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

Comparing libeio/eio.c (file contents):
Revision 1.30 by root, Wed Jun 3 12:24:49 2009 UTC vs.
Revision 1.142 by root, Sun May 1 17:15:45 2016 UTC

1/* 1/*
2 * libeio implementation 2 * libeio implementation
3 * 3 *
4 * Copyright (c) 2007,2008 Marc Alexander Lehmann <libeio@schmorp.de> 4 * Copyright (c) 2007,2008,2009,2010,2011,2012,2013,2016 Marc Alexander Lehmann <libeio@schmorp.de>
5 * All rights reserved. 5 * All rights reserved.
6 * 6 *
7 * Redistribution and use in source and binary forms, with or without modifica- 7 * Redistribution and use in source and binary forms, with or without modifica-
8 * tion, are permitted provided that the following conditions are met: 8 * tion, are permitted provided that the following conditions are met:
9 * 9 *
10 * 1. Redistributions of source code must retain the above copyright notice, 10 * 1. Redistributions of source code must retain the above copyright notice,
11 * this list of conditions and the following disclaimer. 11 * this list of conditions and the following disclaimer.
12 * 12 *
13 * 2. Redistributions in binary form must reproduce the above copyright 13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the 14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution. 15 * documentation and/or other materials provided with the distribution.
16 * 16 *
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR IMPLIED 17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR IMPLIED
18 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MER- 18 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MER-
19 * CHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO 19 * CHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO
20 * EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPE- 20 * EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPE-
21 * CIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, 21 * CIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
… …
35 * and other provisions required by the GPL. If you do not delete the 35 * and other provisions required by the GPL. If you do not delete the
36 * provisions above, a recipient may use your version of this file under 36 * provisions above, a recipient may use your version of this file under
37 * either the BSD or the GPL. 37 * either the BSD or the GPL.
38 */ 38 */
39 39
40#ifndef _WIN32
41# include "config.h"
42#endif
43
40#include "eio.h" 44#include "eio.h"
41
42#ifdef EIO_STACKSIZE
43# define XTHREAD_STACKSIZE EIO_STACKSIZE
44#endif
45#include "xthread.h" 45#include "ecb.h"
46 46
47#include <errno.h> 47#include <errno.h>
48#include <stddef.h> 48#include <stddef.h>
49#include <stdlib.h> 49#include <stdlib.h>
50#include <string.h> 50#include <string.h>
… …
53#include <sys/stat.h> 53#include <sys/stat.h>
54#include <limits.h> 54#include <limits.h>
55#include <fcntl.h> 55#include <fcntl.h>
56#include <assert.h> 56#include <assert.h>
57 57
58/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
59/* intptr_t only comes from stdint.h, says idiot openbsd coder */
60#if HAVE_STDINT_H
61# include <stdint.h>
62#endif
63
64#ifndef ECANCELED
65# define ECANCELED EDOM
66#endif
67#ifndef ELOOP
68# define ELOOP EDOM
69#endif
70
71#if !defined(ENOTSOCK) && defined(WSAENOTSOCK)
72# define ENOTSOCK WSAENOTSOCK
73#endif
74
75static void eio_destroy (eio_req *req);
76
58#ifndef EIO_FINISH 77#ifndef EIO_FINISH
59# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0 78# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
60#endif 79#endif
61 80
62#ifndef EIO_DESTROY 81#ifndef EIO_DESTROY
… …
65 84
66#ifndef EIO_FEED 85#ifndef EIO_FEED
67# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0) 86# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)
68#endif 87#endif
69 88
89#ifndef EIO_FD_TO_WIN32_HANDLE
90# define EIO_FD_TO_WIN32_HANDLE(fd) _get_osfhandle (fd)
91#endif
92#ifndef EIO_WIN32_HANDLE_TO_FD
93# define EIO_WIN32_HANDLE_TO_FD(handle) _open_osfhandle (handle, 0)
94#endif
95
96#define EIO_ERRNO(errval,retval) ((errno = errval), retval)
97
98#define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1)
99
70#ifdef _WIN32 100#ifdef _WIN32
71 101
72 /*doh*/ 102 #undef PAGESIZE
103 #define PAGESIZE 4096 /* GetSystemInfo? */
104
105 /* TODO: look at how perl does stat (non-sloppy), unlink (ro-files), utime, link */
106
107 #ifdef EIO_STRUCT_STATI64
108 /* look at perl's non-sloppy stat */
109 #define stat(path,buf) _stati64 (path,buf)
110 #define fstat(fd,buf) _fstati64 (fd,buf)
111 #endif
112 #define lstat(path,buf) stat (path,buf)
113 #define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
114 #define mkdir(path,mode) _mkdir (path)
115 #define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
116
117 #define chmod(path,mode) _chmod (path, mode)
118 #define dup(fd) _dup (fd)
119 #define dup2(fd1,fd2) _dup2 (fd1, fd2)
120 #define pipe(fds) _pipe (fds, 4096, O_BINARY)
121
122 #define fcntl(fd,cmd,arg) EIO_ENOSYS ()
123 #define ioctl(fd,cmd,arg) EIO_ENOSYS ()
124 #define fchmod(fd,mode) EIO_ENOSYS ()
125 #define chown(path,uid,gid) EIO_ENOSYS ()
