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

Comparing libeio/eio.c (file contents):
Revision 1.30 by root, Wed Jun 3 12:24:49 2009 UTC vs.
Revision 1.141 by root, Tue Feb 23 19:47:44 2016 UTC

1/* 1/*
2 * libeio implementation 2 * libeio implementation
3 * 3 *
4 * Copyright (c) 2007,2008 Marc Alexander Lehmann <libeio@schmorp.de> 4 * Copyright (c) 2007,2008,2009,2010,2011,2012,2013,2016 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 fcntl(fd,cmd,arg) EIO_ENOSYS ()
128 #define ioctl(fd,cmd,arg) EIO_ENOSYS ()
129 #define fchmod(fd,mode) EIO_ENOSYS ()
130 #define chown(path,uid,gid) EIO_ENOSYS ()
131 #define fchown(fd,uid,gid) EIO_ENOSYS ()
132 #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
133 #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
134 #define mknod(path,mode,dev) EIO_ENOSYS ()
135 #define sync() EIO_ENOSYS ()
136 #define readlink(path,buf,s) EIO_ENOSYS ()
137 #define statvfs(path,buf) EIO_ENOSYS ()
138 #define fstatvfs(fd,buf) EIO_ENOSYS ()
139
140 #define pread(fd,buf,count,offset) eio__pread (fd, buf, count, offset)
141 #define pwrite(fd,buf,count,offset) eio__pwrite (fd, buf, count, offset)
142
143 #if __GNUC__
144 typedef long long eio_off_t; /* signed for compatibility to msvc */
145 #else
146 typedef __int64 eio_off_t; /* unsigned not supported by msvc */
147 #endif
148
149 static eio_ssize_t
150 eio__pread (int fd, void *buf, eio_ssize_t count, eio_off_t offset)
151 {
152 OVERLAPPED o = { 0 };
153 DWORD got;
154
155 o.Offset = offset;
156 o.OffsetHigh = offset >> 32;
157
158 return ReadFile ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd), buf, count, &got, &o)
159 ? got : -1;
160 }
161
162 static eio_ssize_t
163 eio__pwrite (int fd, void *buf, eio_ssize_t count, eio_off_t offset)
164 {
165 OVERLAPPED o = { 0 };
166 DWORD got;
167
168 o.Offset = offset;
169 o.OffsetHigh = offset >> 32;
170
171 return WriteFile ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd), buf, count, &got, &o)
172 ? got : -1;
173 }
174
175 /* rename() uses MoveFile, which fails to overwrite */
176 #define rename(old,neu) eio__rename (old, neu)
177
178 static int
179 eio__rename (const char *old, const char *neu)
180 {
181 if (MoveFileEx (old, neu, MOVEFILE_REPLACE_EXISTING))
182 return 0;
183
184 /* should steal _dosmaperr */
185 switch (GetLastError ())
186 {
187 case ERROR_FILE_NOT_FOUND:
188 case ERROR_PATH_NOT_FOUND:
189 case ERROR_INVALID_DRIVE:
190 case ERROR_NO_MORE_FILES:
191 case ERROR_BAD_NETPATH:
192 case ERROR_BAD_NET_NAME:
193 case ERROR_BAD_PATHNAME:
194 case ERROR_FILENAME_EXCED_RANGE:
195 errno = ENOENT;
196 break;
197
198 default:
199 errno = EACCES;
200 break;
201 }
202
203 return -1;
204 }
205
206 /* we could even stat and see if it exists */
207 static int
208 symlink (const char *old, const char *neu)
209 {
210 #if WINVER >= 0x0600
211 if (CreateSymbolicLink (neu, old, 1))
212 return 0;
213
214 if (CreateSymbolicLink (neu, old, 0))
215 return 0;
216 #endif
217
218 return EIO_ERRNO (ENOENT, -1);
219 }
220
221 /* POSIX API only, causing trouble for win32 apps */
222 #define CreateHardLink(neu,old,flags) 0 /* not really creating hardlink, still using relative paths? */
223 #define CreateSymbolicLink(neu,old,flags) 0 /* vista+ only */
224
225 struct statvfs
226 {
227 int dummy;
228 };
229
230 #define DT_DIR EIO_DT_DIR
231 #define DT_REG EIO_DT_REG
232 #define D_NAME(entp) entp.cFileName
233 #define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
73 234
74#else 235#else
75 236
76# include "config.h"
77# include <sys/time.h> 237 #include <sys/time.h>
78# include <sys/select.h> 238 #include <sys/select.h>
79# include <unistd.h> 239 #include <unistd.h>
240 #include <signal.h>
241 #include <dirent.h>
242
243 #ifdef ANDROID
244 #include <sys/vfs.h>
245 #define statvfs statfs
246 #define fstatvfs fstatfs
247 #include <asm/page.h> /* supposedly limits.h does #define PAGESIZE PAGESIZE */
248 #else
249 #include <sys/statvfs.h>
250 #endif
251
252 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
253 #include <sys/mman.h>
254 #endif
255
256 #define D_NAME(entp) entp->d_name
257
258 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
259 #if __FreeBSD__ || __NetBSD__ || __OpenBSD__
260 #define _DIRENT_HAVE_D_TYPE /* sigh */
261 #define D_INO(de) (de)->d_fileno
262 #define D_NAMLEN(de) (de)->d_namlen
263 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
264 #define D_INO(de) (de)->d_ino
265 #endif
266
267 #ifdef _D_EXACT_NAMLEN
268 #undef D_NAMLEN
269 #define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
270 #endif
271
272 #ifdef _DIRENT_HAVE_D_TYPE
273 #define D_TYPE(de) (de)->d_type
274 #endif
275
276 #ifndef EIO_STRUCT_DIRENT
277 #define EIO_STRUCT_DIRENT struct dirent
278 #endif
279
280#endif
281
282#if HAVE_UTIMES
80# include <utime.h> 283# include <utime.h>
81# include <signal.h>
82# include <dirent.h>
83
84# ifndef EIO_STRUCT_DIRENT
85# define EIO_STRUCT_DIRENT struct dirent
86# endif 284#endif
87 285
286#if HAVE_SYS_SYSCALL_H
287# include <sys/syscall.h>
288#endif
289
290#if HAVE_SYS_PRCTL_H
291# include <sys/prctl.h>
88#endif 292#endif
89 293
90#if HAVE_SENDFILE 294#if HAVE_SENDFILE
91# if __linux 295# if __linux
92# include <sys/sendfile.h> 296# include <sys/sendfile.h>
93# elif __freebsd 297# elif __FreeBSD__ || defined __APPLE__
94# include <sys/socket.h> 298# include <sys/socket.h>
95# include <sys/uio.h> 299# include <sys/uio.h>
96# elif __hpux 300# elif __hpux
97# include <sys/socket.h> 301# include <sys/socket.h>
98# elif __solaris /* not yet */ 302# elif __solaris
99# include <sys/sendfile.h> 303# include <sys/sendfile.h>
100# else 304# else
101# error sendfile support requested but not available 305# error sendfile support requested but not available
102# endif 306# endif
103#endif 307#endif
104 308
105/* number of seconds after which an idle threads exit */ 309#ifndef D_TYPE
106#define IDLE_TIMEOUT 10 310# define D_TYPE(de) 0
311#endif
312#ifndef D_INO
313# define D_INO(de) 0
314#endif
315#ifndef D_NAMLEN
316# define D_NAMLEN(entp) strlen (D_NAME (entp))
317#endif
107 318
108/* used for struct dirent, AIX doesn't provide it */ 319/* used for struct dirent, AIX doesn't provide it */
109#ifndef NAME_MAX 320#ifndef NAME_MAX
110# define NAME_MAX 4096 321# define NAME_MAX 4096
111#endif 322#endif
112 323
324/* used for readlink etc. */
325#ifndef PATH_MAX
326# define PATH_MAX 4096
327#endif
328
113/* buffer size for various temporary buffers */ 329/* buffer size for various temporary buffers */
114#define EIO_BUFSIZE 65536 330#define EIO_BUFSIZE 65536
115 331
116#define dBUF \ 332#define dBUF \
117 char *eio_buf; \
118 ETP_WORKER_LOCK (self); \
119 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 333 char *eio_buf = malloc (EIO_BUFSIZE); \
120 ETP_WORKER_UNLOCK (self); \
121 errno = ENOMEM; \ 334 errno = ENOMEM; \
122 if (!eio_buf) \ 335 if (!eio_buf) \
123 return -1; 336 return -1
124 337
125#define EIO_TICKS ((1000000 + 1023) >> 10) 338#define FUBd \
339 free (eio_buf)
126 340
127/*****************************************************************************/ 341/*****************************************************************************/
128 342
129#if __GNUC__ >= 3 343struct etp_tmpbuf;
130# define expect(expr,value) __builtin_expect ((expr),(value)) 344
345#if _POSIX_VERSION >= 200809L
346 #define HAVE_AT 1
347 #define WD2FD(wd) ((wd) ? (wd)->fd : AT_FDCWD)
348 #ifndef O_SEARCH
349 #define O_SEARCH O_RDONLY
350 #endif
131#else 351#else
132# define expect(expr,value) (expr) 352 #define HAVE_AT 0
353 static const char *wd_expand (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path);
133#endif 354#endif
134 355
135#define expect_false(expr) expect ((expr) != 0, 0) 356struct eio_pwd
136#define expect_true(expr) expect ((expr) != 0, 1) 357{
358#if HAVE_AT
359 int fd;
360#endif
361 int len;
362 char str[1]; /* actually, a 0-terminated canonical path */
363};
137 364
138/*****************************************************************************/ 365/*****************************************************************************/
139 366
140#define ETP_PRI_MIN EIO_PRI_MIN 367#define ETP_PRI_MIN EIO_PRI_MIN
141#define ETP_PRI_MAX EIO_PRI_MAX 368#define ETP_PRI_MAX EIO_PRI_MAX
142 369
370#define ETP_TYPE_QUIT -1
371#define ETP_TYPE_GROUP EIO_GROUP
372
373static void eio_nop_callback (void) { }
374static void (*eio_want_poll_cb)(void) = eio_nop_callback;
375static void (*eio_done_poll_cb)(void) = eio_nop_callback;
376
377#define ETP_WANT_POLL(pool) eio_want_poll_cb ()
378#define ETP_DONE_POLL(pool) eio_done_poll_cb ()
379
143struct etp_worker; 380struct etp_worker;
144
145#define ETP_REQ eio_req 381#define ETP_REQ eio_req
146#define ETP_DESTROY(req) eio_destroy (req) 382#define ETP_DESTROY(req) eio_destroy (req)
147static int eio_finish (eio_req *req); 383static int eio_finish (eio_req *req);
148#define ETP_FINISH(req) eio_finish (req) 384#define ETP_FINISH(req) eio_finish (req)
149static void eio_execute (struct etp_worker *self, eio_req *req); 385static void eio_execute (struct etp_worker *self, eio_req *req);
150#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) 386#define ETP_EXECUTE(wrk,req) eio_execute (wrk, req)
151 387
152#define ETP_WORKER_CLEAR(req) \ 388#include "etp.c"
153 if (wrk->dbuf) \ 389
154 { \ 390static struct etp_pool eio_pool;
155 free (wrk->dbuf); \ 391#define EIO_POOL (&eio_pool)
156 wrk->dbuf = 0; \
157 } \
158 \
159 if (wrk->dirp) \
160 { \
161 closedir (wrk->dirp); \
162 wrk->dirp = 0; \
163 }
164#define ETP_WORKER_COMMON \
165 void *dbuf; \
166 DIR *dirp;
167 392
168/*****************************************************************************/ 393/*****************************************************************************/
169 394
170#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
171
172/* calculcate time difference in ~1/EIO_TICKS of a second */
173static int tvdiff (struct timeval *tv1, struct timeval *tv2)
174{
175 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
176 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
177}
178
179static unsigned int started, idle, wanted = 4;
180
181static void (*want_poll_cb) (void);
182static void (*done_poll_cb) (void);
183
184static unsigned int max_poll_time; /* reslock */
185static unsigned int max_poll_reqs; /* reslock */
186
187static volatile unsigned int nreqs; /* reqlock */
188static volatile unsigned int nready; /* reqlock */
189static volatile unsigned int npending; /* reqlock */
190static volatile unsigned int max_idle = 4;
191
192static mutex_t wrklock = X_MUTEX_INIT;
193static mutex_t reslock = X_MUTEX_INIT;
194static mutex_t reqlock = X_MUTEX_INIT;
195static cond_t reqwait = X_COND_INIT;
196
197#if !HAVE_PREADWRITE
198/*
199 * make our pread/pwrite emulation safe against themselves, but not against
200 * normal read/write by using a mutex. slows down execution a lot,
201 * but that's your problem, not mine.
