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

Comparing libeio/eio.c (file contents):
Revision 1.27 by root, Wed Oct 22 18:15:36 2008 UTC vs.
Revision 1.129 by root, Sun Apr 14 09:43:19 2013 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 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"
46
47#ifdef EIO_STACKSIZE
48# define X_STACKSIZE EIO_STACKSIZE
49#endif
41#include "xthread.h" 50#include "xthread.h"
42 51
43#include <errno.h> 52#include <errno.h>
44#include <stddef.h> 53#include <stddef.h>
45#include <stdlib.h> 54#include <stdlib.h>
… …
49#include <sys/stat.h> 58#include <sys/stat.h>
50#include <limits.h> 59#include <limits.h>
51#include <fcntl.h> 60#include <fcntl.h>
52#include <assert.h> 61#include <assert.h>
53 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
54#ifndef EIO_FINISH 82#ifndef EIO_FINISH
55# 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
56#endif 84#endif
57 85
58#ifndef EIO_DESTROY 86#ifndef EIO_DESTROY
… …
61 89
62#ifndef EIO_FEED 90#ifndef EIO_FEED
63# 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)
64#endif 92#endif
65 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
66#ifdef _WIN32 105#ifdef _WIN32
67 106
68 /*doh*/ 107 #undef PAGESIZE
108 #define PAGESIZE 4096 /* GetSystemInfo? */
109
110 /* TODO: look at how perl does stat (non-sloppy), unlink (ro-files), utime, link */
111
112 #ifdef EIO_STRUCT_STATI64
113 /* look at perl's non-sloppy stat */
114 #define stat(path,buf) _stati64 (path,buf)
115 #define fstat(fd,buf) _fstati64 (fd,buf)
116 #endif
117 #define lstat(path,buf) stat (path,buf)
118 #define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
119 #define mkdir(path,mode) _mkdir (path)
120 #define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
121
122 #define chmod(path,mode) _chmod (path, mode)
123 #define dup(fd) _dup (fd)
124 #define dup2(fd1,fd2) _dup2 (fd1, fd2)
125
126 #define fchmod(fd,mode) EIO_ENOSYS ()
127 #define chown(path,uid,gid) EIO_ENOSYS ()
128 #define fchown(fd,uid,gid) EIO_ENOSYS ()
129 #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
130 #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
131 #define mknod(path,mode,dev) EIO_ENOSYS ()
132 #define sync() EIO_ENOSYS ()
133 #define readlink(path,buf,s) EIO_ENOSYS ()
134 #define statvfs(path,buf) EIO_ENOSYS ()
135 #define fstatvfs(fd,buf) EIO_ENOSYS ()
136
137 #define pread(fd,buf,count,offset) eio__pread (fd, buf, count, offset)
138 #define pwrite(fd,buf,count,offset) eio__pwrite (fd, buf, count, offset)
139
140 #if __GNUC__
141 typedef long long eio_off_t; /* signed for compatibility to msvc */
142 #else
143 typedef __int64 eio_off_t; /* unsigned not supported by msvc */
144 #endif
145
146 static eio_ssize_t
147 eio__pread (int fd, void *buf, eio_ssize_t count, eio_off_t offset)
148 {
149 OVERLAPPED o = { 0 };
150 DWORD got;
151
152 o.Offset = offset;
153 o.OffsetHigh = offset >> 32;
154
155 return ReadFile ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd), buf, count, &got, &o)
156 ? got : -1;
157 }
158
159 static eio_ssize_t
160 eio__pwrite (int fd, void *buf, eio_ssize_t count, eio_off_t offset)
161 {
162 OVERLAPPED o = { 0 };
163 DWORD got;
164
165 o.Offset = offset;
166 o.OffsetHigh = offset >> 32;
167
168 return WriteFile ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd), buf, count, &got, &o)
169 ? got : -1;
170 }
171
172 /* rename() uses MoveFile, which fails to overwrite */
173 #define rename(old,neu) eio__rename (old, neu)
174
175 static int
176 eio__rename (const char *old, const char *neu)
177 {
178 if (MoveFileEx (old, neu, MOVEFILE_REPLACE_EXISTING))
179 return 0;
180
181 /* should steal _dosmaperr */
182 switch (GetLastError ())
183 {
184 case ERROR_FILE_NOT_FOUND:
185 case ERROR_PATH_NOT_FOUND:
186 case ERROR_INVALID_DRIVE:
187 case ERROR_NO_MORE_FILES:
188 case ERROR_BAD_NETPATH:
189 case ERROR_BAD_NET_NAME:
190 case ERROR_BAD_PATHNAME:
191 case ERROR_FILENAME_EXCED_RANGE:
192 errno = ENOENT;
193 break;
194
195 default:
196 errno = EACCES;
197 break;
198 }
199
200 return -1;
201 }
202
203 /* we could even stat and see if it exists */
204 static int
205 symlink (const char *old, const char *neu)
206 {
207 #if WINVER >= 0x0600
208 if (CreateSymbolicLink (neu, old, 1))
209 return 0;
210
211 if (CreateSymbolicLink (neu, old, 0))
212 return 0;
213 #endif
214
215 return EIO_ERRNO (ENOENT, -1);
216 }
217
218 /* POSIX API only, causing trouble for win32 apps */
219 #define CreateHardLink(neu,old,flags) 0 /* not really creating hardlink, still using relative paths? */
220 #define CreateSymbolicLink(neu,old,flags) 0 /* vista+ only */
221
222 struct statvfs
223 {
224 int dummy;
225 };
226
227 #define DT_DIR EIO_DT_DIR
228 #define DT_REG EIO_DT_REG
229 #define D_NAME(entp) entp.cFileName
230 #define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
69 231
70#else 232#else
71 233
72# include "config.h"
73# include <sys/time.h> 234 #include <sys/time.h>
74# include <sys/select.h> 235 #include <sys/select.h>
75# include <unistd.h> 236 #include <unistd.h>
237 #include <signal.h>
238 #include <dirent.h>
239
240 #ifdef ANDROID
241 #include <sys/vfs.h>
242 #define statvfs statfs
243 #define fstatvfs fstatfs
244 #include <asm/page.h> /* supposedly limits.h does #define PAGESIZE PAGESIZE */
245 #else
246 #include <sys/statvfs.h>
247 #endif
248
249 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
250 #include <sys/mman.h>
251 #endif
252
253 #define D_NAME(entp) entp->d_name
254
255 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
256 #if __FreeBSD__ || __NetBSD__ || __OpenBSD__
257 #define _DIRENT_HAVE_D_TYPE /* sigh */
258 #define D_INO(de) (de)->d_fileno
259 #define D_NAMLEN(de) (de)->d_namlen
260 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
261 #define D_INO(de) (de)->d_ino
262 #endif
263
264 #ifdef _D_EXACT_NAMLEN
265 #undef D_NAMLEN
266 #define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
267 #endif
268
269 #ifdef _DIRENT_HAVE_D_TYPE
270 #define D_TYPE(de) (de)->d_type
271 #endif
272
273 #ifndef EIO_STRUCT_DIRENT
274 #define EIO_STRUCT_DIRENT struct dirent
275 #endif
276
277#endif
278
279#if HAVE_UTIMES
76# include <utime.h> 280# include <utime.h>
77# include <signal.h>
78# include <dirent.h>
79
80# ifndef EIO_STRUCT_DIRENT
81# define EIO_STRUCT_DIRENT struct dirent
82# endif 281#endif
83 282
283#if HAVE_SYS_SYSCALL_H
284# include <sys/syscall.h>
285#endif
286
287#if HAVE_SYS_PRCTL_H
288# include <sys/prctl.h>
84#endif 289#endif
85 290
86#if HAVE_SENDFILE 291#if HAVE_SENDFILE
87# if __linux 292# if __linux
88# include <sys/sendfile.h> 293# include <sys/sendfile.h>
89# elif __freebsd 294# elif __FreeBSD__ || defined __APPLE__
90# include <sys/socket.h> 295# include <sys/socket.h>
91# include <sys/uio.h> 296# include <sys/uio.h>
92# elif __hpux 297# elif __hpux
93# include <sys/socket.h> 298# include <sys/socket.h>
94# elif __solaris /* not yet */ 299# elif __solaris
95# include <sys/sendfile.h> 300# include <sys/sendfile.h>
96# else 301# else
97# error sendfile support requested but not available 302# error sendfile support requested but not available
98# endif 303# endif
99#endif 304#endif
100 305
101/* number of seconds after which an idle threads exit */ 306#ifndef D_TYPE
102#define IDLE_TIMEOUT 10 307# define D_TYPE(de) 0
308#endif
309#ifndef D_INO
310# define D_INO(de) 0
311#endif
312#ifndef D_NAMLEN
313# define D_NAMLEN(entp) strlen (D_NAME (entp))
314#endif
103 315
104/* used for struct dirent, AIX doesn't provide it */ 316/* used for struct dirent, AIX doesn't provide it */
105#ifndef NAME_MAX 317#ifndef NAME_MAX
106# define NAME_MAX 4096 318# define NAME_MAX 4096
107#endif 319#endif
108 320
321/* used for readlink etc. */
322#ifndef PATH_MAX
323# define PATH_MAX 4096
324#endif
325
109/* buffer size for various temporary buffers */ 326/* buffer size for various temporary buffers */
110#define EIO_BUFSIZE 65536 327#define EIO_BUFSIZE 65536
111 328
112#define dBUF \ 329#define dBUF \
113 char *eio_buf; \
114 ETP_WORKER_LOCK (self); \
115 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 330 char *eio_buf = malloc (EIO_BUFSIZE); \
116 ETP_WORKER_UNLOCK (self); \
117 errno = ENOMEM; \ 331 errno = ENOMEM; \
118 if (!eio_buf) \ 332 if (!eio_buf) \
119 return -1; 333 return -1
120 334
121#define EIO_TICKS ((1000000 + 1023) >> 10) 335#define FUBd \
336 free (eio_buf)
122 337
123/*****************************************************************************/ 338/*****************************************************************************/
124 339
125#if __GNUC__ >= 3 340struct tmpbuf
126# define expect(expr,value) __builtin_expect ((expr),(value)) 341{
342 void *ptr;
343 int len;
344};
345
346static void *
347tmpbuf_get (struct tmpbuf *buf, int len)
348{
349 if (buf->len < len)
350 {
351 free (buf->ptr);
352 buf->ptr = malloc (buf->len = len);
353 }
354
355 return buf->ptr;
356}
357
358struct tmpbuf;
359
360#if _POSIX_VERSION >= 200809L
361 #define HAVE_AT 1
362 #define WD2FD(wd) ((wd) ? (wd)->fd : AT_FDCWD)
363 #ifndef O_SEARCH
364 #define O_SEARCH O_RDONLY
365 #endif
127#else 366#else
128# define expect(expr,value) (expr) 367 #define HAVE_AT 0
368 static const char *wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path);
129#endif 369#endif
130 370
131#define expect_false(expr) expect ((expr) != 0, 0) 371struct eio_pwd
132#define expect_true(expr) expect ((expr) != 0, 1) 372{
373#if HAVE_AT
374 int fd;
375#endif
376 int len;
377 char str[1]; /* actually, a 0-terminated canonical path */
378};
133 379
134/*****************************************************************************/ 380/*****************************************************************************/
135 381
136#define ETP_PRI_MIN EIO_PRI_MIN 382#define ETP_PRI_MIN EIO_PRI_MIN
137#define ETP_PRI_MAX EIO_PRI_MAX 383#define ETP_PRI_MAX EIO_PRI_MAX
384
385#define ETP_TYPE_QUIT -1
386#define ETP_TYPE_GROUP EIO_GROUP
138 387
139struct etp_worker; 388struct etp_worker;
140 389
141#define ETP_REQ eio_req 390#define ETP_REQ eio_req
142#define ETP_DESTROY(req) eio_destroy (req) 391#define ETP_DESTROY(req) eio_destroy (req)
