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

Comparing libeio/eio.c (file contents):
Revision 1.49 by root, Sat Jan 2 14:24:32 2010 UTC vs.
Revision 1.137 by root, Thu Jun 25 17:40:24 2015 UTC

1/* 1/*
2 * libeio implementation 2 * libeio implementation
3 * 3 *
4 * Copyright (c) 2007,2008,2009,2010 Marc Alexander Lehmann <libeio@schmorp.de> 4 * Copyright (c) 2007,2008,2009,2010,2011,2012,2013 Marc Alexander Lehmann <libeio@schmorp.de>
5 * All rights reserved. 5 * All rights reserved.
6 * 6 *
7 * Redistribution and use in source and binary forms, with or without modifica- 7 * Redistribution and use in source and binary forms, with or without modifica-
8 * tion, are permitted provided that the following conditions are met: 8 * tion, are permitted provided that the following conditions are met:
9 * 9 *
10 * 1. Redistributions of source code must retain the above copyright notice, 10 * 1. Redistributions of source code must retain the above copyright notice,
11 * this list of conditions and the following disclaimer. 11 * this list of conditions and the following disclaimer.
12 * 12 *
13 * 2. Redistributions in binary form must reproduce the above copyright 13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the 14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution. 15 * documentation and/or other materials provided with the distribution.
16 * 16 *
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR IMPLIED 17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR IMPLIED
18 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MER- 18 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MER-
19 * CHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO 19 * CHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO
20 * EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPE- 20 * EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPE-
21 * CIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, 21 * CIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
… …
35 * and other provisions required by the GPL. If you do not delete the 35 * and other provisions required by the GPL. If you do not delete the
36 * provisions above, a recipient may use your version of this file under 36 * provisions above, a recipient may use your version of this file under
37 * either the BSD or the GPL. 37 * either the BSD or the GPL.
38 */ 38 */
39 39
40#ifndef _WIN32
41# include "config.h"
42#endif
43
40#include "eio.h" 44#include "eio.h"
45#include "ecb.h"
41 46
42#ifdef EIO_STACKSIZE 47#ifdef EIO_STACKSIZE
43# define XTHREAD_STACKSIZE EIO_STACKSIZE 48# define X_STACKSIZE EIO_STACKSIZE
44#endif 49#endif
45#include "xthread.h" 50#include "xthread.h"
46 51
47#include <errno.h> 52#include <errno.h>
48#include <stddef.h> 53#include <stddef.h>
… …
53#include <sys/stat.h> 58#include <sys/stat.h>
54#include <limits.h> 59#include <limits.h>
55#include <fcntl.h> 60#include <fcntl.h>
56#include <assert.h> 61#include <assert.h>
57 62
63/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
64/* intptr_t only comes from stdint.h, says idiot openbsd coder */
65#if HAVE_STDINT_H
66# include <stdint.h>
67#endif
68
69#ifndef ECANCELED
70# define ECANCELED EDOM
71#endif
72#ifndef ELOOP
73# define ELOOP EDOM
74#endif
75
76#if !defined(ENOTSOCK) && defined(WSAENOTSOCK)
77# define ENOTSOCK WSAENOTSOCK
78#endif
79
80static void eio_destroy (eio_req *req);
81
58#ifndef EIO_FINISH 82#ifndef EIO_FINISH
59# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0 83# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
60#endif 84#endif
61 85
62#ifndef EIO_DESTROY 86#ifndef EIO_DESTROY
… …
65 89
66#ifndef EIO_FEED 90#ifndef EIO_FEED
67# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0) 91# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)
68#endif 92#endif
69 93
94#ifndef EIO_FD_TO_WIN32_HANDLE
95# define EIO_FD_TO_WIN32_HANDLE(fd) _get_osfhandle (fd)
96#endif
97#ifndef EIO_WIN32_HANDLE_TO_FD
98# define EIO_WIN32_HANDLE_TO_FD(handle) _open_osfhandle (handle, 0)
99#endif
100
101#define EIO_ERRNO(errval,retval) ((errno = errval), retval)
102
103#define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1)
104
70#ifdef _WIN32 105#ifdef _WIN32
71 106
72 /*doh*/ 107 #undef PAGESIZE
108 #define PAGESIZE 4096 /* GetSystemInfo? */
109
110 /* TODO: look at how perl does stat (non-sloppy), unlink (ro-files), utime, link */
111
112 #ifdef EIO_STRUCT_STATI64
113 /* look at perl's non-sloppy stat */
114 #define stat(path,buf) _stati64 (path,buf)
115 #define fstat(fd,buf) _fstati64 (fd,buf)
116 #endif
117 #define lstat(path,buf) stat (path,buf)
118 #define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
119 #define mkdir(path,mode) _mkdir (path)
120 #define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
121
122 #define chmod(path,mode) _chmod (path, mode)
123 #define dup(fd) _dup (fd)
124 #define dup2(fd1,fd2) _dup2 (fd1, fd2)
125 #define pipe(fds) _pipe (fds, 4096, O_BINARY)
126
127 #define fchmod(fd,mode) EIO_ENOSYS ()
128 #define chown(path,uid,gid) EIO_ENOSYS ()
129 #define fchown(fd,uid,gid) EIO_ENOSYS ()
130 #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
131 #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
132 #define mknod(path,mode,dev) EIO_ENOSYS ()
133 #define sync() EIO_ENOSYS ()
134 #define readlink(path,buf,s) EIO_ENOSYS ()
135 #define statvfs(path,buf) EIO_ENOSYS ()
136 #define fstatvfs(fd,buf) EIO_ENOSYS ()
137
138 #define pread(fd,buf,count,offset) eio__pread (fd, buf, count, offset)
139 #define pwrite(fd,buf,count,offset) eio__pwrite (fd, buf, count, offset)
140
141 #if __GNUC__
142 typedef long long eio_off_t; /* signed for compatibility to msvc */
143 #else
144 typedef __int64 eio_off_t; /* unsigned not supported by msvc */
145 #endif
146
147 static eio_ssize_t
148 eio__pread (int fd, void *buf, eio_ssize_t count, eio_off_t offset)
149 {
150 OVERLAPPED o = { 0 };
151 DWORD got;
152
153 o.Offset = offset;
154 o.OffsetHigh = offset >> 32;
155
156 return ReadFile ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd), buf, count, &got, &o)
157 ? got : -1;
158 }
159
160 static eio_ssize_t
161 eio__pwrite (int fd, void *buf, eio_ssize_t count, eio_off_t offset)
162 {
163 OVERLAPPED o = { 0 };
164 DWORD got;
165
166 o.Offset = offset;
167 o.OffsetHigh = offset >> 32;
168
169 return WriteFile ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd), buf, count, &got, &o)
170 ? got : -1;
171 }
172
173 /* rename() uses MoveFile, which fails to overwrite */
174 #define rename(old,neu) eio__rename (old, neu)
175
176 static int
177 eio__rename (const char *old, const char *neu)
178 {
179 if (MoveFileEx (old, neu, MOVEFILE_REPLACE_EXISTING))
180 return 0;
181
182 /* should steal _dosmaperr */
183 switch (GetLastError ())
184 {
185 case ERROR_FILE_NOT_FOUND:
186 case ERROR_PATH_NOT_FOUND:
187 case ERROR_INVALID_DRIVE:
188 case ERROR_NO_MORE_FILES:
189 case ERROR_BAD_NETPATH:
190 case ERROR_BAD_NET_NAME:
191 case ERROR_BAD_PATHNAME:
192 case ERROR_FILENAME_EXCED_RANGE:
193 errno = ENOENT;
194 break;
195
196 default:
197 errno = EACCES;
198 break;
199 }
200
201 return -1;
202 }
203
204 /* we could even stat and see if it exists */
205 static int
206 symlink (const char *old, const char *neu)
207 {
208 #if WINVER >= 0x0600
209 if (CreateSymbolicLink (neu, old, 1))
210 return 0;
211
212 if (CreateSymbolicLink (neu, old, 0))
213 return 0;
214 #endif
215
216 return EIO_ERRNO (ENOENT, -1);
217 }
218
219 /* POSIX API only, causing trouble for win32 apps */
220 #define CreateHardLink(neu,old,flags) 0 /* not really creating hardlink, still using relative paths? */
221 #define CreateSymbolicLink(neu,old,flags) 0 /* vista+ only */
222
223 struct statvfs
224 {
225 int dummy;
226 };
227
228 #define DT_DIR EIO_DT_DIR
229 #define DT_REG EIO_DT_REG
230 #define D_NAME(entp) entp.cFileName
231 #define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
232
73#else 233#else
74 234
75# include "config.h"
76# include <sys/time.h> 235 #include <sys/time.h>
77# include <sys/select.h> 236 #include <sys/select.h>
78# include <sys/mman.h>
79# include <unistd.h> 237 #include <unistd.h>
238 #include <signal.h>
239 #include <dirent.h>
240
241 #ifdef ANDROID
242 #include <sys/vfs.h>
243 #define statvfs statfs
244 #define fstatvfs fstatfs
245 #include <asm/page.h> /* supposedly limits.h does #define PAGESIZE PAGESIZE */
246 #else
247 #include <sys/statvfs.h>
248 #endif
249
250 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
251 #include <sys/mman.h>
252 #endif
253
254 #define D_NAME(entp) entp->d_name
255
256 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
257 #if __FreeBSD__ || __NetBSD__ || __OpenBSD__
258 #define _DIRENT_HAVE_D_TYPE /* sigh */
259 #define D_INO(de) (de)->d_fileno
260 #define D_NAMLEN(de) (de)->d_namlen
261 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
262 #define D_INO(de) (de)->d_ino
263 #endif
264
265 #ifdef _D_EXACT_NAMLEN
266 #undef D_NAMLEN
267 #define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
268 #endif
269
270 #ifdef _DIRENT_HAVE_D_TYPE
271 #define D_TYPE(de) (de)->d_type
272 #endif
273
274 #ifndef EIO_STRUCT_DIRENT
275 #define EIO_STRUCT_DIRENT struct dirent
276 #endif
277
278#endif
279
280#if HAVE_UTIMES
80# include <utime.h> 281# include <utime.h>
81# include <signal.h>
82# include <dirent.h>
83
84/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
85# if __freebsd || defined __NetBSD__ || defined __OpenBSD__
86# define _DIRENT_HAVE_D_TYPE /* sigh */
87# define D_INO(de) (de)->d_fileno
88# define D_NAMLEN(de) (de)->d_namlen
89# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
90# define D_INO(de) (de)->d_ino
91# endif 282#endif
92 283
93#ifdef _D_EXACT_NAMLEN 284#if HAVE_SYS_SYSCALL_H
94# undef D_NAMLEN 285# include <sys/syscall.h>
95# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
96#endif
97
98# ifdef _DIRENT_HAVE_D_TYPE
99# define D_TYPE(de) (de)->d_type
100# endif 286#endif
101 287
102# ifndef EIO_STRUCT_DIRENT 288#if HAVE_SYS_PRCTL_H
103# define EIO_STRUCT_DIRENT struct dirent 289# include <sys/prctl.h>
104# endif
105
106#endif 290#endif
107 291
108#if HAVE_SENDFILE 292#if HAVE_SENDFILE
109# if __linux 293# if __linux
110# include <sys/sendfile.h> 294# include <sys/sendfile.h>
111# elif __freebsd || defined __APPLE__ 295# elif __FreeBSD__ || defined __APPLE__
112# include <sys/socket.h> 296# include <sys/socket.h>
113# include <sys/uio.h> 297# include <sys/uio.h>
114# elif __hpux 298# elif __hpux
115# include <sys/socket.h> 299# include <sys/socket.h>
116# elif __solaris 300# elif __solaris
… …
125#endif 309#endif
126#ifndef D_INO 310#ifndef D_INO
127# define D_INO(de) 0 311# define D_INO(de) 0
128#endif 312#endif
129#ifndef D_NAMLEN 313#ifndef D_NAMLEN
130# define D_NAMLEN(de) strlen ((de)->d_name) 314# define D_NAMLEN(entp) strlen (D_NAME (entp))
131#endif 315#endif
132
133/* number of seconds after which an idle threads exit */
134#define IDLE_TIMEOUT 10
135 316
136/* used for struct dirent, AIX doesn't provide it */ 317/* used for struct dirent, AIX doesn't provide it */
137#ifndef NAME_MAX 318#ifndef NAME_MAX
138# define NAME_MAX 4096 319# define NAME_MAX 4096
139#endif 320#endif
140 321
322/* used for readlink etc. */
323#ifndef PATH_MAX
324# define PATH_MAX 4096
325#endif
326