126 #define fchown(fd,uid,gid) EIO_ENOSYS ()
127 #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
128 #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
129 #define mknod(path,mode,dev) EIO_ENOSYS ()
130 #define sync() EIO_ENOSYS ()
131 #define readlink(path,buf,s) EIO_ENOSYS ()
132 #define statvfs(path,buf) EIO_ENOSYS ()
133 #define fstatvfs(fd,buf) EIO_ENOSYS ()
134
135 #define pread(fd,buf,count,offset) eio__pread (fd, buf, count, offset)
136 #define pwrite(fd,buf,count,offset) eio__pwrite (fd, buf, count, offset)
137
138 #if __GNUC__
139 typedef long long eio_off_t; /* signed for compatibility to msvc */
140 #else
141 typedef __int64 eio_off_t; /* unsigned not supported by msvc */
142 #endif
143
144 static eio_ssize_t
145 eio__pread (int fd, void *buf, eio_ssize_t count, eio_off_t offset)
146 {
147 OVERLAPPED o = { 0 };
148 DWORD got;
149
150 o.Offset = offset;
151 o.OffsetHigh = offset >> 32;
152
153 return ReadFile ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd), buf, count, &got, &o)
154 ? got : -1;
155 }
156
157 static eio_ssize_t
158 eio__pwrite (int fd, void *buf, eio_ssize_t count, eio_off_t offset)
159 {
160 OVERLAPPED o = { 0 };
161 DWORD got;
162
163 o.Offset = offset;
164 o.OffsetHigh = offset >> 32;
165
166 return WriteFile ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd), buf, count, &got, &o)
167 ? got : -1;
168 }
169
170 /* rename() uses MoveFile, which fails to overwrite */
171 #define rename(old,neu) eio__rename (old, neu)
172
173 static int
174 eio__rename (const char *old, const char *neu)
175 {
176 if (MoveFileEx (old, neu, MOVEFILE_REPLACE_EXISTING))
177 return 0;
178
179 /* should steal _dosmaperr */
180 switch (GetLastError ())
181 {
182 case ERROR_FILE_NOT_FOUND:
183 case ERROR_PATH_NOT_FOUND:
184 case ERROR_INVALID_DRIVE:
185 case ERROR_NO_MORE_FILES:
186 case ERROR_BAD_NETPATH:
187 case ERROR_BAD_NET_NAME:
188 case ERROR_BAD_PATHNAME:
189 case ERROR_FILENAME_EXCED_RANGE:
190 errno = ENOENT;
191 break;
192
193 default:
194 errno = EACCES;
195 break;
196 }
197
198 return -1;
199 }
200
201 /* we could even stat and see if it exists */
202 static int
203 symlink (const char *old, const char *neu)
204 {
205 #if WINVER >= 0x0600
206 if (CreateSymbolicLink (neu, old, 1))
207 return 0;
208
209 if (CreateSymbolicLink (neu, old, 0))
210 return 0;
211 #endif
212
213 return EIO_ERRNO (ENOENT, -1);
214 }
215
216 /* POSIX API only, causing trouble for win32 apps */
217 #define CreateHardLink(neu,old,flags) 0 /* not really creating hardlink, still using relative paths? */
218 #define CreateSymbolicLink(neu,old,flags) 0 /* vista+ only */
219
220 struct statvfs
221 {
222 int dummy;
223 };
224
225 #define DT_DIR EIO_DT_DIR
226 #define DT_REG EIO_DT_REG
227 #define D_NAME(entp) entp.cFileName
228 #define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
73 229
74#else 230#else
75 231
76# include "config.h"
77# include <sys/time.h> 232 #include <sys/time.h>
78# include <sys/select.h> 233 #include <sys/select.h>
79# include <unistd.h> 234 #include <unistd.h>
235 #include <signal.h>
236 #include <dirent.h>
237
238 #ifdef ANDROID
239 #include <sys/vfs.h>
240 #define statvfs statfs
241 #define fstatvfs fstatfs
242 #include <asm/page.h> /* supposedly limits.h does #define PAGESIZE PAGESIZE */
243 #else
244 #include <sys/statvfs.h>
245 #endif
246
247 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
248 #include <sys/mman.h>
249 #endif
250
251 #define D_NAME(entp) entp->d_name
252
253 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
254 #if __FreeBSD__ || __NetBSD__ || __OpenBSD__
255 #define _DIRENT_HAVE_D_TYPE /* sigh */
256 #define D_INO(de) (de)->d_fileno
257 #define D_NAMLEN(de) (de)->d_namlen
258 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
259 #define D_INO(de) (de)->d_ino
260 #endif
261
262 #ifdef _D_EXACT_NAMLEN
263 #undef D_NAMLEN
264 #define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
265 #endif
266
267 #ifdef _DIRENT_HAVE_D_TYPE
268 #define D_TYPE(de) (de)->d_type
269 #endif
270
271 #ifndef EIO_STRUCT_DIRENT
272 #define EIO_STRUCT_DIRENT struct dirent
273 #endif
274
275#endif
276
277#if HAVE_UTIMES
80# include <utime.h> 278# include <utime.h>
81# include <signal.h>
82# include <dirent.h>
83
84# ifndef EIO_STRUCT_DIRENT
85# define EIO_STRUCT_DIRENT struct dirent
86# endif 279#endif
87 280
281#if HAVE_SYS_SYSCALL_H
282# include <sys/syscall.h>
88#endif 283#endif
89 284
90#if HAVE_SENDFILE 285#if HAVE_SENDFILE
91# if __linux 286# if __linux
92# include <sys/sendfile.h> 287# include <sys/sendfile.h>
93# elif __freebsd 288# elif __FreeBSD__ || defined __APPLE__
94# include <sys/socket.h> 289# include <sys/socket.h>
95# include <sys/uio.h> 290# include <sys/uio.h>
96# elif __hpux 291# elif __hpux
97# include <sys/socket.h> 292# include <sys/socket.h>
98# elif __solaris /* not yet */ 293# elif __solaris
99# include <sys/sendfile.h> 294# include <sys/sendfile.h>
100# else 295# else
101# error sendfile support requested but not available 296# error sendfile support requested but not available
102# endif 297# endif
103#endif 298#endif
104 299