143static int eio_finish (eio_req *req); 392static int eio_finish (eio_req *req);
144#define ETP_FINISH(req) eio_finish (req) 393#define ETP_FINISH(req) eio_finish (req)
145static void eio_execute (struct etp_worker *self, eio_req *req); 394static void eio_execute (struct etp_worker *self, eio_req *req);
146#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) 395#define ETP_EXECUTE(wrk,req) eio_execute (wrk, req)
147 396
148#define ETP_WORKER_CLEAR(req) \ 397#include "etp.c"
149 if (wrk->dbuf) \
150 { \
151 free (wrk->dbuf); \
152 wrk->dbuf = 0; \
153 } \
154 \
155 if (wrk->dirp) \
156 { \
157 closedir (wrk->dirp); \
158 wrk->dirp = 0; \
159 }
160#define ETP_WORKER_COMMON \
161 void *dbuf; \
162 DIR *dirp;
163 398
164/*****************************************************************************/ 399/*****************************************************************************/
165 400
166#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
167
168/* calculcate time difference in ~1/EIO_TICKS of a second */
169static int tvdiff (struct timeval *tv1, struct timeval *tv2)
170{
171 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
172 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
173}
174
175static unsigned int started, idle, wanted = 4;
176
177static void (*want_poll_cb) (void);
178static void (*done_poll_cb) (void);
179
180static unsigned int max_poll_time; /* reslock */
181static unsigned int max_poll_reqs; /* reslock */
182
183static volatile unsigned int nreqs; /* reqlock */
184static volatile unsigned int nready; /* reqlock */
185static volatile unsigned int npending; /* reqlock */
186static volatile unsigned int max_idle = 4;
187
188static mutex_t wrklock = X_MUTEX_INIT;
189static mutex_t reslock = X_MUTEX_INIT;
190static mutex_t reqlock = X_MUTEX_INIT;
191static cond_t reqwait = X_COND_INIT;
192
193#if !HAVE_PREADWRITE
194/*
195 * make our pread/pwrite emulation safe against themselves, but not against
196 * normal read/write by using a mutex. slows down execution a lot,
197 * but that's your problem, not mine.
198 */
199static mutex_t preadwritelock = X_MUTEX_INIT;
200#endif
201
202typedef struct etp_worker
203{
204 /* locked by wrklock */
205 struct etp_worker *prev, *next;
206
207 thread_t tid;
208
209 /* locked by reslock, reqlock or wrklock */
210 ETP_REQ *req; /* currently processed request */
211
212 ETP_WORKER_COMMON
213} etp_worker;
214
215static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */
216
217#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
218#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
219
220/* worker threads management */
221
222static void etp_worker_clear (etp_worker *wrk)
223{
224 ETP_WORKER_CLEAR (wrk);
225}
226
227static void etp_worker_free (etp_worker *wrk)
228{
229 wrk->next->prev = wrk->prev;
230 wrk->prev->next = wrk->next;
231
232 free (wrk);
233}
234
235static unsigned int etp_nreqs (void)
236{
237 int retval;
238 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
239 retval = nreqs;
240 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
241 return retval;
242}
243
244static unsigned int etp_nready (void)
245{
246 unsigned int retval;
247
248 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
249 retval = nready;
250 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
251
252 return retval;
253}
254
255static unsigned int etp_npending (void)
256{
257 unsigned int retval;
258
259 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
260 retval = npending;
261 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
262
263 return retval;
264}
265
266static unsigned int etp_nthreads (void)
267{
268 unsigned int retval;
269
270 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
271 retval = started;
272 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
273
274 return retval;
275}
276
277/*
278 * a somewhat faster data structure might be nice, but
279 * with 8 priorities this actually needs <20 insns
280 * per shift, the most expensive operation.
281 */
282typedef struct {
283 ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */
284 int size;
285} etp_reqq;
286
287static etp_reqq req_queue;
288static etp_reqq res_queue;
289
290static int reqq_push (etp_reqq *q, ETP_REQ *req)
291{
292 int pri = req->pri;
293 req->next = 0;
294
295 if (q->qe[pri])
296 {
297 q->qe[pri]->next = req;
298 q->qe[pri] = req;
299 }
300 else
301 q->qe[pri] = q->qs[pri] = req;
302
303 return q->size++;
304}
305
306static ETP_REQ *reqq_shift (etp_reqq *q)
307{
308 int pri;
309
310 if (!q->size)
311 return 0;
312
313 --q->size;
314
315 for (pri = ETP_NUM_PRI; pri--; )
316 {
317 eio_req *req = q->qs[pri];
318
319 if (req)
320 {
321 if (!(q->qs[pri] = (eio_req *)req->next))
322 q->qe[pri] = 0;
323
324 return req;
325 }
326 }
327
328 abort ();
329}
330
331static void etp_atfork_prepare (void)
332{
333 X_LOCK (wrklock);
334 X_LOCK (reqlock);
335 X_LOCK (reslock);
336#if !HAVE_PREADWRITE
337 X_LOCK (preadwritelock);
338#endif
339}
340
341static void etp_atfork_parent (void)
342{
343#if !HAVE_PREADWRITE
344 X_UNLOCK (preadwritelock);
345#endif
346 X_UNLOCK (reslock);
347 X_UNLOCK (reqlock);
348 X_UNLOCK (wrklock);
349}
350
351static void etp_atfork_child (void)
352{
353 ETP_REQ *prv;
354
355 while ((prv = reqq_shift (&req_queue)))
356 ETP_DESTROY (prv);
357
358 while ((prv = reqq_shift (&res_queue)))
359 ETP_DESTROY (prv);
360
361 while (wrk_first.next != &wrk_first)
362 {
363 etp_worker *wrk = wrk_first.next;
364
365 if (wrk->req)
366 ETP_DESTROY (wrk->req);
367
368 etp_worker_clear (wrk);
369 etp_worker_free (wrk);
370 }
371
372 started = 0;
373 idle = 0;
374 nreqs = 0;
375 nready = 0;
376 npending = 0;
377
378 etp_atfork_parent ();
379}
380
381static void 401static void
382etp_once_init (void)
383{
384 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
385}
386
387static int
388etp_init (void (*want_poll)(void), void (*done_poll)(void))
389{
390 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
391
392 pthread_once (&doinit, etp_once_init);
393
394 want_poll_cb = want_poll;
395 done_poll_cb = done_poll;
396
397 return 0;
398}
399
400X_THREAD_PROC (etp_proc);
401
402static void etp_start_thread (void)
403{
404 etp_worker *wrk = calloc (1, sizeof (etp_worker));
405
406 /*TODO*/
407 assert (("unable to allocate worker thread data", wrk));
408
409 X_LOCK (wrklock);
410
411 if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
412 {
413 wrk->prev = &wrk_first;
414 wrk->next = wrk_first.next;
415 wrk_first.next->prev = wrk;
416 wrk_first.next = wrk;
417 ++started;
418 }
419 else
420 free (wrk);
421
422 X_UNLOCK (wrklock);
423}
424
425static void etp_maybe_start_thread (void)
426{
427 if (expect_true (etp_nthreads () >= wanted))
428 return;
429
430 /* todo: maybe use idle here, but might be less exact */
431 if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
432 return;
433
434 etp_start_thread ();
435}
436
437static void etp_end_thread (void)
438{
439 eio_req *req = calloc (1, sizeof (eio_req));
440
441 req->type = -1;
442 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
443
444 X_LOCK (reqlock);
445 reqq_push (&req_queue, req);
446 X_COND_SIGNAL (reqwait);
447 X_UNLOCK (reqlock);
448
449 X_LOCK (wrklock);
450 --started;
451 X_UNLOCK (wrklock);
452}
453
454static int etp_poll (void)
455{
456 unsigned int maxreqs;
457 unsigned int maxtime;
458 struct timeval tv_start, tv_now;
459
460 X_LOCK (reslock);
461 maxreqs = max_poll_reqs;
462 maxtime = max_poll_time;
463 X_UNLOCK (reslock);
464
465 if (maxtime)
466 gettimeofday (&tv_start, 0);
467
468 for (;;)
469 {
470 ETP_REQ *req;
471
472 etp_maybe_start_thread ();
473
474 X_LOCK (reslock);
475 req = reqq_shift (&res_queue);
476
477 if (req)
478 {
479 --npending;
480
481 if (!res_queue.size && done_poll_cb)
482 done_poll_cb ();
483 }
484
485 X_UNLOCK (reslock);
486
487 if (!req)
488 return 0;
489
490 X_LOCK (reqlock);
491 --nreqs;
492 X_UNLOCK (reqlock);
493
494 if (expect_false (req->type == EIO_GROUP && req->size))
495 {
496 req->int1 = 1; /* mark request as delayed */
497 continue;
498 }
499 else
500 {
501 int res = ETP_FINISH (req);
502 if (expect_false (res))
503 return res;
504 }
505
506 if (expect_false (maxreqs && !--maxreqs))
507 break;
508
509 if (maxtime)
510 {
511 gettimeofday (&tv_now, 0);
512
513 if (tvdiff (&tv_start, &tv_now) >= maxtime)
514 break;
515 }
516 }
517
518 errno = EAGAIN;
519 return -1;
520}
521
522static void etp_cancel (ETP_REQ *req)
523{
524 X_LOCK (wrklock);
525 req->flags |= EIO_FLAG_CANCELLED;
526 X_UNLOCK (wrklock);
527
528 eio_grp_cancel (req);
529}
530
531static void etp_submit (ETP_REQ *req)
532{
533 req->pri -= ETP_PRI_MIN;
534
535 if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
536 if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
537
538 if (expect_false (req->type == EIO_GROUP))
539 {
540 /* I hope this is worth it :/ */
541 X_LOCK (reqlock);
542 ++nreqs;
543 X_UNLOCK (reqlock);
544
545 X_LOCK (reslock);
546
547 ++npending;
548
549 if (!reqq_push (&res_queue, req) && want_poll_cb)
550 want_poll_cb ();
551
552 X_UNLOCK (reslock);
553 }
554 else
555 {
556 X_LOCK (reqlock);
557 ++nreqs;
558 ++nready;
559 reqq_push (&req_queue, req);
560 X_COND_SIGNAL (reqwait);
561 X_UNLOCK (reqlock);
562
563 etp_maybe_start_thread ();
564 }
565}
566
567static void etp_set_max_poll_time (double nseconds)
568{
569 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
570 max_poll_time = nseconds;
571 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
572}
573
574static void etp_set_max_poll_reqs (unsigned int maxreqs)
575{
576 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
577 max_poll_reqs = maxreqs;
578 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
579}
580
581static void etp_set_max_idle (unsigned int nthreads)
582{
583 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
584 max_idle = nthreads <= 0 ? 1 : nthreads;
585 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
586}
587
588static void etp_set_min_parallel (unsigned int nthreads)
589{
590 if (wanted < nthreads)
591 wanted = nthreads;
592}
593
594static void etp_set_max_parallel (unsigned int nthreads)