141/* buffer size for various temporary buffers */ 327/* buffer size for various temporary buffers */
142#define EIO_BUFSIZE 65536 328#define EIO_BUFSIZE 65536
143 329
144#define dBUF \ 330#define dBUF \
145 char *eio_buf; \
146 ETP_WORKER_LOCK (self); \
147 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 331 char *eio_buf = malloc (EIO_BUFSIZE); \
148 ETP_WORKER_UNLOCK (self); \
149 errno = ENOMEM; \ 332 errno = ENOMEM; \
150 if (!eio_buf) \ 333 if (!eio_buf) \
151 return -1; 334 return -1
152 335
153#define EIO_TICKS ((1000000 + 1023) >> 10) 336#define FUBd \
337 free (eio_buf)
154 338
155/*****************************************************************************/ 339/*****************************************************************************/
156 340
157#if __GNUC__ >= 3 341struct etp_tmpbuf;
158# define expect(expr,value) __builtin_expect ((expr),(value)) 342
343#if _POSIX_VERSION >= 200809L
344 #define HAVE_AT 1
345 #define WD2FD(wd) ((wd) ? (wd)->fd : AT_FDCWD)
346 #ifndef O_SEARCH
347 #define O_SEARCH O_RDONLY
348 #endif
159#else 349#else
160# define expect(expr,value) (expr) 350 #define HAVE_AT 0
351 static const char *wd_expand (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path);
161#endif 352#endif
162 353
163#define expect_false(expr) expect ((expr) != 0, 0) 354struct eio_pwd
164#define expect_true(expr) expect ((expr) != 0, 1) 355{
356#if HAVE_AT
357 int fd;
358#endif
359 int len;
360 char str[1]; /* actually, a 0-terminated canonical path */
361};
165 362
166/*****************************************************************************/ 363/*****************************************************************************/
167 364
168#define ETP_PRI_MIN EIO_PRI_MIN 365#define ETP_PRI_MIN EIO_PRI_MIN
169#define ETP_PRI_MAX EIO_PRI_MAX 366#define ETP_PRI_MAX EIO_PRI_MAX
367
368#define ETP_TYPE_QUIT -1
369#define ETP_TYPE_GROUP EIO_GROUP
170 370
171struct etp_worker; 371struct etp_worker;
172 372
173#define ETP_REQ eio_req 373#define ETP_REQ eio_req
174#define ETP_DESTROY(req) eio_destroy (req) 374#define ETP_DESTROY(req) eio_destroy (req)
175static int eio_finish (eio_req *req); 375static int eio_finish (eio_req *req);
176#define ETP_FINISH(req) eio_finish (req) 376#define ETP_FINISH(req) eio_finish (req)
177static void eio_execute (struct etp_worker *self, eio_req *req); 377static void eio_execute (struct etp_worker *self, eio_req *req);
178#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) 378#define ETP_EXECUTE(wrk,req) eio_execute (wrk, req)
179 379
180#define ETP_WORKER_CLEAR(req) \ 380#include "etp.c"
181 if (wrk->dbuf) \
182 { \
183 free (wrk->dbuf); \
184 wrk->dbuf = 0; \
185 } \
186 \
187 if (wrk->dirp) \
188 { \
189 closedir (wrk->dirp); \
190 wrk->dirp = 0; \
191 }
192 381
193#define ETP_WORKER_COMMON \ 382static struct etp_pool eio_pool;
194 void *dbuf; \ 383#define EIO_POOL (&eio_pool)
195 DIR *dirp;
196 384
197/*****************************************************************************/ 385/*****************************************************************************/
198 386
199#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
200
201/* calculate time difference in ~1/EIO_TICKS of a second */
202static int tvdiff (struct timeval *tv1, struct timeval *tv2)
203{
204 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
205 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
206}
207
208static unsigned int started, idle, wanted = 4;
209
210static void (*want_poll_cb) (void);
211static void (*done_poll_cb) (void);
212
213static unsigned int max_poll_time; /* reslock */
214static unsigned int max_poll_reqs; /* reslock */
215
216static volatile unsigned int nreqs; /* reqlock */
217static volatile unsigned int nready; /* reqlock */
218static volatile unsigned int npending; /* reqlock */
219static volatile unsigned int max_idle = 4;
220
221static mutex_t wrklock = X_MUTEX_INIT;
222static mutex_t reslock = X_MUTEX_INIT;
223static mutex_t reqlock = X_MUTEX_INIT;
224static cond_t reqwait = X_COND_INIT;
225
226#if !HAVE_PREADWRITE
227/*
228 * make our pread/pwrite emulation safe against themselves, but not against
229 * normal read/write by using a mutex. slows down execution a lot,
230 * but that's your problem, not mine.
231 */
232static mutex_t preadwritelock = X_MUTEX_INIT;
233#endif
234
235typedef struct etp_worker
236{
237 /* locked by wrklock */
238 struct etp_worker *prev, *next;
239
240 thread_t tid;
241
242 /* locked by reslock, reqlock or wrklock */
243 ETP_REQ *req; /* currently processed request */
244
245 ETP_WORKER_COMMON
246} etp_worker;
247
248static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */
249
250#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
251#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
252
253/* worker threads management */
254
255static void etp_worker_clear (etp_worker *wrk)
256{
257 ETP_WORKER_CLEAR (wrk);
258}
259
260static void etp_worker_free (etp_worker *wrk)
261{
262 wrk->next->prev = wrk->prev;
263 wrk->prev->next = wrk->next;
264
265 free (wrk);
266}
267
268static unsigned int etp_nreqs (void)
269{
270 int retval;
271 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
272 retval = nreqs;
273 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
274 return retval;
275}
276
277static unsigned int etp_nready (void)
278{
279 unsigned int retval;
280
281 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
282 retval = nready;
283 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
284
285 return retval;
286}
287
288static unsigned int etp_npending (void)
289{
290 unsigned int retval;
291
292 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
293 retval = npending;
294 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
295
296 return retval;
297}
298
299static unsigned int etp_nthreads (void)
300{
301 unsigned int retval;
302
303 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
304 retval = started;
305 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
306
307 return retval;
308}
309
310/*
311 * a somewhat faster data structure might be nice, but
312 * with 8 priorities this actually needs <20 insns
313 * per shift, the most expensive operation.
314 */
315typedef struct {
316 ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */
317 int size;
318} etp_reqq;
319
320static etp_reqq req_queue;
321static etp_reqq res_queue;
322
323static int reqq_push (etp_reqq *q, ETP_REQ *req)
324{
325 int pri = req->pri;
326 req->next = 0;
327
328 if (q->qe[pri])
329 {
330 q->qe[pri]->next = req;
331 q->qe[pri] = req;
332 }
333 else
334 q->qe[pri] = q->qs[pri] = req;
335
336 return q->size++;
337}
338
339static ETP_REQ *reqq_shift (etp_reqq *q)
340{
341 int pri;
342
343 if (!q->size)
344 return 0;
345
346 --q->size;
347
348 for (pri = ETP_NUM_PRI; pri--; )
349 {
350 eio_req *req = q->qs[pri];
351
352 if (req)
353 {
354 if (!(q->qs[pri] = (eio_req *)req->next))
355 q->qe[pri] = 0;
356
357 return req;
358 }
359 }
360
361 abort ();
362}
363
364static void etp_atfork_prepare (void)
365{
366 X_LOCK (wrklock);
367 X_LOCK (reqlock);
368 X_LOCK (reslock);
369#if !HAVE_PREADWRITE
370 X_LOCK (preadwritelock);
371#endif
372}
373
374static void etp_atfork_parent (void)
375{
376#if !HAVE_PREADWRITE
377 X_UNLOCK (preadwritelock);
378#endif
379 X_UNLOCK (reslock);
380 X_UNLOCK (reqlock);
381 X_UNLOCK (wrklock);
382}
383
384static void etp_atfork_child (void)
385{
386 ETP_REQ *prv;
387
388 while ((prv = reqq_shift (&req_queue)))
389 ETP_DESTROY (prv);
390
391 while ((prv = reqq_shift (&res_queue)))
392 ETP_DESTROY (prv);
393
394 while (wrk_first.next != &wrk_first)
395 {
396 etp_worker *wrk = wrk_first.next;
397
398 if (wrk->req)
399 ETP_DESTROY (wrk->req);
400
401 etp_worker_clear (wrk);
402 etp_worker_free (wrk);
403 }
404
405 started = 0;
406 idle = 0;
407 nreqs = 0;
408 nready = 0;
409 npending = 0;
410
411 etp_atfork_parent ();
412}
413
414static void 387static void
415etp_once_init (void)
416{
417 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
418}
419
420static int
421etp_init (void (*want_poll)(void), void (*done_poll)(void))
422{
423 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
424
425 pthread_once (&doinit, etp_once_init);
426
427 want_poll_cb = want_poll;
428 done_poll_cb = done_poll;
429
430 return 0;
431}
432
433X_THREAD_PROC (etp_proc);
434
435static void etp_start_thread (void)
436{
437 etp_worker *wrk = calloc (1, sizeof (etp_worker));
438
439 /*TODO*/
440 assert (("unable to allocate worker thread data", wrk));
441
442 X_LOCK (wrklock);
443
444 if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
445 {
446 wrk->prev = &wrk_first;
447 wrk->next = wrk_first.next;
448 wrk_first.next->prev = wrk;
449 wrk_first.next = wrk;
450 ++started;
451 }
452 else
453 free (wrk);
454
455 X_UNLOCK (wrklock);
456}
457
458static void etp_maybe_start_thread (void)
459{
460 if (expect_true (etp_nthreads () >= wanted))
461 return;
462
463 /* todo: maybe use idle here, but might be less exact */
464 if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
465 return;
466
467 etp_start_thread ();
468}
469
470static void etp_end_thread (void)
471{
472 eio_req *req = calloc (1, sizeof (eio_req));
473
474 req->type = -1;
475 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
476
477 X_LOCK (reqlock);
478 reqq_push (&req_queue, req);
479 X_COND_SIGNAL (reqwait);
480 X_UNLOCK (reqlock);
481
482 X_LOCK (wrklock);
483 --started;
484 X_UNLOCK (wrklock);
485}
486
487static int etp_poll (void)
488{
489 unsigned int maxreqs;
490 unsigned int maxtime;
491 struct timeval tv_start, tv_now;
492
493 X_LOCK (reslock);
494 maxreqs = max_poll_reqs;
495 maxtime = max_poll_time;
496 X_UNLOCK (reslock);
497
498 if (maxtime)
499 gettimeofday (&tv_start, 0);
500
501 for (;;)
502 {
503 ETP_REQ *req;
504
505 etp_maybe_start_thread ();
506
507 X_LOCK (reslock);
508 req = reqq_shift (&res_queue);
509
510 if (req)
511 {
512 --npending;
513
514 if (!res_queue.size && done_poll_cb)
515 done_poll_cb ();
516 }
517
518 X_UNLOCK (reslock);
519
520 if (!req)
521 return 0;
522
523 X_LOCK (reqlock);
524 --nreqs;
525 X_UNLOCK (reqlock);
526
527 if (expect_false (req->type == EIO_GROUP && req->size))
528 {
529 req->int1 = 1; /* mark request as delayed */
530 continue;
531 }
532 else
533 {
534 int res = ETP_FINISH (req);
535 if (expect_false (res))
536 return res;
537 }
538
539 if (expect_false (maxreqs && !--maxreqs))
540 break;
541
542 if (maxtime)
543 {
544 gettimeofday (&tv_now, 0);
545
546 if (tvdiff (&tv_start, &tv_now) >= maxtime)
547 break;
548 }
549 }
550
551 errno = EAGAIN;
552 return -1;
553}
554
555static void etp_cancel (ETP_REQ *req)
556{
557 X_LOCK (wrklock);
558 req->flags |= EIO_FLAG_CANCELLED;
559 X_UNLOCK (wrklock);
560
561 eio_grp_cancel (req);
562}
563
564static void etp_submit (ETP_REQ *req)
565{
566 req->pri -= ETP_PRI_MIN;
567
568 if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
569 if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
570
571 if (expect_false (req->type == EIO_GROUP))
572 {
573 /* I hope this is worth it :/ */
574 X_LOCK (reqlock);
575 ++nreqs;
576 X_UNLOCK (reqlock);
577
578 X_LOCK (reslock);
579
580 ++npending;
581
582 if (!reqq_push (&res_queue, req) && want_poll_cb)
583 want_poll_cb ();
584
585 X_UNLOCK (reslock);
586 }
587 else
588 {
589 X_LOCK (reqlock);
590 ++nreqs;
591 ++nready;
