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 1.134 by root, Thu Jun 25 15:59:57 2015 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 #define pipe(fds) _pipe (fds, 4096, O_BINARY)
126
127 #define fchmod(fd,mode) EIO_ENOSYS ()
128 #define chown(path,uid,gid) EIO_ENOSYS ()
129 #define fchown(fd,uid,gid) EIO_ENOSYS ()
130 #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
131 #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
132 #define mknod(path,mode,dev) EIO_ENOSYS ()
133 #define sync() EIO_ENOSYS ()
134 #define readlink(path,buf,s) EIO_ENOSYS ()
135 #define statvfs(path,buf) EIO_ENOSYS ()
136 #define fstatvfs(fd,buf) EIO_ENOSYS ()
137
138 #define pread(fd,buf,count,offset) eio__pread (fd, buf, count, offset)
139 #define pwrite(fd,buf,count,offset) eio__pwrite (fd, buf, count, offset)
140
141 #if __GNUC__
142 typedef long long eio_off_t; /* signed for compatibility to msvc */
143 #else
144 typedef __int64 eio_off_t; /* unsigned not supported by msvc */
145 #endif
146
147 static eio_ssize_t
148 eio__pread (int fd, void *buf, eio_ssize_t count, eio_off_t offset)
149 {
150 OVERLAPPED o = { 0 };
151 DWORD got;
152
153 o.Offset = offset;
154 o.OffsetHigh = offset >> 32;
155
156 return ReadFile ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd), buf, count, &got, &o)
157 ? got : -1;
158 }
159
160 static eio_ssize_t
161 eio__pwrite (int fd, void *buf, eio_ssize_t count, eio_off_t offset)
162 {
163 OVERLAPPED o = { 0 };
164 DWORD got;
165
166 o.Offset = offset;
167 o.OffsetHigh = offset >> 32;
168
169 return WriteFile ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd), buf, count, &got, &o)
170 ? got : -1;
171 }
172
173 /* rename() uses MoveFile, which fails to overwrite */
174 #define rename(old,neu) eio__rename (old, neu)
175
176 static int
177 eio__rename (const char *old, const char *neu)
178 {
179 if (MoveFileEx (old, neu, MOVEFILE_REPLACE_EXISTING))
180 return 0;
181
182 /* should steal _dosmaperr */
183 switch (GetLastError ())
184 {
185 case ERROR_FILE_NOT_FOUND:
186 case ERROR_PATH_NOT_FOUND:
187 case ERROR_INVALID_DRIVE:
188 case ERROR_NO_MORE_FILES:
189 case ERROR_BAD_NETPATH:
190 case ERROR_BAD_NET_NAME:
191 case ERROR_BAD_PATHNAME:
192 case ERROR_FILENAME_EXCED_RANGE:
193 errno = ENOENT;
194 break;
195
196 default:
197 errno = EACCES;
198 break;
199 }
200
201 return -1;
202 }
203
204 /* we could even stat and see if it exists */
205 static int
206 symlink (const char *old, const char *neu)
207 {
208 #if WINVER >= 0x0600
209 if (CreateSymbolicLink (neu, old, 1))
210 return 0;
211
212 if (CreateSymbolicLink (neu, old, 0))
213 return 0;
214 #endif
215
216 return EIO_ERRNO (ENOENT, -1);
217 }
218
219 /* POSIX API only, causing trouble for win32 apps */
220 #define CreateHardLink(neu,old,flags) 0 /* not really creating hardlink, still using relative paths? */
221 #define CreateSymbolicLink(neu,old,flags) 0 /* vista+ only */
222
223 struct statvfs
224 {
225 int dummy;
226 };
227
228 #define DT_DIR EIO_DT_DIR
229 #define DT_REG EIO_DT_REG
230 #define D_NAME(entp) entp.cFileName
231 #define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
69 232
70#else 233#else
71 234
72# include "config.h"
73# include <sys/time.h> 235 #include <sys/time.h>
74# include <sys/select.h> 236 #include <sys/select.h>
75# include <unistd.h> 237 #include <unistd.h>
238 #include <signal.h>
239 #include <dirent.h>
240
241 #ifdef ANDROID
242 #include <sys/vfs.h>
243 #define statvfs statfs
244 #define fstatvfs fstatfs
245 #include <asm/page.h> /* supposedly limits.h does #define PAGESIZE PAGESIZE */
246 #else
247 #include <sys/statvfs.h>
248 #endif
249
250 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
251 #include <sys/mman.h>
252 #endif
253
254 #define D_NAME(entp) entp->d_name
255
256 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
257 #if __FreeBSD__ || __NetBSD__ || __OpenBSD__
258 #define _DIRENT_HAVE_D_TYPE /* sigh */
259 #define D_INO(de) (de)->d_fileno
260 #define D_NAMLEN(de) (de)->d_namlen
261 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
262 #define D_INO(de) (de)->d_ino
263 #endif
264
265 #ifdef _D_EXACT_NAMLEN
266 #undef D_NAMLEN
267 #define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
268 #endif
269
270 #ifdef _DIRENT_HAVE_D_TYPE
271 #define D_TYPE(de) (de)->d_type
272 #endif
273
274 #ifndef EIO_STRUCT_DIRENT
275 #define EIO_STRUCT_DIRENT struct dirent
276 #endif
277
278#endif
279
280#if HAVE_UTIMES
76# include <utime.h> 281# 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 282#endif
83 283
284#if HAVE_SYS_SYSCALL_H
285# include <sys/syscall.h>
286#endif
287
288#if HAVE_SYS_PRCTL_H
289# include <sys/prctl.h>
84#endif 290#endif
85 291
86#if HAVE_SENDFILE 292#if HAVE_SENDFILE
87# if __linux 293# if __linux
88# include <sys/sendfile.h> 294# include <sys/sendfile.h>
89# elif __freebsd 295# elif __FreeBSD__ || defined __APPLE__
90# include <sys/socket.h> 296# include <sys/socket.h>
91# include <sys/uio.h> 297# include <sys/uio.h>
92# elif __hpux 298# elif __hpux
93# include <sys/socket.h> 299# include <sys/socket.h>
94# elif __solaris /* not yet */ 300# elif __solaris
95# include <sys/sendfile.h> 301# include <sys/sendfile.h>
96# else 302# else
97# error sendfile support requested but not available 303# error sendfile support requested but not available
98# endif 304# endif
99#endif 305#endif
100 306
101/* number of seconds after which an idle threads exit */ 307#ifndef D_TYPE
102#define IDLE_TIMEOUT 10 308# define D_TYPE(de) 0
309#endif
310#ifndef D_INO
311# define D_INO(de) 0
312#endif
313#ifndef D_NAMLEN
314# define D_NAMLEN(entp) strlen (D_NAME (entp))
315#endif
103 316
104/* used for struct dirent, AIX doesn't provide it */ 317/* used for struct dirent, AIX doesn't provide it */
105#ifndef NAME_MAX 318#ifndef NAME_MAX
106# define NAME_MAX 4096 319# define NAME_MAX 4096
107#endif 320#endif
108 321
322/* used for readlink etc. */
323#ifndef PATH_MAX
324# define PATH_MAX 4096
325#endif
326
109/* buffer size for various temporary buffers */ 327/* buffer size for various temporary buffers */
110#define EIO_BUFSIZE 65536 328#define EIO_BUFSIZE 65536
111 329
112#define dBUF \ 330#define dBUF \
113 char *eio_buf; \
114 ETP_WORKER_LOCK (self); \
115 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 331 char *eio_buf = malloc (EIO_BUFSIZE); \
116 ETP_WORKER_UNLOCK (self); \
117 errno = ENOMEM; \ 332 errno = ENOMEM; \
118 if (!eio_buf) \ 333 if (!eio_buf) \
119 return -1; 334 return -1
120 335
121#define EIO_TICKS ((1000000 + 1023) >> 10) 336#define FUBd \
337 free (eio_buf)
122 338
123/*****************************************************************************/ 339/*****************************************************************************/
124 340
125#if __GNUC__ >= 3 341struct etp_tmpbuf;
126# define expect(expr,value) __builtin_expect ((expr),(value)) 342
343#if _POSIX_VERSION >= 200809L
344 #define HAVE_AT 1
345 #define WD2FD(wd) ((wd) ? (wd)->fd : AT_FDCWD)
346 #ifndef O_SEARCH
347 #define O_SEARCH O_RDONLY
348 #endif
127#else 349#else
128# define expect(expr,value) (expr) 350 #define HAVE_AT 0
351 static const char *wd_expand (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path);
129#endif 352#endif
130 353
131#define expect_false(expr) expect ((expr) != 0, 0) 354struct eio_pwd
132#define expect_true(expr) expect ((expr) != 0, 1) 355{
356#if HAVE_AT
357 int fd;
358#endif
359 int len;