595{
596 if (wanted > nthreads)
597 wanted = nthreads;
598
599 while (started > wanted)
600 etp_end_thread ();
601}
602
603/*****************************************************************************/
604
605static void grp_try_feed (eio_req *grp) 402grp_try_feed (eio_req *grp)
606{ 403{
607 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 404 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
608 { 405 {
609 grp->flags &= ~EIO_FLAG_GROUPADD; 406 grp->flags &= ~EIO_FLAG_GROUPADD;
610 407
… …
617 break; 414 break;
618 } 415 }
619 } 416 }
620} 417}
621 418
419static int
622static int grp_dec (eio_req *grp) 420grp_dec (eio_req *grp)
623{ 421{
624 --grp->size; 422 --grp->size;
625 423
626 /* call feeder, if applicable */ 424 /* call feeder, if applicable */
627 grp_try_feed (grp); 425 grp_try_feed (grp);
… …
631 return eio_finish (grp); 429 return eio_finish (grp);
632 else 430 else
633 return 0; 431 return 0;
634} 432}
635 433
434static void
636void eio_destroy (eio_req *req) 435eio_destroy (eio_req *req)
637{ 436{
638 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); 437 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
639 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); 438 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
640 439
641 EIO_DESTROY (req); 440 EIO_DESTROY (req);
642} 441}
643 442
443static int
644static int eio_finish (eio_req *req) 444eio_finish (eio_req *req)
645{ 445{
646 int res = EIO_FINISH (req); 446 int res = EIO_FINISH (req);
647 447
648 if (req->grp) 448 if (req->grp)
649 { 449 {
… …
657 if (grp->grp_first == req) 457 if (grp->grp_first == req)
658 grp->grp_first = req->grp_next; 458 grp->grp_first = req->grp_next;
659 459
660 res2 = grp_dec (grp); 460 res2 = grp_dec (grp);
661 461
662 if (!res && res2) 462 if (!res)
663 res = res2; 463 res = res2;
664 } 464 }
665 465
666 eio_destroy (req); 466 eio_destroy (req);
667 467
668 return res; 468 return res;
669} 469}
670 470
471void
671void eio_grp_cancel (eio_req *grp) 472eio_grp_cancel (eio_req *grp)
672{ 473{
673 for (grp = grp->grp_first; grp; grp = grp->grp_next)
674 eio_cancel (grp); 474 etp_grp_cancel (grp);
675} 475}
676 476
477void
677void eio_cancel (eio_req *req) 478eio_cancel (eio_req *req)
678{ 479{
679 etp_cancel (req); 480 etp_cancel (req);
680} 481}
681 482
483void
682void eio_submit (eio_req *req) 484eio_submit (eio_req *req)
683{ 485{
684 etp_submit (req); 486 etp_submit (req);
685} 487}
686 488
687unsigned int eio_nreqs (void) 489unsigned int
490eio_nreqs (void)
688{ 491{
689 return etp_nreqs (); 492 return etp_nreqs ();
690} 493}
691 494
692unsigned int eio_nready (void) 495unsigned int
496eio_nready (void)
693{ 497{
694 return etp_nready (); 498 return etp_nready ();
695} 499}
696 500
697unsigned int eio_npending (void) 501unsigned int
502eio_npending (void)
698{ 503{
699 return etp_npending (); 504 return etp_npending ();
700} 505}
701 506
702unsigned int eio_nthreads (void) 507unsigned int ecb_cold
508eio_nthreads (void)
703{ 509{
704 return etp_nthreads (); 510 return etp_nthreads ();
705} 511}
706 512
513void ecb_cold
707void eio_set_max_poll_time (double nseconds) 514eio_set_max_poll_time (double nseconds)
708{ 515{
709 etp_set_max_poll_time (nseconds); 516 etp_set_max_poll_time (nseconds);
710} 517}
711 518
519void ecb_cold
712void eio_set_max_poll_reqs (unsigned int maxreqs) 520eio_set_max_poll_reqs (unsigned int maxreqs)
713{ 521{
714 etp_set_max_poll_reqs (maxreqs); 522 etp_set_max_poll_reqs (maxreqs);
715} 523}
716 524
525void ecb_cold
717void eio_set_max_idle (unsigned int nthreads) 526eio_set_max_idle (unsigned int nthreads)
718{ 527{
719 etp_set_max_idle (nthreads); 528 etp_set_max_idle (nthreads);
720} 529}
721 530
531void ecb_cold
532eio_set_idle_timeout (unsigned int seconds)
533{
534 etp_set_idle_timeout (seconds);
535}
536
537void ecb_cold
722void eio_set_min_parallel (unsigned int nthreads) 538eio_set_min_parallel (unsigned int nthreads)
723{ 539{
724 etp_set_min_parallel (nthreads); 540 etp_set_min_parallel (nthreads);
725} 541}
726 542
543void ecb_cold
727void eio_set_max_parallel (unsigned int nthreads) 544eio_set_max_parallel (unsigned int nthreads)
728{ 545{
729 etp_set_max_parallel (nthreads); 546 etp_set_max_parallel (nthreads);
730} 547}
731 548
732int eio_poll (void) 549int eio_poll (void)
… …
735} 552}
736 553
737/*****************************************************************************/ 554/*****************************************************************************/
738/* work around various missing functions */ 555/* work around various missing functions */
739 556
740#if !HAVE_PREADWRITE
741# undef pread
742# undef pwrite
743# define pread eio__pread
744# define pwrite eio__pwrite
745
746static ssize_t
747eio__pread (int fd, void *buf, size_t count, off_t offset)
748{
749 ssize_t res;
750 off_t ooffset;
751
752 X_LOCK (preadwritelock);
753 ooffset = lseek (fd, 0, SEEK_CUR);
754 lseek (fd, offset, SEEK_SET);
755 res = read (fd, buf, count);
756 lseek (fd, ooffset, SEEK_SET);
757 X_UNLOCK (preadwritelock);
758
759 return res;
760}
761
762static ssize_t
763eio__pwrite (int fd, void *buf, size_t count, off_t offset)
764{
765 ssize_t res;
766 off_t ooffset;
767
768 X_LOCK (preadwritelock);
769 ooffset = lseek (fd, 0, SEEK_CUR);
770 lseek (fd, offset, SEEK_SET);
771 res = write (fd, buf, count);
772 lseek (fd, offset, SEEK_SET);
773 X_UNLOCK (preadwritelock);
774
775 return res;
776}
777#endif
778
779#ifndef HAVE_FUTIMES 557#ifndef HAVE_UTIMES
780 558
781# undef utimes 559# undef utimes
782# undef futimes
783# define utimes(path,times) eio__utimes (path, times) 560# define utimes(path,times) eio__utimes (path, times)
784# define futimes(fd,times) eio__futimes (fd, times)
785 561
786static int 562static int
787eio__utimes (const char *filename, const struct timeval times[2]) 563eio__utimes (const char *filename, const struct timeval times[2])
788{ 564{
789 if (times) 565 if (times)
… …
797 } 573 }
798 else 574 else
799 return utime (filename, 0); 575 return utime (filename, 0);
800} 576}
801 577
578#endif
579
580#ifndef HAVE_FUTIMES
581
582# undef futimes
583# define futimes(fd,times) eio__futimes (fd, times)
584
585static int
802static int eio__futimes (int fd, const struct timeval tv[2]) 586eio__futimes (int fd, const struct timeval tv[2])
803{ 587{
804 errno = ENOSYS; 588 errno = ENOSYS;
805 return -1; 589 return -1;
806} 590}
807 591
… …
810#if !HAVE_FDATASYNC 594#if !HAVE_FDATASYNC
811# undef fdatasync 595# undef fdatasync
812# define fdatasync(fd) fsync (fd) 596# define fdatasync(fd) fsync (fd)
813#endif 597#endif
814 598
599static int
600eio__syncfs (int fd)
601{
602 int res;
603
604#if HAVE_SYS_SYNCFS
605 res = (int)syscall (__NR_syncfs, (int)(fd));
606#else
607 res = EIO_ENOSYS ();
608#endif
609
610 if (res < 0 && errno == ENOSYS && fd >= 0)
611 sync ();
612
613 return res;
614}
615
815/* sync_file_range always needs emulation */ 616/* sync_file_range always needs emulation */
816int 617static int
817eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags) 618eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
818{ 619{
819#if HAVE_SYNC_FILE_RANGE 620#if HAVE_SYNC_FILE_RANGE
820 int res; 621 int res;
821 622
… …
829 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0); 630 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
830 } 631 }
831 632
832 res = sync_file_range (fd, offset, nbytes, flags); 633 res = sync_file_range (fd, offset, nbytes, flags);
833 634
834 if (res != ENOSYS) 635 if (!res || errno != ENOSYS)
835 return res; 636 return res;
836#endif 637#endif
837 638
838 /* even though we could play tricks with the flags, it's better to always 639 /* even though we could play tricks with the flags, it's better to always
839 * call fdatasync, as thta matches the expectation of it's users best */ 640 * call fdatasync, as that matches the expectation of its users best */
840 return fdatasync (fd); 641 return fdatasync (fd);
642}
643
644static int
645eio__fallocate (int fd, int mode, off_t offset, size_t len)
646{
647#if HAVE_LINUX_FALLOCATE
648 return fallocate (fd, mode, offset, len);
649#else
650 return EIO_ENOSYS ();
651#endif
841} 652}
842 653
843#if !HAVE_READAHEAD 654#if !HAVE_READAHEAD
844# undef readahead 655# undef readahead
845# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 656# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
846 657
847static ssize_t 658static eio_ssize_t
848eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) 659eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
849{ 660{
850 size_t todo = count; 661 size_t todo = count;
851 dBUF; 662 dBUF;
852 663
… …
857 pread (fd, eio_buf, len, offset); 668 pread (fd, eio_buf, len, offset);
858 offset += len; 669 offset += len;
859 todo -= len; 670 todo -= len;
860 } 671 }
861 672
862 errno = 0; 673 FUBd;
674
675 /* linux's readahead basically only fails for EBADF or EINVAL (not mmappable) */
676 /* but not for e.g. EIO or eof, so we also never fail */
863 return count; 677 return 0;
864} 678}
865 679
866#endif 680#endif
867 681
868/* sendfile always needs emulation */ 682/* sendfile always needs emulation */
869static ssize_t 683static eio_ssize_t
870eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) 684eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
871{ 685{
686 eio_ssize_t written = 0;
872 ssize_t res; 687 eio_ssize_t res;
873 688
874 if (!count) 689 if (!count)
875 return 0; 690 return 0;
876 691
692 for (;;)
693 {
694#ifdef __APPLE__
695# undef HAVE_SENDFILE /* broken, as everything on os x */
696#endif
877#if HAVE_SENDFILE 697#if HAVE_SENDFILE
878# if __linux 698# if __linux
699 off_t soffset = offset;
879 res = sendfile (ofd, ifd, &offset, count); 700 res = sendfile (ofd, ifd, &soffset, count);
880 701
881# elif __freebsd 702# elif __FreeBSD__
882 /* 703 /*
883 * Of course, the freebsd sendfile is a dire hack with no thoughts 704 * Of course, the freebsd sendfile is a dire hack with no thoughts
884 * wasted on making it similar to other I/O functions. 705 * wasted on making it similar to other I/O functions.