105/* number of seconds after which an idle threads exit */ 300#ifndef D_TYPE
106#define IDLE_TIMEOUT 10 301# define D_TYPE(de) 0
302#endif
303#ifndef D_INO
304# define D_INO(de) 0
305#endif
306#ifndef D_NAMLEN
307# define D_NAMLEN(entp) strlen (D_NAME (entp))
308#endif
107 309
108/* used for struct dirent, AIX doesn't provide it */ 310/* used for struct dirent, AIX doesn't provide it */
109#ifndef NAME_MAX 311#ifndef NAME_MAX
110# define NAME_MAX 4096 312# define NAME_MAX 4096
111#endif 313#endif
112 314
315/* used for readlink etc. */
316#ifndef PATH_MAX
317# define PATH_MAX 4096
318#endif
319
113/* buffer size for various temporary buffers */ 320/* buffer size for various temporary buffers */
114#define EIO_BUFSIZE 65536 321#define EIO_BUFSIZE 65536
115 322
116#define dBUF \ 323#define dBUF \
117 char *eio_buf; \
118 ETP_WORKER_LOCK (self); \
119 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 324 char *eio_buf = malloc (EIO_BUFSIZE); \
120 ETP_WORKER_UNLOCK (self); \
121 errno = ENOMEM; \ 325 errno = ENOMEM; \
122 if (!eio_buf) \ 326 if (!eio_buf) \
123 return -1; 327 return -1
124 328
125#define EIO_TICKS ((1000000 + 1023) >> 10) 329#define FUBd \
330 free (eio_buf)
126 331
127/*****************************************************************************/ 332/*****************************************************************************/
128 333
129#if __GNUC__ >= 3 334struct etp_tmpbuf;
130# define expect(expr,value) __builtin_expect ((expr),(value)) 335
336#if _POSIX_VERSION >= 200809L
337 #define HAVE_AT 1
338 #define WD2FD(wd) ((wd) ? (wd)->fd : AT_FDCWD)
339 #ifndef O_SEARCH
340 #define O_SEARCH O_RDONLY
341 #endif
131#else 342#else
132# define expect(expr,value) (expr) 343 #define HAVE_AT 0
344 static const char *wd_expand (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path);
133#endif 345#endif
134 346
135#define expect_false(expr) expect ((expr) != 0, 0) 347struct eio_pwd
136#define expect_true(expr) expect ((expr) != 0, 1) 348{
349#if HAVE_AT
350 int fd;
351#endif
352 int len;
353 char str[1]; /* actually, a 0-terminated canonical path */
354};
137 355
138/*****************************************************************************/ 356/*****************************************************************************/
139 357
140#define ETP_PRI_MIN EIO_PRI_MIN 358#define ETP_PRI_MIN EIO_PRI_MIN
141#define ETP_PRI_MAX EIO_PRI_MAX 359#define ETP_PRI_MAX EIO_PRI_MAX
142 360
361#define ETP_TYPE_QUIT -1
362#define ETP_TYPE_GROUP EIO_GROUP
363
364static void eio_nop_callback (void) { }
365static void (*eio_want_poll_cb)(void) = eio_nop_callback;
366static void (*eio_done_poll_cb)(void) = eio_nop_callback;
367
368#define ETP_WANT_POLL(pool) eio_want_poll_cb ()
369#define ETP_DONE_POLL(pool) eio_done_poll_cb ()
370
143struct etp_worker; 371struct etp_worker;
144
145#define ETP_REQ eio_req 372#define ETP_REQ eio_req
146#define ETP_DESTROY(req) eio_destroy (req) 373#define ETP_DESTROY(req) eio_destroy (req)
147static int eio_finish (eio_req *req); 374static int eio_finish (eio_req *req);
148#define ETP_FINISH(req) eio_finish (req) 375#define ETP_FINISH(req) eio_finish (req)
149static void eio_execute (struct etp_worker *self, eio_req *req); 376static void eio_execute (struct etp_worker *self, eio_req *req);
150#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) 377#define ETP_EXECUTE(wrk,req) eio_execute (wrk, req)
151 378
152#define ETP_WORKER_CLEAR(req) \ 379#include "etp.c"
153 if (wrk->dbuf) \ 380
154 { \ 381static struct etp_pool eio_pool;
155 free (wrk->dbuf); \ 382#define EIO_POOL (&eio_pool)
156 wrk->dbuf = 0; \
157 } \
158 \
159 if (wrk->dirp) \
160 { \
161 closedir (wrk->dirp); \
162 wrk->dirp = 0; \
163 }
164#define ETP_WORKER_COMMON \
165 void *dbuf; \
166 DIR *dirp;
167 383
168/*****************************************************************************/ 384/*****************************************************************************/
169 385
170#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
171
172/* calculcate time difference in ~1/EIO_TICKS of a second */
173static int tvdiff (struct timeval *tv1, struct timeval *tv2)
174{
175 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
176 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
177}
178
179static unsigned int started, idle, wanted = 4;
180
181static void (*want_poll_cb) (void);
182static void (*done_poll_cb) (void);
183
184static unsigned int max_poll_time; /* reslock */
185static unsigned int max_poll_reqs; /* reslock */
186
187static volatile unsigned int nreqs; /* reqlock */
188static volatile unsigned int nready; /* reqlock */
189static volatile unsigned int npending; /* reqlock */
190static volatile unsigned int max_idle = 4;
191
192static mutex_t wrklock = X_MUTEX_INIT;
193static mutex_t reslock = X_MUTEX_INIT;
194static mutex_t reqlock = X_MUTEX_INIT;
195static cond_t reqwait = X_COND_INIT;
196
197#if !HAVE_PREADWRITE
198/*
199 * make our pread/pwrite emulation safe against themselves, but not against
200 * normal read/write by using a mutex. slows down execution a lot,
201 * but that's your problem, not mine.