592 reqq_push (&req_queue, req);
593 X_COND_SIGNAL (reqwait);
594 X_UNLOCK (reqlock);
595
596 etp_maybe_start_thread ();
597 }
598}
599
600static void etp_set_max_poll_time (double nseconds)
601{
602 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
603 max_poll_time = nseconds * EIO_TICKS;
604 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
605}
606
607static void etp_set_max_poll_reqs (unsigned int maxreqs)
608{
609 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
610 max_poll_reqs = maxreqs;
611 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
612}
613
614static void etp_set_max_idle (unsigned int nthreads)
615{
616 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
617 max_idle = nthreads <= 0 ? 1 : nthreads;
618 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
619}
620
621static void etp_set_min_parallel (unsigned int nthreads)
622{
623 if (wanted < nthreads)
624 wanted = nthreads;
625}
626
627static void etp_set_max_parallel (unsigned int nthreads)
628{
629 if (wanted > nthreads)
630 wanted = nthreads;
631
632 while (started > wanted)
633 etp_end_thread ();
634}
635
636/*****************************************************************************/
637
638static void grp_try_feed (eio_req *grp) 388grp_try_feed (eio_req *grp)
639{ 389{
640 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 390 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
641 { 391 {
642 grp->flags &= ~EIO_FLAG_GROUPADD; 392 grp->flags &= ~ETP_FLAG_GROUPADD;
643 393
644 EIO_FEED (grp); 394 EIO_FEED (grp);
645 395
646 /* stop if no progress has been made */ 396 /* stop if no progress has been made */
647 if (!(grp->flags & EIO_FLAG_GROUPADD)) 397 if (!(grp->flags & ETP_FLAG_GROUPADD))
648 { 398 {
649 grp->feed = 0; 399 grp->feed = 0;
650 break; 400 break;
651 } 401 }
652 } 402 }
653} 403}
654 404
405static int
655static int grp_dec (eio_req *grp) 406grp_dec (eio_req *grp)
656{ 407{
657 --grp->size; 408 --grp->size;
658 409
659 /* call feeder, if applicable */ 410 /* call feeder, if applicable */
660 grp_try_feed (grp); 411 grp_try_feed (grp);
661 412
662 /* finish, if done */ 413 /* finish, if done */
663 if (!grp->size && grp->int1) 414 if (!grp->size && grp->flags & ETP_FLAG_DELAYED)
664 return eio_finish (grp); 415 return eio_finish (grp);
665 else 416 else
666 return 0; 417 return 0;
667} 418}
668 419
420static void
669void eio_destroy (eio_req *req) 421eio_destroy (eio_req *req)
670{ 422{
671 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); 423 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
672 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); 424 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
673 425
674 EIO_DESTROY (req); 426 EIO_DESTROY (req);
675} 427}
676 428
429static int
677static int eio_finish (eio_req *req) 430eio_finish (eio_req *req)
678{ 431{
679 int res = EIO_FINISH (req); 432 int res = EIO_FINISH (req);
680 433
681 if (req->grp) 434 if (req->grp)
682 { 435 {
… …
690 if (grp->grp_first == req) 443 if (grp->grp_first == req)
691 grp->grp_first = req->grp_next; 444 grp->grp_first = req->grp_next;
692 445
693 res2 = grp_dec (grp); 446 res2 = grp_dec (grp);
694 447
695 if (!res && res2) 448 if (!res)
696 res = res2; 449 res = res2;
697 } 450 }
698 451
699 eio_destroy (req); 452 eio_destroy (req);
700 453
701 return res; 454 return res;
702} 455}
703 456
457void
704void eio_grp_cancel (eio_req *grp) 458eio_grp_cancel (eio_req *grp)
705{ 459{
706 for (grp = grp->grp_first; grp; grp = grp->grp_next) 460 etp_grp_cancel (EIO_POOL, grp);
707 eio_cancel (grp);
708} 461}
709 462
463void
710void eio_cancel (eio_req *req) 464eio_cancel (eio_req *req)
711{ 465{
712 etp_cancel (req); 466 etp_cancel (EIO_POOL, req);
713} 467}
714 468
469void
715void eio_submit (eio_req *req) 470eio_submit (eio_req *req)
716{ 471{
717 etp_submit (req); 472 etp_submit (EIO_POOL, req);
718} 473}
719 474
720unsigned int eio_nreqs (void) 475unsigned int
476eio_nreqs (void)
721{ 477{
722 return etp_nreqs (); 478 return etp_nreqs (EIO_POOL);
723} 479}
724 480
725unsigned int eio_nready (void) 481unsigned int
482eio_nready (void)
726{ 483{
727 return etp_nready (); 484 return etp_nready (EIO_POOL);
728} 485}
729 486
730unsigned int eio_npending (void) 487unsigned int
488eio_npending (void)
731{ 489{
732 return etp_npending (); 490 return etp_npending (EIO_POOL);
733} 491}
734 492
735unsigned int eio_nthreads (void) 493unsigned int ecb_cold
494eio_nthreads (void)
736{ 495{
737 return etp_nthreads (); 496 return etp_nthreads (EIO_POOL);
738} 497}
739 498
499void ecb_cold
740void eio_set_max_poll_time (double nseconds) 500eio_set_max_poll_time (double nseconds)
741{ 501{
742 etp_set_max_poll_time (nseconds); 502 etp_set_max_poll_time (EIO_POOL, nseconds);
743} 503}
744 504
505void ecb_cold
745void eio_set_max_poll_reqs (unsigned int maxreqs) 506eio_set_max_poll_reqs (unsigned int maxreqs)
746{ 507{
747 etp_set_max_poll_reqs (maxreqs); 508 etp_set_max_poll_reqs (EIO_POOL, maxreqs);
748} 509}
749 510
511void ecb_cold
750void eio_set_max_idle (unsigned int nthreads) 512eio_set_max_idle (unsigned int nthreads)
751{ 513{
752 etp_set_max_idle (nthreads); 514 etp_set_max_idle (EIO_POOL, nthreads);
753} 515}
754 516
517void ecb_cold
518eio_set_idle_timeout (unsigned int seconds)
519{
520 etp_set_idle_timeout (EIO_POOL, seconds);
521}
522
523void ecb_cold
755void eio_set_min_parallel (unsigned int nthreads) 524eio_set_min_parallel (unsigned int nthreads)
756{ 525{
757 etp_set_min_parallel (nthreads); 526 etp_set_min_parallel (EIO_POOL, nthreads);
758} 527}
759 528
529void ecb_cold
760void eio_set_max_parallel (unsigned int nthreads) 530eio_set_max_parallel (unsigned int nthreads)
761{ 531{
762 etp_set_max_parallel (nthreads); 532 etp_set_max_parallel (EIO_POOL, nthreads);
763} 533}
764 534
765int eio_poll (void) 535int eio_poll (void)
766{ 536{
767 return etp_poll (); 537 return etp_poll (EIO_POOL);
768} 538}
769 539
770/*****************************************************************************/ 540/*****************************************************************************/
771/* work around various missing functions */ 541/* work around various missing functions */
772 542
773#if !HAVE_PREADWRITE
774# undef pread
775# undef pwrite
776# define pread eio__pread
777# define pwrite eio__pwrite
778
779static ssize_t
780eio__pread (int fd, void *buf, size_t count, off_t offset)
781{
782 ssize_t res;
783 off_t ooffset;
784
785 X_LOCK (preadwritelock);
786 ooffset = lseek (fd, 0, SEEK_CUR);
787 lseek (fd, offset, SEEK_SET);
788 res = read (fd, buf, count);
789 lseek (fd, ooffset, SEEK_SET);
790 X_UNLOCK (preadwritelock);
791
792 return res;
793}
794
795static ssize_t
796eio__pwrite (int fd, void *buf, size_t count, off_t offset)
797{
798 ssize_t res;
799 off_t ooffset;
800
801 X_LOCK (preadwritelock);
802 ooffset = lseek (fd, 0, SEEK_CUR);
803 lseek (fd, offset, SEEK_SET);
804 res = write (fd, buf, count);
805 lseek (fd, ooffset, SEEK_SET);
806 X_UNLOCK (preadwritelock);
807
808 return res;
809}
810#endif
811
812#ifndef HAVE_FUTIMES 543#ifndef HAVE_UTIMES
813 544
814# undef utimes 545# undef utimes
815# undef futimes
816# define utimes(path,times) eio__utimes (path, times) 546# define utimes(path,times) eio__utimes (path, times)
817# define futimes(fd,times) eio__futimes (fd, times)
818 547
819static int 548static int
820eio__utimes (const char *filename, const struct timeval times[2]) 549eio__utimes (const char *filename, const struct timeval times[2])
821{ 550{
822 if (times) 551 if (times)
… …
830 } 559 }
831 else 560 else
832 return utime (filename, 0); 561 return utime (filename, 0);
833} 562}
834 563
564#endif
565
566#ifndef HAVE_FUTIMES
567
568# undef futimes
569# define futimes(fd,times) eio__futimes (fd, times)
570
571static int
835static int eio__futimes (int fd, const struct timeval tv[2]) 572eio__futimes (int fd, const struct timeval tv[2])
836{ 573{
837 errno = ENOSYS; 574 errno = ENOSYS;
838 return -1; 575 return -1;
839} 576}
840 577
… …
843#if !HAVE_FDATASYNC 580#if !HAVE_FDATASYNC
844# undef fdatasync 581# undef fdatasync
845# define fdatasync(fd) fsync (fd) 582# define fdatasync(fd) fsync (fd)
846#endif 583#endif
847 584
585static int
586eio__syncfs (int fd)
587{
588 int res;
589
590#if HAVE_SYS_SYNCFS
591 res = (int)syscall (__NR_syncfs, (int)(fd));
592#else
593 res = EIO_ENOSYS ();
594#endif
595
596 if (res < 0 && errno == ENOSYS && fd >= 0)
597 sync ();
598
599 return res;
600}
601
848/* sync_file_range always needs emulation */ 602/* sync_file_range always needs emulation */
849int 603static int
850eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags) 604eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
851{ 605{
852#if HAVE_SYNC_FILE_RANGE 606#if HAVE_SYNC_FILE_RANGE
853 int res; 607 int res;
854 608
… …
871 /* even though we could play tricks with the flags, it's better to always 625 /* even though we could play tricks with the flags, it's better to always
872 * call fdatasync, as that matches the expectation of its users best */ 626 * call fdatasync, as that matches the expectation of its users best */
873 return fdatasync (fd); 627 return fdatasync (fd);
874} 628}
875 629
630static int
631eio__fallocate (int fd, int mode, off_t offset, size_t len)
632{
633#if HAVE_LINUX_FALLOCATE
634 return fallocate (fd, mode, offset, len);
635#else
636 return EIO_ENOSYS ();
637#endif
638}
639
876#if !HAVE_READAHEAD 640#if !HAVE_READAHEAD
877# undef readahead 641# undef readahead
878# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 642# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
879 643
880static ssize_t 644static eio_ssize_t
881eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) 645eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
882{ 646{
883 size_t todo = count; 647 size_t todo = count;
884 dBUF; 648 dBUF;
885 649
… …
890 pread (fd, eio_buf, len, offset); 654 pread (fd, eio_buf, len, offset);
891 offset += len; 655 offset += len;
892 todo -= len; 656 todo -= len;
893 } 657 }
894 658
895 errno = 0; 659 FUBd;
660
661 /* linux's readahead basically only fails for EBADF or EINVAL (not mmappable) */
662 /* but not for e.g. EIO or eof, so we also never fail */
896 return count; 663 return 0;
897} 664}
898 665
899#endif 666#endif
900 667
901/* sendfile always needs emulation */ 668/* sendfile always needs emulation */
902static ssize_t 669static eio_ssize_t
903eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) 670eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
904{ 671{
672 eio_ssize_t written = 0;
905 ssize_t res; 673 eio_ssize_t res;
906 674
907 if (!count) 675 if (!count)
908 return 0; 676 return 0;
909 677
678 for (;;)
679 {
680#ifdef __APPLE__
681# undef HAVE_SENDFILE /* broken, as everything on os x */
682#endif
910#if HAVE_SENDFILE 683#if HAVE_SENDFILE
911# if __linux 684# if __linux
685 off_t soffset = offset;
912 res = sendfile (ofd, ifd, &offset, count); 686 res = sendfile (ofd, ifd, &soffset, count);
913 687
914# elif __freebsd 688# elif __FreeBSD__
915 /* 689 /*
916 * Of course, the freebsd sendfile is a dire hack with no thoughts 690 * Of course, the freebsd sendfile is a dire hack with no thoughts
917 * wasted on making it similar to other I/O functions. 691 * wasted on making it similar to other I/O functions.