202 */
203static mutex_t preadwritelock = X_MUTEX_INIT;
204#endif
205
206typedef struct etp_worker
207{
208 /* locked by wrklock */
209 struct etp_worker *prev, *next;
210
211 thread_t tid;
212
213 /* locked by reslock, reqlock or wrklock */
214 ETP_REQ *req; /* currently processed request */
215
216 ETP_WORKER_COMMON
217} etp_worker;
218
219static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */
220
221#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
222#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
223
224/* worker threads management */
225
226static void etp_worker_clear (etp_worker *wrk)
227{
228 ETP_WORKER_CLEAR (wrk);
229}
230
231static void etp_worker_free (etp_worker *wrk)
232{
233 wrk->next->prev = wrk->prev;
234 wrk->prev->next = wrk->next;
235
236 free (wrk);
237}
238
239static unsigned int etp_nreqs (void)
240{
241 int retval;
242 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
243 retval = nreqs;
244 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
245 return retval;
246}
247
248static unsigned int etp_nready (void)
249{
250 unsigned int retval;
251
252 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
253 retval = nready;
254 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
255
256 return retval;
257}
258
259static unsigned int etp_npending (void)
260{
261 unsigned int retval;
262
263 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
264 retval = npending;
265 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
266
267 return retval;
268}
269
270static unsigned int etp_nthreads (void)
271{
272 unsigned int retval;
273
274 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
275 retval = started;
276 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
277
278 return retval;
279}
280
281/*
282 * a somewhat faster data structure might be nice, but
283 * with 8 priorities this actually needs <20 insns
284 * per shift, the most expensive operation.
285 */
286typedef struct {
287 ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */
288 int size;
289} etp_reqq;
290
291static etp_reqq req_queue;
292static etp_reqq res_queue;
293
294static int reqq_push (etp_reqq *q, ETP_REQ *req)
295{
296 int pri = req->pri;
297 req->next = 0;
298
299 if (q->qe[pri])
300 {
301 q->qe[pri]->next = req;
302 q->qe[pri] = req;
303 }
304 else
305 q->qe[pri] = q->qs[pri] = req;
306
307 return q->size++;
308}
309
310static ETP_REQ *reqq_shift (etp_reqq *q)
311{
312 int pri;
313
314 if (!q->size)
315 return 0;
316
317 --q->size;
318
319 for (pri = ETP_NUM_PRI; pri--; )
320 {
321 eio_req *req = q->qs[pri];
322
323 if (req)
324 {
325 if (!(q->qs[pri] = (eio_req *)req->next))
326 q->qe[pri] = 0;
327
328 return req;
329 }
330 }
331
332 abort ();
333}
334
335static void etp_atfork_prepare (void)
336{
337 X_LOCK (wrklock);
338 X_LOCK (reqlock);
339 X_LOCK (reslock);
340#if !HAVE_PREADWRITE
341 X_LOCK (preadwritelock);
342#endif
343}
344
345static void etp_atfork_parent (void)
346{
347#if !HAVE_PREADWRITE
348 X_UNLOCK (preadwritelock);
349#endif
350 X_UNLOCK (reslock);
351 X_UNLOCK (reqlock);
352 X_UNLOCK (wrklock);
353}
354
355static void etp_atfork_child (void)
356{
357 ETP_REQ *prv;
358
359 while ((prv = reqq_shift (&req_queue)))
360 ETP_DESTROY (prv);
361
362 while ((prv = reqq_shift (&res_queue)))
363 ETP_DESTROY (prv);
364
365 while (wrk_first.next != &wrk_first)
366 {
367 etp_worker *wrk = wrk_first.next;
368
369 if (wrk->req)
370 ETP_DESTROY (wrk->req);
371
372 etp_worker_clear (wrk);
373 etp_worker_free (wrk);
374 }
375
376 started = 0;
377 idle = 0;
378 nreqs = 0;
379 nready = 0;
380 npending = 0;
381
382 etp_atfork_parent ();
383}
384
385static void 395static void
386etp_once_init (void)
387{
388 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
389}
390
391static int
392etp_init (void (*want_poll)(void), void (*done_poll)(void))
393{
394 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
395
396 pthread_once (&doinit, etp_once_init);
397
398 want_poll_cb = want_poll;
399 done_poll_cb = done_poll;
400
401 return 0;
402}
403
404X_THREAD_PROC (etp_proc);
405
406static void etp_start_thread (void)
407{
408 etp_worker *wrk = calloc (1, sizeof (etp_worker));
409
410 /*TODO*/
411 assert (("unable to allocate worker thread data", wrk));
412
413 X_LOCK (wrklock);
414
415 if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
416 {
417 wrk->prev = &wrk_first;
418 wrk->next = wrk_first.next;
419 wrk_first.next->prev = wrk;
420 wrk_first.next = wrk;
421 ++started;
422 }
423 else
424 free (wrk);
425
426 X_UNLOCK (wrklock);
427}
428
429static void etp_maybe_start_thread (void)
430{
431 if (expect_true (etp_nthreads () >= wanted))
432 return;
433
434 /* todo: maybe use idle here, but might be less exact */
435 if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
436 return;
437
438 etp_start_thread ();
439}
440
441static void etp_end_thread (void)
442{
443 eio_req *req = calloc (1, sizeof (eio_req));
444
445 req->type = -1;
446 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
447
448 X_LOCK (reqlock);
449 reqq_push (&req_queue, req);
450 X_COND_SIGNAL (reqwait);
451 X_UNLOCK (reqlock);
452
453 X_LOCK (wrklock);
454 --started;
455 X_UNLOCK (wrklock);
456}
457
458static int etp_poll (void)
459{
460 unsigned int maxreqs;
461 unsigned int maxtime;
462 struct timeval tv_start, tv_now;
463
464 X_LOCK (reslock);
465 maxreqs = max_poll_reqs;
466 maxtime = max_poll_time;
467 X_UNLOCK (reslock);
468
469 if (maxtime)
470 gettimeofday (&tv_start, 0);
471
472 for (;;)
473 {
474 ETP_REQ *req;
475
476 etp_maybe_start_thread ();
477
478 X_LOCK (reslock);
479 req = reqq_shift (&res_queue);
480
481 if (req)
482 {
483 --npending;
484
485 if (!res_queue.size && done_poll_cb)
486 done_poll_cb ();
487 }
488
489 X_UNLOCK (reslock);
490
491 if (!req)
492 return 0;
493
494 X_LOCK (reqlock);
495 --nreqs;
496 X_UNLOCK (reqlock);
497
498 if (expect_false (req->type == EIO_GROUP && req->size))
499 {
500 req->int1 = 1; /* mark request as delayed */
501 continue;
502 }
503 else
504 {
505 int res = ETP_FINISH (req);
506 if (expect_false (res))
507 return res;
508 }
509
510 if (expect_false (maxreqs && !--maxreqs))
511 break;
512
513 if (maxtime)
514 {
515 gettimeofday (&tv_now, 0);
516
517 if (tvdiff (&tv_start, &tv_now) >= maxtime)
518 break;
519 }
520 }
521
522 errno = EAGAIN;
523 return -1;
524}
525
526static void etp_cancel (ETP_REQ *req)
527{
528 X_LOCK (wrklock);
529 req->flags |= EIO_FLAG_CANCELLED;
530 X_UNLOCK (wrklock);
531
532 eio_grp_cancel (req);
533}
534
535static void etp_submit (ETP_REQ *req)
536{
537 req->pri -= ETP_PRI_MIN;
538
539 if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
540 if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
541
542 if (expect_false (req->type == EIO_GROUP))
543 {
544 /* I hope this is worth it :/ */
545 X_LOCK (reqlock);
546 ++nreqs;
547 X_UNLOCK (reqlock);
548
549 X_LOCK (reslock);
550
551 ++npending;
552
553 if (!reqq_push (&res_queue, req) && want_poll_cb)
554 want_poll_cb ();
555
556 X_UNLOCK (reslock);
557 }
558 else
559 {
560 X_LOCK (reqlock);
561 ++nreqs;
562 ++nready;
563 reqq_push (&req_queue, req);
564 X_COND_SIGNAL (reqwait);
565 X_UNLOCK (reqlock);
566
567 etp_maybe_start_thread ();
568 }
569}
570
571static void etp_set_max_poll_time (double nseconds)
572{
573 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
574 max_poll_time = nseconds;
575 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
576}
577
578static void etp_set_max_poll_reqs (unsigned int maxreqs)
579{
580 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
581 max_poll_reqs = maxreqs;
582 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
583}
584
585static void etp_set_max_idle (unsigned int nthreads)
586{
587 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
588 max_idle = nthreads <= 0 ? 1 : nthreads;
589 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
590}
591
592static void etp_set_min_parallel (unsigned int nthreads)
593{
594 if (wanted < nthreads)
595 wanted = nthreads;
596}
597
598static void etp_set_max_parallel (unsigned int nthreads)
599{
600 if (wanted > nthreads)
601 wanted = nthreads;
602
603 while (started > wanted)
604 etp_end_thread ();
605}
606
607/*****************************************************************************/
608
609static void grp_try_feed (eio_req *grp) 396grp_try_feed (eio_req *grp)
610{ 397{
611 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 398 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
612 { 399 {
613 grp->flags &= ~EIO_FLAG_GROUPADD; 400 grp->flags &= ~ETP_FLAG_GROUPADD;
614 401
615 EIO_FEED (grp); 402 EIO_FEED (grp);
616 403
617 /* stop if no progress has been made */ 404 /* stop if no progress has been made */
618 if (!(grp->flags & EIO_FLAG_GROUPADD)) 405 if (!(grp->flags & ETP_FLAG_GROUPADD))
619 { 406 {
620 grp->feed = 0; 407 grp->feed = 0;
621 break; 408 break;
622 } 409 }
623 } 410 }
624} 411}
625 412
413static int
626static int grp_dec (eio_req *grp) 414grp_dec (eio_req *grp)
627{ 415{
628 --grp->size; 416 --grp->size;
629 417
630 /* call feeder, if applicable */ 418 /* call feeder, if applicable */
631 grp_try_feed (grp); 419 grp_try_feed (grp);
632 420
633 /* finish, if done */ 421 /* finish, if done */