360 char str[1]; /* actually, a 0-terminated canonical path */
361};
133 362
134/*****************************************************************************/ 363/*****************************************************************************/
135 364
136#define ETP_PRI_MIN EIO_PRI_MIN 365#define ETP_PRI_MIN EIO_PRI_MIN
137#define ETP_PRI_MAX EIO_PRI_MAX 366#define ETP_PRI_MAX EIO_PRI_MAX
367
368#define ETP_TYPE_QUIT -1
369#define ETP_TYPE_GROUP EIO_GROUP
138 370
139struct etp_worker; 371struct etp_worker;
140 372
141#define ETP_REQ eio_req 373#define ETP_REQ eio_req
142#define ETP_DESTROY(req) eio_destroy (req) 374#define ETP_DESTROY(req) eio_destroy (req)
143static int eio_finish (eio_req *req); 375static int eio_finish (eio_req *req);
144#define ETP_FINISH(req) eio_finish (req) 376#define ETP_FINISH(req) eio_finish (req)
145static void eio_execute (struct etp_worker *self, eio_req *req); 377static void eio_execute (struct etp_worker *self, eio_req *req);
146#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) 378#define ETP_EXECUTE(wrk,req) eio_execute (wrk, req)
147 379
148#define ETP_WORKER_CLEAR(req) \ 380#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;
163 381
164/*****************************************************************************/ 382/*****************************************************************************/
165 383
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 384static 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) 385grp_try_feed (eio_req *grp)
606{ 386{
607 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 387 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
608 { 388 {
609 grp->flags &= ~EIO_FLAG_GROUPADD; 389 grp->flags &= ~EIO_FLAG_GROUPADD;
610 390
… …
617 break; 397 break;
618 } 398 }
619 } 399 }
620} 400}
621 401
402static int
622static int grp_dec (eio_req *grp) 403grp_dec (eio_req *grp)
623{ 404{
624 --grp->size; 405 --grp->size;
625 406
626 /* call feeder, if applicable */ 407 /* call feeder, if applicable */
627 grp_try_feed (grp); 408 grp_try_feed (grp);
… …
631 return eio_finish (grp); 412 return eio_finish (grp);
632 else 413 else
633 return 0; 414 return 0;
634} 415}
635 416
417static void
636void eio_destroy (eio_req *req) 418eio_destroy (eio_req *req)
637{ 419{
638 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); 420 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
639 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); 421 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
640 422
641 EIO_DESTROY (req); 423 EIO_DESTROY (req);
642} 424}
643 425
426static int
644static int eio_finish (eio_req *req) 427eio_finish (eio_req *req)
645{ 428{
646 int res = EIO_FINISH (req); 429 int res = EIO_FINISH (req);
647 430
648 if (req->grp) 431 if (req->grp)
649 { 432 {
… …
657 if (grp->grp_first == req) 440 if (grp->grp_first == req)
658 grp->grp_first = req->grp_next; 441 grp->grp_first = req->grp_next;
659 442
660 res2 = grp_dec (grp); 443 res2 = grp_dec (grp);
661 444
662 if (!res && res2) 445 if (!res)
663 res = res2; 446 res = res2;
664 } 447 }
665 448
666 eio_destroy (req); 449 eio_destroy (req);
667 450
668 return res; 451 return res;
669} 452}
670 453
454void
671void eio_grp_cancel (eio_req *grp) 455eio_grp_cancel (eio_req *grp)
672{ 456{
673 for (grp = grp->grp_first; grp; grp = grp->grp_next)
674 eio_cancel (grp); 457 etp_grp_cancel (grp);
675} 458}
676 459
460void
677void eio_cancel (eio_req *req) 461eio_cancel (eio_req *req)
678{ 462{
679 etp_cancel (req); 463 etp_cancel (req);
680} 464}
681 465
466void
682void eio_submit (eio_req *req) 467eio_submit (eio_req *req)
683{ 468{
684 etp_submit (req); 469 etp_submit (req);
685} 470}
686 471
687unsigned int eio_nreqs (void) 472unsigned int
473eio_nreqs (void)
688{ 474{
689 return etp_nreqs (); 475 return etp_nreqs ();
690} 476}
691 477
692unsigned int eio_nready (void) 478unsigned int
479eio_nready (void)
693{ 480{
694 return etp_nready (); 481 return etp_nready ();
695} 482}
696 483
697unsigned int eio_npending (void) 484unsigned int
485eio_npending (void)
698{ 486{
699 return etp_npending (); 487 return etp_npending ();
700} 488}
701 489
702unsigned int eio_nthreads (void) 490unsigned int ecb_cold
491eio_nthreads (void)
703{ 492{
704 return etp_nthreads (); 493 return etp_nthreads ();
705} 494}
706 495
496void ecb_cold
707void eio_set_max_poll_time (double nseconds) 497eio_set_max_poll_time (double nseconds)
708{ 498{
709 etp_set_max_poll_time (nseconds); 499 etp_set_max_poll_time (nseconds);
710} 500}
711 501
502void ecb_cold
712void eio_set_max_poll_reqs (unsigned int maxreqs) 503eio_set_max_poll_reqs (unsigned int maxreqs)
713{ 504{
714 etp_set_max_poll_reqs (maxreqs); 505 etp_set_max_poll_reqs (maxreqs);
715} 506}
716 507
508void ecb_cold
717void eio_set_max_idle (unsigned int nthreads) 509eio_set_max_idle (unsigned int nthreads)
718{ 510{
719 etp_set_max_idle (nthreads); 511 etp_set_max_idle (nthreads);
720} 512}
721 513
514void ecb_cold
515eio_set_idle_timeout (unsigned int seconds)
516{
517 etp_set_idle_timeout (seconds);
518}
519
520void ecb_cold
722void eio_set_min_parallel (unsigned int nthreads) 521eio_set_min_parallel (unsigned int nthreads)
723{ 522{
724 etp_set_min_parallel (nthreads); 523 etp_set_min_parallel (nthreads);
725} 524}
726 525
526void ecb_cold
727void eio_set_max_parallel (unsigned int nthreads) 527eio_set_max_parallel (unsigned int nthreads)
728{ 528{
729 etp_set_max_parallel (nthreads); 529 etp_set_max_parallel (nthreads);
730} 530}
731 531
732int eio_poll (void) 532int eio_poll (void)
… …
735} 535}
736 536
737/*****************************************************************************/ 537/*****************************************************************************/
738/* work around various missing functions */ 538/* work around various missing functions */
739 539
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 540#ifndef HAVE_UTIMES
780 541
781# undef utimes 542# undef utimes
782# undef futimes
783# define utimes(path,times) eio__utimes (path, times) 543# define utimes(path,times) eio__utimes (path, times)
784# define futimes(fd,times) eio__futimes (fd, times)
785 544
786static int 545static int
787eio__utimes (const char *filename, const struct timeval times[2]) 546eio__utimes (const char *filename, const struct timeval times[2])
788{ 547{
789 if (times) 548 if (times)
… …
797 } 556 }
798 else 557 else
799 return utime (filename, 0); 558 return utime (filename, 0);
800} 559}
801 560
561#endif
562
563#ifndef HAVE_FUTIMES
564
565# undef futimes
566# define futimes(fd,times) eio__futimes (fd, times)
567
568static int
802static int eio__futimes (int fd, const struct timeval tv[2]) 569eio__futimes (int fd, const struct timeval tv[2])
803{ 570{
804 errno = ENOSYS; 571 errno = ENOSYS;
805 return -1; 572 return -1;
806} 573}
807 574
… …
810#if !HAVE_FDATASYNC 577#if !HAVE_FDATASYNC
811# undef fdatasync 578# undef fdatasync
812# define fdatasync(fd) fsync (fd) 579# define fdatasync(fd) fsync (fd)
813#endif 580#endif
814 581
582static int
583eio__syncfs (int fd)
584{
585 int res;
586
587#if HAVE_SYS_SYNCFS
588 res = (int)syscall (__NR_syncfs, (int)(fd));
589#else
590 res = EIO_ENOSYS ();
591#endif
592
593 if (res < 0 && errno == ENOSYS && fd >= 0)
594 sync ();
595
596 return res;
597}
598
815/* sync_file_range always needs emulation */ 599/* sync_file_range always needs emulation */
816int 600static int
817eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags) 601eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
818{ 602{
819#if HAVE_SYNC_FILE_RANGE 603#if HAVE_SYNC_FILE_RANGE
820 int res; 604 int res;
821 605
… …
829 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0); 613 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
830 } 614 }
831 615
832 res = sync_file_range (fd, offset, nbytes, flags); 616 res = sync_file_range (fd, offset, nbytes, flags);
833 617
834 if (res != ENOSYS) 618 if (!res || errno != ENOSYS)
835 return res; 619 return res;
836#endif 620#endif
837 621
838 /* even though we could play tricks with the flags, it's better to always 622 /* 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 */ 623 * call fdatasync, as that matches the expectation of its users best */
840 return fdatasync (fd); 624 return fdatasync (fd);
625}
626
627static int
628eio__fallocate (int fd, int mode, off_t offset, size_t len)
629{
630#if HAVE_LINUX_FALLOCATE
631 return fallocate (fd, mode, offset, len);
632#else
633 return EIO_ENOSYS ();
634#endif
841} 635}
842 636
843#if !HAVE_READAHEAD 637#if !HAVE_READAHEAD
844# undef readahead 638# undef readahead
845# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 639# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
846 640
847static ssize_t 641static eio_ssize_t
848eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) 642eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
849{ 643{
850 size_t todo = count; 644 size_t todo = count;
851 dBUF; 645 dBUF;
852 646
… …
857 pread (fd, eio_buf, len, offset); 651 pread (fd, eio_buf, len, offset);
858 offset += len; 652 offset += len;
859 todo -= len; 653 todo -= len;
860 } 654 }
861 655
862 errno = 0; 656 FUBd;
657
658 /* linux's readahead basically only fails for EBADF or EINVAL (not mmappable) */
659 /* but not for e.g. EIO or eof, so we also never fail */
863 return count; 660 return 0;
864} 661}
865 662
866#endif 663#endif
867 664
868/* sendfile always needs emulation */ 665/* sendfile always needs emulation */
869static ssize_t 666static eio_ssize_t
870eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) 667eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
871{ 668{
669 eio_ssize_t written = 0;
872 ssize_t res; 670 eio_ssize_t res;
873 671
874 if (!count) 672 if (!count)
875 return 0; 673 return 0;
876 674
675 for (;;)
676 {
677#ifdef __APPLE__
678# undef HAVE_SENDFILE /* broken, as everything on os x */
679#endif
877#if HAVE_SENDFILE 680#if HAVE_SENDFILE
878# if __linux 681# if __linux
682 off_t soffset = offset;
879 res = sendfile (ofd, ifd, &offset, count); 683 res = sendfile (ofd, ifd, &soffset, count);
880 684
881# elif __freebsd 685# elif __FreeBSD__
882 /* 686 /*
883 * Of course, the freebsd sendfile is a dire hack with no thoughts 687 * Of course, the freebsd sendfile is a dire hack with no thoughts
884 * wasted on making it similar to other I/O functions. 688 * wasted on making it similar to other I/O functions.
885 */ 689 */
886 {
887 off_t sbytes; 690 off_t sbytes;
888 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 691 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
889 692
890 if (res < 0 && sbytes) 693 #if 0 /* according to the manpage, this is correct, but broken behaviour */
891 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 694 /* freebsd' sendfile will return 0 on success */
695 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
696 /* not on e.g. EIO or EPIPE - sounds broken */
697 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
892 res = sbytes; 698 res = sbytes;
893 } 699 #endif
700
701 /* according to source inspection, this is correct, and useful behaviour */
702 if (sbytes)
703 res = sbytes;
704
705# elif defined __APPLE__
706 off_t sbytes = count;
707 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
708
709 /* according to the manpage, sbytes is always valid */
710 if (sbytes)
711 res = sbytes;
894 712
895# elif __hpux 713# elif __hpux
896 res = sendfile (ofd, ifd, offset, count, 0, 0); 714 res = sendfile (ofd, ifd, offset, count, 0, 0);
897 715
898# elif __solaris 716# elif __solaris
899 {
900 struct sendfilevec vec; 717 struct sendfilevec vec;
901 size_t sbytes; 718 size_t sbytes;
902 719
903 vec.sfv_fd = ifd; 720 vec.sfv_fd = ifd;
904 vec.sfv_flag = 0; 721 vec.sfv_flag = 0;
905 vec.sfv_off = offset; 722 vec.sfv_off = offset;
906 vec.sfv_len = count; 723 vec.sfv_len = count;
907 724
908 res = sendfilev (ofd, &vec, 1, &sbytes); 725 res = sendfilev (ofd, &vec, 1, &sbytes);
909 726
910 if (res < 0 && sbytes) 727 if (res < 0 && sbytes)
911 res = sbytes; 728 res = sbytes;
912 }
913 729
914# endif 730# endif
731
732#elif defined (_WIN32) && 0
733 /* does not work, just for documentation of what would need to be done */
734 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
735 /* libeio guarantees that the file offset does not change, and windows */
736 /* has no way to get an independent handle to the same file description */
737 HANDLE h = TO_SOCKET (ifd);
738 SetFilePointer (h, offset, 0, FILE_BEGIN);
739 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
740
915#else 741#else
916 res = -1; 742 res = EIO_ENOSYS ();
917 errno = ENOSYS;
918#endif 743#endif
919 744
745 /* we assume sendfile can copy at least 128mb in one go */
746 if (res <= 128 * 1024 * 1024)
747 {
748 if (res > 0)
749 written += res;
750
751 if (written)
752 return written;
753
754 break;
755 }
756 else
757 {
758 /* if we requested more, then probably the kernel was lazy */
759 written += res;
760 offset += res;
761 count -= res;
762
763 if (!count)
764 return written;
765 }
766 }
767
920 if (res < 0 768 if (res < 0
921 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 769 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
770 /* BSDs */
771#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
772 || errno == ENOTSUP
773#endif
774#ifdef EOPNOTSUPP /* windows */
775 || errno == EOPNOTSUPP /* BSDs */
776#endif
922#if __solaris 777#if __solaris
923 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 778 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
924#endif 779#endif
925 ) 780 )
926 ) 781 )
… …
930 785
931 res = 0; 786 res = 0;
932 787
933 while (count) 788 while (count)
934 { 789 {
935 ssize_t cnt; 790 eio_ssize_t cnt;
936 791
937 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 792 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
938 793
939 if (cnt <= 0) 794 if (cnt <= 0)
940 { 795 {
… …
952 807
953 offset += cnt; 808 offset += cnt;
954 res += cnt; 809 res += cnt;
955 count -= cnt; 810 count -= cnt;
956 } 811 }
812
813 FUBd;
957 } 814 }
958 815
959 return res; 816 return res;
817}
818
819#ifdef PAGESIZE
820# define eio_pagesize() PAGESIZE
821#else
822static intptr_t
823eio_pagesize (void)
824{
825 static intptr_t page;
826
827 if (!page)
828 page = sysconf (_SC_PAGESIZE);
829
830 return page;
831}
832#endif
833
834static void
835eio_page_align (void **addr, size_t *length)
836{
837 intptr_t mask = eio_pagesize () - 1;
838
839 /* round down addr */
840 intptr_t adj = mask & (intptr_t)*addr;
841
842 *addr = (void *)((intptr_t)*addr - adj);
843 *length += adj;
844
845 /* round up length */
846 *length = (*length + mask) & ~mask;
847}
848
849#if !_POSIX_MEMLOCK
850# define eio__mlockall(a) EIO_ENOSYS ()
851#else
852
853static int
854eio__mlockall (int flags)
855{
856 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