918 */ 692 */
919 {
920 off_t sbytes; 693 off_t sbytes;
921 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 694 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
922 695
923 if (res < 0 && sbytes) 696 #if 0 /* according to the manpage, this is correct, but broken behaviour */
924 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 697 /* freebsd' sendfile will return 0 on success */
698 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
699 /* not on e.g. EIO or EPIPE - sounds broken */
700 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
925 res = sbytes; 701 res = sbytes;
926 } 702 #endif
927 703
704 /* according to source inspection, this is correct, and useful behaviour */
705 if (sbytes)
706 res = sbytes;
707
928# elif defined (__APPLE__) 708# elif defined __APPLE__
929
930 {
931 off_t sbytes = count; 709 off_t sbytes = count;
932 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0); 710 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
933 711
934 if (res < 0 && errno == EAGAIN && sbytes) 712 /* according to the manpage, sbytes is always valid */
713 if (sbytes)
935 res = sbytes; 714 res = sbytes;
936 }
937 715
938# elif __hpux 716# elif __hpux
939 res = sendfile (ofd, ifd, offset, count, 0, 0); 717 res = sendfile (ofd, ifd, offset, count, 0, 0);
940 718
941# elif __solaris 719# elif __solaris
942 {
943 struct sendfilevec vec; 720 struct sendfilevec vec;
944 size_t sbytes; 721 size_t sbytes;
945 722
946 vec.sfv_fd = ifd; 723 vec.sfv_fd = ifd;
947 vec.sfv_flag = 0; 724 vec.sfv_flag = 0;
948 vec.sfv_off = offset; 725 vec.sfv_off = offset;
949 vec.sfv_len = count; 726 vec.sfv_len = count;
950 727
951 res = sendfilev (ofd, &vec, 1, &sbytes); 728 res = sendfilev (ofd, &vec, 1, &sbytes);
952 729
953 if (res < 0 && sbytes) 730 if (res < 0 && sbytes)
954 res = sbytes; 731 res = sbytes;
955 }
956 732
957# endif 733# endif
958 734
959#elif defined (_WIN32) 735#elif defined (_WIN32) && 0
960
961 /* does not work, just for documentation of what would need to be done */ 736 /* does not work, just for documentation of what would need to be done */
962 { 737 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
738 /* libeio guarantees that the file offset does not change, and windows */
739 /* has no way to get an independent handle to the same file description */
963 HANDLE h = TO_SOCKET (ifd); 740 HANDLE h = TO_SOCKET (ifd);
964 SetFilePointer (h, offset, 0, FILE_BEGIN); 741 SetFilePointer (h, offset, 0, FILE_BEGIN);
965 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0); 742 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
966 }
967 743
968#else 744#else
969 res = -1; 745 res = EIO_ENOSYS ();
970 errno = ENOSYS;
971#endif 746#endif
972 747
748 /* we assume sendfile can copy at least 128mb in one go */
749 if (res <= 128 * 1024 * 1024)
750 {
751 if (res > 0)
752 written += res;
753
754 if (written)
755 return written;
756
757 break;
758 }
759 else
760 {
761 /* if we requested more, then probably the kernel was lazy */
762 written += res;
763 offset += res;
764 count -= res;
765
766 if (!count)
767 return written;
768 }
769 }
770
973 if (res < 0 771 if (res < 0
974 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 772 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
975 /* BSDs */ 773 /* BSDs */
976#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */ 774#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
977 || errno == ENOTSUP 775 || errno == ENOTSUP
978#endif 776#endif
777#ifdef EOPNOTSUPP /* windows */
979 || errno == EOPNOTSUPP /* BSDs */ 778 || errno == EOPNOTSUPP /* BSDs */
779#endif
980#if __solaris 780#if __solaris
981 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 781 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
982#endif 782#endif
983 ) 783 )
984 ) 784 )
… …
988 788
989 res = 0; 789 res = 0;
990 790
991 while (count) 791 while (count)
992 { 792 {
993 ssize_t cnt; 793 eio_ssize_t cnt;
994 794
995 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 795 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
996 796
997 if (cnt <= 0) 797 if (cnt <= 0)
998 { 798 {
… …
1010 810
1011 offset += cnt; 811 offset += cnt;
1012 res += cnt; 812 res += cnt;
1013 count -= cnt; 813 count -= cnt;
1014 } 814 }
815
816 FUBd;
1015 } 817 }
1016 818
1017 return res; 819 return res;
820}
821
822#ifdef PAGESIZE
823# define eio_pagesize() PAGESIZE
824#else
825static intptr_t
826eio_pagesize (void)
827{
828 static intptr_t page;
829
830 if (!page)
831 page = sysconf (_SC_PAGESIZE);
832
833 return page;
834}
835#endif
836
837static void
838eio_page_align (void **addr, size_t *length)
839{
840 intptr_t mask = eio_pagesize () - 1;
841
842 /* round down addr */
843 intptr_t adj = mask & (intptr_t)*addr;
844
845 *addr = (void *)((intptr_t)*addr - adj);
846 *length += adj;
847
848 /* round up length */
849 *length = (*length + mask) & ~mask;
850}
851
852#if !_POSIX_MEMLOCK
853# define eio__mlockall(a) EIO_ENOSYS ()
854#else
855
856static int
857eio__mlockall (int flags)
858{
859 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
860 extern int mallopt (int, int);
861 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
862 #endif
863
864 if (EIO_MCL_CURRENT != MCL_CURRENT
865 || EIO_MCL_FUTURE != MCL_FUTURE)
866 {
867 flags = 0
868 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
869 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
870 }
871
872 return mlockall (flags);
873}
874#endif
875
876#if !_POSIX_MEMLOCK_RANGE
877# define eio__mlock(a,b) EIO_ENOSYS ()
878#else
879
880static int
881eio__mlock (void *addr, size_t length)
882{
883 eio_page_align (&addr, &length);
884
885 return mlock (addr, length);
886}
887
888#endif
889
890#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
891# define eio__msync(a,b,c) EIO_ENOSYS ()
892#else
893
894static int
895eio__msync (void *mem, size_t len, int flags)
896{
897 eio_page_align (&mem, &len);
898
899 if (EIO_MS_ASYNC != MS_SYNC
900 || EIO_MS_INVALIDATE != MS_INVALIDATE
901 || EIO_MS_SYNC != MS_SYNC)
902 {
903 flags = 0
904 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
905 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
906 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
907 }
908
909 return msync (mem, len, flags);
910}
911
912#endif
913
914static int
915eio__mtouch (eio_req *req)
916{
917 void *mem = req->ptr2;
918 size_t len = req->size;
919 int flags = req->int1;
920
921 eio_page_align (&mem, &len);
922
923 {
924 intptr_t addr = (intptr_t)mem;
925 intptr_t end = addr + len;
926 intptr_t page = eio_pagesize ();
927
928 if (addr < end)
929 if (flags & EIO_MT_MODIFY) /* modify */
930 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
931 else
932 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
933 }
934
935 return 0;
936}
937
938/*****************************************************************************/
939/* requests implemented outside eio_execute, because they are so large */
940
941static void
942eio__lseek (eio_req *req)
943{
944 /* this usually gets optimised away completely, or your compiler sucks, */
945 /* or the whence constants really are not 0, 1, 2 */
946 int whence = req->int2 == EIO_SEEK_SET ? SEEK_SET
947 : req->int2 == EIO_SEEK_CUR ? SEEK_CUR
948 : req->int2 == EIO_SEEK_END ? SEEK_END
949 : req->int2;
950
951 req->offs = lseek (req->int1, req->offs, whence);
952 req->result = req->offs == (off_t)-1 ? -1 : 0;
953}
954
955/* result will always end up in tmpbuf, there is always space for adding a 0-byte */
956static int
957eio__realpath (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
958{
959 char *res;
960 const char *rel = path;
961 char *tmp1, *tmp2;
962#if SYMLOOP_MAX > 32
963 int symlinks = SYMLOOP_MAX;
964#else
965 int symlinks = 32;
966#endif
967
968 errno = EINVAL;
969 if (!rel)
970 return -1;
971
972 errno = ENOENT;
973 if (!*rel)
974 return -1;
975
976 res = etp_tmpbuf_get (tmpbuf, PATH_MAX * 3);
977#ifdef _WIN32
978 if (_access (rel, 4) != 0)
979 return -1;
980
981 symlinks = GetFullPathName (rel, PATH_MAX * 3, res, 0);
982
983 errno = ENAMETOOLONG;
984 if (symlinks >= PATH_MAX * 3)
985 return -1;
986
987 errno = EIO;
988 if (symlinks <= 0)
989 return -1;
990
991 return symlinks;
992
993#else
994 tmp1 = res + PATH_MAX;
995 tmp2 = tmp1 + PATH_MAX;
996
997#if 0 /* disabled, the musl way to do things is just too racy */
998#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
999 /* on linux we may be able to ask the kernel */
1000 {
1001 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1002
1003 if (fd >= 0)
1004 {
1005 sprintf (tmp1, "/proc/self/fd/%d", fd);
1006 req->result = readlink (tmp1, res, PATH_MAX);
1007 /* here we should probably stat the open file and the disk file, to make sure they still match */
1008 close (fd);
1009
1010 if (req->result > 0)
1011 goto done;
1012 }
1013 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1014 return -1;
1015 }
1016#endif
1017#endif
1018
1019 if (*rel != '/')
1020 {
1021 int len;
1022
1023 errno = ENOENT;
1024 if (wd == EIO_INVALID_WD)
1025 return -1;
1026
1027 if (wd == EIO_CWD)
1028 {
1029 if (!getcwd (res, PATH_MAX))
1030 return -1;
1031
1032 len = strlen (res);
1033 }
1034 else
1035 memcpy (res, wd->str, len = wd->len);
1036
1037 if (res [1]) /* only use if not / */
1038 res += len;
1039 }
1040
1041 while (*rel)
1042 {
1043 eio_ssize_t len, linklen;
1044 const char *beg = rel;
1045
1046 while (*rel && *rel != '/')
1047 ++rel;
1048
1049 len = rel - beg;
1050
1051 if (!len) /* skip slashes */
1052 {
1053 ++rel;
1054 continue;
1055 }
1056
1057 if (beg [0] == '.')