885 */ 706 */
886 {
887 off_t sbytes; 707 off_t sbytes;
888 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 708 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
889 709
890 if (res < 0 && sbytes) 710 #if 0 /* according to the manpage, this is correct, but broken behaviour */
891 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 711 /* freebsd' sendfile will return 0 on success */
712 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
713 /* not on e.g. EIO or EPIPE - sounds broken */
714 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
892 res = sbytes; 715 res = sbytes;
893 } 716 #endif
717
718 /* according to source inspection, this is correct, and useful behaviour */
719 if (sbytes)
720 res = sbytes;
721
722# elif defined __APPLE__
723 off_t sbytes = count;
724 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
725
726 /* according to the manpage, sbytes is always valid */
727 if (sbytes)
728 res = sbytes;
894 729
895# elif __hpux 730# elif __hpux
896 res = sendfile (ofd, ifd, offset, count, 0, 0); 731 res = sendfile (ofd, ifd, offset, count, 0, 0);
897 732
898# elif __solaris 733# elif __solaris
899 {
900 struct sendfilevec vec; 734 struct sendfilevec vec;
901 size_t sbytes; 735 size_t sbytes;
902 736
903 vec.sfv_fd = ifd; 737 vec.sfv_fd = ifd;
904 vec.sfv_flag = 0; 738 vec.sfv_flag = 0;
905 vec.sfv_off = offset; 739 vec.sfv_off = offset;
906 vec.sfv_len = count; 740 vec.sfv_len = count;
907 741
908 res = sendfilev (ofd, &vec, 1, &sbytes); 742 res = sendfilev (ofd, &vec, 1, &sbytes);
909 743
910 if (res < 0 && sbytes) 744 if (res < 0 && sbytes)
911 res = sbytes; 745 res = sbytes;
912 }
913 746
914# endif 747# endif
748
749#elif defined (_WIN32) && 0
750 /* does not work, just for documentation of what would need to be done */
751 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
752 /* libeio guarantees that the file offset does not change, and windows */
753 /* has no way to get an independent handle to the same file description */
754 HANDLE h = TO_SOCKET (ifd);
755 SetFilePointer (h, offset, 0, FILE_BEGIN);
756 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
757
915#else 758#else
916 res = -1; 759 res = EIO_ENOSYS ();
917 errno = ENOSYS;
918#endif 760#endif
919 761
762 /* we assume sendfile can copy at least 128mb in one go */
763 if (res <= 128 * 1024 * 1024)
764 {
765 if (res > 0)
766 written += res;
767
768 if (written)
769 return written;
770
771 break;
772 }
773 else
774 {
775 /* if we requested more, then probably the kernel was lazy */
776 written += res;
777 offset += res;
778 count -= res;
779
780 if (!count)
781 return written;
782 }
783 }
784
920 if (res < 0 785 if (res < 0
921 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 786 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
787 /* BSDs */
788#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
789 || errno == ENOTSUP
790#endif
791#ifdef EOPNOTSUPP /* windows */
792 || errno == EOPNOTSUPP /* BSDs */
793#endif
922#if __solaris 794#if __solaris
923 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 795 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
924#endif 796#endif
925 ) 797 )
926 ) 798 )
… …
930 802
931 res = 0; 803 res = 0;
932 804
933 while (count) 805 while (count)
934 { 806 {
935 ssize_t cnt; 807 eio_ssize_t cnt;
936 808
937 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 809 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
938 810
939 if (cnt <= 0) 811 if (cnt <= 0)
940 { 812 {
… …
952 824
953 offset += cnt; 825 offset += cnt;
954 res += cnt; 826 res += cnt;
955 count -= cnt; 827 count -= cnt;
956 } 828 }
829
830 FUBd;
957 } 831 }
958 832
959 return res; 833 return res;
834}
835
836#ifdef PAGESIZE
837# define eio_pagesize() PAGESIZE
838#else
839static intptr_t
840eio_pagesize (void)
841{
842 static intptr_t page;
843
844 if (!page)
845 page = sysconf (_SC_PAGESIZE);
846
847 return page;
848}
849#endif
850
851static void
852eio_page_align (void **addr, size_t *length)
853{
854 intptr_t mask = eio_pagesize () - 1;
855
856 /* round down addr */
857 intptr_t adj = mask & (intptr_t)*addr;
858
859 *addr = (void *)((intptr_t)*addr - adj);
860 *length += adj;
861
862 /* round up length */
863 *length = (*length + mask) & ~mask;
864}
865
866#if !_POSIX_MEMLOCK
867# define eio__mlockall(a) EIO_ENOSYS ()
868#else
869
870static int
871eio__mlockall (int flags)
872{
873 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
874 extern int mallopt (int, int);
875 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
876 #endif
877
878 if (EIO_MCL_CURRENT != MCL_CURRENT
879 || EIO_MCL_FUTURE != MCL_FUTURE)
880 {
881 flags = 0
882 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
883 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
884 }
885
886 return mlockall (flags);
887}
888#endif
889
890#if !_POSIX_MEMLOCK_RANGE
891# define eio__mlock(a,b) EIO_ENOSYS ()
892#else
893
894static int
895eio__mlock (void *addr, size_t length)
896{
897 eio_page_align (&addr, &length);
898
899 return mlock (addr, length);
900}
901
902#endif
903
904#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
905# define eio__msync(a,b,c) EIO_ENOSYS ()
906#else
907
908static int
909eio__msync (void *mem, size_t len, int flags)
910{
911 eio_page_align (&mem, &len);
912
913 if (EIO_MS_ASYNC != MS_SYNC
914 || EIO_MS_INVALIDATE != MS_INVALIDATE
915 || EIO_MS_SYNC != MS_SYNC)
916 {
917 flags = 0
918 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
919 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
920 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
921 }
922
923 return msync (mem, len, flags);
924}
925
926#endif
927
928static int
929eio__mtouch (eio_req *req)
930{
931 void *mem = req->ptr2;
932 size_t len = req->size;
933 int flags = req->int1;
934
935 eio_page_align (&mem, &len);
936
937 {
938 intptr_t addr = (intptr_t)mem;
939 intptr_t end = addr + len;
940 intptr_t page = eio_pagesize ();
941
942 if (addr < end)
943 if (flags & EIO_MT_MODIFY) /* modify */
944 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
945 else
946 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
947 }
948
949 return 0;
950}
951
952/*****************************************************************************/
953/* requests implemented outside eio_execute, because they are so large */
954
955static void
956eio__lseek (eio_req *req)
957{
958 /* this usually gets optimised away completely, or your compiler sucks, */
959 /* or the whence constants really are not 0, 1, 2 */
960 int whence = req->int2 == EIO_SEEK_SET ? SEEK_SET
961 : req->int2 == EIO_SEEK_CUR ? SEEK_CUR
962 : req->int2 == EIO_SEEK_END ? SEEK_END
963 : req->int2;
964
965 req->offs = lseek (req->int1, req->offs, whence);
966 req->result = req->offs == (off_t)-1 ? -1 : 0;
967}
968
969/* result will always end up in tmpbuf, there is always space for adding a 0-byte */
970static int
971eio__realpath (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
972{
973 const char *rel = path;
974 char *res;
975 char *tmp1, *tmp2;
976#if SYMLOOP_MAX > 32
977 int symlinks = SYMLOOP_MAX;
978#else
979 int symlinks = 32;
980#endif
981
982 errno = EINVAL;
983 if (!rel)
984 return -1;
985
986 errno = ENOENT;
987 if (!*rel)
988 return -1;
989
990 res = tmpbuf_get (tmpbuf, PATH_MAX * 3);
991 tmp1 = res + PATH_MAX;
992 tmp2 = tmp1 + PATH_MAX;
993
994#if 0 /* disabled, the musl way to do things is just too racy */
995#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
996 /* on linux we may be able to ask the kernel */
997 {
998 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
999
1000 if (fd >= 0)
1001 {
1002 sprintf (tmp1, "/proc/self/fd/%d", fd);
1003 req->result = readlink (tmp1, res, PATH_MAX);
1004 /* here we should probably stat the open file and the disk file, to make sure they still match */
1005 close (fd);
1006
1007 if (req->result > 0)
1008 goto done;
1009 }
1010 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1011 return -1;
1012 }
1013#endif
1014#endif
1015
1016 if (*rel != '/')
1017 {
1018 int len;
1019
1020 errno = ENOENT;
1021 if (wd == EIO_INVALID_WD)
1022 return -1;
1023
1024 if (wd == EIO_CWD)
1025 {
1026 if (!getcwd (res, PATH_MAX))
1027 return -1;
1028
1029 len = strlen (res);
1030 }
1031 else
1032 memcpy (res, wd->str, len = wd->len);
1033
1034 if (res [1]) /* only use if not / */
1035 res += len;
1036 }
1037
1038 while (*rel)
1039 {
1040 eio_ssize_t len, linklen;
1041 const char *beg = rel;
1042
1043 while (*rel && *rel != '/')
1044 ++rel;
1045
1046 len = rel - beg;
1047
1048 if (!len) /* skip slashes */
1049 {
1050 ++rel;
1051 continue;
1052 }
1053
1054 if (beg [0] == '.')