857 extern int mallopt (int, int);
858 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
859 #endif
860
861 if (EIO_MCL_CURRENT != MCL_CURRENT
862 || EIO_MCL_FUTURE != MCL_FUTURE)
863 {
864 flags = 0
865 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
866 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
867 }
868
869 return mlockall (flags);
870}
871#endif
872
873#if !_POSIX_MEMLOCK_RANGE
874# define eio__mlock(a,b) EIO_ENOSYS ()
875#else
876
877static int
878eio__mlock (void *addr, size_t length)
879{
880 eio_page_align (&addr, &length);
881
882 return mlock (addr, length);
883}
884
885#endif
886
887#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
888# define eio__msync(a,b,c) EIO_ENOSYS ()
889#else
890
891static int
892eio__msync (void *mem, size_t len, int flags)
893{
894 eio_page_align (&mem, &len);
895
896 if (EIO_MS_ASYNC != MS_SYNC
897 || EIO_MS_INVALIDATE != MS_INVALIDATE
898 || EIO_MS_SYNC != MS_SYNC)
899 {
900 flags = 0
901 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
902 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
903 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
904 }
905
906 return msync (mem, len, flags);
907}
908
909#endif
910
911static int
912eio__mtouch (eio_req *req)
913{
914 void *mem = req->ptr2;
915 size_t len = req->size;
916 int flags = req->int1;
917
918 eio_page_align (&mem, &len);
919
920 {
921 intptr_t addr = (intptr_t)mem;
922 intptr_t end = addr + len;
923 intptr_t page = eio_pagesize ();
924
925 if (addr < end)
926 if (flags & EIO_MT_MODIFY) /* modify */
927 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
928 else
929 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
930 }
931
932 return 0;
933}
934
935/*****************************************************************************/
936/* requests implemented outside eio_execute, because they are so large */
937
938static void
939eio__lseek (eio_req *req)
940{
941 /* this usually gets optimised away completely, or your compiler sucks, */
942 /* or the whence constants really are not 0, 1, 2 */
943 int whence = req->int2 == EIO_SEEK_SET ? SEEK_SET
944 : req->int2 == EIO_SEEK_CUR ? SEEK_CUR
945 : req->int2 == EIO_SEEK_END ? SEEK_END
946 : req->int2;
947
948 req->offs = lseek (req->int1, req->offs, whence);
949 req->result = req->offs == (off_t)-1 ? -1 : 0;
950}
951
952/* result will always end up in tmpbuf, there is always space for adding a 0-byte */
953static int
954eio__realpath (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
955{
956 char *res;
957 const char *rel = path;
958 char *tmp1, *tmp2;
959#if SYMLOOP_MAX > 32
960 int symlinks = SYMLOOP_MAX;
961#else
962 int symlinks = 32;
963#endif
964
965 errno = EINVAL;
966 if (!rel)
967 return -1;
968
969 errno = ENOENT;
970 if (!*rel)
971 return -1;
972
973 res = etp_tmpbuf_get (tmpbuf, PATH_MAX * 3);
974#ifdef _WIN32
975 if (_access (rel, 4) != 0)
976 return -1;
977
978 symlinks = GetFullPathName (rel, PATH_MAX * 3, res, 0);
979
980 errno = ENAMETOOLONG;
981 if (symlinks >= PATH_MAX * 3)
982 return -1;
983
984 errno = EIO;
985 if (symlinks <= 0)
986 return -1;
987
988 return symlinks;
989
990#else
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#endif
1124}
1125
1126static signed char
1127eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1128{
1129 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1130 : a->inode < b->inode ? -1
1131 : a->inode > b->inode ? 1
1132 : 0;
1133}
1134
1135#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1136
1137#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1138#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1139
1140static void
1141eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1142{
1143 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1144 unsigned char *bit = bits;
1145
1146 assert (CHAR_BIT == 8);
1147 assert (sizeof (eio_dirent) * 8 < 256);
1148 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1149 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1150
1151 if (size <= EIO_SORT_FAST)
1152 return;
1153
1154 /* first prepare an array of bits to test in our radix sort */
1155 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1156 /* inode_bits must contain all inodes ORed together */
1157 /* which is used to skip bits that are 0 everywhere, which is very common */
1158 {
1159 eio_ino_t endianness;
1160 int i, j;
1161
1162 /* we store the byte offset of byte n into byte n of "endianness" */
1163 for (i = 0; i < sizeof (eio_ino_t); ++i)
1164 ((unsigned char *)&endianness)[i] = i;
1165
1166 *bit++ = 0;
1167
1168 for (i = 0; i < sizeof (eio_ino_t); ++i)
1169 {
1170 /* shifting off the byte offsets out of "endianness" */
1171 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1172 endianness >>= 8;
1173
1174 for (j = 0; j < 8; ++j)
1175 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1176 *bit++ = offs + j;
1177 }
1178
1179 for (j = 0; j < 8; ++j)
1180 if (score_bits & (1 << j))
1181 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1182 }
1183
1184 /* now actually do the sorting (a variant of MSD radix sort) */
1185 {
1186 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1187 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1188 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1189 int stk_idx = 0;
1190
1191 base_stk [stk_idx] = dents;
1192 end_stk [stk_idx] = dents + size;
1193 bit_stk [stk_idx] = bit - 1;
1194
1195 do
1196 {
1197 base = base_stk [stk_idx];
1198 end = end_stk [stk_idx];
1199 bit = bit_stk [stk_idx];
1200
1201 for (;;)
1202 {
1203 unsigned char O = *bit >> 3;
1204 unsigned char M = 1 << (*bit & 7);
1205
1206 eio_dirent *a = base;
1207 eio_dirent *b = end;
1208
1209 if (b - a < EIO_SORT_CUTOFF)
1210 break;
1211
1212 /* now bit-partition the array on the bit */
1213 /* this ugly asymmetric loop seems to perform much better than typical */
1214 /* partition algos found in the literature */
1215 do
1216 if (!(((unsigned char *)a)[O] & M))
1217 ++a;
1218 else if (!(((unsigned char *)--b)[O] & M))
1219 {
1220 eio_dirent tmp = *a; *a = *b; *b = tmp;
1221 ++a;
1222 }
1223 while (b > a);
1224
1225 /* next bit, or stop, if no bits left in this path */
1226 if (!*--bit)
1227 break;
1228
1229 base_stk [stk_idx] = a;
1230 end_stk [stk_idx] = end;
1231 bit_stk [stk_idx] = bit;
1232 ++stk_idx;
1233
1234 end = a;
1235 }
1236 }
1237 while (stk_idx--);
1238 }
1239}
1240
1241static void
1242eio_dent_insertion_sort (eio_dirent *dents, int size)
1243{
1244 /* first move the smallest element to the front, to act as a sentinel */
1245 {
1246 int i;
1247 eio_dirent *min = dents;
1248
1249 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1250 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1251 if (EIO_DENT_CMP (dents [i], <, *min))
1252 min = &dents [i];
1253
1254 /* swap elements 0 and j (minimum) */
1255 {
1256 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1257 }
1258 }
1259
1260 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1261 {
1262 eio_dirent *i, *j;
1263
1264 for (i = dents + 1; i < dents + size; ++i)
1265 {
1266 eio_dirent value = *i;
1267
1268 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1269 j [1] = j [0];
1270
1271 j [1] = value;
1272 }
1273 }
1274}
1275
1276static void
1277eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1278{
1279 if (size <= 1)
1280 return; /* our insertion sort relies on size > 0 */
1281