1058 {
1059 if (len == 1)
1060 continue; /* . - nop */
1061
1062 if (beg [1] == '.' && len == 2)
1063 {
1064 /* .. - back up one component, if possible */
1065
1066 while (res != tmpbuf->ptr)
1067 if (*--res == '/')
1068 break;
1069
1070 continue;
1071 }
1072 }
1073
1074 errno = ENAMETOOLONG;
1075 if (res + 1 + len + 1 >= tmp1)
1076 return -1;
1077
1078 /* copy one component */
1079 *res = '/';
1080 memcpy (res + 1, beg, len);
1081
1082 /* zero-terminate, for readlink */
1083 res [len + 1] = 0;
1084
1085 /* now check if it's a symlink */
1086 linklen = readlink (tmpbuf->ptr, tmp1, PATH_MAX);
1087
1088 if (linklen < 0)
1089 {
1090 if (errno != EINVAL)
1091 return -1;
1092
1093 /* it's a normal directory. hopefully */
1094 res += len + 1;
1095 }
1096 else
1097 {
1098 /* yay, it was a symlink - build new path in tmp2 */
1099 int rellen = strlen (rel);
1100
1101 errno = ENAMETOOLONG;
1102 if (linklen + 1 + rellen >= PATH_MAX)
1103 return -1;
1104
1105 errno = ELOOP;
1106 if (!--symlinks)
1107 return -1;
1108
1109 if (*tmp1 == '/')
1110 res = tmpbuf->ptr; /* symlink resolves to an absolute path */
1111
1112 /* we need to be careful, as rel might point into tmp2 already */
1113 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1114 tmp2 [linklen] = '/';
1115 memcpy (tmp2, tmp1, linklen);
1116
1117 rel = tmp2;
1118 }
1119 }
1120
1121 /* special case for the lone root path */
1122 if (res == tmpbuf->ptr)
1123 *res++ = '/';
1124
1125 return res - (char *)tmpbuf->ptr;
1126#endif
1018} 1127}
1019 1128
1020static signed char 1129static signed char
1021eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) 1130eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1022{ 1131{
1023 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */ 1132 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1024 : a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0; 1133 : a->inode < b->inode ? -1
1134 : a->inode > b->inode ? 1
1135 : 0;
1025} 1136}
1026 1137
1027#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0 1138#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1028 1139
1029#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */ 1140#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1030#define EIO_SORT_FAST 60 /* when to only use insertion sort */ 1141#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1031 1142
1032static void 1143static void
1033eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1144eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1034{ 1145{
1035 unsigned char bits [9 + sizeof (ino_t) * 8]; 1146 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1036 unsigned char *bit = bits; 1147 unsigned char *bit = bits;
1037 1148
1038 assert (CHAR_BIT == 8); 1149 assert (CHAR_BIT == 8);
1039 assert (sizeof (eio_dirent) * 8 < 256); 1150 assert (sizeof (eio_dirent) * 8 < 256);
1040 assert (offsetof (eio_dirent, inode)); /* we use 0 as sentinel */ 1151 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1041 assert (offsetof (eio_dirent, score)); /* we use 0 as sentinel */ 1152 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1042 1153
1043 if (size <= EIO_SORT_FAST) 1154 if (size <= EIO_SORT_FAST)
1044 return; 1155 return;
1045 1156
1046 /* first prepare an array of bits to test in our radix sort */ 1157 /* first prepare an array of bits to test in our radix sort */
1047 /* try to take endianness into account, as well as differences in ino_t sizes */ 1158 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1048 /* inode_bits must contain all inodes ORed together */ 1159 /* inode_bits must contain all inodes ORed together */
1049 /* which is used to skip bits that are 0 everywhere, which is very common */ 1160 /* which is used to skip bits that are 0 everywhere, which is very common */
1050 { 1161 {
1051 ino_t endianness; 1162 eio_ino_t endianness;
1052 int i, j; 1163 int i, j;
1053 1164
1054 /* we store the byte offset of byte n into byte n of "endianness" */ 1165 /* we store the byte offset of byte n into byte n of "endianness" */
1055 for (i = 0; i < sizeof (ino_t); ++i) 1166 for (i = 0; i < sizeof (eio_ino_t); ++i)
1056 ((unsigned char *)&endianness)[i] = i; 1167 ((unsigned char *)&endianness)[i] = i;
1057 1168
1058 *bit++ = 0; 1169 *bit++ = 0;
1059 1170
1060 for (i = 0; i < sizeof (ino_t); ++i) 1171 for (i = 0; i < sizeof (eio_ino_t); ++i)
1061 { 1172 {
1062 /* shifting off the byte offsets out of "endianness" */ 1173 /* shifting off the byte offsets out of "endianness" */
1063 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8; 1174 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1064 endianness >>= 8; 1175 endianness >>= 8;
1065 1176
1066 for (j = 0; j < 8; ++j) 1177 for (j = 0; j < 8; ++j)
1067 if (inode_bits & (((ino_t)1) << (i * 8 + j))) 1178 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1068 *bit++ = offs + j; 1179 *bit++ = offs + j;
1069 } 1180 }
1070 1181
1071 for (j = 0; j < 8; ++j) 1182 for (j = 0; j < 8; ++j)
1072 if (score_bits & (1 << j)) 1183 if (score_bits & (1 << j))
1073 *bit++ = offsetof (eio_dirent, score) * 8 + j; 1184 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1074 } 1185 }
1075 1186
1076 /* now actually do the sorting (a variant of MSD radix sort) */ 1187 /* now actually do the sorting (a variant of MSD radix sort) */
1077 { 1188 {
1078 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base; 1189 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1079 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end; 1190 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1080 unsigned char *bit_stk [9 + sizeof (ino_t) * 8]; 1191 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1081 int stk_idx = 0; 1192 int stk_idx = 0;
1082 1193
1083 base_stk [stk_idx] = dents; 1194 base_stk [stk_idx] = dents;
1084 end_stk [stk_idx] = dents + size; 1195 end_stk [stk_idx] = dents + size;
1085 bit_stk [stk_idx] = bit - 1; 1196 bit_stk [stk_idx] = bit - 1;
… …
1164 } 1275 }
1165 } 1276 }
1166} 1277}
1167 1278
1168static void 1279static void
1169eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1280eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1170{ 1281{
1171 if (size <= 1) 1282 if (size <= 1)
1172 return; /* our insertion sort relies on size > 0 */ 1283 return; /* our insertion sort relies on size > 0 */
1173 1284
1174 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */ 1285 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
… …
1182 1293
1183/* read a full directory */ 1294/* read a full directory */
1184static void 1295static void
1185eio__scandir (eio_req *req, etp_worker *self) 1296eio__scandir (eio_req *req, etp_worker *self)
1186{ 1297{
1187 DIR *dirp;
1188 EIO_STRUCT_DIRENT *entp;
1189 char *name, *names; 1298 char *name, *names;
1190 int namesalloc = 4096; 1299 int namesalloc = 4096 - sizeof (void *) * 4;
1191 int namesoffs = 0; 1300 int namesoffs = 0;
1192 int flags = req->int1; 1301 int flags = req->int1;
1193 eio_dirent *dents = 0; 1302 eio_dirent *dents = 0;
1194 int dentalloc = 128; 1303 int dentalloc = 128;
1195 int dentoffs = 0; 1304 int dentoffs = 0;
1196 ino_t inode_bits = 0; 1305 eio_ino_t inode_bits = 0;
1306#ifdef _WIN32
1307 HANDLE dirp;
1308 WIN32_FIND_DATA entp;
1309#else
1310 DIR *dirp;
1311 EIO_STRUCT_DIRENT *entp;
1312#endif
1197 1313
1198 req->result = -1; 1314 req->result = -1;
1199 1315
1200 if (!(flags & EIO_READDIR_DENTS)) 1316 if (!(flags & EIO_READDIR_DENTS))
1201 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); 1317 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1202 1318
1203 X_LOCK (wrklock); 1319#ifdef _WIN32
1204 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 1320 {
1321 int len = strlen ((const char *)req->ptr1);
1322 char *path = malloc (MAX_PATH);
1323 const char *fmt;
1324 const char *reqpath = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1325
1326 if (!len)
1327 fmt = "./*";
1328 else if (reqpath[len - 1] == '/' || reqpath[len - 1] == '\\')
1329 fmt = "%s*";
1330 else
1331 fmt = "%s/*";
1332
1333 _snprintf (path, MAX_PATH, fmt, reqpath);
1334 dirp = FindFirstFile (path, &entp);
1335 free (path);
1336
1337 if (dirp == INVALID_HANDLE_VALUE)
1338 {
1339 /* should steal _dosmaperr */
1340 switch (GetLastError ())
1341 {
1342 case ERROR_FILE_NOT_FOUND:
1343 req->result = 0;
1344 break;
1345
1346 case ERROR_INVALID_NAME:
1347 case ERROR_PATH_NOT_FOUND:
1348 case ERROR_NO_MORE_FILES:
1349 errno = ENOENT;
1350 break;
1351
1352 case ERROR_NOT_ENOUGH_MEMORY:
1353 errno = ENOMEM;
1354 break;
1355
1356 default:
1357 errno = EINVAL;
1358 break;
1359 }
1360
1361 return;
1362 }
1363 }
1364#else
1365 #if HAVE_AT
1366 if (req->wd)
1367 {
1368 int fd = openat (WD2FD (req->wd), req->ptr1, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1369
1370 if (fd < 0)
1371 return;
1372
1373 dirp = fdopendir (fd);
1374
1375 if (!dirp)
1376 close (fd);
1377 }
1378 else
1205 self->dirp = dirp = opendir (req->ptr1); 1379 dirp = opendir (req->ptr1);
1380 #else
1381 dirp = opendir (wd_expand (&self->tmpbuf, req->wd, req->ptr1));
1382 #endif
1383
1384 if (!dirp)
1385 return;
1386#endif
1387
1388 if (req->flags & EIO_FLAG_PTR1_FREE)
1389 free (req->ptr1);
1390
1206 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1391 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1207 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; 1392 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1208 req->ptr2 = names = malloc (namesalloc); 1393 req->ptr2 = names = malloc (namesalloc);
1209 X_UNLOCK (wrklock);
1210 1394
1211 if (dirp && names && (!flags || dents)) 1395 if (!names || (flags && !dents))
1396 return;
1397
1212 for (;;) 1398 for (;;)
1213 { 1399 {
1400 int done;
1401
1402#ifdef _WIN32
1403 done = !dirp;
1404#else
1214 errno = 0; 1405 errno = 0;
1215 entp = readdir (dirp); 1406 entp = readdir (dirp);
1407 done = !entp;
1408#endif
1216 1409
1217 if (!entp) 1410 if (done)
1218 { 1411 {
1412#ifndef _WIN32
1413 int old_errno = errno;
1414 closedir (dirp);
1415 errno = old_errno;
1416
1219 if (errno) 1417 if (errno)
1220 break; 1418 break;
1419#endif
1221 1420
1222 /* sort etc. */ 1421 /* sort etc. */