634 if (!grp->size && grp->int1) 422 if (!grp->size && grp->flags & ETP_FLAG_DELAYED)
635 return eio_finish (grp); 423 return eio_finish (grp);
636 else 424 else
637 return 0; 425 return 0;
638} 426}
639 427
428static void
640void eio_destroy (eio_req *req) 429eio_destroy (eio_req *req)
641{ 430{
642 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); 431 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
643 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); 432 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
644 433
645 EIO_DESTROY (req); 434 EIO_DESTROY (req);
646} 435}
647 436
437static int
648static int eio_finish (eio_req *req) 438eio_finish (eio_req *req)
649{ 439{
650 int res = EIO_FINISH (req); 440 int res = EIO_FINISH (req);
651 441
652 if (req->grp) 442 if (req->grp)
653 { 443 {
… …
661 if (grp->grp_first == req) 451 if (grp->grp_first == req)
662 grp->grp_first = req->grp_next; 452 grp->grp_first = req->grp_next;
663 453
664 res2 = grp_dec (grp); 454 res2 = grp_dec (grp);
665 455
666 if (!res && res2) 456 if (!res)
667 res = res2; 457 res = res2;
668 } 458 }
669 459
670 eio_destroy (req); 460 eio_destroy (req);
671 461
672 return res; 462 return res;
673} 463}
674 464
465void
675void eio_grp_cancel (eio_req *grp) 466eio_grp_cancel (eio_req *grp)
676{ 467{
677 for (grp = grp->grp_first; grp; grp = grp->grp_next) 468 etp_grp_cancel (EIO_POOL, grp);
678 eio_cancel (grp);
679} 469}
680 470
471void
681void eio_cancel (eio_req *req) 472eio_cancel (eio_req *req)
682{ 473{
683 etp_cancel (req); 474 etp_cancel (EIO_POOL, req);
684} 475}
685 476
477void
686void eio_submit (eio_req *req) 478eio_submit (eio_req *req)
687{ 479{
688 etp_submit (req); 480 etp_submit (EIO_POOL, req);
689} 481}
690 482
691unsigned int eio_nreqs (void) 483unsigned int
484eio_nreqs (void)
692{ 485{
693 return etp_nreqs (); 486 return etp_nreqs (EIO_POOL);
694} 487}
695 488
696unsigned int eio_nready (void) 489unsigned int
490eio_nready (void)
697{ 491{
698 return etp_nready (); 492 return etp_nready (EIO_POOL);
699} 493}
700 494
701unsigned int eio_npending (void) 495unsigned int
496eio_npending (void)
702{ 497{
703 return etp_npending (); 498 return etp_npending (EIO_POOL);
704} 499}
705 500
706unsigned int eio_nthreads (void) 501unsigned int ecb_cold
502eio_nthreads (void)
707{ 503{
708 return etp_nthreads (); 504 return etp_nthreads (EIO_POOL);
709} 505}
710 506
507void ecb_cold
711void eio_set_max_poll_time (double nseconds) 508eio_set_max_poll_time (double nseconds)
712{ 509{
713 etp_set_max_poll_time (nseconds); 510 etp_set_max_poll_time (EIO_POOL, nseconds);
714} 511}
715 512
513void ecb_cold
716void eio_set_max_poll_reqs (unsigned int maxreqs) 514eio_set_max_poll_reqs (unsigned int maxreqs)
717{ 515{
718 etp_set_max_poll_reqs (maxreqs); 516 etp_set_max_poll_reqs (EIO_POOL, maxreqs);
719} 517}
720 518
519void ecb_cold
721void eio_set_max_idle (unsigned int nthreads) 520eio_set_max_idle (unsigned int nthreads)
722{ 521{
723 etp_set_max_idle (nthreads); 522 etp_set_max_idle (EIO_POOL, nthreads);
724} 523}
725 524
525void ecb_cold
526eio_set_idle_timeout (unsigned int seconds)
527{
528 etp_set_idle_timeout (EIO_POOL, seconds);
529}
530
531void ecb_cold
726void eio_set_min_parallel (unsigned int nthreads) 532eio_set_min_parallel (unsigned int nthreads)
727{ 533{
728 etp_set_min_parallel (nthreads); 534 etp_set_min_parallel (EIO_POOL, nthreads);
729} 535}
730 536
537void ecb_cold
731void eio_set_max_parallel (unsigned int nthreads) 538eio_set_max_parallel (unsigned int nthreads)
732{ 539{
733 etp_set_max_parallel (nthreads); 540 etp_set_max_parallel (EIO_POOL, nthreads);
734} 541}
735 542
736int eio_poll (void) 543int eio_poll (void)
737{ 544{
738 return etp_poll (); 545 return etp_poll (EIO_POOL);
739} 546}
740 547
741/*****************************************************************************/ 548/*****************************************************************************/
742/* work around various missing functions */ 549/* work around various missing functions */
743 550
744#if !HAVE_PREADWRITE
745# undef pread
746# undef pwrite
747# define pread eio__pread
748# define pwrite eio__pwrite
749
750static ssize_t
751eio__pread (int fd, void *buf, size_t count, off_t offset)
752{
753 ssize_t res;
754 off_t ooffset;
755
756 X_LOCK (preadwritelock);
757 ooffset = lseek (fd, 0, SEEK_CUR);
758 lseek (fd, offset, SEEK_SET);
759 res = read (fd, buf, count);
760 lseek (fd, ooffset, SEEK_SET);
761 X_UNLOCK (preadwritelock);
762
763 return res;
764}
765
766static ssize_t
767eio__pwrite (int fd, void *buf, size_t count, off_t offset)
768{
769 ssize_t res;
770 off_t ooffset;
771
772 X_LOCK (preadwritelock);
773 ooffset = lseek (fd, 0, SEEK_CUR);
774 lseek (fd, offset, SEEK_SET);
775 res = write (fd, buf, count);
776 lseek (fd, ooffset, SEEK_SET);
777 X_UNLOCK (preadwritelock);
778
779 return res;
780}
781#endif
782
783#ifndef HAVE_FUTIMES 551#ifndef HAVE_UTIMES
784 552
785# undef utimes 553# undef utimes
786# undef futimes
787# define utimes(path,times) eio__utimes (path, times) 554# define utimes(path,times) eio__utimes (path, times)
788# define futimes(fd,times) eio__futimes (fd, times)
789 555
790static int 556static int
791eio__utimes (const char *filename, const struct timeval times[2]) 557eio__utimes (const char *filename, const struct timeval times[2])
792{ 558{
793 if (times) 559 if (times)
… …
801 } 567 }
802 else 568 else
803 return utime (filename, 0); 569 return utime (filename, 0);
804} 570}
805 571
572#endif
573
574#ifndef HAVE_FUTIMES
575
576# undef futimes
577# define futimes(fd,times) eio__futimes (fd, times)
578
579static int
806static int eio__futimes (int fd, const struct timeval tv[2]) 580eio__futimes (int fd, const struct timeval tv[2])
807{ 581{
808 errno = ENOSYS; 582 errno = ENOSYS;
809 return -1; 583 return -1;
810} 584}
811 585
… …
814#if !HAVE_FDATASYNC 588#if !HAVE_FDATASYNC
815# undef fdatasync 589# undef fdatasync
816# define fdatasync(fd) fsync (fd) 590# define fdatasync(fd) fsync (fd)
817#endif 591#endif
818 592
593static int
594eio__syncfs (int fd)
595{
596 int res;
597
598#if HAVE_SYS_SYNCFS
599 res = (int)syscall (__NR_syncfs, (int)(fd));
600#else
601 res = EIO_ENOSYS ();
602#endif
603
604 if (res < 0 && errno == ENOSYS && fd >= 0)
605 sync ();
606
607 return res;
608}
609
819/* sync_file_range always needs emulation */ 610/* sync_file_range always needs emulation */
820int 611static int
821eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags) 612eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
822{ 613{
823#if HAVE_SYNC_FILE_RANGE 614#if HAVE_SYNC_FILE_RANGE
824 int res; 615 int res;
825 616
… …
838 if (!res || errno != ENOSYS) 629 if (!res || errno != ENOSYS)
839 return res; 630 return res;
840#endif 631#endif
841 632
842 /* even though we could play tricks with the flags, it's better to always 633 /* even though we could play tricks with the flags, it's better to always
843 * call fdatasync, as thta matches the expectation of it's users best */ 634 * call fdatasync, as that matches the expectation of its users best */
844 return fdatasync (fd); 635 return fdatasync (fd);
636}
637
638static int
639eio__fallocate (int fd, int mode, off_t offset, size_t len)
640{
641#if HAVE_LINUX_FALLOCATE
642 return fallocate (fd, mode, offset, len);
643#else
644 return EIO_ENOSYS ();
645#endif
845} 646}
846 647
847#if !HAVE_READAHEAD 648#if !HAVE_READAHEAD
848# undef readahead 649# undef readahead
849# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 650# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
850 651
851static ssize_t 652static eio_ssize_t
852eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) 653eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
853{ 654{
854 size_t todo = count; 655 size_t todo = count;
855 dBUF; 656 dBUF;
856 657
… …
861 pread (fd, eio_buf, len, offset); 662 pread (fd, eio_buf, len, offset);
862 offset += len; 663 offset += len;
863 todo -= len; 664 todo -= len;
864 } 665 }
865 666
866 errno = 0; 667 FUBd;
668
669 /* linux's readahead basically only fails for EBADF or EINVAL (not mmappable) */
670 /* but not for e.g. EIO or eof, so we also never fail */
867 return count; 671 return 0;
868} 672}
869 673
870#endif 674#endif
871 675
872/* sendfile always needs emulation */ 676/* sendfile always needs emulation */
873static ssize_t 677static eio_ssize_t
874eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) 678eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
875{ 679{
680 eio_ssize_t written = 0;
876 ssize_t res; 681 eio_ssize_t res;
877 682
878 if (!count) 683 if (!count)
879 return 0; 684 return 0;
880 685
686 for (;;)
687 {
688#ifdef __APPLE__
689# undef HAVE_SENDFILE /* broken, as everything on os x */
690#endif
881#if HAVE_SENDFILE 691#if HAVE_SENDFILE
882# if __linux 692# if __linux
693 off_t soffset = offset;
883 res = sendfile (ofd, ifd, &offset, count); 694 res = sendfile (ofd, ifd, &soffset, count);
884 695
885# elif __freebsd 696# elif __FreeBSD__
886 /* 697 /*
887 * Of course, the freebsd sendfile is a dire hack with no thoughts 698 * Of course, the freebsd sendfile is a dire hack with no thoughts
888 * wasted on making it similar to other I/O functions. 699 * wasted on making it similar to other I/O functions.