1282 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1283 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1284 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1285
1286 /* use an insertion sort at the end, or for small arrays, */
1287 /* as insertion sort is more efficient for small partitions */
1288 eio_dent_insertion_sort (dents, size);
960} 1289}
961 1290
962/* read a full directory */ 1291/* read a full directory */
963static void 1292static void
964eio__scandir (eio_req *req, etp_worker *self) 1293eio__scandir (eio_req *req, etp_worker *self)
965{ 1294{
1295 char *name, *names;
1296 int namesalloc = 4096 - sizeof (void *) * 4;
1297 int namesoffs = 0;
1298 int flags = req->int1;
1299 eio_dirent *dents = 0;
1300 int dentalloc = 128;
1301 int dentoffs = 0;
1302 eio_ino_t inode_bits = 0;
1303#ifdef _WIN32
1304 HANDLE dirp;
1305 WIN32_FIND_DATA entp;
1306#else
966 DIR *dirp; 1307 DIR *dirp;
967 EIO_STRUCT_DIRENT *entp; 1308 EIO_STRUCT_DIRENT *entp;
968 char *name, *names; 1309#endif
969 int memlen = 4096;
970 int memofs = 0;
971 int res = 0;
972 1310
973 X_LOCK (wrklock); 1311 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 1312
980 if (dirp && names) 1313 if (!(flags & EIO_READDIR_DENTS))
981 for (;;) 1314 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1315
1316#ifdef _WIN32
1317 {
1318 int len = strlen ((const char *)req->ptr1);
1319 char *path = malloc (MAX_PATH);
1320 const char *fmt;
1321 const char *reqpath = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1322
1323 if (!len)
1324 fmt = "./*";
1325 else if (reqpath[len - 1] == '/' || reqpath[len - 1] == '\\')
1326 fmt = "%s*";
1327 else
1328 fmt = "%s/*";
1329
1330 _snprintf (path, MAX_PATH, fmt, reqpath);
1331 dirp = FindFirstFile (path, &entp);
1332 free (path);
1333
1334 if (dirp == INVALID_HANDLE_VALUE)
1335 {
1336 /* should steal _dosmaperr */
1337 switch (GetLastError ())
1338 {
1339 case ERROR_FILE_NOT_FOUND:
1340 req->result = 0;
1341 break;
1342
1343 case ERROR_INVALID_NAME:
1344 case ERROR_PATH_NOT_FOUND:
1345 case ERROR_NO_MORE_FILES:
1346 errno = ENOENT;
1347 break;
1348
1349 case ERROR_NOT_ENOUGH_MEMORY:
1350 errno = ENOMEM;
1351 break;
1352
1353 default:
1354 errno = EINVAL;
1355 break;
1356 }
1357
1358 return;
1359 }
1360 }
1361#else
1362 #if HAVE_AT
1363 if (req->wd)
982 { 1364 {
1365 int fd = openat (WD2FD (req->wd), req->ptr1, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1366
1367 if (fd < 0)
1368 return;
1369
1370 dirp = fdopendir (fd);
1371
1372 if (!dirp)
1373 close (fd);
1374 }
1375 else
1376 dirp = opendir (req->ptr1);
1377 #else
1378 dirp = opendir (wd_expand (&self->tmpbuf, req->wd, req->ptr1));
1379 #endif
1380
1381 if (!dirp)
1382 return;
1383#endif
1384
1385 if (req->flags & EIO_FLAG_PTR1_FREE)
1386 free (req->ptr1);
1387
1388 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1389 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1390 req->ptr2 = names = malloc (namesalloc);
1391
1392 if (!names || (flags && !dents))
1393 return;
1394
1395 for (;;)
1396 {
1397 int done;
1398
1399#ifdef _WIN32
1400 done = !dirp;
1401#else
983 errno = 0; 1402 errno = 0;
984 entp = readdir (dirp); 1403 entp = readdir (dirp);
1404 done = !entp;
1405#endif
985 1406
1407 if (done)
1408 {
1409#ifndef _WIN32
1410 int old_errno = errno;
1411 closedir (dirp);
1412 errno = old_errno;
1413
986 if (!entp) 1414 if (errno)
1415 break;
1416#endif
1417
1418 /* sort etc. */
1419 req->int1 = flags;
1420 req->result = dentoffs;
1421
1422 if (flags & EIO_READDIR_STAT_ORDER)
1423 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1424 else if (flags & EIO_READDIR_DIRS_FIRST)
1425 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1426 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1427 else
1428 {
1429 /* in this case, all is known, and we just put dirs first and sort them */
1430 eio_dirent *oth = dents + dentoffs;
1431 eio_dirent *dir = dents;
1432
1433 /* now partition dirs to the front, and non-dirs to the back */
1434 /* by walking from both sides and swapping if necessary */
1435 while (oth > dir)
1436 {
1437 if (dir->type == EIO_DT_DIR)
1438 ++dir;
1439 else if ((--oth)->type == EIO_DT_DIR)
1440 {
1441 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1442
1443 ++dir;
1444 }
1445 }
1446
1447 /* now sort the dirs only (dirs all have the same score) */
1448 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1449 }
1450
987 break; 1451 break;
1452 }
988 1453
989 name = entp->d_name; 1454 /* now add the entry to our list(s) */
1455 name = D_NAME (entp);
990 1456
1457 /* skip . and .. entries */
991 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1458 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
992 { 1459 {
993 int len = strlen (name) + 1; 1460 int len = D_NAMLEN (entp) + 1;
994 1461
995 res++; 1462 while (ecb_expect_false (namesoffs + len > namesalloc))
996
997 while (memofs + len > memlen)
998 { 1463 {
999 memlen *= 2; 1464 namesalloc *= 2;
1000 X_LOCK (wrklock);
1001 req->ptr2 = names = realloc (names, memlen); 1465 req->ptr2 = names = realloc (names, namesalloc);
1002 X_UNLOCK (wrklock);
1003 1466
1004 if (!names) 1467 if (!names)
1468 break;
1469 }
1470
1471 memcpy (names + namesoffs, name, len);
1472
1473 if (dents)
1474 {
1475 struct eio_dirent *ent;
1476
1477 if (ecb_expect_false (dentoffs == dentalloc))
1478 {
1479 dentalloc *= 2;
1480 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1481
1482 if (!dents)
1005 break; 1483 break;
1006 } 1484 }
1007 1485
1008 memcpy (names + memofs, name, len); 1486 ent = dents + dentoffs;
1009 memofs += len; 1487
1488 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1489 ent->namelen = len - 1;
1490 ent->inode = D_INO (entp);
1491
1492 inode_bits |= ent->inode;
1493
1494 switch (D_TYPE (entp))
1495 {
1496 default:
1497 ent->type = EIO_DT_UNKNOWN;
1498 flags |= EIO_READDIR_FOUND_UNKNOWN;
1499 break;
1500
1501 #ifdef DT_FIFO
1502 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1503 #endif
1504 #ifdef DT_CHR
1505 case DT_CHR: ent->type = EIO_DT_CHR; break;
1506 #endif
1507 #ifdef DT_MPC
1508 case DT_MPC: ent->type = EIO_DT_MPC; break;
1509 #endif
1510 #ifdef DT_DIR
1511 case DT_DIR: ent->type = EIO_DT_DIR; break;
1512 #endif
1513 #ifdef DT_NAM
1514 case DT_NAM: ent->type = EIO_DT_NAM; break;
1515 #endif
1516 #ifdef DT_BLK
1517 case DT_BLK: ent->type = EIO_DT_BLK; break;
1518 #endif
1519 #ifdef DT_MPB
1520 case DT_MPB: ent->type = EIO_DT_MPB; break;
1521 #endif
1522 #ifdef DT_REG
1523 case DT_REG: ent->type = EIO_DT_REG; break;
1524 #endif
1525 #ifdef DT_NWK
1526 case DT_NWK: ent->type = EIO_DT_NWK; break;
1527 #endif
1528 #ifdef DT_CMP
1529 case DT_CMP: ent->type = EIO_DT_CMP; break;
1530 #endif
1531 #ifdef DT_LNK
1532 case DT_LNK: ent->type = EIO_DT_LNK; break;
1533 #endif
1534 #ifdef DT_SOCK
1535 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1536 #endif
1537 #ifdef DT_DOOR
1538 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1539 #endif
1540 #ifdef DT_WHT
1541 case DT_WHT: ent->type = EIO_DT_WHT; break;
1542 #endif
1543 }
1544
1545 ent->score = 7;
1546
1547 if (flags & EIO_READDIR_DIRS_FIRST)
1548 {
1549 if (ent->type == EIO_DT_UNKNOWN)
1550 {
1551 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1552 ent->score = 1;
1553 else if (!strchr (name, '.')) /* absence of dots indicate likely dirs */
1554 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1555 }
1556 else if (ent->type == EIO_DT_DIR)