202 */
203static mutex_t preadwritelock = X_MUTEX_INIT;
204#endif
205
206typedef struct etp_worker
207{
208 /* locked by wrklock */
209 struct etp_worker *prev, *next;
210
211 thread_t tid;
212
213 /* locked by reslock, reqlock or wrklock */
214 ETP_REQ *req; /* currently processed request */
215
216 ETP_WORKER_COMMON
217} etp_worker;
218
219static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */
220
221#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
222#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
223
224/* worker threads management */
225
226static void etp_worker_clear (etp_worker *wrk)
227{
228 ETP_WORKER_CLEAR (wrk);
229}
230
231static void etp_worker_free (etp_worker *wrk)
232{
233 wrk->next->prev = wrk->prev;
234 wrk->prev->next = wrk->next;
235
236 free (wrk);
237}
238
239static unsigned int etp_nreqs (void)
240{
241 int retval;
242 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
243 retval = nreqs;
244 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
245 return retval;
246}
247
248static unsigned int etp_nready (void)
249{
250 unsigned int retval;
251
252 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
253 retval = nready;
254 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
255
256 return retval;
257}
258
259static unsigned int etp_npending (void)
260{
261 unsigned int retval;
262
263 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
264 retval = npending;
265 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
266
267 return retval;
268}
269
270static unsigned int etp_nthreads (void)
271{
272 unsigned int retval;
273
274 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
275 retval = started;
276 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
277
278 return retval;
279}
280
281/*
282 * a somewhat faster data structure might be nice, but
283 * with 8 priorities this actually needs <20 insns
284 * per shift, the most expensive operation.
285 */
286typedef struct {
287 ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */
288 int size;
289} etp_reqq;
290
291static etp_reqq req_queue;
292static etp_reqq res_queue;
293
294static int reqq_push (etp_reqq *q, ETP_REQ *req)
295{
296 int pri = req->pri;
297 req->next = 0;
298
299 if (q->qe[pri])
300 {
301 q->qe[pri]->next = req;
302 q->qe[pri] = req;
303 }
304 else
305 q->qe[pri] = q->qs[pri] = req;
306
307 return q->size++;
308}
309
310static ETP_REQ *reqq_shift (etp_reqq *q)
311{
312 int pri;
313
314 if (!q->size)
315 return 0;
316
317 --q->size;
318
319 for (pri = ETP_NUM_PRI; pri--; )
320 {
321 eio_req *req = q->qs[pri];
322
323 if (req)
324 {
325 if (!(q->qs[pri] = (eio_req *)req->next))
326 q->qe[pri] = 0;
327
328 return req;
329 }
330 }
331
332 abort ();
333}
334
335static void etp_atfork_prepare (void)
336{
337 X_LOCK (wrklock);
338 X_LOCK (reqlock);
339 X_LOCK (reslock);
340#if !HAVE_PREADWRITE
341 X_LOCK (preadwritelock);
342#endif
343}
344
345static void etp_atfork_parent (void)
346{
347#if !HAVE_PREADWRITE
348 X_UNLOCK (preadwritelock);
349#endif
350 X_UNLOCK (reslock);
351 X_UNLOCK (reqlock);
352 X_UNLOCK (wrklock);
353}
354
355static void etp_atfork_child (void)
356{
357 ETP_REQ *prv;
358
359 while ((prv = reqq_shift (&req_queue)))
360 ETP_DESTROY (prv);
361
362 while ((prv = reqq_shift (&res_queue)))
363 ETP_DESTROY (prv);
364
365 while (wrk_first.next != &wrk_first)
366 {
367 etp_worker *wrk = wrk_first.next;
368
369 if (wrk->req)
370 ETP_DESTROY (wrk->req);
371
372 etp_worker_clear (wrk);
373 etp_worker_free (wrk);
374 }
375
376 started = 0;
377 idle = 0;
378 nreqs = 0;
379 nready = 0;
380 npending = 0;
381
382 etp_atfork_parent ();
383}
384
385static void 386static void
386etp_once_init (void)
387{
388 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
389}
390
391static int
392etp_init (void (*want_poll)(void), void (*done_poll)(void))
393{
394 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
395
396 pthread_once (&doinit, etp_once_init);
397
398 want_poll_cb = want_poll;
399 done_poll_cb = done_poll;
400
401 return 0;
402}
403
404X_THREAD_PROC (etp_proc);
405
406static void etp_start_thread (void)
407{
408 etp_worker *wrk = calloc (1, sizeof (etp_worker));
409
410 /*TODO*/
411 assert (("unable to allocate worker thread data", wrk));
412
413 X_LOCK (wrklock);
414
415 if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
416 {
417 wrk->prev = &wrk_first;
418 wrk->next = wrk_first.next;
419 wrk_first.next->prev = wrk;
420 wrk_first.next = wrk;
421 ++started;
422 }
423 else
424 free (wrk);
425
426 X_UNLOCK (wrklock);
427}
428
429static void etp_maybe_start_thread (void)
430{