1055 {
1056 if (len == 1)
1057 continue; /* . - nop */
1058
1059 if (beg [1] == '.' && len == 2)
1060 {
1061 /* .. - back up one component, if possible */
1062
1063 while (res != tmpbuf->ptr)
1064 if (*--res == '/')
1065 break;
1066
1067 continue;
1068 }
1069 }
1070
1071 errno = ENAMETOOLONG;
1072 if (res + 1 + len + 1 >= tmp1)
1073 return -1;
1074
1075 /* copy one component */
1076 *res = '/';
1077 memcpy (res + 1, beg, len);
1078
1079 /* zero-terminate, for readlink */
1080 res [len + 1] = 0;
1081
1082 /* now check if it's a symlink */
1083 linklen = readlink (tmpbuf->ptr, tmp1, PATH_MAX);
1084
1085 if (linklen < 0)
1086 {
1087 if (errno != EINVAL)
1088 return -1;
1089
1090 /* it's a normal directory. hopefully */
1091 res += len + 1;
1092 }
1093 else
1094 {
1095 /* yay, it was a symlink - build new path in tmp2 */
1096 int rellen = strlen (rel);
1097
1098 errno = ENAMETOOLONG;
1099 if (linklen + 1 + rellen >= PATH_MAX)
1100 return -1;
1101
1102 errno = ELOOP;
1103 if (!--symlinks)
1104 return -1;
1105
1106 if (*tmp1 == '/')
1107 res = tmpbuf->ptr; /* symlink resolves to an absolute path */
1108
1109 /* we need to be careful, as rel might point into tmp2 already */
1110 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1111 tmp2 [linklen] = '/';
1112 memcpy (tmp2, tmp1, linklen);
1113
1114 rel = tmp2;
1115 }
1116 }
1117
1118 /* special case for the lone root path */
1119 if (res == tmpbuf->ptr)
1120 *res++ = '/';
1121
1122 return res - (char *)tmpbuf->ptr;
1123}
1124
1125static signed char
1126eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1127{
1128 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1129 : a->inode < b->inode ? -1
1130 : a->inode > b->inode ? 1
1131 : 0;
1132}
1133
1134#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1135
1136#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1137#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1138
1139static void
1140eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1141{
1142 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1143 unsigned char *bit = bits;
1144
1145 assert (CHAR_BIT == 8);
1146 assert (sizeof (eio_dirent) * 8 < 256);
1147 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1148 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1149
1150 if (size <= EIO_SORT_FAST)
1151 return;
1152
1153 /* first prepare an array of bits to test in our radix sort */
1154 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1155 /* inode_bits must contain all inodes ORed together */
1156 /* which is used to skip bits that are 0 everywhere, which is very common */
1157 {
1158 eio_ino_t endianness;
1159 int i, j;
1160
1161 /* we store the byte offset of byte n into byte n of "endianness" */
1162 for (i = 0; i < sizeof (eio_ino_t); ++i)
1163 ((unsigned char *)&endianness)[i] = i;
1164
1165 *bit++ = 0;
1166
1167 for (i = 0; i < sizeof (eio_ino_t); ++i)
1168 {
1169 /* shifting off the byte offsets out of "endianness" */
1170 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1171 endianness >>= 8;
1172
1173 for (j = 0; j < 8; ++j)
1174 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1175 *bit++ = offs + j;
1176 }
1177
1178 for (j = 0; j < 8; ++j)
1179 if (score_bits & (1 << j))
1180 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1181 }
1182
1183 /* now actually do the sorting (a variant of MSD radix sort) */
1184 {
1185 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1186 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1187 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1188 int stk_idx = 0;
1189
1190 base_stk [stk_idx] = dents;
1191 end_stk [stk_idx] = dents + size;
1192 bit_stk [stk_idx] = bit - 1;
1193
1194 do
1195 {
1196 base = base_stk [stk_idx];
1197 end = end_stk [stk_idx];
1198 bit = bit_stk [stk_idx];
1199
1200 for (;;)
1201 {
1202 unsigned char O = *bit >> 3;
1203 unsigned char M = 1 << (*bit & 7);
1204
1205 eio_dirent *a = base;
1206 eio_dirent *b = end;
1207
1208 if (b - a < EIO_SORT_CUTOFF)
1209 break;
1210
1211 /* now bit-partition the array on the bit */
1212 /* this ugly asymmetric loop seems to perform much better than typical */
1213 /* partition algos found in the literature */
1214 do
1215 if (!(((unsigned char *)a)[O] & M))
1216 ++a;
1217 else if (!(((unsigned char *)--b)[O] & M))
1218 {
1219 eio_dirent tmp = *a; *a = *b; *b = tmp;
1220 ++a;
1221 }
1222 while (b > a);
1223
1224 /* next bit, or stop, if no bits left in this path */
1225 if (!*--bit)
1226 break;
1227
1228 base_stk [stk_idx] = a;
1229 end_stk [stk_idx] = end;
1230 bit_stk [stk_idx] = bit;
1231 ++stk_idx;
1232
1233 end = a;
1234 }
1235 }
1236 while (stk_idx--);
1237 }
1238}
1239
1240static void
1241eio_dent_insertion_sort (eio_dirent *dents, int size)
1242{
1243 /* first move the smallest element to the front, to act as a sentinel */
1244 {
1245 int i;
1246 eio_dirent *min = dents;
1247
1248 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1249 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1250 if (EIO_DENT_CMP (dents [i], <, *min))
1251 min = &dents [i];
1252
1253 /* swap elements 0 and j (minimum) */
1254 {
1255 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1256 }
1257 }
1258
1259 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1260 {
1261 eio_dirent *i, *j;
1262
1263 for (i = dents + 1; i < dents + size; ++i)
1264 {
1265 eio_dirent value = *i;
1266
1267 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1268 j [1] = j [0];
1269
1270 j [1] = value;
1271 }
1272 }
1273}
1274
1275static void
1276eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1277{
1278 if (size <= 1)
1279 return; /* our insertion sort relies on size > 0 */
1280
1281 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1282 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1283 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1284
1285 /* use an insertion sort at the end, or for small arrays, */
1286 /* as insertion sort is more efficient for small partitions */
1287 eio_dent_insertion_sort (dents, size);
960} 1288}
961 1289
962/* read a full directory */ 1290/* read a full directory */
963static void 1291static void
964eio__scandir (eio_req *req, etp_worker *self) 1292eio__scandir (eio_req *req, etp_worker *self)
965{ 1293{
1294 char *name, *names;
1295 int namesalloc = 4096 - sizeof (void *) * 4;
1296 int namesoffs = 0;
1297 int flags = req->int1;
1298 eio_dirent *dents = 0;
1299 int dentalloc = 128;
1300 int dentoffs = 0;
1301 eio_ino_t inode_bits = 0;
1302#ifdef _WIN32
1303 HANDLE dirp;
1304 WIN32_FIND_DATA entp;
1305#else
966 DIR *dirp; 1306 DIR *dirp;
967 EIO_STRUCT_DIRENT *entp; 1307 EIO_STRUCT_DIRENT *entp;
968 char *name, *names; 1308#endif
969 int memlen = 4096;
970 int memofs = 0;
971 int res = 0;
972 1309
973 X_LOCK (wrklock); 1310 req->result = -1;
974 /* the corresponding closedir is in ETP_WORKER_CLEAR */
975 self->dirp = dirp = opendir (req->ptr1);
976 req->flags |= EIO_FLAG_PTR2_FREE;
977 req->ptr2 = names = malloc (memlen);
978 X_UNLOCK (wrklock);
979 1311
980 if (dirp && names) 1312 if (!(flags & EIO_READDIR_DENTS))
981 for (;;) 1313 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1314
1315#ifdef _WIN32
1316 {
1317 int len = strlen ((const char *)req->ptr1);
1318 char *path = malloc (MAX_PATH);
1319 const char *fmt;
1320 const char *reqpath = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1321
1322 if (!len)
1323 fmt = "./*";
1324 else if (reqpath[len - 1] == '/' || reqpath[len - 1] == '\\')
1325 fmt = "%s*";
1326 else
1327 fmt = "%s/*";
1328
1329 _snprintf (path, MAX_PATH, fmt, reqpath);
1330 dirp = FindFirstFile (path, &entp);
1331 free (path);
1332
1333 if (dirp == INVALID_HANDLE_VALUE)
1334 {
1335 /* should steal _dosmaperr */
1336 switch (GetLastError ())
1337 {
1338 case ERROR_FILE_NOT_FOUND:
1339 req->result = 0;
1340 break;
1341
1342 case ERROR_INVALID_NAME:
1343 case ERROR_PATH_NOT_FOUND:
1344 case ERROR_NO_MORE_FILES:
1345 errno = ENOENT;
1346 break;
1347
1348 case ERROR_NOT_ENOUGH_MEMORY:
1349 errno = ENOMEM;
1350 break;
1351
1352 default:
1353 errno = EINVAL;
1354 break;
1355 }
1356
1357 return;
1358 }
1359 }
1360#else
1361 #if HAVE_AT
1362 if (req->wd)
982 { 1363 {
1364 int fd = openat (WD2FD (req->wd), req->ptr1, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1365
1366 if (fd < 0)
1367 return;
1368
1369 dirp = fdopendir (fd);
1370
1371 if (!dirp)
1372 close (fd);
1373 }
1374 else
1375 dirp = opendir (req->ptr1);
1376 #else
1377 dirp = opendir (wd_expand (&self->tmpbuf, req->wd, req->ptr1));
1378 #endif
1379
1380 if (!dirp)
1381 return;
1382#endif
1383
1384 if (req->flags & EIO_FLAG_PTR1_FREE)
1385 free (req->ptr1);
1386
1387 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1388 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1389 req->ptr2 = names = malloc (namesalloc);
1390
1391 if (!names || (flags && !dents))
1392 return;
1393
1394 for (;;)
1395 {
1396 int done;
1397
1398#ifdef _WIN32
1399 done = !dirp;
1400#else
983 errno = 0; 1401 errno = 0;
984 entp = readdir (dirp); 1402 entp = readdir (dirp);
1403 done = !entp;
1404#endif
985 1405
1406 if (done)
1407 {
1408#ifndef _WIN32
1409 int old_errno = errno;
1410 closedir (dirp);
1411 errno = old_errno;
1412
986 if (!entp) 1413 if (errno)
1414 break;
1415#endif
1416
1417 /* sort etc. */
1418 req->int1 = flags;
1419 req->result = dentoffs;
1420
1421 if (flags & EIO_READDIR_STAT_ORDER)
1422 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1423 else if (flags & EIO_READDIR_DIRS_FIRST)
1424 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1425 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1426 else
1427 {
1428 /* in this case, all is known, and we just put dirs first and sort them */
1429 eio_dirent *oth = dents + dentoffs;
1430 eio_dirent *dir = dents;
1431
1432 /* now partition dirs to the front, and non-dirs to the back */
1433 /* by walking from both sides and swapping if necessary */