1557 ent->score = 0;
1558 }
1010 } 1559 }
1560
1561 namesoffs += len;
1562 ++dentoffs;
1011 } 1563 }
1012 1564
1013 if (errno) 1565 if (EIO_CANCELLED (req))
1014 res = -1; 1566 {
1567 errno = ECANCELED;
1568 break;
1569 }
1570
1571#ifdef _WIN32
1572 if (!FindNextFile (dirp, &entp))
1573 {
1574 FindClose (dirp);
1575 dirp = 0;
1576 }
1577#endif
1578 }
1579}
1580
1581/*****************************************************************************/
1582/* working directory stuff */
1583/* various deficiencies in the posix 2008 api force us to */
1584/* keep the absolute path in string form at all times */
1585/* fuck yeah. */
1586
1587#if !HAVE_AT
1588
1589/* a bit like realpath, but usually faster because it doesn'T have to return */
1590/* an absolute or canonical path */
1591static const char *
1592wd_expand (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
1593{
1594 if (!wd || *path == '/')
1595 return path;
1596
1597 if (path [0] == '.' && !path [1])
1598 return wd->str;
1599
1015 1600 {
1016 req->result = res; 1601 int l1 = wd->len;
1017} 1602 int l2 = strlen (path);
1018 1603
1019#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) 1604 char *res = etp_tmpbuf_get (tmpbuf, l1 + l2 + 2);
1020# undef msync
1021# define msync(a,b,c) ENOSYS
1022#endif
1023 1605
1024int 1606 memcpy (res, wd->str, l1);
1025eio__mtouch (void *mem, size_t len, int flags) 1607 res [l1] = '/';
1026{ 1608 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 1609
1034 if (!page) 1610 return res;
1035 page = sysconf (_SC_PAGESIZE); 1611 }
1036#endif 1612}
1037 1613
1038 addr &= ~(page - 1); /* assume page size is always a power of two */ 1614#endif
1039 1615
1040 if (addr < end) 1616static eio_wd
1041 if (flags) /* modify */ 1617eio__wd_open_sync (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
1042 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len); 1618{
1043 else 1619 int fd;
1044 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len); 1620 eio_wd res;
1621 int len = eio__realpath (tmpbuf, wd, path);
1045 1622
1623 if (len < 0)
1624 return EIO_INVALID_WD;
1625
1626#if HAVE_AT
1627 fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1628
1629 if (fd < 0)
1630 return EIO_INVALID_WD;
1631#endif
1632
1633 res = malloc (sizeof (*res) + len); /* one extra 0-byte */
1634
1635#if HAVE_AT
1636 res->fd = fd;
1637#endif
1638
1639 res->len = len;
1640 memcpy (res->str, tmpbuf->ptr, len);
1641 res->str [len] = 0;
1642
1643 return res;
1644}
1645
1646eio_wd
1647eio_wd_open_sync (eio_wd wd, const char *path)
1648{
1649 struct etp_tmpbuf tmpbuf = { };
1650 wd = eio__wd_open_sync (&tmpbuf, wd, path);
1651 free (tmpbuf.ptr);
1652
1046 return 0; 1653 return wd;
1047} 1654}
1655
1656void
1657eio_wd_close_sync (eio_wd wd)
1658{
1659 if (wd != EIO_INVALID_WD && wd != EIO_CWD)
1660 {
1661 #if HAVE_AT
1662 close (wd->fd);
1663 #endif
1664 free (wd);
1665 }
1666}
1667
1668#if HAVE_AT
1669
1670/* they forgot these */
1671
1672static int
1673eio__truncateat (int dirfd, const char *path, off_t length)
1674{
1675 int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
1676 int res;
1677
1678 if (fd < 0)
1679 return fd;
1680
1681 res = ftruncate (fd, length);
1682 close (fd);
1683 return res;
1684}
1685
1686static int
1687eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
1688{
1689 int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
1690 int res;
1691
1692 if (fd < 0)
1693 return fd;
1694
1695 res = fstatvfs (fd, buf);
1696 close (fd);
1697 return res;
1698
1699}
1700
1701#endif
1048 1702
1049/*****************************************************************************/ 1703/*****************************************************************************/
1050 1704
1051#define ALLOC(len) \ 1705#define ALLOC(len) \
1052 if (!req->ptr2) \ 1706 if (!req->ptr2) \
… …
1061 req->result = -1; \ 1715 req->result = -1; \
1062 break; \ 1716 break; \
1063 } \ 1717 } \
1064 } 1718 }
1065 1719
1720static void ecb_noinline ecb_cold
1721etp_proc_init (void)
1722{
1723#if HAVE_PRCTL_SET_NAME
1724 /* provide a more sensible "thread name" */
1725 char name[16 + 1];
1726 const int namelen = sizeof (name) - 1;
1727 int len;
1728
1729 prctl (PR_GET_NAME, (unsigned long)name, 0, 0, 0);
1730 name [namelen] = 0;
1731 len = strlen (name);
1732 strcpy (name + (len <= namelen - 4 ? len : namelen - 4), "/eio");
1733 prctl (PR_SET_NAME, (unsigned long)name, 0, 0, 0);
1734#endif
1735}
1736
1737/* TODO: move somehow to etp.c */
1066X_THREAD_PROC (etp_proc) 1738X_THREAD_PROC (etp_proc)
1067{ 1739{
1068 ETP_REQ *req; 1740 ETP_REQ *req;
1069 struct timespec ts; 1741 struct timespec ts;
1070 etp_worker *self = (etp_worker *)thr_arg; 1742 etp_worker *self = (etp_worker *)thr_arg;
1071 1743
1744 etp_proc_init ();
1745
1072 /* try to distribute timeouts somewhat randomly */ 1746 /* try to distribute timeouts somewhat evenly */
1073 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 1747 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1074 1748
1075 for (;;) 1749 for (;;)
1076 { 1750 {
1751 ts.tv_sec = 0;
1752
1077 X_LOCK (reqlock); 1753 X_LOCK (reqlock);
1078 1754
1079 for (;;) 1755 for (;;)
1080 { 1756 {
1081 self->req = req = reqq_shift (&req_queue); 1757 req = reqq_shift (&req_queue);
1082 1758
1083 if (req) 1759 if (req)
1084 break; 1760 break;
1085 1761
1762 if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
1763 {
1764 X_UNLOCK (reqlock);
1765 X_LOCK (wrklock);
1766 --started;
1767 X_UNLOCK (wrklock);
1768 goto quit;
1769 }
1770
1086 ++idle; 1771 ++idle;
1087 1772
1088 ts.tv_sec = time (0) + IDLE_TIMEOUT; 1773 if (idle <= max_idle)
1089 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 1774 /* we are allowed to idle, so do so without any timeout */
1775 X_COND_WAIT (reqwait, reqlock);
1776 else
1090 { 1777 {
1091 if (idle > max_idle) 1778 /* initialise timeout once */
1092 { 1779 if (!ts.tv_sec)
1093 --idle; 1780 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 1781
1101 /* we are allowed to idle, so do so without any timeout */
1102 X_COND_WAIT (reqwait, reqlock); 1782 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
1783 ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1103 } 1784 }
1104 1785
1105 --idle; 1786 --idle;
1106 } 1787 }
1107 1788
1108 --nready; 1789 --nready;
1109 1790
1110 X_UNLOCK (reqlock); 1791 X_UNLOCK (reqlock);
1111 1792
1112 if (req->type < 0) 1793 if (req->type == ETP_TYPE_QUIT)
1113 goto quit; 1794 goto quit;
1114 1795
1115 if (!EIO_CANCELLED (req))
1116 ETP_EXECUTE (self, req); 1796 ETP_EXECUTE (self, req);
1117 1797
1118 X_LOCK (reslock); 1798 X_LOCK (reslock);
1119 1799
1120 ++npending; 1800 ++npending;
1121 1801
1122 if (!reqq_push (&res_queue, req) && want_poll_cb) 1802 if (!reqq_push (&res_queue, req) && want_poll_cb)
1123 want_poll_cb (); 1803 want_poll_cb ();
1124 1804
1125 self->req = 0;
1126 etp_worker_clear (self); 1805 etp_worker_clear (self);
1127 1806
1128 X_UNLOCK (reslock); 1807 X_UNLOCK (reslock);
1129 } 1808 }
1130 1809
1131quit: 1810quit:
1811 free (req);
1812
1132 X_LOCK (wrklock); 1813 X_LOCK (wrklock);
1133 etp_worker_free (self); 1814 etp_worker_free (self);
1134 X_UNLOCK (wrklock); 1815 X_UNLOCK (wrklock);
1135 1816
1136 return 0; 1817 return 0;
1137} 1818}
1138 1819
1139/*****************************************************************************/ 1820/*****************************************************************************/