889 */ 700 */
890 {
891 off_t sbytes; 701 off_t sbytes;
892 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 702 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
893 703
894 if (res < 0 && sbytes) 704 #if 0 /* according to the manpage, this is correct, but broken behaviour */
895 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 705 /* freebsd' sendfile will return 0 on success */
706 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
707 /* not on e.g. EIO or EPIPE - sounds broken */
708 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
896 res = sbytes; 709 res = sbytes;
897 } 710 #endif
711
712 /* according to source inspection, this is correct, and useful behaviour */
713 if (sbytes)
714 res = sbytes;
715
716# elif defined __APPLE__
717 off_t sbytes = count;
718 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
719
720 /* according to the manpage, sbytes is always valid */
721 if (sbytes)
722 res = sbytes;
898 723
899# elif __hpux 724# elif __hpux
900 res = sendfile (ofd, ifd, offset, count, 0, 0); 725 res = sendfile (ofd, ifd, offset, count, 0, 0);
901 726
902# elif __solaris 727# elif __solaris
903 {
904 struct sendfilevec vec; 728 struct sendfilevec vec;
905 size_t sbytes; 729 size_t sbytes;
906 730
907 vec.sfv_fd = ifd; 731 vec.sfv_fd = ifd;
908 vec.sfv_flag = 0; 732 vec.sfv_flag = 0;
909 vec.sfv_off = offset; 733 vec.sfv_off = offset;
910 vec.sfv_len = count; 734 vec.sfv_len = count;
911 735
912 res = sendfilev (ofd, &vec, 1, &sbytes); 736 res = sendfilev (ofd, &vec, 1, &sbytes);
913 737
914 if (res < 0 && sbytes) 738 if (res < 0 && sbytes)
915 res = sbytes; 739 res = sbytes;
916 }
917 740
918# endif 741# endif
742
743#elif defined (_WIN32) && 0
744 /* does not work, just for documentation of what would need to be done */
745 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
746 /* libeio guarantees that the file offset does not change, and windows */
747 /* has no way to get an independent handle to the same file description */
748 HANDLE h = TO_SOCKET (ifd);
749 SetFilePointer (h, offset, 0, FILE_BEGIN);
750 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
751
919#else 752#else
920 res = -1; 753 res = EIO_ENOSYS ();
921 errno = ENOSYS;
922#endif 754#endif
923 755
756 /* we assume sendfile can copy at least 128mb in one go */
757 if (res <= 128 * 1024 * 1024)
758 {
759 if (res > 0)
760 written += res;
761
762 if (written)
763 return written;
764
765 break;
766 }
767 else
768 {
769 /* if we requested more, then probably the kernel was lazy */
770 written += res;
771 offset += res;
772 count -= res;
773
774 if (!count)
775 return written;
776 }
777 }
778
924 if (res < 0 779 if (res < 0
925 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 780 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
781 /* BSDs */
782#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
783 || errno == ENOTSUP
784#endif
785#ifdef EOPNOTSUPP /* windows */
786 || errno == EOPNOTSUPP /* BSDs */
787#endif
926#if __solaris 788#if __solaris
927 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 789 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
928#endif 790#endif
929 ) 791 )
930 ) 792 )
… …
934 796
935 res = 0; 797 res = 0;
936 798
937 while (count) 799 while (count)
938 { 800 {
939 ssize_t cnt; 801 eio_ssize_t cnt;
940 802
941 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 803 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
942 804
943 if (cnt <= 0) 805 if (cnt <= 0)
944 { 806 {
… …
956 818
957 offset += cnt; 819 offset += cnt;
958 res += cnt; 820 res += cnt;
959 count -= cnt; 821 count -= cnt;
960 } 822 }
823
824 FUBd;
961 } 825 }
962 826
963 return res; 827 return res;
828}
829
830#ifdef PAGESIZE
831# define eio_pagesize() PAGESIZE
832#else
833static intptr_t
834eio_pagesize (void)
835{
836 static intptr_t page;
837
838 if (!page)
839 page = sysconf (_SC_PAGESIZE);
840
841 return page;
842}
843#endif
844
845static void
846eio_page_align (void **addr, size_t *length)
847{
848 intptr_t mask = eio_pagesize () - 1;
849
850 /* round down addr */
851 intptr_t adj = mask & (intptr_t)*addr;
852
853 *addr = (void *)((intptr_t)*addr - adj);
854 *length += adj;
855
856 /* round up length */
857 *length = (*length + mask) & ~mask;
858}
859
860#if !_POSIX_MEMLOCK
861# define eio__mlockall(a) EIO_ENOSYS ()
862#else
863
864static int
865eio__mlockall (int flags)
866{
867 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
868 extern int mallopt (int, int);
869 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
870 #endif
871
872 if (EIO_MCL_CURRENT != MCL_CURRENT
873 || EIO_MCL_FUTURE != MCL_FUTURE)
874 {
875 flags = 0
876 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
877 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
878 }
879
880 return mlockall (flags);
881}
882#endif
883
884#if !_POSIX_MEMLOCK_RANGE
885# define eio__mlock(a,b) EIO_ENOSYS ()
886#else
887
888static int
889eio__mlock (void *addr, size_t length)
890{
891 eio_page_align (&addr, &length);
892
893 return mlock (addr, length);
894}
895
896#endif
897
898#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
899# define eio__msync(a,b,c) EIO_ENOSYS ()
900#else
901
902static int
903eio__msync (void *mem, size_t len, int flags)
904{
905 eio_page_align (&mem, &len);
906
907 if (EIO_MS_ASYNC != MS_SYNC
908 || EIO_MS_INVALIDATE != MS_INVALIDATE
909 || EIO_MS_SYNC != MS_SYNC)
910 {
911 flags = 0
912 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
913 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
914 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
915 }
916
917 return msync (mem, len, flags);
918}
919
920#endif
921
922static int
923eio__mtouch (eio_req *req)
924{
925 void *mem = req->ptr2;
926 size_t len = req->size;
927 int flags = req->int1;
928
929 eio_page_align (&mem, &len);
930
931 {
932 intptr_t addr = (intptr_t)mem;
933 intptr_t end = addr + len;
934 intptr_t page = eio_pagesize ();
935
936 if (addr < end)
937 if (flags & EIO_MT_MODIFY) /* modify */
938 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
939 else
940 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
941 }
942
943 return 0;
944}
945
946/*****************************************************************************/
947/* requests implemented outside eio_execute, because they are so large */
948
949static void
950eio__lseek (eio_req *req)
951{
952 /* this usually gets optimised away completely, or your compiler sucks, */
953 /* or the whence constants really are not 0, 1, 2 */
954 int whence = req->int2 == EIO_SEEK_SET ? SEEK_SET
955 : req->int2 == EIO_SEEK_CUR ? SEEK_CUR
956 : req->int2 == EIO_SEEK_END ? SEEK_END
957 : req->int2;
958
959 req->offs = lseek (req->int1, req->offs, whence);
960 req->result = req->offs == (off_t)-1 ? -1 : 0;
961}
962
963/* result will always end up in tmpbuf, there is always space for adding a 0-byte */
964static int
965eio__realpath (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
966{
967 char *res;
968 const char *rel = path;
969 char *tmp1, *tmp2;
970#if SYMLOOP_MAX > 32
971 int symlinks = SYMLOOP_MAX;
972#else
973 int symlinks = 32;
974#endif
975
976 errno = EINVAL;
977 if (!rel)
978 return -1;
979
980 errno = ENOENT;
981 if (!*rel)
982 return -1;
983
984 res = etp_tmpbuf_get (tmpbuf, PATH_MAX * 3);
985#ifdef _WIN32
986 if (_access (rel, 4) != 0)
987 return -1;
988
989 symlinks = GetFullPathName (rel, PATH_MAX * 3, res, 0);
990
991 errno = ENAMETOOLONG;
992 if (symlinks >= PATH_MAX * 3)
993 return -1;
994
995 errno = EIO;
996 if (symlinks <= 0)
997 return -1;
998
999 return symlinks;
1000
1001#else
1002 tmp1 = res + PATH_MAX;
1003 tmp2 = tmp1 + PATH_MAX;
1004
1005#if 0 /* disabled, the musl way to do things is just too racy */
1006#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1007 /* on linux we may be able to ask the kernel */
1008 {
1009 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1010
1011 if (fd >= 0)
1012 {
1013 sprintf (tmp1, "/proc/self/fd/%d", fd);
1014 req->result = readlink (tmp1, res, PATH_MAX);
1015 /* here we should probably stat the open file and the disk file, to make sure they still match */
1016 close (fd);
1017
1018 if (req->result > 0)
1019 goto done;
1020 }
1021 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1022 return -1;
1023 }
1024#endif
1025#endif
1026
1027 if (*rel != '/')
1028 {
1029 int len;
1030
1031 errno = ENOENT;
1032 if (wd == EIO_INVALID_WD)
1033 return -1;
1034
1035 if (wd == EIO_CWD)
1036 {
1037 if (!getcwd (res, PATH_MAX))
1038 return -1;
1039
1040 len = strlen (res);
1041 }
1042 else
1043 memcpy (res, wd->str, len = wd->len);
1044
1045 if (res [1]) /* only use if not / */
1046 res += len;
1047 }
1048
1049 while (*rel)
1050 {
1051 eio_ssize_t len, linklen;
1052 const char *beg = rel;
1053
1054 while (*rel && *rel != '/')
1055 ++rel;
1056
1057 len = rel - beg;
1058
1059 if (!len) /* skip slashes */
1060 {
1061 ++rel;
1062 continue;
1063 }
1064
1065 if (beg [0] == '.')