1434 while (oth > dir)
1435 {
1436 if (dir->type == EIO_DT_DIR)
1437 ++dir;
1438 else if ((--oth)->type == EIO_DT_DIR)
1439 {
1440 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1441
1442 ++dir;
1443 }
1444 }
1445
1446 /* now sort the dirs only (dirs all have the same score) */
1447 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1448 }
1449
987 break; 1450 break;
1451 }
988 1452
989 name = entp->d_name; 1453 /* now add the entry to our list(s) */
1454 name = D_NAME (entp);
990 1455
1456 /* skip . and .. entries */
991 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1457 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
992 { 1458 {
993 int len = strlen (name) + 1; 1459 int len = D_NAMLEN (entp) + 1;
994 1460
995 res++; 1461 while (ecb_expect_false (namesoffs + len > namesalloc))
996
997 while (memofs + len > memlen)
998 { 1462 {
999 memlen *= 2; 1463 namesalloc *= 2;
1000 X_LOCK (wrklock);
1001 req->ptr2 = names = realloc (names, memlen); 1464 req->ptr2 = names = realloc (names, namesalloc);
1002 X_UNLOCK (wrklock);
1003 1465
1004 if (!names) 1466 if (!names)
1467 break;
1468 }
1469
1470 memcpy (names + namesoffs, name, len);
1471
1472 if (dents)
1473 {
1474 struct eio_dirent *ent;
1475
1476 if (ecb_expect_false (dentoffs == dentalloc))
1477 {
1478 dentalloc *= 2;
1479 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1480
1481 if (!dents)
1005 break; 1482 break;
1006 } 1483 }
1007 1484
1008 memcpy (names + memofs, name, len); 1485 ent = dents + dentoffs;
1009 memofs += len; 1486
1487 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1488 ent->namelen = len - 1;
1489 ent->inode = D_INO (entp);
1490
1491 inode_bits |= ent->inode;
1492
1493 switch (D_TYPE (entp))
1494 {
1495 default:
1496 ent->type = EIO_DT_UNKNOWN;
1497 flags |= EIO_READDIR_FOUND_UNKNOWN;
1498 break;
1499
1500 #ifdef DT_FIFO
1501 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1502 #endif
1503 #ifdef DT_CHR
1504 case DT_CHR: ent->type = EIO_DT_CHR; break;
1505 #endif
1506 #ifdef DT_MPC
1507 case DT_MPC: ent->type = EIO_DT_MPC; break;
1508 #endif
1509 #ifdef DT_DIR
1510 case DT_DIR: ent->type = EIO_DT_DIR; break;
1511 #endif
1512 #ifdef DT_NAM
1513 case DT_NAM: ent->type = EIO_DT_NAM; break;
1514 #endif
1515 #ifdef DT_BLK
1516 case DT_BLK: ent->type = EIO_DT_BLK; break;
1517 #endif
1518 #ifdef DT_MPB
1519 case DT_MPB: ent->type = EIO_DT_MPB; break;
1520 #endif
1521 #ifdef DT_REG
1522 case DT_REG: ent->type = EIO_DT_REG; break;
1523 #endif
1524 #ifdef DT_NWK
1525 case DT_NWK: ent->type = EIO_DT_NWK; break;
1526 #endif
1527 #ifdef DT_CMP
1528 case DT_CMP: ent->type = EIO_DT_CMP; break;
1529 #endif
1530 #ifdef DT_LNK
1531 case DT_LNK: ent->type = EIO_DT_LNK; break;
1532 #endif
1533 #ifdef DT_SOCK
1534 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1535 #endif
1536 #ifdef DT_DOOR
1537 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1538 #endif
1539 #ifdef DT_WHT
1540 case DT_WHT: ent->type = EIO_DT_WHT; break;
1541 #endif
1542 }
1543
1544 ent->score = 7;
1545
1546 if (flags & EIO_READDIR_DIRS_FIRST)
1547 {
1548 if (ent->type == EIO_DT_UNKNOWN)
1549 {
1550 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1551 ent->score = 1;
1552 else if (!strchr (name, '.')) /* absence of dots indicate likely dirs */
1553 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1554 }
1555 else if (ent->type == EIO_DT_DIR)
1556 ent->score = 0;
1557 }
1010 } 1558 }
1559
1560 namesoffs += len;
1561 ++dentoffs;
1011 } 1562 }
1012 1563
1013 if (errno) 1564 if (EIO_CANCELLED (req))
1014 res = -1; 1565 {
1566 errno = ECANCELED;
1567 break;
1568 }
1569
1570#ifdef _WIN32
1571 if (!FindNextFile (dirp, &entp))
1572 {
1573 FindClose (dirp);
1574 dirp = 0;
1575 }
1576#endif
1577 }
1578}
1579
1580/*****************************************************************************/
1581/* working directory stuff */
1582/* various deficiencies in the posix 2008 api force us to */
1583/* keep the absolute path in string form at all times */
1584/* fuck yeah. */
1585
1586#if !HAVE_AT
1587
1588/* a bit like realpath, but usually faster because it doesn'T have to return */
1589/* an absolute or canonical path */
1590static const char *
1591wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
1592{
1593 if (!wd || *path == '/')
1594 return path;
1595
1596 if (path [0] == '.' && !path [1])
1597 return wd->str;
1598
1015 1599 {
1016 req->result = res; 1600 int l1 = wd->len;
1017} 1601 int l2 = strlen (path);
1018 1602
1019#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) 1603 char *res = tmpbuf_get (tmpbuf, l1 + l2 + 2);
1020# undef msync
1021# define msync(a,b,c) ENOSYS
1022#endif
1023 1604
1024int 1605 memcpy (res, wd->str, l1);
1025eio__mtouch (void *mem, size_t len, int flags) 1606 res [l1] = '/';
1026{ 1607 memcpy (res + l1 + 1, path, l2 + 1);
1027 intptr_t addr = (intptr_t)mem;
1028 intptr_t end = addr + len;
1029#ifdef PAGESIZE
1030 const intptr_t page = PAGESIZE;
1031#else
1032 static intptr_t page;
1033 1608
1034 if (!page) 1609 return res;
1035 page = sysconf (_SC_PAGESIZE); 1610 }
1036#endif 1611}
1037 1612
1038 addr &= ~(page - 1); /* assume page size is always a power of two */ 1613#endif
1039 1614
1040 if (addr < end) 1615static eio_wd
1041 if (flags) /* modify */ 1616eio__wd_open_sync (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
1042 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len); 1617{
1043 else 1618 int fd;
1044 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len); 1619 eio_wd res;
1620 int len = eio__realpath (tmpbuf, wd, path);
1045 1621
1622 if (len < 0)
1623 return EIO_INVALID_WD;
1624
1625#if HAVE_AT
1626 fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1627
1628 if (fd < 0)
1629 return EIO_INVALID_WD;
1630#endif
1631
1632 res = malloc (sizeof (*res) + len); /* one extra 0-byte */
1633
1634#if HAVE_AT
1635 res->fd = fd;
1636#endif
1637
1638 res->len = len;
1639 memcpy (res->str, tmpbuf->ptr, len);
1640 res->str [len] = 0;
1641
1642 return res;
1643}
1644
1645eio_wd
1646eio_wd_open_sync (eio_wd wd, const char *path)
1647{
1648 struct tmpbuf tmpbuf = { 0 };
1649 wd = eio__wd_open_sync (&tmpbuf, wd, path);
1650 free (tmpbuf.ptr);
1651
1046 return 0; 1652 return wd;
1047} 1653}
1654
1655void
1656eio_wd_close_sync (eio_wd wd)
1657{
1658 if (wd != EIO_INVALID_WD && wd != EIO_CWD)
1659 {
1660 #if HAVE_AT
1661 close (wd->fd);
1662 #endif
1663 free (wd);
1664 }
1665}
1666
1667#if HAVE_AT
1668
1669/* they forgot these */
1670
1671static int
1672eio__truncateat (int dirfd, const char *path, off_t length)
1673{
1674 int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
1675 int res;
1676
1677 if (fd < 0)
1678 return fd;
1679
1680 res = ftruncate (fd, length);
1681 close (fd);
1682 return res;
1683}
1684
1685static int
1686eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
1687{
1688 int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
1689 int res;
1690
1691 if (fd < 0)
1692 return fd;
1693
1694 res = fstatvfs (fd, buf);
1695 close (fd);
1696 return res;
1697
1698}
1699
1700#endif
1048 1701
1049/*****************************************************************************/ 1702/*****************************************************************************/
1050 1703
1051#define ALLOC(len) \ 1704#define ALLOC(len) \
1052 if (!req->ptr2) \ 1705 if (!req->ptr2) \
… …
1061 req->result = -1; \ 1714 req->result = -1; \
1062 break; \ 1715 break; \
1063 } \ 1716 } \
1064 } 1717 }
1065 1718
1719static void ecb_noinline ecb_cold
1720etp_proc_init (void)
1721{
1722#if HAVE_PRCTL_SET_NAME
1723 /* provide a more sensible "thread name" */
1724 char name[16 + 1];
1725 const int namelen = sizeof (name) - 1;
1726 int len;
1727
1728 prctl (PR_GET_NAME, (unsigned long)name, 0, 0, 0);
1729 name [namelen] = 0;
1730 len = strlen (name);
1731 strcpy (name + (len <= namelen - 4 ? len : namelen - 4), "/eio");
1732 prctl (PR_SET_NAME, (unsigned long)name, 0, 0, 0);
1733#endif
1734}
1735
1736/* TODO: move somehow to etp.c */
1066X_THREAD_PROC (etp_proc) 1737X_THREAD_PROC (etp_proc)
1067{ 1738{
1068 ETP_REQ *req; 1739 ETP_REQ *req;
1069 struct timespec ts; 1740 struct timespec ts;
1070 etp_worker *self = (etp_worker *)thr_arg; 1741 etp_worker *self = (etp_worker *)thr_arg;
1071 1742
1743 etp_proc_init ();
1744
1072 /* try to distribute timeouts somewhat randomly */ 1745 /* try to distribute timeouts somewhat evenly */
1073 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 1746 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1074 1747
1075 for (;;) 1748 for (;;)
1076 { 1749 {
1750 ts.tv_sec = 0;
1751
1077 X_LOCK (reqlock); 1752 X_LOCK (reqlock);
1078 1753
1079 for (;;) 1754 for (;;)
1080 { 1755 {
1081 self->req = req = reqq_shift (&req_queue); 1756 req = reqq_shift (&req_queue);
1082 1757
1083 if (req) 1758 if (req)
1084 break; 1759 break;
1085 1760
1761 if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
1762 {
1763 X_UNLOCK (reqlock);
1764 X_LOCK (wrklock);
1765 --started;
1766 X_UNLOCK (wrklock);
1767 goto quit;
1768 }
1769
1086 ++idle; 1770 ++idle;
1087 1771
1088 ts.tv_sec = time (0) + IDLE_TIMEOUT; 1772 if (idle <= max_idle)
1089 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 1773 /* we are allowed to idle, so do so without any timeout */
1774 X_COND_WAIT (reqwait, reqlock);
1775 else
1090 { 1776 {
1091 if (idle > max_idle) 1777 /* initialise timeout once */
1092 { 1778 if (!ts.tv_sec)
1093 --idle; 1779 ts.tv_sec = time (0) + idle_timeout;
1094 X_UNLOCK (reqlock);
1095 X_LOCK (wrklock);
1096 --started;
1097 X_UNLOCK (wrklock);
1098 goto quit;
1099 }
1100 1780
1101 /* we are allowed to idle, so do so without any timeout */
1102 X_COND_WAIT (reqwait, reqlock); 1781 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
1782 ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1103 } 1783 }
1104 1784
1105 --idle; 1785 --idle;
1106 } 1786 }
1107 1787
1108 --nready; 1788 --nready;
1109 1789
