 2642{
1919 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2643 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1920} 2644}
1921 2645
1922eio_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)
1923{ 2647{
1924 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;
1925} 2649}
1926 2650
1927eio_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)
… …
1941} 2665}
1942 2666
1943eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data) 2667eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1944{ 2668{
1945 return eio__1path (EIO_READLINK, path, pri, cb, data); 2669 return eio__1path (EIO_READLINK, path, pri, cb, data);
2670}
2671
2672eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2673{
2674 return eio__1path (EIO_REALPATH, path, pri, cb, data);
1946} 2675}
1947 2676
1948eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) 2677eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1949{ 2678{
1950 return eio__1path (EIO_STAT, path, pri, cb, data); 2679 return eio__1path (EIO_STAT, path, pri, cb, data);
… …
2009eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2738eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2010{ 2739{
2011 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2740 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
2012} 2741}
2013 2742
2014eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) 2743eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
2015{ 2744{
2016 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND; 2745 REQ (EIO_CUSTOM); req->feed = execute; SEND;
2017} 2746}
2018 2747
2019#endif 2748#endif
2020 2749
2021eio_req *eio_grp (eio_cb cb, void *data) 2750eio_req *eio_grp (eio_cb cb, void *data)
… …
2030#undef SEND 2759#undef SEND
2031 2760
2032/*****************************************************************************/ 2761/*****************************************************************************/
2033/* grp functions */ 2762/* grp functions */
2034 2763
2764void
2035void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2765eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
2036{ 2766{
2037 grp->int2 = limit; 2767 grp->int2 = limit;
2038 grp->feed = feed; 2768 grp->feed = feed;
2039 2769
2040 grp_try_feed (grp); 2770 grp_try_feed (grp);
2041} 2771}
2042 2772
2773void
2043void eio_grp_limit (eio_req *grp, int limit) 2774eio_grp_limit (eio_req *grp, int limit)
2044{ 2775{
2045 grp->int2 = limit; 2776 grp->int2 = limit;
2046 2777
2047 grp_try_feed (grp); 2778 grp_try_feed (grp);
2048} 2779}
2049 2780
2781void
2050void eio_grp_add (eio_req *grp, eio_req *req) 2782eio_grp_add (eio_req *grp, eio_req *req)
2051{ 2783{
2052 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2784 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
2053 2785
2054 grp->flags |= EIO_FLAG_GROUPADD; 2786 grp->flags |= EIO_FLAG_GROUPADD;
2055 2787
… …
2066} 2798}
2067 2799
2068/*****************************************************************************/ 2800/*****************************************************************************/
2069/* misc garbage */ 2801/* misc garbage */
2070 2802
2803eio_ssize_t
2071ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2804eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2072{ 2805{
2073 etp_worker wrk;
2074 ssize_t ret;
2075
2076 wrk.dbuf = 0;
2077
2078 ret = eio__sendfile (ofd, ifd, offset, count, &wrk); 2806 return eio__sendfile (ofd, ifd, offset, count);
2079
2080 if (wrk.dbuf)
2081 free (wrk.dbuf);
2082
2083 return ret;
2084} 2807}
2085 2808

Diff Legend

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