431 if (expect_true (etp_nthreads () >= wanted))
432 return;
433
434 /* todo: maybe use idle here, but might be less exact */
435 if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
436 return;
437
438 etp_start_thread ();
439}
440
441static void etp_end_thread (void)
442{
443 eio_req *req = calloc (1, sizeof (eio_req));
444
445 req->type = -1;
446 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
447
448 X_LOCK (reqlock);
449 reqq_push (&req_queue, req);
450 X_COND_SIGNAL (reqwait);
451 X_UNLOCK (reqlock);
452
453 X_LOCK (wrklock);
454 --started;
455 X_UNLOCK (wrklock);
456}
457
458static int etp_poll (void)
459{
460 unsigned int maxreqs;
461 unsigned int maxtime;
462 struct timeval tv_start, tv_now;
463
464 X_LOCK (reslock);
465 maxreqs = max_poll_reqs;
466 maxtime = max_poll_time;
467 X_UNLOCK (reslock);
468
469 if (maxtime)
470 gettimeofday (&tv_start, 0);
471
472 for (;;)
473 {
474 ETP_REQ *req;
475
476 etp_maybe_start_thread ();
477
478 X_LOCK (reslock);
479 req = reqq_shift (&res_queue);
480
481 if (req)
482 {
483 --npending;
484
485 if (!res_queue.size && done_poll_cb)
486 done_poll_cb ();
487 }
488
489 X_UNLOCK (reslock);
490
491 if (!req)
492 return 0;
493
494 X_LOCK (reqlock);
495 --nreqs;
496 X_UNLOCK (reqlock);
497
498 if (expect_false (req->type == EIO_GROUP && req->size))
499 {
500 req->int1 = 1; /* mark request as delayed */
501 continue;
502 }
503 else
504 {
505 int res = ETP_FINISH (req);
506 if (expect_false (res))
507 return res;
508 }
509
510 if (expect_false (maxreqs && !--maxreqs))
511 break;
512
513 if (maxtime)
514 {
515 gettimeofday (&tv_now, 0);
516
517 if (tvdiff (&tv_start, &tv_now) >= maxtime)
518 break;
519 }
520 }
521
522 errno = EAGAIN;
523 return -1;
524}
525
526static void etp_cancel (ETP_REQ *req)
527{
528 X_LOCK (wrklock);
529 req->flags |= EIO_FLAG_CANCELLED;
530 X_UNLOCK (wrklock);
531
532 eio_grp_cancel (req);
533}
534
535static void etp_submit (ETP_REQ *req)
536{
537 req->pri -= ETP_PRI_MIN;
538
539 if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
540 if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
541
542 if (expect_false (req->type == EIO_GROUP))
543 {
544 /* I hope this is worth it :/ */
545 X_LOCK (reqlock);
546 ++nreqs;
547 X_UNLOCK (reqlock);
548
549 X_LOCK (reslock);
550
551 ++npending;
552
553 if (!reqq_push (&res_queue, req) && want_poll_cb)
554 want_poll_cb ();
555
556 X_UNLOCK (reslock);
557 }
558 else
559 {
560 X_LOCK (reqlock);
561 ++nreqs;
562 ++nready;
563 reqq_push (&req_queue, req);
564 X_COND_SIGNAL (reqwait);
565 X_UNLOCK (reqlock);
566
567 etp_maybe_start_thread ();
568 }
569}
570
571static void etp_set_max_poll_time (double nseconds)
572{
573 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
574 max_poll_time = nseconds;
575 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
576}
577
578static void etp_set_max_poll_reqs (unsigned int maxreqs)
579{
580 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
581 max_poll_reqs = maxreqs;
582 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
583}
584
585static void etp_set_max_idle (unsigned int nthreads)
586{
587 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
588 max_idle = nthreads <= 0 ? 1 : nthreads;
589 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
590}
591
592static void etp_set_min_parallel (unsigned int nthreads)
593{
594 if (wanted < nthreads)
595 wanted = nthreads;
596}
597
598static void etp_set_max_parallel (unsigned int nthreads)
599{
600 if (wanted > nthreads)
601 wanted = nthreads;
602
603 while (started > wanted)
604 etp_end_thread ();
605}
606
607/*****************************************************************************/
608
609static void grp_try_feed (eio_req *grp) 387grp_try_feed (eio_req *grp)
610{ 388{
611 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 389 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
612 { 390 {
613 grp->flags &= ~EIO_FLAG_GROUPADD; 391 grp->flags &= ~ETP_FLAG_GROUPADD;
614 392
615 EIO_FEED (grp); 393 EIO_FEED (grp);
616 394
617 /* stop if no progress has been made */ 395 /* stop if no progress has been made */
618 if (!(grp->flags & EIO_FLAG_GROUPADD)) 396 if (!(grp->flags & ETP_FLAG_GROUPADD))
619 { 397 {
620 grp->feed = 0; 398 grp->feed = 0;
621 break; 399 break;
622 } 400 }
623 } 401 }
624} 402}
625 403
404static int
626static int grp_dec (eio_req *grp) 405grp_dec (eio_req *grp)
627{ 406{
628 --grp->size; 407 --grp->size;
629 408
630 /* call feeder, if applicable */ 409 /* call feeder, if applicable */
631 grp_try_feed (grp); 410 grp_try_feed (grp);
632 411
633 /* finish, if done */ 412 /* finish, if done */