1110 X_UNLOCK (reqlock); 1790 X_UNLOCK (reqlock);
1111 1791
1112 if (req->type < 0) 1792 if (req->type == ETP_TYPE_QUIT)
1113 goto quit; 1793 goto quit;
1114 1794
1115 if (!EIO_CANCELLED (req))
1116 ETP_EXECUTE (self, req); 1795 ETP_EXECUTE (self, req);
1117 1796
1118 X_LOCK (reslock); 1797 X_LOCK (reslock);
1119 1798
1120 ++npending; 1799 ++npending;
1121 1800
1122 if (!reqq_push (&res_queue, req) && want_poll_cb) 1801 if (!reqq_push (&res_queue, req) && want_poll_cb)
1123 want_poll_cb (); 1802 want_poll_cb ();
1124 1803
1125 self->req = 0;
1126 etp_worker_clear (self); 1804 etp_worker_clear (self);
1127 1805
1128 X_UNLOCK (reslock); 1806 X_UNLOCK (reslock);
1129 } 1807 }
1130 1808
1131quit: 1809quit:
1810 free (req);
1811
1132 X_LOCK (wrklock); 1812 X_LOCK (wrklock);
1133 etp_worker_free (self); 1813 etp_worker_free (self);
1134 X_UNLOCK (wrklock); 1814 X_UNLOCK (wrklock);
1135 1815
1136 return 0; 1816 return 0;
1137} 1817}
1138 1818
1139/*****************************************************************************/ 1819/*****************************************************************************/
1140 1820
1821int ecb_cold
1141int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1822eio_init (void (*want_poll)(void), void (*done_poll)(void))
1142{ 1823{
1143 return etp_init (want_poll, done_poll); 1824 return etp_init (want_poll, done_poll);
1144} 1825}
1145 1826
1827ecb_inline void
1146static void eio_api_destroy (eio_req *req) 1828eio_api_destroy (eio_req *req)
1147{ 1829{
1148 free (req); 1830 free (req);
1149} 1831}
1150 1832
1151#define REQ(rtype) \ 1833#define REQ(rtype) \
1152 eio_req *req; \ 1834 eio_req *req; \
1153 \ 1835 \
1154 req = (eio_req *)calloc (1, sizeof *req); \ 1836 req = (eio_req *)calloc (1, sizeof *req); \
1155 if (!req) \ 1837 if (!req) \
1156 return 0; \ 1838 return 0; \
… …
1170 { \ 1852 { \
1171 eio_api_destroy (req); \ 1853 eio_api_destroy (req); \
1172 return 0; \ 1854 return 0; \
1173 } 1855 }
1174 1856
1857#define SINGLEDOT(ptr) (0[(char *)(ptr)] == '.' && !1[(char *)(ptr)])
1858
1859static void
1175static void eio_execute (etp_worker *self, eio_req *req) 1860eio_execute (etp_worker *self, eio_req *req)
1176{ 1861{
1177 errno = 0; 1862#if HAVE_AT
1863 int dirfd;
1864#else
1865 const char *path;
1866#endif
1867
1868 if (ecb_expect_false (EIO_CANCELLED (req)))
1869 {
1870 req->result = -1;
1871 req->errorno = ECANCELED;
1872 return;
1873 }
1874
1875 if (ecb_expect_false (req->wd == EIO_INVALID_WD))
1876 {
1877 req->result = -1;
1878 req->errorno = ENOENT;
1879 return;
1880 }
1881
1882 if (req->type >= EIO_OPEN)
1883 {
1884 #if HAVE_AT
1885 dirfd = WD2FD (req->wd);
1886 #else
1887 path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1888 #endif
1889 }
1178 1890
1179 switch (req->type) 1891 switch (req->type)
1180 { 1892 {
1893 case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
1894 req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
1895 break;
1896 case EIO_WD_CLOSE: req->result = 0;
1897 eio_wd_close_sync (req->wd); break;
1898
1899 case EIO_SEEK: eio__lseek (req); break;
1181 case EIO_READ: ALLOC (req->size); 1900 case EIO_READ: ALLOC (req->size);
1182 req->result = req->offs >= 0 1901 req->result = req->offs >= 0
1183 ? pread (req->int1, req->ptr2, req->size, req->offs) 1902 ? pread (req->int1, req->ptr2, req->size, req->offs)
1184 : read (req->int1, req->ptr2, req->size); break; 1903 : read (req->int1, req->ptr2, req->size); break;
1185 case EIO_WRITE: req->result = req->offs >= 0 1904 case EIO_WRITE: req->result = req->offs >= 0
1186 ? pwrite (req->int1, req->ptr2, req->size, req->offs) 1905 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1187 : write (req->int1, req->ptr2, req->size); break; 1906 : write (req->int1, req->ptr2, req->size); break;
1188 1907
1189 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break; 1908 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1190 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break; 1909 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
1910
1911#if HAVE_AT
1191 1912
1192 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1913 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1193 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 1914 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break;
1194 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1915 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1195 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 1916 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, AT_SYMLINK_NOFOLLOW); break;
1196 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1197 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1198
1199 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; 1917 case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break;
1200 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1201 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break; 1918 case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break;
1202 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1203 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break; 1919 case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break;
1204 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1205
1206 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break; 1920 case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break;
1207 case EIO_CLOSE: req->result = close (req->int1); break; 1921
1208 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1209 case EIO_UNLINK: req->result = unlink (req->ptr1); break; 1922 case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
1210 case EIO_RMDIR: req->result = rmdir (req->ptr1); break; 1923 case EIO_RMDIR: /* complications arise because "." cannot be removed, so we might have to expand */
1924 req->result = req->wd && SINGLEDOT (req->ptr1)
1925 ? rmdir (req->wd->str)
1926 : unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break;
1211 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; 1927 case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); break;
1212 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; 1928 case EIO_RENAME: /* complications arise because "." cannot be renamed, so we might have to expand */
1929 req->result = req->wd && SINGLEDOT (req->ptr1)
1930 ? rename (req->wd->str, req->ptr2)
1931 : renameat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2); break;
1213 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; 1932 case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2, 0); break;
1214 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; 1933 case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
1215 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break; 1934 case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1216
1217 case EIO_READLINK: ALLOC (NAME_MAX); 1935 case EIO_READLINK: ALLOC (PATH_MAX);
1218 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; 1936 req->result = readlinkat (dirfd, req->ptr1, req->ptr2, PATH_MAX); break;
1219 1937 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1220 case EIO_SYNC: req->result = 0; sync (); break; 1938 req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1221 case EIO_FSYNC: req->result = fsync (req->int1); break;
1222 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1223 case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break;
1224 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;
1225 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1226
1227 case EIO_READDIR: eio__scandir (req, self); break;
1228
1229 case EIO_BUSY: 1939 case EIO_UTIME:
1230#ifdef _WIN32 1940 case EIO_FUTIME:
1231 Sleep (req->nv1 * 1000.); 1941 {
1942 struct timespec ts[2];
1943 struct timespec *times;
1944
1945 if (req->nv1 != -1. || req->nv2 != -1.)
1946 {
1947 ts[0].tv_sec = req->nv1;
1948 ts[0].tv_nsec = (req->nv1 - ts[0].tv_sec) * 1e9;
1949 ts[1].tv_sec = req->nv2;
1950 ts[1].tv_nsec = (req->nv2 - ts[1].tv_sec) * 1e9;
1951
1952 times = ts;
1953 }
1954 else
1955 times = 0;
1956
1957 req->result = req->type == EIO_FUTIME
1958 ? futimens (req->int1, times)
1959 : utimensat (dirfd, req->ptr1, times, 0);
1960 }
1961 break;
1962
1232#else 1963#else
1233 {
1234 struct timeval tv;
1235 1964
1236 tv.tv_sec = req->nv1; 1965 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1237 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; 1966 req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
1967 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1968 req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
1969 case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
1970 case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
1971 case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
1972 case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
1238 1973
1239 req->result = select (0, 0, 0, 0, &tv); 1974 case EIO_UNLINK: req->result = unlink (path ); break;
1240 } 1975 case EIO_RMDIR: req->result = rmdir (path ); break;
1241#endif 1976 case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
1242 break; 1977 case EIO_RENAME: req->result = rename (path , req->ptr2); break;
1978 case EIO_LINK: req->result = link (path , req->ptr2); break;
1979 case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
1980 case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
1981 case EIO_READLINK: ALLOC (PATH_MAX);
1982 req->result = readlink (path, req->ptr2, PATH_MAX); break;
1983 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1984 req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
1243 1985
1244 case EIO_UTIME: 1986 case EIO_UTIME:
1245 case EIO_FUTIME: 1987 case EIO_FUTIME:
1246 { 1988 {
1247 struct timeval tv[2]; 1989 struct timeval tv[2];
1248 struct timeval *times; 1990 struct timeval *times;
1249 1991
1250 if (req->nv1 != -1. || req->nv2 != -1.) 1992 if (req->nv1 != -1. || req->nv2 != -1.)