1066 {
1067 if (len == 1)
1068 continue; /* . - nop */
1069
1070 if (beg [1] == '.' && len == 2)
1071 {
1072 /* .. - back up one component, if possible */
1073
1074 while (res != tmpbuf->ptr)
1075 if (*--res == '/')
1076 break;
1077
1078 continue;
1079 }
1080 }
1081
1082 errno = ENAMETOOLONG;
1083 if (res + 1 + len + 1 >= tmp1)
1084 return -1;
1085
1086 /* copy one component */
1087 *res = '/';
1088 memcpy (res + 1, beg, len);
1089
1090 /* zero-terminate, for readlink */
1091 res [len + 1] = 0;
1092
1093 /* now check if it's a symlink */
1094 linklen = readlink (tmpbuf->ptr, tmp1, PATH_MAX);
1095
1096 if (linklen < 0)
1097 {
1098 if (errno != EINVAL)
1099 return -1;
1100
1101 /* it's a normal directory. hopefully */
1102 res += len + 1;
1103 }
1104 else
1105 {
1106 /* yay, it was a symlink - build new path in tmp2 */
1107 int rellen = strlen (rel);
1108
1109 errno = ENAMETOOLONG;
1110 if (linklen + 1 + rellen >= PATH_MAX)
1111 return -1;
1112
1113 errno = ELOOP;
1114 if (!--symlinks)
1115 return -1;
1116
1117 if (*tmp1 == '/')
1118 res = tmpbuf->ptr; /* symlink resolves to an absolute path */
1119
1120 /* we need to be careful, as rel might point into tmp2 already */
1121 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1122 tmp2 [linklen] = '/';
1123 memcpy (tmp2, tmp1, linklen);
1124
1125 rel = tmp2;
1126 }
1127 }
1128
1129 /* special case for the lone root path */
1130 if (res == tmpbuf->ptr)
1131 *res++ = '/';
1132
1133 return res - (char *)tmpbuf->ptr;
1134#endif
1135}
1136
1137static signed char
1138eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1139{
1140 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1141 : a->inode < b->inode ? -1
1142 : a->inode > b->inode ? 1
1143 : 0;
1144}
1145
1146#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1147
1148#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1149#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1150
1151static void
1152eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1153{
1154 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1155 unsigned char *bit = bits;
1156
1157 assert (CHAR_BIT == 8);
1158 assert (sizeof (eio_dirent) * 8 < 256);
1159 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1160 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1161
1162 if (size <= EIO_SORT_FAST)
1163 return;
1164
1165 /* first prepare an array of bits to test in our radix sort */
1166 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1167 /* inode_bits must contain all inodes ORed together */
1168 /* which is used to skip bits that are 0 everywhere, which is very common */
1169 {
1170 eio_ino_t endianness;
1171 int i, j;
1172
1173 /* we store the byte offset of byte n into byte n of "endianness" */
1174 for (i = 0; i < sizeof (eio_ino_t); ++i)
1175 ((unsigned char *)&endianness)[i] = i;
1176
1177 *bit++ = 0;
1178
1179 for (i = 0; i < sizeof (eio_ino_t); ++i)
1180 {
1181 /* shifting off the byte offsets out of "endianness" */
1182 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1183 endianness >>= 8;
1184
1185 for (j = 0; j < 8; ++j)
1186 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1187 *bit++ = offs + j;
1188 }
1189
1190 for (j = 0; j < 8; ++j)
1191 if (score_bits & (1 << j))
1192 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1193 }
1194
1195 /* now actually do the sorting (a variant of MSD radix sort) */
1196 {
1197 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1198 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1199 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1200 int stk_idx = 0;
1201
1202 base_stk [stk_idx] = dents;
1203 end_stk [stk_idx] = dents + size;
1204 bit_stk [stk_idx] = bit - 1;
1205
1206 do
1207 {
1208 base = base_stk [stk_idx];
1209 end = end_stk [stk_idx];
1210 bit = bit_stk [stk_idx];
1211
1212 for (;;)
1213 {
1214 unsigned char O = *bit >> 3;
1215 unsigned char M = 1 << (*bit & 7);
1216
1217 eio_dirent *a = base;
1218 eio_dirent *b = end;
1219
1220 if (b - a < EIO_SORT_CUTOFF)
1221 break;
1222
1223 /* now bit-partition the array on the bit */
1224 /* this ugly asymmetric loop seems to perform much better than typical */
1225 /* partition algos found in the literature */
1226 do
1227 if (!(((unsigned char *)a)[O] & M))
1228 ++a;
1229 else if (!(((unsigned char *)--b)[O] & M))
1230 {
1231 eio_dirent tmp = *a; *a = *b; *b = tmp;
1232 ++a;
1233 }
1234 while (b > a);
1235
1236 /* next bit, or stop, if no bits left in this path */
1237 if (!*--bit)
1238 break;
1239
1240 base_stk [stk_idx] = a;
1241 end_stk [stk_idx] = end;
1242 bit_stk [stk_idx] = bit;
1243 ++stk_idx;
1244
1245 end = a;
1246 }
1247 }
1248 while (stk_idx--);
1249 }
1250}
1251
1252static void
1253eio_dent_insertion_sort (eio_dirent *dents, int size)
1254{
1255 /* first move the smallest element to the front, to act as a sentinel */
1256 {
1257 int i;
1258 eio_dirent *min = dents;
1259
1260 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1261 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1262 if (EIO_DENT_CMP (dents [i], <, *min))
1263 min = &dents [i];
1264
1265 /* swap elements 0 and j (minimum) */
1266 {
1267 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1268 }
1269 }
1270
1271 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1272 {
1273 eio_dirent *i, *j;
1274
1275 for (i = dents + 1; i < dents + size; ++i)
1276 {
1277 eio_dirent value = *i;
1278
1279 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1280 j [1] = j [0];
1281
1282 j [1] = value;
1283 }
1284 }
1285}
1286
1287static void
1288eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1289{
1290 if (size <= 1)
1291 return; /* our insertion sort relies on size > 0 */
1292
1293 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1294 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1295 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1296
1297 /* use an insertion sort at the end, or for small arrays, */
1298 /* as insertion sort is more efficient for small partitions */
1299 eio_dent_insertion_sort (dents, size);
964} 1300}
965 1301
966/* read a full directory */ 1302/* read a full directory */
967static void 1303static void
968eio__scandir (eio_req *req, etp_worker *self) 1304eio__scandir (eio_req *req, etp_worker *self)
969{ 1305{
1306 char *name, *names;
1307 int namesalloc = 4096 - sizeof (void *) * 4;
1308 int namesoffs = 0;
1309 int flags = req->int1;
1310 eio_dirent *dents = 0;
1311 int dentalloc = 128;
1312 int dentoffs = 0;
1313 eio_ino_t inode_bits = 0;
1314#ifdef _WIN32
1315 HANDLE dirp;
1316 WIN32_FIND_DATA entp;
1317#else
970 DIR *dirp; 1318 DIR *dirp;
971 EIO_STRUCT_DIRENT *entp; 1319 EIO_STRUCT_DIRENT *entp;
972 char *name, *names; 1320#endif
973 int memlen = 4096;
974 int memofs = 0;
975 int res = 0;
976 1321
977 X_LOCK (wrklock); 1322 req->result = -1;
978 /* the corresponding closedir is in ETP_WORKER_CLEAR */
979 self->dirp = dirp = opendir (req->ptr1);
980 req->flags |= EIO_FLAG_PTR2_FREE;
981 req->ptr2 = names = malloc (memlen);
982 X_UNLOCK (wrklock);
983 1323
984 if (dirp && names) 1324 if (!(flags & EIO_READDIR_DENTS))
985 for (;;) 1325 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1326
1327#ifdef _WIN32
1328 {
1329 int len = strlen ((const char *)req->ptr1);
1330 char *path = malloc (MAX_PATH);
1331 const char *fmt;
1332 const char *reqpath = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1333
1334 if (!len)
1335 fmt = "./*";
1336 else if (reqpath[len - 1] == '/' || reqpath[len - 1] == '\\')
1337 fmt = "%s*";
1338 else
1339 fmt = "%s/*";
1340
1341 _snprintf (path, MAX_PATH, fmt, reqpath);
1342 dirp = FindFirstFile (path, &entp);
1343 free (path);
1344
1345 if (dirp == INVALID_HANDLE_VALUE)
1346 {
1347 /* should steal _dosmaperr */
1348 switch (GetLastError ())
1349 {
1350 case ERROR_FILE_NOT_FOUND:
1351 req->result = 0;
1352 break;
1353
1354 case ERROR_INVALID_NAME:
1355 case ERROR_PATH_NOT_FOUND:
1356 case ERROR_NO_MORE_FILES:
1357 errno = ENOENT;
1358 break;
1359
1360 case ERROR_NOT_ENOUGH_MEMORY:
1361 errno = ENOMEM;
1362 break;
1363
1364 default:
1365 errno = EINVAL;
1366 break;
1367 }
1368
1369 return;
1370 }
1371 }
1372#else
1373 #if HAVE_AT
1374 if (req->wd)
986 { 1375 {
1376 int fd = openat (WD2FD (req->wd), req->ptr1, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1377
1378 if (fd < 0)
1379 return;
1380
1381 dirp = fdopendir (fd);
1382
1383 if (!dirp)
1384 close (fd);
1385 }
1386 else
1387 dirp = opendir (req->ptr1);
1388 #else
1389 dirp = opendir (wd_expand (&self->tmpbuf, req->wd, req->ptr1));
1390 #endif
1391
1392 if (!dirp)
1393 return;
1394#endif
1395
1396 if (req->flags & EIO_FLAG_PTR1_FREE)
1397 free (req->ptr1);
1398
1399 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1400 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1401 req->ptr2 = names = malloc (namesalloc);
1402
1403 if (!names || (flags && !dents))
1404 return;
1405
1406 for (;;)
1407 {
1408 int done;
1409
1410#ifdef _WIN32
1411 done = !dirp;
1412#else
987 errno = 0; 1413 errno = 0;
988 entp = readdir (dirp); 1414 entp = readdir (dirp);
1415 done = !entp;
1416#endif
989 1417
1418 if (done)
1419 {
1420#ifndef _WIN32
1421 int old_errno = errno;
1422 closedir (dirp);
1423 errno = old_errno;
1424
990 if (!entp) 1425 if (errno)
1426 break;
1427#endif
1428
1429 /* sort etc. */
1430 req->int1 = flags;
1431 req->result = dentoffs;
1432
1433 if (flags & EIO_READDIR_STAT_ORDER)
1434 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1435 else if (flags & EIO_READDIR_DIRS_FIRST)