1140 1821
1822int ecb_cold
1141int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1823eio_init (void (*want_poll)(void), void (*done_poll)(void))
1142{ 1824{
1143 return etp_init (want_poll, done_poll); 1825 return etp_init (want_poll, done_poll);
1144} 1826}
1145 1827
1828ecb_inline void
1146static void eio_api_destroy (eio_req *req) 1829eio_api_destroy (eio_req *req)
1147{ 1830{
1148 free (req); 1831 free (req);
1149} 1832}
1150 1833
1151#define REQ(rtype) \ 1834#define REQ(rtype) \
1152 eio_req *req; \ 1835 eio_req *req; \
1153 \ 1836 \
1154 req = (eio_req *)calloc (1, sizeof *req); \ 1837 req = (eio_req *)calloc (1, sizeof *req); \
1155 if (!req) \ 1838 if (!req) \
1156 return 0; \ 1839 return 0; \
… …
1170 { \ 1853 { \
1171 eio_api_destroy (req); \ 1854 eio_api_destroy (req); \
1172 return 0; \ 1855 return 0; \
1173 } 1856 }
1174 1857
1858#define SINGLEDOT(ptr) (0[(char *)(ptr)] == '.' && !1[(char *)(ptr)])
1859
1860static void
1175static void eio_execute (etp_worker *self, eio_req *req) 1861eio_execute (etp_worker *self, eio_req *req)
1176{ 1862{
1177 errno = 0; 1863#if HAVE_AT
1864 int dirfd;
1865#else
1866 const char *path;
1867#endif
1868
1869 if (ecb_expect_false (EIO_CANCELLED (req)))
1870 {
1871 req->result = -1;
1872 req->errorno = ECANCELED;
1873 return;
1874 }
1875
1876 if (ecb_expect_false (req->wd == EIO_INVALID_WD))
1877 {
1878 req->result = -1;
1879 req->errorno = ENOENT;
1880 return;
1881 }
1882
1883 if (req->type >= EIO_OPEN)
1884 {
1885 #if HAVE_AT
1886 dirfd = WD2FD (req->wd);
1887 #else
1888 path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1889 #endif
1890 }
1178 1891
1179 switch (req->type) 1892 switch (req->type)
1180 { 1893 {
1894 case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
1895 req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
1896 break;
1897 case EIO_WD_CLOSE: req->result = 0;
1898 eio_wd_close_sync (req->wd); break;
1899
1900 case EIO_SEEK: eio__lseek (req); break;
1181 case EIO_READ: ALLOC (req->size); 1901 case EIO_READ: ALLOC (req->size);
1182 req->result = req->offs >= 0 1902 req->result = req->offs >= 0
1183 ? pread (req->int1, req->ptr2, req->size, req->offs) 1903 ? pread (req->int1, req->ptr2, req->size, req->offs)
1184 : read (req->int1, req->ptr2, req->size); break; 1904 : read (req->int1, req->ptr2, req->size); break;
1185 case EIO_WRITE: req->result = req->offs >= 0 1905 case EIO_WRITE: req->result = req->offs >= 0
1186 ? pwrite (req->int1, req->ptr2, req->size, req->offs) 1906 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1187 : write (req->int1, req->ptr2, req->size); break; 1907 : write (req->int1, req->ptr2, req->size); break;
1188 1908
1189 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break; 1909 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; 1910 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
1911
1912#if HAVE_AT
1191 1913
1192 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1914 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1193 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 1915 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break;
1194 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1916 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1195 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 1917 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; 1918 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; 1919 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; 1920 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; 1921 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; 1922
1208 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1209 case EIO_UNLINK: req->result = unlink (req->ptr1); break; 1923 case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
1210 case EIO_RMDIR: req->result = rmdir (req->ptr1); break; 1924 case EIO_RMDIR: /* complications arise because "." cannot be removed, so we might have to expand */
1925 req->result = req->wd && SINGLEDOT (req->ptr1)
1926 ? rmdir (req->wd->str)
1927 : unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break;
1211 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; 1928 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; 1929 case EIO_RENAME: /* complications arise because "." cannot be renamed, so we might have to expand */
1930 req->result = req->wd && SINGLEDOT (req->ptr1)
1931 ? rename (req->wd->str, req->ptr2)
1932 : renameat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2); break;
1213 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; 1933 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; 1934 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; 1935 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); 1936 case EIO_READLINK: ALLOC (PATH_MAX);
1218 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; 1937 req->result = readlinkat (dirfd, req->ptr1, req->ptr2, PATH_MAX); break;
1219 1938 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1220 case EIO_SYNC: req->result = 0; sync (); break; 1939 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: 1940 case EIO_UTIME:
1230#ifdef _WIN32 1941 case EIO_FUTIME:
1231 Sleep (req->nv1 * 1000.); 1942 {
1943 struct timespec ts[2];
1944 struct timespec *times;
1945
1946 if (req->nv1 != -1. || req->nv2 != -1.)
1947 {
1948 ts[0].tv_sec = req->nv1;
1949 ts[0].tv_nsec = (req->nv1 - ts[0].tv_sec) * 1e9;
1950 ts[1].tv_sec = req->nv2;
1951 ts[1].tv_nsec = (req->nv2 - ts[1].tv_sec) * 1e9;
1952
1953 times = ts;
1954 }
1955 else
1956 times = 0;
1957
1958 req->result = req->type == EIO_FUTIME
1959 ? futimens (req->int1, times)
1960 : utimensat (dirfd, req->ptr1, times, 0);
1961 }
1962 break;
1963
1232#else 1964#else
1233 {
1234 struct timeval tv;
1235 1965
1236 tv.tv_sec = req->nv1; 1966 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1237 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; 1967 req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
1968 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1969 req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
1970 case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
1971 case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
1972 case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
1973 case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
1238 1974
1239 req->result = select (0, 0, 0, 0, &tv); 1975 case EIO_UNLINK: req->result = unlink (path ); break;
1240 } 1976 case EIO_RMDIR: req->result = rmdir (path ); break;
1241#endif 1977 case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
1242 break; 1978 case EIO_RENAME: req->result = rename (path , req->ptr2); break;
1979 case EIO_LINK: req->result = link (path , req->ptr2); break;
1980 case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
1981 case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
1982 case EIO_READLINK: ALLOC (PATH_MAX);
1983 req->result = readlink (path, req->ptr2, PATH_MAX); break;
1984 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1985 req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
1243 1986
1244 case EIO_UTIME: 1987 case EIO_UTIME:
1245 case EIO_FUTIME: 1988 case EIO_FUTIME:
1246 { 1989 {
1247 struct timeval tv[2]; 1990 struct timeval tv[2];
1248 struct timeval *times; 1991 struct timeval *times;
1249 1992
1250 if (req->nv1 != -1. || req->nv2 != -1.) 1993 if (req->nv1 != -1. || req->nv2 != -1.)