1251 { 1993 {
1252 tv[0].tv_sec = req->nv1; 1994 tv[0].tv_sec = req->nv1;
1253 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.; 1995 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
1254 tv[1].tv_sec = req->nv2; 1996 tv[1].tv_sec = req->nv2;
1255 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.; 1997 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
1256 1998
1257 times = tv; 1999 times = tv;
1258 } 2000 }
1259 else 2001 else
1260 times = 0; 2002 times = 0;
1261
1262 2003
1263 req->result = req->type == EIO_FUTIME 2004 req->result = req->type == EIO_FUTIME
1264 ? futimes (req->int1, times) 2005 ? futimes (req->int1, times)
1265 : utimes (req->ptr1, times); 2006 : utimes (req->ptr1, times);
1266 } 2007 }
1267 break; 2008 break;
1268 2009
2010#endif
2011
2012 case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
2013 {
2014 ALLOC (req->result);
2015 memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
2016 }
2017 break;
2018
2019 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2020 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
2021
2022 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2023 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2024
2025 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
2026 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
2027 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
2028
2029 case EIO_CLOSE: req->result = close (req->int1); break;
2030 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
2031 case EIO_SYNC: req->result = 0; sync (); break;
2032 case EIO_FSYNC: req->result = fsync (req->int1); break;
2033 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
2034 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
2035 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
2036 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
2037 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
2038 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
2039 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
2040 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
2041
2042 case EIO_READDIR: eio__scandir (req, self); break;
2043
2044 case EIO_BUSY:
2045#ifdef _WIN32
2046 Sleep (req->nv1 * 1e3);
2047#else
2048 {
2049 struct timeval tv;
2050
2051 tv.tv_sec = req->nv1;
2052 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
2053
2054 req->result = select (0, 0, 0, 0, &tv);
2055 }
2056#endif
2057 break;
2058
1269 case EIO_GROUP: 2059 case EIO_GROUP:
1270 abort (); /* handled in eio_request */ 2060 abort (); /* handled in eio_request */
1271 2061
1272 case EIO_NOP: 2062 case EIO_NOP:
1273 req->result = 0; 2063 req->result = 0;
1274 break; 2064 break;
1275 2065
1276 case EIO_CUSTOM: 2066 case EIO_CUSTOM:
1277 ((void (*)(eio_req *))req->feed) (req); 2067 req->feed (req);
1278 break; 2068 break;
1279 2069
1280 default: 2070 default:
1281 req->result = -1; 2071 req->result = EIO_ENOSYS ();
1282 break; 2072 break;
1283 } 2073 }
1284 2074
1285 req->errorno = errno; 2075 req->errorno = errno;
1286} 2076}
1287 2077
1288#ifndef EIO_NO_WRAPPERS 2078#ifndef EIO_NO_WRAPPERS
1289 2079
2080eio_req *eio_wd_open (const char *path, int pri, eio_cb cb, void *data)
2081{
2082 REQ (EIO_WD_OPEN); PATH; SEND;
2083}
2084
2085eio_req *eio_wd_close (eio_wd wd, int pri, eio_cb cb, void *data)
2086{
2087 REQ (EIO_WD_CLOSE); req->wd = wd; SEND;
2088}
2089
1290eio_req *eio_nop (int pri, eio_cb cb, void *data) 2090eio_req *eio_nop (int pri, eio_cb cb, void *data)
1291{ 2091{
1292 REQ (EIO_NOP); SEND; 2092 REQ (EIO_NOP); SEND;
1293} 2093}
1294 2094
… …
1310eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2110eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1311{ 2111{
1312 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2112 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1313} 2113}
1314 2114
2115eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2116{
2117 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2118}
2119
2120eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
2121{
2122 REQ (EIO_SYNCFS); req->int1 = fd; SEND;
2123}
2124
2125eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2126{
2127 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2128}
2129
1315eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2130eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1316{ 2131{
1317 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2132 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1318} 2133}
1319 2134
1320eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) 2135eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
1321{ 2136{
1322 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; 2137 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
1323} 2138}
1324 2139
1325eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) 2140eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
1326{ 2141{
1327 REQ (EIO_FDATASYNC); req->int1 = fd; SEND; 2142 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2143}
2144
2145eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2146{
2147 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
1328} 2148}
1329 2149
1330eio_req *eio_close (int fd, int pri, eio_cb cb, void *data) 2150eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
1331{ 2151{
1332 REQ (EIO_CLOSE); req->int1 = fd; SEND; 2152 REQ (EIO_CLOSE); req->int1 = fd; SEND;
… …
1335eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data) 2155eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
1336{ 2156{
1337 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND; 2157 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
1338} 2158}
1339 2159
2160eio_req *eio_seek (int fd, off_t offset, int whence, int pri, eio_cb cb, void *data)
2161{
2162 REQ (EIO_SEEK); req->int1 = fd; req->offs = offset; req->int2 = whence; SEND;
2163}
2164
1340eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data) 2165eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1341{ 2166{
1342 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND; 2167 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1343} 2168}
1344 2169
… …
1350eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) 2175eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1351{ 2176{
1352 REQ (EIO_FSTAT); req->int1 = fd; SEND; 2177 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1353} 2178}
1354 2179
2180eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
2181{
2182 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
2183}
2184
1355eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) 2185eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1356{ 2186{
1357 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; 2187 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1358} 2188}
1359 2189
… …
1365eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) 2195eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1366{ 2196{
1367 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2197 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1368} 2198}
1369 2199
1370eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2200eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1371{ 2201{
1372 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2202 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1373} 2203}
1374 2204
1375eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) 2205eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
… …
1395eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data) 2225eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1396{ 2226{
1397 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2227 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1398} 2228}
1399 2229
1400eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2230eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1401{ 2231{
1402 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2232 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1403} 2233}
1404 2234
1405eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) 2235eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
… …
1421eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data) 2251eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1422{ 2252{
1423 return eio__1path (EIO_READLINK, path, pri, cb, data); 2253 return eio__1path (EIO_READLINK, path, pri, cb, data);
1424} 2254}
1425 2255
2256eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2257{
2258 return eio__1path (EIO_REALPATH, path, pri, cb, data);
2259}
2260
1426eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) 2261eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1427{ 2262{
1428 return eio__1path (EIO_STAT, path, pri, cb, data); 2263 return eio__1path (EIO_STAT, path, pri, cb, data);
1429} 2264}
1430 2265
1431eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) 2266eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1432{ 2267{
1433 return eio__1path (EIO_LSTAT, path, pri, cb, data); 2268 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1434} 2269}
1435 2270
2271eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
2272{
2273 return eio__1path (EIO_STATVFS, path, pri, cb, data);
2274}
2275
1436eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) 2276eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1437{ 2277{
1438 return eio__1path (EIO_UNLINK, path, pri, cb, data); 2278 return eio__1path (EIO_UNLINK, path, pri, cb, data);
1439} 2279}
1440 2280
1441eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data) 2281eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1442{ 2282{
1443 return eio__1path (EIO_RMDIR, path, pri, cb, data); 2283 return eio__1path (EIO_RMDIR, path, pri, cb, data);
1444} 2284}
1445 2285
1446eio_req *eio_readdir (const char *path, int pri, eio_cb cb, void *data) 2286eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1447{ 2287{
1448 return eio__1path (EIO_READDIR, path, pri, cb, data); 2288 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1449} 2289}
1450 2290
1451eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) 2291eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1452{ 2292{
1453 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; 2293 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1454} 2294}
1455 2295
1456static eio_req * 2296static eio_req *
1457eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2297eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1458{ 2298{
… …
1482eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2322eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1483{ 2323{
1484 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2324 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1485} 2325}
1486 2326
1487eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) 2327eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
1488{ 2328{
1489 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND; 2329 REQ (EIO_CUSTOM); req->feed = execute; SEND;
1490} 2330}
1491 2331
1492#endif 2332#endif
1493 2333
1494eio_req *eio_grp (eio_cb cb, void *data) 2334eio_req *eio_grp (eio_cb cb, void *data)
… …
1503#undef SEND 2343#undef SEND
1504 2344
1505/*****************************************************************************/ 2345/*****************************************************************************/
1506/* grp functions */ 2346/* grp functions */
1507 2347
2348void
1508void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2349eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1509{ 2350{
1510 grp->int2 = limit; 2351 grp->int2 = limit;
1511 grp->feed = feed; 2352 grp->feed = feed;
1512 2353
1513 grp_try_feed (grp); 2354 grp_try_feed (grp);
1514} 2355}
1515 2356
2357void
1516void eio_grp_limit (eio_req *grp, int limit) 2358eio_grp_limit (eio_req *grp, int limit)
1517{ 2359{
1518 grp->int2 = limit; 2360 grp->int2 = limit;
1519 2361
1520 grp_try_feed (grp); 2362 grp_try_feed (grp);
1521} 2363}
1522 2364
2365void
1523void eio_grp_add (eio_req *grp, eio_req *req) 2366eio_grp_add (eio_req *grp, eio_req *req)
1524{ 2367{
1525 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2368 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1526 2369
1527 grp->flags |= EIO_FLAG_GROUPADD; 2370 grp->flags |= EIO_FLAG_GROUPADD;
1528 2371
… …
1539} 2382}
1540 2383
1541/*****************************************************************************/ 2384/*****************************************************************************/
1542/* misc garbage */ 2385/* misc garbage */
1543 2386
2387eio_ssize_t
1544ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2388eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1545{ 2389{
1546 etp_worker wrk;
1547
1548 wrk.dbuf = 0;
1549
1550 eio__sendfile (ofd, ifd, offset, count, &wrk); 2390 return eio__sendfile (ofd, ifd, offset, count);
1551
1552 if (wrk.dbuf)
1553 free (wrk.dbuf);
1554} 2391}
1555 2392

Diff Legend

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