1436 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1437 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1438 else
1439 {
1440 /* in this case, all is known, and we just put dirs first and sort them */
1441 eio_dirent *oth = dents + dentoffs;
1442 eio_dirent *dir = dents;
1443
1444 /* now partition dirs to the front, and non-dirs to the back */
1445 /* by walking from both sides and swapping if necessary */
1446 while (oth > dir)
1447 {
1448 if (dir->type == EIO_DT_DIR)
1449 ++dir;
1450 else if ((--oth)->type == EIO_DT_DIR)
1451 {
1452 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1453
1454 ++dir;
1455 }
1456 }
1457
1458 /* now sort the dirs only (dirs all have the same score) */
1459 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1460 }
1461
991 break; 1462 break;
1463 }
992 1464
993 name = entp->d_name; 1465 /* now add the entry to our list(s) */
1466 name = D_NAME (entp);
994 1467
1468 /* skip . and .. entries */
995 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1469 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
996 { 1470 {
997 int len = strlen (name) + 1; 1471 int len = D_NAMLEN (entp) + 1;
998 1472
999 res++; 1473 while (ecb_expect_false (namesoffs + len > namesalloc))
1000
1001 while (memofs + len > memlen)
1002 { 1474 {
1003 memlen *= 2; 1475 namesalloc *= 2;
1004 X_LOCK (wrklock);
1005 req->ptr2 = names = realloc (names, memlen); 1476 req->ptr2 = names = realloc (names, namesalloc);
1006 X_UNLOCK (wrklock);
1007 1477
1008 if (!names) 1478 if (!names)
1479 break;
1480 }
1481
1482 memcpy (names + namesoffs, name, len);
1483
1484 if (dents)
1485 {
1486 struct eio_dirent *ent;
1487
1488 if (ecb_expect_false (dentoffs == dentalloc))
1489 {
1490 dentalloc *= 2;
1491 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1492
1493 if (!dents)
1009 break; 1494 break;
1010 } 1495 }
1011 1496
1012 memcpy (names + memofs, name, len); 1497 ent = dents + dentoffs;
1013 memofs += len; 1498
1499 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1500 ent->namelen = len - 1;
1501 ent->inode = D_INO (entp);
1502
1503 inode_bits |= ent->inode;
1504
1505 switch (D_TYPE (entp))
1506 {
1507 default:
1508 ent->type = EIO_DT_UNKNOWN;
1509 flags |= EIO_READDIR_FOUND_UNKNOWN;
1510 break;
1511
1512 #ifdef DT_FIFO
1513 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1514 #endif
1515 #ifdef DT_CHR
1516 case DT_CHR: ent->type = EIO_DT_CHR; break;
1517 #endif
1518 #ifdef DT_MPC
1519 case DT_MPC: ent->type = EIO_DT_MPC; break;
1520 #endif
1521 #ifdef DT_DIR
1522 case DT_DIR: ent->type = EIO_DT_DIR; break;
1523 #endif
1524 #ifdef DT_NAM
1525 case DT_NAM: ent->type = EIO_DT_NAM; break;
1526 #endif
1527 #ifdef DT_BLK
1528 case DT_BLK: ent->type = EIO_DT_BLK; break;
1529 #endif
1530 #ifdef DT_MPB
1531 case DT_MPB: ent->type = EIO_DT_MPB; break;
1532 #endif
1533 #ifdef DT_REG
1534 case DT_REG: ent->type = EIO_DT_REG; break;
1535 #endif
1536 #ifdef DT_NWK
1537 case DT_NWK: ent->type = EIO_DT_NWK; break;
1538 #endif
1539 #ifdef DT_CMP
1540 case DT_CMP: ent->type = EIO_DT_CMP; break;
1541 #endif
1542 #ifdef DT_LNK
1543 case DT_LNK: ent->type = EIO_DT_LNK; break;
1544 #endif
1545 #ifdef DT_SOCK
1546 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1547 #endif
1548 #ifdef DT_DOOR
1549 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1550 #endif
1551 #ifdef DT_WHT
1552 case DT_WHT: ent->type = EIO_DT_WHT; break;
1553 #endif
1554 }
1555
1556 ent->score = 7;
1557
1558 if (flags & EIO_READDIR_DIRS_FIRST)
1559 {
1560 if (ent->type == EIO_DT_UNKNOWN)
1561 {
1562 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1563 ent->score = 1;
1564 else if (!strchr (name, '.')) /* absence of dots indicate likely dirs */
1565 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1566 }
1567 else if (ent->type == EIO_DT_DIR)
1568 ent->score = 0;
1569 }
1014 } 1570 }
1571
1572 namesoffs += len;
1573 ++dentoffs;
1015 } 1574 }
1016 1575
1017 if (errno) 1576 if (EIO_CANCELLED (req))
1018 res = -1; 1577 {
1578 errno = ECANCELED;
1579 break;
1580 }
1581
1582#ifdef _WIN32
1583 if (!FindNextFile (dirp, &entp))
1584 {
1585 FindClose (dirp);
1586 dirp = 0;
1587 }
1588#endif
1589 }
1590}
1591
1592/*****************************************************************************/
1593/* working directory stuff */
1594/* various deficiencies in the posix 2008 api force us to */
1595/* keep the absolute path in string form at all times */
1596/* fuck yeah. */
1597
1598#if !HAVE_AT
1599
1600/* a bit like realpath, but usually faster because it doesn'T have to return */
1601/* an absolute or canonical path */
1602static const char *
1603wd_expand (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
1604{
1605 if (!wd || *path == '/')
1606 return path;
1607
1608 if (path [0] == '.' && !path [1])
1609 return wd->str;
1610
1019 1611 {
1020 req->result = res; 1612 int l1 = wd->len;
1021} 1613 int l2 = strlen (path);
1022 1614
1023#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) 1615 char *res = etp_tmpbuf_get (tmpbuf, l1 + l2 + 2);
1024# undef msync
1025# define msync(a,b,c) ((errno = ENOSYS), -1)
1026#endif
1027 1616
1028int 1617 memcpy (res, wd->str, l1);
1029eio__mtouch (void *mem, size_t len, int flags) 1618 res [l1] = '/';
1030{ 1619 memcpy (res + l1 + 1, path, l2 + 1);
1031 intptr_t addr = (intptr_t)mem;
1032 intptr_t end = addr + len;
1033#ifdef PAGESIZE
1034 const intptr_t page = PAGESIZE;
1035#else
1036 static intptr_t page;
1037 1620
1038 if (!page) 1621 return res;
1039 page = sysconf (_SC_PAGESIZE); 1622 }
1040#endif 1623}
1041 1624
1042 addr &= ~(page - 1); /* assume page size is always a power of two */ 1625#endif
1043 1626
1044 if (addr < end) 1627static eio_wd
1045 if (flags) /* modify */ 1628eio__wd_open_sync (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
1046 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len); 1629{
1047 else 1630 int fd;
1048 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len); 1631 eio_wd res;
1632 int len = eio__realpath (tmpbuf, wd, path);
1049 1633
1634 if (len < 0)
1635 return EIO_INVALID_WD;
1636
1637#if HAVE_AT
1638 fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1639
1640 if (fd < 0)
1641 return EIO_INVALID_WD;
1642#endif
1643
1644 res = malloc (sizeof (*res) + len); /* one extra 0-byte */
1645
1646#if HAVE_AT
1647 res->fd = fd;
1648#endif
1649
1650 res->len = len;
1651 memcpy (res->str, tmpbuf->ptr, len);
1652 res->str [len] = 0;
1653
1654 return res;
1655}
1656
1657eio_wd
1658eio_wd_open_sync (eio_wd wd, const char *path)
1659{
1660 struct etp_tmpbuf tmpbuf = { };
1661 wd = eio__wd_open_sync (&tmpbuf, wd, path);
1662 free (tmpbuf.ptr);
1663
1050 return 0; 1664 return wd;
1051} 1665}
1666
1667void
1668eio_wd_close_sync (eio_wd wd)
1669{
1670 if (wd != EIO_INVALID_WD && wd != EIO_CWD)
1671 {
1672 #if HAVE_AT
1673 close (wd->fd);
1674 #endif
1675 free (wd);
1676 }
1677}
1678
1679#if HAVE_AT
1680
1681/* they forgot these */
1682
1683static int
1684eio__truncateat (int dirfd, const char *path, off_t length)
1685{
1686 int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
1687 int res;
1688
1689 if (fd < 0)
1690 return fd;
1691
1692 res = ftruncate (fd, length);
1693 close (fd);
1694 return res;
1695}
1696
1697static int
1698eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
1699{
1700 int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
1701 int res;
1702
1703 if (fd < 0)
1704 return fd;
1705
1706 res = fstatvfs (fd, buf);
1707 close (fd);
1708 return res;
1709
1710}
1711
1712#endif
1052 1713
1053/*****************************************************************************/ 1714/*****************************************************************************/
1054 1715
1055#define ALLOC(len) \ 1716#define ALLOC(len) \
1056 if (!req->ptr2) \ 1717 if (!req->ptr2) \
1057 { \ 1718 { \
1058 X_LOCK (wrklock); \ 1719 X_LOCK (EIO_POOL->wrklock); \
1059 req->flags |= EIO_FLAG_PTR2_FREE; \ 1720 req->flags |= EIO_FLAG_PTR2_FREE; \
1060 X_UNLOCK (wrklock); \ 1721 X_UNLOCK (EIO_POOL->wrklock); \
1061 req->ptr2 = malloc (len); \ 1722 req->ptr2 = malloc (len); \
1062 if (!req->ptr2) \ 1723 if (!req->ptr2) \
1063 { \ 1724 { \
1064 errno = ENOMEM; \ 1725 errno = ENOMEM; \
1065 req->result = -1; \ 1726 req->result = -1; \
1066 break; \ 1727 break; \
1067 } \ 1728 } \
1068 } 1729 }
1069 1730
1070X_THREAD_PROC (etp_proc)
1071{
1072 ETP_REQ *req;
1073 struct timespec ts;
1074 etp_worker *self = (etp_worker *)thr_arg;
1075
1076 /* try to distribute timeouts somewhat randomly */
1077 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1078
1079 for (;;)
1080 {
1081 X_LOCK (reqlock);
1082
1083 for (;;)
1084 {
1085 self->req = req = reqq_shift (&req_queue);
1086
1087 if (req)
1088 break;
1089
1090 ++idle;
1091
1092 ts.tv_sec = time (0) + IDLE_TIMEOUT;
1093 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
1094 {
1095 if (idle > max_idle)
1096 {
1097 --idle;
1098 X_UNLOCK (reqlock);
1099 X_LOCK (wrklock);
1100 --started;
1101 X_UNLOCK (wrklock);
1102 goto quit;
1103 }
1104
1105 /* we are allowed to idle, so do so without any timeout */
1106 X_COND_WAIT (reqwait, reqlock);
1107 }
1108
1109 --idle;
1110 }
1111
1112 --nready;
1113
1114 X_UNLOCK (reqlock);
1115
1116 if (req->type < 0)
1117 goto quit;
1118
1119 if (!EIO_CANCELLED (req))
1120 ETP_EXECUTE (self, req);
1121
1122 X_LOCK (reslock);
1123
1124 ++npending;
1125
1126 if (!reqq_push (&res_queue, req) && want_poll_cb)
1127 want_poll_cb ();