634 if (!grp->size && grp->int1) 413 if (!grp->size && grp->flags & ETP_FLAG_DELAYED)
635 return eio_finish (grp); 414 return eio_finish (grp);
636 else 415 else
637 return 0; 416 return 0;
638} 417}
639 418
419static void
640void eio_destroy (eio_req *req) 420eio_destroy (eio_req *req)
641{ 421{
642 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); 422 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
643 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); 423 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
644 424
645 EIO_DESTROY (req); 425 EIO_DESTROY (req);
646} 426}
647 427
428static int
648static int eio_finish (eio_req *req) 429eio_finish (eio_req *req)
649{ 430{
650 int res = EIO_FINISH (req); 431 int res = EIO_FINISH (req);
651 432
652 if (req->grp) 433 if (req->grp)
653 { 434 {
… …
661 if (grp->grp_first == req) 442 if (grp->grp_first == req)
662 grp->grp_first = req->grp_next; 443 grp->grp_first = req->grp_next;
663 444
664 res2 = grp_dec (grp); 445 res2 = grp_dec (grp);
665 446
666 if (!res && res2) 447 if (!res)
667 res = res2; 448 res = res2;
668 } 449 }
669 450
670 eio_destroy (req); 451 eio_destroy (req);
671 452
672 return res; 453 return res;
673} 454}
674 455
456void
675void eio_grp_cancel (eio_req *grp) 457eio_grp_cancel (eio_req *grp)
676{ 458{
677 for (grp = grp->grp_first; grp; grp = grp->grp_next) 459 etp_grp_cancel (EIO_POOL, grp);
678 eio_cancel (grp);
679} 460}
680 461
462void
681void eio_cancel (eio_req *req) 463eio_cancel (eio_req *req)
682{ 464{
683 etp_cancel (req); 465 etp_cancel (EIO_POOL, req);
684} 466}
685 467
468void
686void eio_submit (eio_req *req) 469eio_submit (eio_req *req)
687{ 470{
688 etp_submit (req); 471 etp_submit (EIO_POOL, req);
689} 472}
690 473
691unsigned int eio_nreqs (void) 474unsigned int
475eio_nreqs (void)
692{ 476{
693 return etp_nreqs (); 477 return etp_nreqs (EIO_POOL);
694} 478}
695 479
696unsigned int eio_nready (void) 480unsigned int
481eio_nready (void)
697{ 482{
698 return etp_nready (); 483 return etp_nready (EIO_POOL);
699} 484}
700 485
701unsigned int eio_npending (void) 486unsigned int
487eio_npending (void)
702{ 488{
703 return etp_npending (); 489 return etp_npending (EIO_POOL);
704} 490}
705 491
706unsigned int eio_nthreads (void) 492unsigned int ecb_cold
493eio_nthreads (void)
707{ 494{
708 return etp_nthreads (); 495 return etp_nthreads (EIO_POOL);
709} 496}
710 497
498void ecb_cold
711void eio_set_max_poll_time (double nseconds) 499eio_set_max_poll_time (double nseconds)
712{ 500{
713 etp_set_max_poll_time (nseconds); 501 etp_set_max_poll_time (EIO_POOL, nseconds);
714} 502}
715 503
504void ecb_cold
716void eio_set_max_poll_reqs (unsigned int maxreqs) 505eio_set_max_poll_reqs (unsigned int maxreqs)
717{ 506{
718 etp_set_max_poll_reqs (maxreqs); 507 etp_set_max_poll_reqs (EIO_POOL, maxreqs);
719} 508}
720 509
510void ecb_cold
721void eio_set_max_idle (unsigned int nthreads) 511eio_set_max_idle (unsigned int nthreads)
722{ 512{
723 etp_set_max_idle (nthreads); 513 etp_set_max_idle (EIO_POOL, nthreads);
724} 514}
725 515
516void ecb_cold
517eio_set_idle_timeout (unsigned int seconds)
518{
519 etp_set_idle_timeout (EIO_POOL, seconds);
520}
521
522void ecb_cold
726void eio_set_min_parallel (unsigned int nthreads) 523eio_set_min_parallel (unsigned int nthreads)
727{ 524{
728 etp_set_min_parallel (nthreads); 525 etp_set_min_parallel (EIO_POOL, nthreads);
729} 526}
730 527
528void ecb_cold
731void eio_set_max_parallel (unsigned int nthreads) 529eio_set_max_parallel (unsigned int nthreads)
732{ 530{
733 etp_set_max_parallel (nthreads); 531 etp_set_max_parallel (EIO_POOL, nthreads);
734} 532}
735 533
736int eio_poll (void) 534int eio_poll (void)
737{ 535{
738 return etp_poll (); 536 return etp_poll (EIO_POOL);
739} 537}
740 538
741/*****************************************************************************/ 539/*****************************************************************************/
742/* work around various missing functions */ 540/* work around various missing functions */
743 541
744#if !HAVE_PREADWRITE
745# undef pread
746# undef pwrite
747# define pread eio__pread
748# define pwrite eio__pwrite
749
750static ssize_t
751eio__pread (int fd, void *buf, size_t count, off_t offset)
752{
753 ssize_t res;
754 off_t ooffset;
755
756 X_LOCK (preadwritelock);
757 ooffset = lseek (fd, 0, SEEK_CUR);
758 lseek (fd, offset, SEEK_SET);
759 res = read (fd, buf, count);
760 lseek (fd, ooffset, SEEK_SET);
761 X_UNLOCK (preadwritelock);
762
763 return res;
764}
765
766static ssize_t
767eio__pwrite (int fd, void *buf, size_t count, off_t offset)
768{
769 ssize_t res;
770 off_t ooffset;
771