1251 { 1994 {
1252 tv[0].tv_sec = req->nv1; 1995 tv[0].tv_sec = req->nv1;
1253 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.; 1996 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
1254 tv[1].tv_sec = req->nv2; 1997 tv[1].tv_sec = req->nv2;
1255 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.; 1998 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
1256 1999
1257 times = tv; 2000 times = tv;
1258 } 2001 }
1259 else 2002 else
1260 times = 0; 2003 times = 0;
1261
1262 2004
1263 req->result = req->type == EIO_FUTIME 2005 req->result = req->type == EIO_FUTIME
1264 ? futimes (req->int1, times) 2006 ? futimes (req->int1, times)
1265 : utimes (req->ptr1, times); 2007 : utimes (req->ptr1, times);
1266 } 2008 }
1267 break; 2009 break;
1268 2010
2011#endif
2012
2013 case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
2014 {
2015 ALLOC (req->result);
2016 memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
2017 }
2018 break;
2019
2020 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2021 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
2022
2023 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2024 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2025
2026 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
2027 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
2028 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
2029
2030 case EIO_CLOSE: req->result = close (req->int1); break;
2031 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
2032 case EIO_SYNC: req->result = 0; sync (); break;
2033 case EIO_FSYNC: req->result = fsync (req->int1); break;
2034 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
2035 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
2036 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
2037 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
2038 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
2039 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
2040 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
2041 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
2042
2043 case EIO_READDIR: eio__scandir (req, self); break;
2044
2045 case EIO_BUSY:
2046#ifdef _WIN32
2047 Sleep (req->nv1 * 1e3);
2048#else
2049 {
2050 struct timeval tv;
2051
2052 tv.tv_sec = req->nv1;
2053 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
2054
2055 req->result = select (0, 0, 0, 0, &tv);
2056 }
2057#endif
2058 break;
2059
1269 case EIO_GROUP: 2060 case EIO_GROUP:
1270 abort (); /* handled in eio_request */ 2061 abort (); /* handled in eio_request */
1271 2062
1272 case EIO_NOP: 2063 case EIO_NOP:
1273 req->result = 0; 2064 req->result = 0;
1274 break; 2065 break;
1275 2066
1276 case EIO_CUSTOM: 2067 case EIO_CUSTOM:
1277 ((void (*)(eio_req *))req->feed) (req); 2068 req->feed (req);
1278 break; 2069 break;
1279 2070
1280 default: 2071 default:
1281 req->result = -1; 2072 req->result = EIO_ENOSYS ();
1282 break; 2073 break;
1283 } 2074 }
1284 2075
1285 req->errorno = errno; 2076 req->errorno = errno;
1286} 2077}
1287 2078
1288#ifndef EIO_NO_WRAPPERS 2079#ifndef EIO_NO_WRAPPERS
1289 2080
2081eio_req *eio_wd_open (const char *path, int pri, eio_cb cb, void *data)
2082{
2083 REQ (EIO_WD_OPEN); PATH; SEND;
2084}
2085
2086eio_req *eio_wd_close (eio_wd wd, int pri, eio_cb cb, void *data)
2087{
2088 REQ (EIO_WD_CLOSE); req->wd = wd; SEND;
2089}
2090
1290eio_req *eio_nop (int pri, eio_cb cb, void *data) 2091eio_req *eio_nop (int pri, eio_cb cb, void *data)
1291{ 2092{
1292 REQ (EIO_NOP); SEND; 2093 REQ (EIO_NOP); SEND;
1293} 2094}
1294 2095
… …
1310eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2111eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1311{ 2112{
1312 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2113 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1313} 2114}
1314 2115
2116eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2117{
2118 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2119}
2120
2121eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
2122{
2123 REQ (EIO_SYNCFS); req->int1 = fd; SEND;
2124}
2125
2126eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2127{
2128 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2129}
2130
1315eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2131eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1316{ 2132{
1317 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2133 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1318} 2134}
1319 2135
1320eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) 2136eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
1321{ 2137{
1322 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; 2138 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
1323} 2139}
1324 2140
1325eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) 2141eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
1326{ 2142{
1327 REQ (EIO_FDATASYNC); req->int1 = fd; SEND; 2143 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2144}
2145
2146eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2147{
2148 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
1328} 2149}
1329 2150
1330eio_req *eio_close (int fd, int pri, eio_cb cb, void *data) 2151eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
1331{ 2152{
1332 REQ (EIO_CLOSE); req->int1 = fd; SEND; 2153 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) 2156eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
1336{ 2157{
1337 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND; 2158 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
1338} 2159}
1339 2160
2161eio_req *eio_seek (int fd, off_t offset, int whence, int pri, eio_cb cb, void *data)
2162{
2163 REQ (EIO_SEEK); req->int1 = fd; req->offs = offset; req->int2 = whence; SEND;
2164}
2165
1340eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data) 2166eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1341{ 2167{
1342 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND; 2168 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1343} 2169}
1344 2170
… …
1350eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) 2176eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1351{ 2177{
1352 REQ (EIO_FSTAT); req->int1 = fd; SEND; 2178 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1353} 2179}
1354 2180
2181eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
2182{
2183 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
2184}
2185
1355eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) 2186eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1356{ 2187{
1357 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; 2188 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1358} 2189}
1359 2190
… …
1365eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) 2196eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1366{ 2197{
1367 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2198 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1368} 2199}
1369 2200
1370eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2201eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1371{ 2202{
1372 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2203 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1373} 2204}
1374 2205
1375eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) 2206eio_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) 2226eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1396{ 2227{
1397 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2228 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1398} 2229}
1399 2230
1400eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2231eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1401{ 2232{
1402 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2233 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1403} 2234}
1404 2235
1405eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) 2236eio_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) 2252eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1422{ 2253{
1423 return eio__1path (EIO_READLINK, path, pri, cb, data); 2254 return eio__1path (EIO_READLINK, path, pri, cb, data);
1424} 2255}
1425 2256
2257eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2258{
2259 return eio__1path (EIO_REALPATH, path, pri, cb, data);
2260}
2261
1426eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) 2262eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1427{ 2263{
1428 return eio__1path (EIO_STAT, path, pri, cb, data); 2264 return eio__1path (EIO_STAT, path, pri, cb, data);
1429} 2265}
1430 2266
1431eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) 2267eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1432{ 2268{
1433 return eio__1path (EIO_LSTAT, path, pri, cb, data); 2269 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1434} 2270}
1435 2271
2272eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
2273{
2274 return eio__1path (EIO_STATVFS, path, pri, cb, data);
2275}
2276
1436eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) 2277eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1437{ 2278{
1438 return eio__1path (EIO_UNLINK, path, pri, cb, data); 2279 return eio__1path (EIO_UNLINK, path, pri, cb, data);
1439} 2280}
1440 2281
1441eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data) 2282eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1442{ 2283{
1443 return eio__1path (EIO_RMDIR, path, pri, cb, data); 2284 return eio__1path (EIO_RMDIR, path, pri, cb, data);
1444} 2285}
1445 2286
1446eio_req *eio_readdir (const char *path, int pri, eio_cb cb, void *data) 2287eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1447{ 2288{
1448 return eio__1path (EIO_READDIR, path, pri, cb, data); 2289 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1449} 2290}
1450 2291
1451eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) 2292eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1452{ 2293{
1453 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; 2294 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1454} 2295}
1455 2296
1456static eio_req * 2297static eio_req *
1457eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2298eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1458{ 2299{
… …
1482eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2323eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1483{ 2324{
1484 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2325 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1485} 2326}
1486 2327
1487eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) 2328eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
1488{ 2329{
1489 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND; 2330 REQ (EIO_CUSTOM); req->feed = execute; SEND;
1490} 2331}
1491 2332
1492#endif 2333#endif
1493 2334
1494eio_req *eio_grp (eio_cb cb, void *data) 2335eio_req *eio_grp (eio_cb cb, void *data)
… …
1503#undef SEND 2344#undef SEND
1504 2345
1505/*****************************************************************************/ 2346/*****************************************************************************/
1506/* grp functions */ 2347/* grp functions */
1507 2348
2349void
1508void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2350eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1509{ 2351{
1510 grp->int2 = limit; 2352 grp->int2 = limit;
1511 grp->feed = feed; 2353 grp->feed = feed;
1512 2354
1513 grp_try_feed (grp); 2355 grp_try_feed (grp);
1514} 2356}
1515 2357
2358void
1516void eio_grp_limit (eio_req *grp, int limit) 2359eio_grp_limit (eio_req *grp, int limit)
1517{ 2360{
1518 grp->int2 = limit; 2361 grp->int2 = limit;
1519 2362
1520 grp_try_feed (grp); 2363 grp_try_feed (grp);
1521} 2364}
1522 2365
2366void
1523void eio_grp_add (eio_req *grp, eio_req *req) 2367eio_grp_add (eio_req *grp, eio_req *req)
1524{ 2368{
1525 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2369 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1526 2370
1527 grp->flags |= EIO_FLAG_GROUPADD; 2371 grp->flags |= EIO_FLAG_GROUPADD;
1528 2372
… …
1539} 2383}
1540 2384
1541/*****************************************************************************/ 2385/*****************************************************************************/
1542/* misc garbage */ 2386/* misc garbage */
1543 2387
2388eio_ssize_t
1544ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2389eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1545{ 2390{
1546 etp_worker wrk;
1547
1548 wrk.dbuf = 0;
1549
1550 eio__sendfile (ofd, ifd, offset, count, &wrk); 2391 return eio__sendfile (ofd, ifd, offset, count);
1551
1552 if (wrk.dbuf)
1553 free (wrk.dbuf);
1554} 2392}
1555 2393

Diff Legend

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