1128
1129 self->req = 0;
1130 etp_worker_clear (self);
1131
1132 X_UNLOCK (reslock);
1133 }
1134
1135quit:
1136 X_LOCK (wrklock);
1137 etp_worker_free (self);
1138 X_UNLOCK (wrklock);
1139
1140 return 0;
1141}
1142
1143/*****************************************************************************/ 1731/*****************************************************************************/
1144 1732
1733int ecb_cold
1145int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1734eio_init (void (*want_poll)(void), void (*done_poll)(void))
1146{ 1735{
1147 return etp_init (want_poll, done_poll); 1736 eio_want_poll_cb = want_poll;
1148} 1737 eio_done_poll_cb = done_poll;
1149 1738
1739 return etp_init (EIO_POOL, 0, 0, 0);
1740}
1741
1742ecb_inline void
1150static void eio_api_destroy (eio_req *req) 1743eio_api_destroy (eio_req *req)
1151{ 1744{
1152 free (req); 1745 free (req);
1153} 1746}
1154 1747
1155#define REQ(rtype) \ 1748#define REQ(rtype) \
1156 eio_req *req; \ 1749 eio_req *req; \
1157 \ 1750 \
1158 req = (eio_req *)calloc (1, sizeof *req); \ 1751 req = (eio_req *)calloc (1, sizeof *req); \
1159 if (!req) \ 1752 if (!req) \
1160 return 0; \ 1753 return 0; \
… …
1174 { \ 1767 { \
1175 eio_api_destroy (req); \ 1768 eio_api_destroy (req); \
1176 return 0; \ 1769 return 0; \
1177 } 1770 }
1178 1771
1772#define SINGLEDOT(ptr) (0[(char *)(ptr)] == '.' && !1[(char *)(ptr)])
1773
1774static void
1179static void eio_execute (etp_worker *self, eio_req *req) 1775eio_execute (etp_worker *self, eio_req *req)
1180{ 1776{
1181 errno = 0; 1777#if HAVE_AT
1778 int dirfd;
1779#else
1780 const char *path;
1781#endif
1782
1783 if (ecb_expect_false (EIO_CANCELLED (req)))
1784 {
1785 req->result = -1;
1786 req->errorno = ECANCELED;
1787 return;
1788 }
1789
1790 if (ecb_expect_false (req->wd == EIO_INVALID_WD))
1791 {
1792 req->result = -1;
1793 req->errorno = ENOENT;
1794 return;
1795 }
1796
1797 if (req->type >= EIO_OPEN)
1798 {
1799 #if HAVE_AT
1800 dirfd = WD2FD (req->wd);
1801 #else
1802 path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1803 #endif
1804 }
1182 1805
1183 switch (req->type) 1806 switch (req->type)
1184 { 1807 {
1808 case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
1809 req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
1810 break;
1811 case EIO_WD_CLOSE: req->result = 0;
1812 eio_wd_close_sync (req->wd); break;
1813
1814 case EIO_SEEK: eio__lseek (req); break;
1185 case EIO_READ: ALLOC (req->size); 1815 case EIO_READ: ALLOC (req->size);
1186 req->result = req->offs >= 0 1816 req->result = req->offs >= 0
1187 ? pread (req->int1, req->ptr2, req->size, req->offs) 1817 ? pread (req->int1, req->ptr2, req->size, req->offs)
1188 : read (req->int1, req->ptr2, req->size); break; 1818 : read (req->int1, req->ptr2, req->size); break;
1189 case EIO_WRITE: req->result = req->offs >= 0 1819 case EIO_WRITE: req->result = req->offs >= 0
1190 ? pwrite (req->int1, req->ptr2, req->size, req->offs) 1820 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1191 : write (req->int1, req->ptr2, req->size); break; 1821 : write (req->int1, req->ptr2, req->size); break;
1192 1822
1823 case EIO_FCNTL: req->result = fcntl (req->int1, (int) req->int2, req->ptr2); break;
1824 case EIO_IOCTL: req->result = ioctl (req->int1, (unsigned long)req->int2, req->ptr2); break;
1825
1193 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break; 1826 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1194 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break; 1827 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
1828
1829#if HAVE_AT
1195 1830
1196 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1831 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1197 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 1832 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break;
1198 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1833 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1199 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 1834 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, AT_SYMLINK_NOFOLLOW); break;
1200 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1201 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1202
1203 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; 1835 case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break;
1204 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1205 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break; 1836 case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break;
1206 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1207 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break; 1837 case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break;
1208 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1209
1210 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break; 1838 case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break;
1211 case EIO_CLOSE: req->result = close (req->int1); break; 1839
1212 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1213 case EIO_UNLINK: req->result = unlink (req->ptr1); break; 1840 case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
1214 case EIO_RMDIR: req->result = rmdir (req->ptr1); break; 1841 case EIO_RMDIR: /* complications arise because "." cannot be removed, so we might have to expand */
1842 req->result = req->wd && SINGLEDOT (req->ptr1)
1843 ? rmdir (req->wd->str)
1844 : unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break;
1215 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; 1845 case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); break;
1216 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; 1846 case EIO_RENAME: /* complications arise because "." cannot be renamed, so we might have to expand */
1847 req->result = req->wd && SINGLEDOT (req->ptr1)
1848 ? rename (req->wd->str, req->ptr2)
1849 : renameat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2); break;
1217 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; 1850 case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2, 0); break;
1218 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; 1851 case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
1219 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break; 1852 case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1220
1221 case EIO_READLINK: ALLOC (NAME_MAX); 1853 case EIO_READLINK: ALLOC (PATH_MAX);
1222 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; 1854 req->result = readlinkat (dirfd, req->ptr1, req->ptr2, PATH_MAX); break;
1223 1855 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1224 case EIO_SYNC: req->result = 0; sync (); break; 1856 req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1225 case EIO_FSYNC: req->result = fsync (req->int1); break;
1226 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1227 case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break;
1228 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;
1229 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1230
1231 case EIO_READDIR: eio__scandir (req, self); break;
1232
1233 case EIO_BUSY: 1857 case EIO_UTIME:
1234#ifdef _WIN32 1858 case EIO_FUTIME:
1235 Sleep (req->nv1 * 1000.); 1859 {
1860 struct timespec ts[2];
1861 struct timespec *times;
1862
1863 if (req->nv1 != -1. || req->nv2 != -1.)
1864 {
1865 ts[0].tv_sec = req->nv1;
1866 ts[0].tv_nsec = (req->nv1 - ts[0].tv_sec) * 1e9;
1867 ts[1].tv_sec = req->nv2;
1868 ts[1].tv_nsec = (req->nv2 - ts[1].tv_sec) * 1e9;
1869
1870 times = ts;
1871 }
1872 else
1873 times = 0;
1874
1875 req->result = req->type == EIO_FUTIME
1876 ? futimens (req->int1, times)
1877 : utimensat (dirfd, req->ptr1, times, 0);
1878 }
1879 break;
1880
1236#else 1881#else
1237 {
1238 struct timeval tv;
1239 1882
1240 tv.tv_sec = req->nv1; 1883 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1241 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; 1884 req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
1885 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1886 req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
1887 case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
1888 case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
1889 case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
1890 case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
1242 1891
1243 req->result = select (0, 0, 0, 0, &tv); 1892 case EIO_UNLINK: req->result = unlink (path ); break;
1244 } 1893 case EIO_RMDIR: req->result = rmdir (path ); break;
1245#endif 1894 case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
1246 break; 1895 case EIO_RENAME: req->result = rename (path , req->ptr2); break;
1896 case EIO_LINK: req->result = link (path , req->ptr2); break;
1897 case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
1898 case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
1899 case EIO_READLINK: ALLOC (PATH_MAX);
1900 req->result = readlink (path, req->ptr2, PATH_MAX); break;
1901 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1902 req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
1247 1903
1248 case EIO_UTIME: 1904 case EIO_UTIME:
1249 case EIO_FUTIME: 1905 case EIO_FUTIME:
1250 { 1906 {
1251 struct timeval tv[2]; 1907 struct timeval tv[2];
1252 struct timeval *times; 1908 struct timeval *times;
1253 1909
1254 if (req->nv1 != -1. || req->nv2 != -1.) 1910 if (req->nv1 != -1. || req->nv2 != -1.)