772 X_LOCK (preadwritelock);
773 ooffset = lseek (fd, 0, SEEK_CUR);
774 lseek (fd, offset, SEEK_SET);
775 res = write (fd, buf, count);
776 lseek (fd, ooffset, SEEK_SET);
777 X_UNLOCK (preadwritelock);
778
779 return res;
780}
781#endif
782
783#ifndef HAVE_FUTIMES 542#ifndef HAVE_UTIMES
784 543
785# undef utimes 544# undef utimes
786# undef futimes
787# define utimes(path,times) eio__utimes (path, times) 545# define utimes(path,times) eio__utimes (path, times)
788# define futimes(fd,times) eio__futimes (fd, times)
789 546
790static int 547static int
791eio__utimes (const char *filename, const struct timeval times[2]) 548eio__utimes (const char *filename, const struct timeval times[2])
792{ 549{
793 if (times) 550 if (times)
… …
801 } 558 }
802 else 559 else
803 return utime (filename, 0); 560 return utime (filename, 0);
804} 561}
805 562
563#endif
564
565#ifndef HAVE_FUTIMES
566
567# undef futimes
568# define futimes(fd,times) eio__futimes (fd, times)
569
570static int
806static int eio__futimes (int fd, const struct timeval tv[2]) 571eio__futimes (int fd, const struct timeval tv[2])
807{ 572{
808 errno = ENOSYS; 573 errno = ENOSYS;
809 return -1; 574 return -1;
810} 575}
811 576
… …
814#if !HAVE_FDATASYNC 579#if !HAVE_FDATASYNC
815# undef fdatasync 580# undef fdatasync
816# define fdatasync(fd) fsync (fd) 581# define fdatasync(fd) fsync (fd)
817#endif 582#endif
818 583
584static int
585eio__syncfs (int fd)
586{
587 int res;
588
589#if HAVE_SYS_SYNCFS
590 res = (int)syscall (__NR_syncfs, (int)(fd));
591#else
592 res = EIO_ENOSYS ();
593#endif
594
595 if (res < 0 && errno == ENOSYS && fd >= 0)
596 sync ();
597
598 return res;
599}
600
819/* sync_file_range always needs emulation */ 601/* sync_file_range always needs emulation */
820int 602static int
821eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags) 603eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
822{ 604{
823#if HAVE_SYNC_FILE_RANGE 605#if HAVE_SYNC_FILE_RANGE
824 int res; 606 int res;
825 607
… …
838 if (!res || errno != ENOSYS) 620 if (!res || errno != ENOSYS)
839 return res; 621 return res;
840#endif 622#endif
841 623
842 /* even though we could play tricks with the flags, it's better to always 624 /* even though we could play tricks with the flags, it's better to always
843 * call fdatasync, as thta matches the expectation of it's users best */ 625 * call fdatasync, as that matches the expectation of its users best */
844 return fdatasync (fd); 626 return fdatasync (fd);
627}
628
629static int
630eio__fallocate (int fd, int mode, off_t offset, size_t len)
631{
632#if HAVE_LINUX_FALLOCATE
633 return fallocate (fd, mode, offset, len);
634#else
635 return EIO_ENOSYS ();
636#endif
845} 637}
846 638
847#if !HAVE_READAHEAD 639#if !HAVE_READAHEAD
848# undef readahead 640# undef readahead
849# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 641# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
850 642
851static ssize_t 643static eio_ssize_t
852eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) 644eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
853{ 645{
854 size_t todo = count; 646 size_t todo = count;
855 dBUF; 647 dBUF;
856 648
… …
861 pread (fd, eio_buf, len, offset); 653 pread (fd, eio_buf, len, offset);
862 offset += len; 654 offset += len;
863 todo -= len; 655 todo -= len;
864 } 656 }
865 657
866 errno = 0; 658 FUBd;
659
660 /* linux's readahead basically only fails for EBADF or EINVAL (not mmappable) */
661 /* but not for e.g. EIO or eof, so we also never fail */
867 return count; 662 return 0;
868} 663}
869 664
870#endif 665#endif
871 666
872/* sendfile always needs emulation */ 667/* sendfile always needs emulation */
873static ssize_t 668static eio_ssize_t
874eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) 669eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
875{ 670{
671 eio_ssize_t written = 0;
876 ssize_t res; 672 eio_ssize_t res;
877 673
878 if (!count) 674 if (!count)
879 return 0; 675 return 0;
880 676
677 for (;;)
678 {
679#ifdef __APPLE__
680# undef HAVE_SENDFILE /* broken, as everything on os x */
681#endif
881#if HAVE_SENDFILE 682#if HAVE_SENDFILE
882# if __linux 683# if __linux
684 off_t soffset = offset;
883 res = sendfile (ofd, ifd, &offset, count); 685 res = sendfile (ofd, ifd, &soffset, count);
884 686
885# elif __freebsd 687# elif __FreeBSD__
886 /* 688 /*
887 * Of course, the freebsd sendfile is a dire hack with no thoughts 689 * Of course, the freebsd sendfile is a dire hack with no thoughts
888 * wasted on making it similar to other I/O functions. 690 * wasted on making it similar to other I/O functions.