1223 req->int1 = flags; 1422 req->int1 = flags;
1224 req->result = dentoffs; 1423 req->result = dentoffs;
1225 1424
1226 if (flags & EIO_READDIR_STAT_ORDER) 1425 if (flags & EIO_READDIR_STAT_ORDER)
1227 eio_dent_sort (dents, dentoffs, 0, inode_bits); /* sort by inode exclusively */ 1426 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1228 else if (flags & EIO_READDIR_DIRS_FIRST) 1427 else if (flags & EIO_READDIR_DIRS_FIRST)
1229 if (flags & EIO_READDIR_FOUND_UNKNOWN) 1428 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1230 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */ 1429 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1231 else 1430 else
1431 {
1432 /* in this case, all is known, and we just put dirs first and sort them */
1433 eio_dirent *oth = dents + dentoffs;
1434 eio_dirent *dir = dents;
1435
1436 /* now partition dirs to the front, and non-dirs to the back */
1437 /* by walking from both sides and swapping if necessary */
1438 while (oth > dir)
1439 {
1440 if (dir->type == EIO_DT_DIR)
1441 ++dir;
1442 else if ((--oth)->type == EIO_DT_DIR)
1443 {
1444 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1445
1446 ++dir;
1447 }
1448 }
1449
1450 /* now sort the dirs only (dirs all have the same score) */
1451 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1452 }
1453
1454 break;
1455 }
1456
1457 /* now add the entry to our list(s) */
1458 name = D_NAME (entp);
1459
1460 /* skip . and .. entries */
1461 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1462 {
1463 int len = D_NAMLEN (entp) + 1;
1464
1465 while (ecb_expect_false (namesoffs + len > namesalloc))
1466 {
1467 namesalloc *= 2;
1468 req->ptr2 = names = realloc (names, namesalloc);
1469
1470 if (!names)
1471 break;
1472 }
1473
1474 memcpy (names + namesoffs, name, len);
1475
1476 if (dents)
1477 {
1478 struct eio_dirent *ent;
1479
1480 if (ecb_expect_false (dentoffs == dentalloc))
1232 { 1481 {
1233 /* in this case, all is known, and we just put dirs first and sort them */ 1482 dentalloc *= 2;
1483 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1484
1485 if (!dents)
1486 break;
1487 }
1488
1234 eio_dirent *oth = dents + dentoffs; 1489 ent = dents + dentoffs;
1235 eio_dirent *dir = dents;
1236 1490
1237 /* now partition dirs to the front, and non-dirs to the back */ 1491 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1238 /* by walking from both sides and swapping if necessary */ 1492 ent->namelen = len - 1;
1239 /* also clear score, so it doesn't influence sorting */ 1493 ent->inode = D_INO (entp);
1240 while (oth > dir) 1494
1495 inode_bits |= ent->inode;
1496
1497 switch (D_TYPE (entp))
1498 {
1499 default:
1500 ent->type = EIO_DT_UNKNOWN;
1501 flags |= EIO_READDIR_FOUND_UNKNOWN;
1502 break;
1503
1504 #ifdef DT_FIFO
1505 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1506 #endif
1507 #ifdef DT_CHR
1508 case DT_CHR: ent->type = EIO_DT_CHR; break;
1509 #endif
1510 #ifdef DT_MPC
1511 case DT_MPC: ent->type = EIO_DT_MPC; break;
1512 #endif
1513 #ifdef DT_DIR
1514 case DT_DIR: ent->type = EIO_DT_DIR; break;
1515 #endif
1516 #ifdef DT_NAM
1517 case DT_NAM: ent->type = EIO_DT_NAM; break;
1518 #endif
1519 #ifdef DT_BLK
1520 case DT_BLK: ent->type = EIO_DT_BLK; break;
1521 #endif
1522 #ifdef DT_MPB
1523 case DT_MPB: ent->type = EIO_DT_MPB; break;
1524 #endif
1525 #ifdef DT_REG
1526 case DT_REG: ent->type = EIO_DT_REG; break;
1527 #endif
1528 #ifdef DT_NWK
1529 case DT_NWK: ent->type = EIO_DT_NWK; break;
1530 #endif
1531 #ifdef DT_CMP
1532 case DT_CMP: ent->type = EIO_DT_CMP; break;
1533 #endif
1534 #ifdef DT_LNK
1535 case DT_LNK: ent->type = EIO_DT_LNK; break;
1536 #endif
1537 #ifdef DT_SOCK
1538 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1539 #endif
1540 #ifdef DT_DOOR
1541 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1542 #endif
1543 #ifdef DT_WHT
1544 case DT_WHT: ent->type = EIO_DT_WHT; break;
1545 #endif
1546 }
1547
1548 ent->score = 7;
1549
1550 if (flags & EIO_READDIR_DIRS_FIRST)
1551 {
1552 if (ent->type == EIO_DT_UNKNOWN)
1241 { 1553 {
1242 if (dir->type == EIO_DT_DIR) 1554 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1243 ++dir; 1555 ent->score = 1;
1244 else if ((--oth)->type == EIO_DT_DIR) 1556 else if (!strchr (name, '.')) /* absence of dots indicate likely dirs */
1245 { 1557 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1246 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1247
1248 ++dir;
1249 }
1250 } 1558 }
1251 1559 else if (ent->type == EIO_DT_DIR)
1252 /* now sort the dirs only */ 1560 ent->score = 0;
1253 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1254 } 1561 }
1255
1256 break;
1257 }
1258
1259 /* now add the entry to our list(s) */
1260 name = entp->d_name;
1261
1262 /* skip . and .. entries */
1263 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1264 {
1265 int len = D_NAMLEN (entp) + 1;
1266
1267 while (expect_false (namesoffs + len > namesalloc))
1268 {
1269 namesalloc *= 2;
1270 X_LOCK (wrklock);
1271 req->ptr2 = names = realloc (names, namesalloc);
1272 X_UNLOCK (wrklock);
1273
1274 if (!names)
1275 break;
1276 } 1562 }
1277 1563
1278 memcpy (names + namesoffs, name, len);
1279
1280 if (dents)
1281 {
1282 struct eio_dirent *ent;
1283
1284 if (expect_false (dentoffs == dentalloc))
1285 {
1286 dentalloc *= 2;
1287 X_LOCK (wrklock);
1288 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1289 X_UNLOCK (wrklock);
1290
1291 if (!dents)
1292 break;
1293 }
1294
1295 ent = dents + dentoffs;
1296
1297 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1298 ent->namelen = len - 1;
1299 ent->inode = D_INO (entp);
1300
1301 inode_bits |= ent->inode;
1302
1303 switch (D_TYPE (entp))
1304 {
1305 default:
1306 ent->type = EIO_DT_UNKNOWN;
1307 flags |= EIO_READDIR_FOUND_UNKNOWN;
1308 break;
1309
1310 #ifdef DT_FIFO
1311 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1312 #endif
1313 #ifdef DT_CHR
1314 case DT_CHR: ent->type = EIO_DT_CHR; break;
1315 #endif
1316 #ifdef DT_MPC
1317 case DT_MPC: ent->type = EIO_DT_MPC; break;
1318 #endif
1319 #ifdef DT_DIR
1320 case DT_DIR: ent->type = EIO_DT_DIR; break;
1321 #endif
1322 #ifdef DT_NAM
1323 case DT_NAM: ent->type = EIO_DT_NAM; break;
1324 #endif
1325 #ifdef DT_BLK
1326 case DT_BLK: ent->type = EIO_DT_BLK; break;
1327 #endif
1328 #ifdef DT_MPB
1329 case DT_MPB: ent->type = EIO_DT_MPB; break;
1330 #endif
1331 #ifdef DT_REG
1332 case DT_REG: ent->type = EIO_DT_REG; break;
1333 #endif
1334 #ifdef DT_NWK
1335 case DT_NWK: ent->type = EIO_DT_NWK; break;
1336 #endif
1337 #ifdef DT_CMP
1338 case DT_CMP: ent->type = EIO_DT_CMP; break;
1339 #endif
1340 #ifdef DT_LNK
1341 case DT_LNK: ent->type = EIO_DT_LNK; break;
1342 #endif
1343 #ifdef DT_SOCK
1344 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1345 #endif
1346 #ifdef DT_DOOR
1347 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1348 #endif
1349 #ifdef DT_WHT
1350 case DT_WHT: ent->type = EIO_DT_WHT; break;
1351 #endif
1352 }
1353
1354 ent->score = 7;
1355
1356 if (flags & EIO_READDIR_DIRS_FIRST)
1357 {
1358 if (ent->type == EIO_DT_UNKNOWN)
1359 {
1360 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1361 ent->score = 1;
1362 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1363 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1364 }
1365 else if (ent->type == EIO_DT_DIR)
1366 ent->score = 0;
1367 }
1368 }
1369
1370 namesoffs += len; 1564 namesoffs += len;
1371 ++dentoffs; 1565 ++dentoffs;
1372 } 1566 }
1373 1567
1374 if (EIO_CANCELLED (req)) 1568 if (EIO_CANCELLED (req))
1375 { 1569 {
1376 errno = ECANCELED; 1570 errno = ECANCELED;
1377 break; 1571 break;
1378 } 1572 }
1573
1574#ifdef _WIN32
1575 if (!FindNextFile (dirp, &entp))
1576 {
1577 FindClose (dirp);
1578 dirp = 0;
1379 } 1579 }
1580#endif
1581 }
1380} 1582}
1381 1583
1382#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) 1584/*****************************************************************************/
1383# define eio__msync(a,b,c) ((errno = ENOSYS), -1) 1585/* working directory stuff */
1384#else 1586/* various deficiencies in the posix 2008 api force us to */
1587/* keep the absolute path in string form at all times */
1588/* fuck yeah. */
1385 1589
1386int 1590#if !HAVE_AT
1387eio__msync (void *mem, size_t len, int flags) 1591
1592/* a bit like realpath, but usually faster because it doesn'T have to return */
1593/* an absolute or canonical path */
1594static const char *
1595wd_expand (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
1388{ 1596{
1389 if (EIO_MS_ASYNC != MS_SYNC 1597 if (!wd || *path == '/')
1390 || EIO_MS_INVALIDATE != MS_INVALIDATE 1598 return path;
1391 || EIO_MS_SYNC != MS_SYNC) 1599
1600 if (path [0] == '.' && !path [1])
1601 return wd->str;
1602
1603 {
1604 int l1 = wd->len;
1605 int l2 = strlen (path);
1606
1607 char *res = etp_tmpbuf_get (tmpbuf, l1 + l2 + 2);
1608
1609 memcpy (res, wd->str, l1);
1610 res [l1] = '/';
1611 memcpy (res + l1 + 1, path, l2 + 1);
1612
1613 return res;
1614 }
1615}
1616
1617#endif
1618
1619static eio_wd
1620eio__wd_open_sync (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
1621{
1622 int fd;
1623 eio_wd res;
1624 int len = eio__realpath (tmpbuf, wd, path);
1625
1626 if (len < 0)
1627 return EIO_INVALID_WD;
1628
1629#if HAVE_AT
1630 fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1631
1632 if (fd < 0)
1633 return EIO_INVALID_WD;
1634#endif
1635
1636 res = malloc (sizeof (*res) + len); /* one extra 0-byte */
1637
1638#if HAVE_AT
1639 res->fd = fd;
1640#endif
1641
1642 res->len = len;
1643 memcpy (res->str, tmpbuf->ptr, len);
1644 res->str [len] = 0;
1645
1646 return res;
1647}
1648
1649eio_wd
1650eio_wd_open_sync (eio_wd wd, const char *path)
1651{
1652 struct etp_tmpbuf tmpbuf = { };
1653 wd = eio__wd_open_sync (&tmpbuf, wd, path);
1654 free (tmpbuf.ptr);
1655
1656 return wd;
1657}
1658
1659void
1660eio_wd_close_sync (eio_wd wd)
1661{
1662 if (wd != EIO_INVALID_WD && wd != EIO_CWD)
1392 { 1663 {
1393 flags = 0 1664 #if HAVE_AT
1394 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0) 1665 close (wd->fd);
1395 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0) 1666 #endif