1255 { 1911 {
1256 tv[0].tv_sec = req->nv1; 1912 tv[0].tv_sec = req->nv1;
1257 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.; 1913 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
1258 tv[1].tv_sec = req->nv2; 1914 tv[1].tv_sec = req->nv2;
1259 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.; 1915 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
1260 1916
1261 times = tv; 1917 times = tv;
1262 } 1918 }
1263 else 1919 else
1264 times = 0; 1920 times = 0;
1265
1266 1921
1267 req->result = req->type == EIO_FUTIME 1922 req->result = req->type == EIO_FUTIME
1268 ? futimes (req->int1, times) 1923 ? futimes (req->int1, times)
1269 : utimes (req->ptr1, times); 1924 : utimes (req->ptr1, times);
1270 } 1925 }
1271 break; 1926 break;
1272 1927
1928#endif
1929
1930 case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
1931 {
1932 ALLOC (req->result);
1933 memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
1934 }
1935 break;
1936
1937 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1938 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1939
1940 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1941 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1942
1943 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1944 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1945 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1946
1947 case EIO_CLOSE: req->result = close (req->int1); break;
1948 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1949 case EIO_SYNC: req->result = 0; sync (); break;
1950 case EIO_FSYNC: req->result = fsync (req->int1); break;
1951 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1952 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
1953 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1954 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1955 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1956 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1957 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1958 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
1959
1960 case EIO_READDIR: eio__scandir (req, self); break;
1961
1962 case EIO_BUSY:
1963#ifdef _WIN32
1964 Sleep (req->nv1 * 1e3);
1965#else
1966 {
1967 struct timeval tv;
1968
1969 tv.tv_sec = req->nv1;
1970 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1971
1972 req->result = select (0, 0, 0, 0, &tv);
1973 }
1974#endif
1975 break;
1976
1977#if 0
1273 case EIO_GROUP: 1978 case EIO_GROUP:
1274 abort (); /* handled in eio_request */ 1979 abort (); /* handled in eio_request */
1980#endif
1275 1981
1276 case EIO_NOP: 1982 case EIO_NOP:
1277 req->result = 0; 1983 req->result = 0;
1278 break; 1984 break;
1279 1985
1280 case EIO_CUSTOM: 1986 case EIO_CUSTOM:
1281 ((void (*)(eio_req *))req->feed) (req); 1987 req->feed (req);
1282 break; 1988 break;
1283 1989
1284 default: 1990 default:
1285 req->result = -1; 1991 req->result = EIO_ENOSYS ();
1286 break; 1992 break;
1287 } 1993 }
1288 1994
1289 req->errorno = errno; 1995 req->errorno = errno;
1290} 1996}
1291 1997
1292#ifndef EIO_NO_WRAPPERS 1998#ifndef EIO_NO_WRAPPERS
1293 1999
2000eio_req *eio_wd_open (const char *path, int pri, eio_cb cb, void *data)
2001{
2002 REQ (EIO_WD_OPEN); PATH; SEND;
2003}
2004
2005eio_req *eio_wd_close (eio_wd wd, int pri, eio_cb cb, void *data)
2006{
2007 REQ (EIO_WD_CLOSE); req->wd = wd; SEND;
2008}
2009
1294eio_req *eio_nop (int pri, eio_cb cb, void *data) 2010eio_req *eio_nop (int pri, eio_cb cb, void *data)
1295{ 2011{
1296 REQ (EIO_NOP); SEND; 2012 REQ (EIO_NOP); SEND;
1297} 2013}
1298 2014
… …
1314eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2030eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1315{ 2031{
1316 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2032 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1317} 2033}
1318 2034
2035eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2036{
2037 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2038}
2039
2040eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
2041{
2042 REQ (EIO_SYNCFS); req->int1 = fd; SEND;
2043}
2044
2045eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2046{
2047 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2048}
2049
1319eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2050eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1320{ 2051{
1321 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2052 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1322} 2053}
1323 2054
1324eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) 2055eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
1325{ 2056{
1326 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; 2057 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
1327} 2058}
1328 2059
1329eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) 2060eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
1330{ 2061{
1331 REQ (EIO_FDATASYNC); req->int1 = fd; SEND; 2062 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2063}
2064
2065eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2066{
2067 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
1332} 2068}
1333 2069
1334eio_req *eio_close (int fd, int pri, eio_cb cb, void *data) 2070eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
1335{ 2071{
1336 REQ (EIO_CLOSE); req->int1 = fd; SEND; 2072 REQ (EIO_CLOSE); req->int1 = fd; SEND;
… …
1339eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data) 2075eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
1340{ 2076{
1341 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND; 2077 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
1342} 2078}
1343 2079
2080eio_req *eio_seek (int fd, off_t offset, int whence, int pri, eio_cb cb, void *data)
2081{
2082 REQ (EIO_SEEK); req->int1 = fd; req->offs = offset; req->int2 = whence; SEND;
2083}
2084
1344eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data) 2085eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1345{ 2086{
1346 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND; 2087 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1347} 2088}
1348 2089
1349eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data) 2090eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1350{ 2091{
1351 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND; 2092 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1352} 2093}
1353 2094
2095eio_req *eio_fcntl (int fd, int cmd, void *arg, int pri, eio_cb cb, void *data)
2096{
2097 REQ (EIO_IOCTL); req->int1 = fd; req->int2 = cmd; req->ptr2 = arg; SEND;
2098}
2099
2100eio_req *eio_ioctl (int fd, unsigned long request, void *buf, int pri, eio_cb cb, void *data)
2101{
2102 REQ (EIO_IOCTL); req->int1 = fd; req->int2 = request; req->ptr2 = buf; SEND;
2103}
2104
1354eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) 2105eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1355{ 2106{
1356 REQ (EIO_FSTAT); req->int1 = fd; SEND; 2107 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1357} 2108}
1358 2109
2110eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
2111{
2112 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
2113}
2114
1359eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) 2115eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1360{ 2116{
1361 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; 2117 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1362} 2118}
1363 2119
… …
1369eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) 2125eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1370{ 2126{
1371 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2127 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1372} 2128}
1373 2129
1374eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2130eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1375{ 2131{
1376 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2132 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1377} 2133}
1378 2134
1379eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) 2135eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
… …
1399eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data) 2155eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1400{ 2156{
1401 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2157 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1402} 2158}
1403 2159
1404eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2160eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1405{ 2161{
1406 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2162 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1407} 2163}
1408 2164
1409eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) 2165eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
… …
1425eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data) 2181eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1426{ 2182{
1427 return eio__1path (EIO_READLINK, path, pri, cb, data); 2183 return eio__1path (EIO_READLINK, path, pri, cb, data);
1428} 2184}
1429 2185
2186eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2187{
2188 return eio__1path (EIO_REALPATH, path, pri, cb, data);
2189}
2190
1430eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) 2191eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1431{ 2192{
1432 return eio__1path (EIO_STAT, path, pri, cb, data); 2193 return eio__1path (EIO_STAT, path, pri, cb, data);
1433} 2194}
1434 2195
1435eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) 2196eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1436{ 2197{
1437 return eio__1path (EIO_LSTAT, path, pri, cb, data); 2198 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1438} 2199}
1439 2200
2201eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
2202{
2203 return eio__1path (EIO_STATVFS, path, pri, cb, data);
2204}
2205
1440eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) 2206eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1441{ 2207{
1442 return eio__1path (EIO_UNLINK, path, pri, cb, data); 2208 return eio__1path (EIO_UNLINK, path, pri, cb, data);
1443} 2209}
1444 2210
1445eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data) 2211eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1446{ 2212{
1447 return eio__1path (EIO_RMDIR, path, pri, cb, data); 2213 return eio__1path (EIO_RMDIR, path, pri, cb, data);
1448} 2214}
1449 2215
1450eio_req *eio_readdir (const char *path, int pri, eio_cb cb, void *data) 2216eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1451{ 2217{
1452 return eio__1path (EIO_READDIR, path, pri, cb, data); 2218 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1453} 2219}
1454 2220
1455eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) 2221eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1456{ 2222{
1457 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; 2223 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1458} 2224}
1459 2225
1460static eio_req * 2226static eio_req *
1461eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2227eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1462{ 2228{
… …
1486eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2252eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1487{ 2253{
1488 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2254 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1489} 2255}
1490 2256
1491eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) 2257eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
1492{ 2258{
1493 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND; 2259 REQ (EIO_CUSTOM); req->feed = execute; SEND;
1494} 2260}
1495 2261
1496#endif 2262#endif
1497 2263
1498eio_req *eio_grp (eio_cb cb, void *data) 2264eio_req *eio_grp (eio_cb cb, void *data)
… …
1507#undef SEND 2273#undef SEND
1508 2274
1509/*****************************************************************************/ 2275/*****************************************************************************/
1510/* grp functions */ 2276/* grp functions */
1511 2277
2278void
1512void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2279eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1513{ 2280{
1514 grp->int2 = limit; 2281 grp->int2 = limit;
1515 grp->feed = feed; 2282 grp->feed = feed;
1516 2283
1517 grp_try_feed (grp); 2284 grp_try_feed (grp);
1518} 2285}
1519 2286
2287void
1520void eio_grp_limit (eio_req *grp, int limit) 2288eio_grp_limit (eio_req *grp, int limit)
1521{ 2289{
1522 grp->int2 = limit; 2290 grp->int2 = limit;
1523 2291
1524 grp_try_feed (grp); 2292 grp_try_feed (grp);
1525} 2293}
1526 2294
2295void
1527void eio_grp_add (eio_req *grp, eio_req *req) 2296eio_grp_add (eio_req *grp, eio_req *req)
1528{ 2297{
1529 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2298 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1530 2299
1531 grp->flags |= EIO_FLAG_GROUPADD; 2300 grp->flags |= ETP_FLAG_GROUPADD;
1532 2301
1533 ++grp->size; 2302 ++grp->size;
1534 req->grp = grp; 2303 req->grp = grp;
1535 2304
1536 req->grp_prev = 0; 2305 req->grp_prev = 0;
… …
1543} 2312}
1544 2313
1545/*****************************************************************************/ 2314/*****************************************************************************/
1546/* misc garbage */ 2315/* misc garbage */
1547 2316
2317eio_ssize_t
1548ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2318eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1549{ 2319{
1550 etp_worker wrk;
1551
1552 wrk.dbuf = 0;
1553
1554 eio__sendfile (ofd, ifd, offset, count, &wrk); 2320 return eio__sendfile (ofd, ifd, offset, count);
1555
1556 if (wrk.dbuf)
1557 free (wrk.dbuf);
1558} 2321}
1559 2322

Diff Legend

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