1396 | (flags & EIO_MS_SYNC ? MS_SYNC : 0); 1667 free (wd);
1397 } 1668 }
1398
1399 return msync (mem, len, flags);
1400} 1669}
1401 1670
1402#endif 1671#if HAVE_AT
1403 1672
1404int 1673/* they forgot these */
1405eio__mtouch (void *mem, size_t len, int flags)
1406{
1407 intptr_t addr = (intptr_t)mem;
1408 intptr_t end = addr + len;
1409#ifdef PAGESIZE
1410 const intptr_t page = PAGESIZE;
1411#else
1412 static intptr_t page;
1413 1674
1414 if (!page) 1675static int
1415 page = sysconf (_SC_PAGESIZE); 1676eio__truncateat (int dirfd, const char *path, off_t length)
1416#endif 1677{
1678 int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
1679 int res;
1417 1680
1418 /* round down to start of page, although this is probably useless */ 1681 if (fd < 0)
1419 addr &= ~(page - 1); /* assume page size is always a power of two */ 1682 return fd;
1420 1683
1421 if (addr < end) 1684 res = ftruncate (fd, length);
1422 if (flags & EIO_MT_MODIFY) /* modify */ 1685 close (fd);
1423 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
1424 else
1425 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
1426
1427 return 0; 1686 return res;
1428} 1687}
1688
1689static int
1690eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
1691{
1692 int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
1693 int res;
1694
1695 if (fd < 0)
1696 return fd;
1697
1698 res = fstatvfs (fd, buf);
1699 close (fd);
1700 return res;
1701
1702}
1703
1704#endif
1429 1705
1430/*****************************************************************************/ 1706/*****************************************************************************/
1431 1707
1432#define ALLOC(len) \ 1708#define ALLOC(len) \
1433 if (!req->ptr2) \ 1709 if (!req->ptr2) \
… …
1442 req->result = -1; \ 1718 req->result = -1; \
1443 break; \ 1719 break; \
1444 } \ 1720 } \
1445 } 1721 }
1446 1722
1447X_THREAD_PROC (etp_proc)
1448{
1449 ETP_REQ *req;
1450 struct timespec ts;
1451 etp_worker *self = (etp_worker *)thr_arg;
1452
1453 /* try to distribute timeouts somewhat randomly */
1454 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1455
1456 for (;;)
1457 {
1458 X_LOCK (reqlock);
1459
1460 for (;;)
1461 {
1462 self->req = req = reqq_shift (&req_queue);
1463
1464 if (req)
1465 break;
1466
1467 ++idle;
1468
1469 ts.tv_sec = time (0) + IDLE_TIMEOUT;
1470 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
1471 {
1472 if (idle > max_idle)
1473 {
1474 --idle;
1475 X_UNLOCK (reqlock);
1476 X_LOCK (wrklock);
1477 --started;
1478 X_UNLOCK (wrklock);
1479 goto quit;
1480 }
1481
1482 /* we are allowed to idle, so do so without any timeout */
1483 X_COND_WAIT (reqwait, reqlock);
1484 }
1485
1486 --idle;
1487 }
1488
1489 --nready;
1490
1491 X_UNLOCK (reqlock);
1492
1493 if (req->type < 0)
1494 goto quit;
1495
1496 if (!EIO_CANCELLED (req))
1497 ETP_EXECUTE (self, req);
1498
1499 X_LOCK (reslock);
1500
1501 ++npending;
1502
1503 if (!reqq_push (&res_queue, req) && want_poll_cb)
1504 want_poll_cb ();
1505
1506 self->req = 0;
1507 etp_worker_clear (self);
1508
1509 X_UNLOCK (reslock);
1510 }
1511
1512quit:
1513 X_LOCK (wrklock);
1514 etp_worker_free (self);
1515 X_UNLOCK (wrklock);
1516
1517 return 0;
1518}
1519
1520/*****************************************************************************/ 1723/*****************************************************************************/
1521 1724
1725int ecb_cold
1522int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1726eio_init (void (*want_poll)(void), void (*done_poll)(void))
1523{ 1727{
1524 return etp_init (want_poll, done_poll); 1728 return etp_init (EIO_POOL, want_poll, done_poll);
1525} 1729}
1526 1730
1731ecb_inline void
1527static void eio_api_destroy (eio_req *req) 1732eio_api_destroy (eio_req *req)
1528{ 1733{
1529 free (req); 1734 free (req);
1530} 1735}
1531 1736
1532#define REQ(rtype) \ 1737#define REQ(rtype) \
1533 eio_req *req; \ 1738 eio_req *req; \
1534 \ 1739 \
1535 req = (eio_req *)calloc (1, sizeof *req); \ 1740 req = (eio_req *)calloc (1, sizeof *req); \
1536 if (!req) \ 1741 if (!req) \
1537 return 0; \ 1742 return 0; \
… …
1551 { \ 1756 { \
1552 eio_api_destroy (req); \ 1757 eio_api_destroy (req); \
1553 return 0; \ 1758 return 0; \
1554 } 1759 }
1555 1760
1761#define SINGLEDOT(ptr) (0[(char *)(ptr)] == '.' && !1[(char *)(ptr)])
1762
1763static void
1556static void eio_execute (etp_worker *self, eio_req *req) 1764eio_execute (etp_worker *self, eio_req *req)
1557{ 1765{
1558 errno = 0; 1766#if HAVE_AT
1767 int dirfd;
1768#else
1769 const char *path;
1770#endif
1771
1772 if (ecb_expect_false (EIO_CANCELLED (req)))
1773 {
1774 req->result = -1;
1775 req->errorno = ECANCELED;
1776 return;
1777 }
1778
1779 if (ecb_expect_false (req->wd == EIO_INVALID_WD))
1780 {
1781 req->result = -1;
1782 req->errorno = ENOENT;
1783 return;
1784 }
1785
1786 if (req->type >= EIO_OPEN)
1787 {
1788 #if HAVE_AT
1789 dirfd = WD2FD (req->wd);
1790 #else
1791 path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1792 #endif
1793 }
1559 1794
1560 switch (req->type) 1795 switch (req->type)
1561 { 1796 {
1797 case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
1798 req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
1799 break;
1800 case EIO_WD_CLOSE: req->result = 0;
1801 eio_wd_close_sync (req->wd); break;
1802
1803 case EIO_SEEK: eio__lseek (req); break;
1562 case EIO_READ: ALLOC (req->size); 1804 case EIO_READ: ALLOC (req->size);
1563 req->result = req->offs >= 0 1805 req->result = req->offs >= 0
1564 ? pread (req->int1, req->ptr2, req->size, req->offs) 1806 ? pread (req->int1, req->ptr2, req->size, req->offs)
1565 : read (req->int1, req->ptr2, req->size); break; 1807 : read (req->int1, req->ptr2, req->size); break;
1566 case EIO_WRITE: req->result = req->offs >= 0 1808 case EIO_WRITE: req->result = req->offs >= 0
1567 ? pwrite (req->int1, req->ptr2, req->size, req->offs) 1809 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1568 : write (req->int1, req->ptr2, req->size); break; 1810 : write (req->int1, req->ptr2, req->size); break;
1569 1811
1570 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break; 1812 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1571 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break; 1813 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
1814
1815#if HAVE_AT
1572 1816
1573 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1817 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1574 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 1818 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break;
1575 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1819 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1576 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 1820 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, AT_SYMLINK_NOFOLLOW); break;
1577 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1578 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1579
1580 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; 1821 case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break;
1581 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1582 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break; 1822 case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break;
1583 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1584 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break; 1823 case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break;
1585 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1586
1587 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break; 1824 case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break;
1588 case EIO_CLOSE: req->result = close (req->int1); break; 1825
1589 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1590 case EIO_UNLINK: req->result = unlink (req->ptr1); break; 1826 case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
1591 case EIO_RMDIR: req->result = rmdir (req->ptr1); break; 1827 case EIO_RMDIR: /* complications arise because "." cannot be removed, so we might have to expand */
1828 req->result = req->wd && SINGLEDOT (req->ptr1)
1829 ? rmdir (req->wd->str)
1830 : unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break;
1592 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; 1831 case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); break;
1593 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; 1832 case EIO_RENAME: /* complications arise because "." cannot be renamed, so we might have to expand */
1833 req->result = req->wd && SINGLEDOT (req->ptr1)
1834 ? rename (req->wd->str, req->ptr2)
1835 : renameat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2); break;
1594 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; 1836 case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2, 0); break;
1595 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; 1837 case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
1596 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break; 1838 case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1597
1598 case EIO_READLINK: ALLOC (NAME_MAX); 1839 case EIO_READLINK: ALLOC (PATH_MAX);
1599 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; 1840 req->result = readlinkat (dirfd, req->ptr1, req->ptr2, PATH_MAX); break;
1600 1841 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1601 case EIO_SYNC: req->result = 0; sync (); break; 1842 req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1602 case EIO_FSYNC: req->result = fsync (req->int1); break;
1603 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1604 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1605 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;
1606 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1607
1608 case EIO_READDIR: eio__scandir (req, self); break;
1609
1610 case EIO_BUSY: 1843 case EIO_UTIME:
1611#ifdef _WIN32 1844 case EIO_FUTIME:
1612 Sleep (req->nv1 * 1000.); 1845 {
1846 struct timespec ts[2];
1847 struct timespec *times;
1848
1849 if (req->nv1 != -1. || req->nv2 != -1.)
1850 {
1851 ts[0].tv_sec = req->nv1;
1852 ts[0].tv_nsec = (req->nv1 - ts[0].tv_sec) * 1e9;
1853 ts[1].tv_sec = req->nv2;
1854 ts[1].tv_nsec = (req->nv2 - ts[1].tv_sec) * 1e9;
1855
1856 times = ts;
1857 }
1858 else
1859 times = 0;
1860
1861 req->result = req->type == EIO_FUTIME
1862 ? futimens (req->int1, times)
1863 : utimensat (dirfd, req->ptr1, times, 0);
1864 }
1865 break;
1866
1613#else 1867#else
1614 {
1615 struct timeval tv;
1616 1868
1617 tv.tv_sec = req->nv1; 1869 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1618 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; 1870 req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
1871 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1872 req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
1873 case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
1874 case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
1875 case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
1876 case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
1619 1877
1620 req->result = select (0, 0, 0, 0, &tv); 1878 case EIO_UNLINK: req->result = unlink (path ); break;
1621 } 1879 case EIO_RMDIR: req->result = rmdir (path ); break;
1622#endif 1880 case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
1623 break; 1881 case EIO_RENAME: req->result = rename (path , req->ptr2); break;
1882 case EIO_LINK: req->result = link (path , req->ptr2); break;
1883 case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
1884 case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
1885 case EIO_READLINK: ALLOC (PATH_MAX);
1886 req->result = readlink (path, req->ptr2, PATH_MAX); break;
1887 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1888 req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
1624 1889
1625 case EIO_UTIME: 1890 case EIO_UTIME:
1626 case EIO_FUTIME: 1891 case EIO_FUTIME:
1627 { 1892 {
1628 struct timeval tv[2]; 1893 struct timeval tv[2];
1629 struct timeval *times; 1894 struct timeval *times;
1630 1895
1631 if (req->nv1 != -1. || req->nv2 != -1.) 1896 if (req->nv1 != -1. || req->nv2 != -1.)
1632 { 1897 {
1633 tv[0].tv_sec = req->nv1; 1898 tv[0].tv_sec = req->nv1;
1634 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.; 1899 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
1635 tv[1].tv_sec = req->nv2; 1900 tv[1].tv_sec = req->nv2;
1636 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.; 1901 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
1637 1902
1638 times = tv; 1903 times = tv;
1639 } 1904 }
1640 else 1905 else
1641 times = 0; 1906 times = 0;
1642
1643 1907
1644 req->result = req->type == EIO_FUTIME 1908 req->result = req->type == EIO_FUTIME
1645 ? futimes (req->int1, times) 1909 ? futimes (req->int1, times)
1646 : utimes (req->ptr1, times); 1910 : utimes (req->ptr1, times);
1647 } 1911 }
1648 break; 1912 break;
1649 1913
1914#endif
1915
1916 case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
1917 {
1918 ALLOC (req->result);
1919 memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
1920 }
1921 break;
1922
1923 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1924 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1925
1926 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1927 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1928
1929 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1930 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1931 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1932
1933 case EIO_CLOSE: req->result = close (req->int1); break;
1934 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1935 case EIO_SYNC: req->result = 0; sync (); break;
1936 case EIO_FSYNC: req->result = fsync (req->int1); break;
1937 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1938 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
1939 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1940 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1941 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1942 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1943 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1944 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
1945
1946 case EIO_READDIR: eio__scandir (req, self); break;
1947
1948 case EIO_BUSY:
1949#ifdef _WIN32
1950 Sleep (req->nv1 * 1e3);
1951#else
1952 {
1953 struct timeval tv;
1954
1955 tv.tv_sec = req->nv1;
1956 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1957
1958 req->result = select (0, 0, 0, 0, &tv);
1959 }
1960#endif
1961 break;
1962
1963#if 0
1650 case EIO_GROUP: 1964 case EIO_GROUP:
1651 abort (); /* handled in eio_request */ 1965 abort (); /* handled in eio_request */
1966#endif
1652 1967
1653 case EIO_NOP: 1968 case EIO_NOP:
1654 req->result = 0; 1969 req->result = 0;
1655 break; 1970 break;
1656 1971
1657 case EIO_CUSTOM: 1972 case EIO_CUSTOM:
1658 ((void (*)(eio_req *))req->feed) (req); 1973 req->feed (req);
1659 break; 1974 break;
1660 1975
1661 default: 1976 default:
1662 req->result = -1; 1977 req->result = EIO_ENOSYS ();
1663 break; 1978 break;
1664 } 1979 }
1665 1980
1666 req->errorno = errno; 1981 req->errorno = errno;
1667} 1982}
1668 1983
1669#ifndef EIO_NO_WRAPPERS 1984#ifndef EIO_NO_WRAPPERS
1670 1985
1986eio_req *eio_wd_open (const char *path, int pri, eio_cb cb, void *data)
1987{
1988 REQ (EIO_WD_OPEN); PATH; SEND;
1989}
1990
1991eio_req *eio_wd_close (eio_wd wd, int pri, eio_cb cb, void *data)
1992{
1993 REQ (EIO_WD_CLOSE); req->wd = wd; SEND;
1994}
1995
1671eio_req *eio_nop (int pri, eio_cb cb, void *data) 1996eio_req *eio_nop (int pri, eio_cb cb, void *data)
1672{ 1997{
1673 REQ (EIO_NOP); SEND; 1998 REQ (EIO_NOP); SEND;
1674} 1999}
1675 2000
… …
1691eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2016eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1692{ 2017{
1693 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2018 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1694} 2019}
1695 2020
2021eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2022{
2023 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2024}
2025
2026eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
2027{
2028 REQ (EIO_SYNCFS); req->int1 = fd; SEND;
2029}
2030
2031eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2032{
2033 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2034}
2035
1696eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2036eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1697{ 2037{
1698 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2038 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1699} 2039}
1700 2040
1701eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) 2041eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
1702{ 2042{
1703 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; 2043 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
1704} 2044}
1705 2045
1706eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) 2046eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
1707{ 2047{
1708 REQ (EIO_FDATASYNC); req->int1 = fd; SEND; 2048 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2049}
2050
2051eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2052{
2053 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
1709} 2054}
1710 2055
1711eio_req *eio_close (int fd, int pri, eio_cb cb, void *data) 2056eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
1712{ 2057{
1713 REQ (EIO_CLOSE); req->int1 = fd; SEND; 2058 REQ (EIO_CLOSE); req->int1 = fd; SEND;
… …
1716eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data) 2061eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
1717{ 2062{
1718 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND; 2063 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
1719} 2064}
1720 2065
2066eio_req *eio_seek (int fd, off_t offset, int whence, int pri, eio_cb cb, void *data)
2067{
2068 REQ (EIO_SEEK); req->int1 = fd; req->offs = offset; req->int2 = whence; SEND;
2069}
2070
1721eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data) 2071eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1722{ 2072{
1723 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND; 2073 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1724} 2074}
1725 2075
… …
1731eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) 2081eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1732{ 2082{
1733 REQ (EIO_FSTAT); req->int1 = fd; SEND; 2083 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1734} 2084}
1735 2085
2086eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
2087{
2088 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
2089}
2090
1736eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) 2091eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1737{ 2092{
1738 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; 2093 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1739} 2094}
1740 2095
… …
1746eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) 2101eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1747{ 2102{
1748 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2103 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1749} 2104}
1750 2105
1751eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2106eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1752{ 2107{
1753 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2108 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1754} 2109}
1755 2110
1756eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) 2111eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
… …
1776eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data) 2131eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1777{ 2132{
1778 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2133 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1779} 2134}
1780 2135
1781eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2136eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1782{ 2137{
1783 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2138 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1784} 2139}
1785 2140
1786eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) 2141eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
… …
1802eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data) 2157eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1803{ 2158{
1804 return eio__1path (EIO_READLINK, path, pri, cb, data); 2159 return eio__1path (EIO_READLINK, path, pri, cb, data);
1805} 2160}
1806 2161
2162eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2163{
2164 return eio__1path (EIO_REALPATH, path, pri, cb, data);
2165}
2166
1807eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) 2167eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1808{ 2168{
1809 return eio__1path (EIO_STAT, path, pri, cb, data); 2169 return eio__1path (EIO_STAT, path, pri, cb, data);
1810} 2170}
1811 2171
1812eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) 2172eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1813{ 2173{
1814 return eio__1path (EIO_LSTAT, path, pri, cb, data); 2174 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1815} 2175}
1816 2176
2177eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
2178{
2179 return eio__1path (EIO_STATVFS, path, pri, cb, data);
2180}
2181
1817eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) 2182eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1818{ 2183{
1819 return eio__1path (EIO_UNLINK, path, pri, cb, data); 2184 return eio__1path (EIO_UNLINK, path, pri, cb, data);
1820} 2185}
1821 2186
… …
1829 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND; 2194 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1830} 2195}
1831 2196
1832eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) 2197eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1833{ 2198{
1834 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; 2199 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1835} 2200}
1836 2201
1837static eio_req * 2202static eio_req *
1838eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2203eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1839{ 2204{
… …
1863eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2228eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1864{ 2229{
1865 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2230 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1866} 2231}
1867 2232
1868eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) 2233eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
1869{ 2234{
1870 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND; 2235 REQ (EIO_CUSTOM); req->feed = execute; SEND;
1871} 2236}
1872 2237
1873#endif 2238#endif
1874 2239
1875eio_req *eio_grp (eio_cb cb, void *data) 2240eio_req *eio_grp (eio_cb cb, void *data)
… …
1884#undef SEND 2249#undef SEND
1885 2250
1886/*****************************************************************************/ 2251/*****************************************************************************/
1887/* grp functions */ 2252/* grp functions */
1888 2253
2254void
1889void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2255eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1890{ 2256{
1891 grp->int2 = limit; 2257 grp->int2 = limit;
1892 grp->feed = feed; 2258 grp->feed = feed;
1893 2259
1894 grp_try_feed (grp); 2260 grp_try_feed (grp);
1895} 2261}
1896 2262
2263void
1897void eio_grp_limit (eio_req *grp, int limit) 2264eio_grp_limit (eio_req *grp, int limit)
1898{ 2265{
1899 grp->int2 = limit; 2266 grp->int2 = limit;
1900 2267
1901 grp_try_feed (grp); 2268 grp_try_feed (grp);
1902} 2269}
1903 2270
2271void
1904void eio_grp_add (eio_req *grp, eio_req *req) 2272eio_grp_add (eio_req *grp, eio_req *req)
1905{ 2273{
1906 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2274 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1907 2275
1908 grp->flags |= EIO_FLAG_GROUPADD; 2276 grp->flags |= ETP_FLAG_GROUPADD;
1909 2277
1910 ++grp->size; 2278 ++grp->size;
1911 req->grp = grp; 2279 req->grp = grp;
1912 2280
1913 req->grp_prev = 0; 2281 req->grp_prev = 0;
… …
1920} 2288}
1921 2289
1922/*****************************************************************************/ 2290/*****************************************************************************/
1923/* misc garbage */ 2291/* misc garbage */
1924 2292
2293eio_ssize_t
1925ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2294eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1926{ 2295{
1927 etp_worker wrk;
1928 ssize_t ret;
1929
1930 wrk.dbuf = 0;
1931
1932 ret = eio__sendfile (ofd, ifd, offset, count, &wrk); 2296 return eio__sendfile (ofd, ifd, offset, count);
1933
1934 if (wrk.dbuf)
1935 free (wrk.dbuf);
1936
1937 return ret;
1938} 2297}
1939 2298

Diff Legend

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