889 */ 691 */
890 {
891 off_t sbytes; 692 off_t sbytes;
892 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 693 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
893 694
894 if (res < 0 && sbytes) 695 #if 0 /* according to the manpage, this is correct, but broken behaviour */
895 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 696 /* freebsd' sendfile will return 0 on success */
697 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
698 /* not on e.g. EIO or EPIPE - sounds broken */
699 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
896 res = sbytes; 700 res = sbytes;
897 } 701 #endif
702
703 /* according to source inspection, this is correct, and useful behaviour */
704 if (sbytes)
705 res = sbytes;
706
707# elif defined __APPLE__
708 off_t sbytes = count;
709 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
710
711 /* according to the manpage, sbytes is always valid */
712 if (sbytes)
713 res = sbytes;
898 714
899# elif __hpux 715# elif __hpux
900 res = sendfile (ofd, ifd, offset, count, 0, 0); 716 res = sendfile (ofd, ifd, offset, count, 0, 0);
901 717
902# elif __solaris 718# elif __solaris
903 {
904 struct sendfilevec vec; 719 struct sendfilevec vec;
905 size_t sbytes; 720 size_t sbytes;
906 721
907 vec.sfv_fd = ifd; 722 vec.sfv_fd = ifd;
908 vec.sfv_flag = 0; 723 vec.sfv_flag = 0;
909 vec.sfv_off = offset; 724 vec.sfv_off = offset;
910 vec.sfv_len = count; 725 vec.sfv_len = count;
911 726
912 res = sendfilev (ofd, &vec, 1, &sbytes); 727 res = sendfilev (ofd, &vec, 1, &sbytes);
913 728
914 if (res < 0 && sbytes) 729 if (res < 0 && sbytes)
915 res = sbytes; 730 res = sbytes;
916 }
917 731
918# endif 732# endif
733
734#elif defined (_WIN32) && 0
735 /* does not work, just for documentation of what would need to be done */
736 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
737 /* libeio guarantees that the file offset does not change, and windows */
738 /* has no way to get an independent handle to the same file description */
739 HANDLE h = TO_SOCKET (ifd);
740 SetFilePointer (h, offset, 0, FILE_BEGIN);
741 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
742
919#else 743#else
920 res = -1; 744 res = EIO_ENOSYS ();
921 errno = ENOSYS;
922#endif 745#endif
923 746
747 /* we assume sendfile can copy at least 128mb in one go */
748 if (res <= 128 * 1024 * 1024)
749 {
750 if (res > 0)
751 written += res;
752
753 if (written)
754 return written;
755
756 break;
757 }
758 else
759 {
760 /* if we requested more, then probably the kernel was lazy */
761 written += res;
762 offset += res;
763 count -= res;
764
765 if (!count)
766 return written;
767 }
768 }
769
924 if (res < 0 770 if (res < 0
925 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 771 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
772 /* BSDs */
773#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
774 || errno == ENOTSUP
775#endif
776#ifdef EOPNOTSUPP /* windows */
777 || errno == EOPNOTSUPP /* BSDs */
778#endif
926#if __solaris 779#if __solaris
927 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 780 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
928#endif 781#endif
929 ) 782 )
930 ) 783 )
… …
934 787
935 res = 0; 788 res = 0;
936 789
937 while (count) 790 while (count)
938 { 791 {
939 ssize_t cnt; 792 eio_ssize_t cnt;
940 793
941 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 794 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
942 795
943 if (cnt <= 0) 796 if (cnt <= 0)
944 { 797 {
… …
956 809
957 offset += cnt; 810 offset += cnt;
958 res += cnt; 811 res += cnt;
959 count -= cnt; 812 count -= cnt;
960 } 813 }
814
815 FUBd;
961 } 816 }
962 817
963 return res; 818 return res;
819}
820
821#ifdef PAGESIZE
822# define eio_pagesize() PAGESIZE
823#else
824static intptr_t
825eio_pagesize (void)
826{
827 static intptr_t page;
828
829 if (!page)
830 page = sysconf (_SC_PAGESIZE);
831
832 return page;
833}
834#endif
835
836static void
837eio_page_align (void **addr, size_t *length)
838{
839 intptr_t mask = eio_pagesize () - 1;
840
841 /* round down addr */
842 intptr_t adj = mask & (intptr_t)*addr;
843
844 *addr = (void *)((intptr_t)*addr - adj);
845 *length += adj;
846
847 /* round up length */
848 *length = (*length + mask) & ~mask;
849}
850
851#if !_POSIX_MEMLOCK
852# define eio__mlockall(a) EIO_ENOSYS ()
853#else
854
855static int
856eio__mlockall (int flags)
857{
858 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
859 extern int mallopt (int, int);
860 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
861 #endif
862
863 if (EIO_MCL_CURRENT != MCL_CURRENT
864 || EIO_MCL_FUTURE != MCL_FUTURE)
865 {
866 flags = 0
867 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
868 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
869 }
870
871 return mlockall (flags);
872}
873#endif
874
875#if !_POSIX_MEMLOCK_RANGE
876# define eio__mlock(a,b) EIO_ENOSYS ()
877#else
878
879static int
880eio__mlock (void *addr, size_t length)
881{
882 eio_page_align (&addr, &length);
883
884 return mlock (addr, length);
885}
886
887#endif
888
889#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
890# define eio__msync(a,b,c) EIO_ENOSYS ()
891#else
892
893static int
894eio__msync (void *mem, size_t len, int flags)
895{
896 eio_page_align (&mem, &len);
897
898 if (EIO_MS_ASYNC != MS_SYNC
899 || EIO_MS_INVALIDATE != MS_INVALIDATE
900 || EIO_MS_SYNC != MS_SYNC)
901 {
902 flags = 0
903 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
904 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
905 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
906 }
907
908 return msync (mem, len, flags);
909}
910
911#endif
912
913static int
914eio__mtouch (eio_req *req)
915{
916 void *mem = req->ptr2;
917 size_t len = req->size;
918 int flags = req->int1;
919
920 eio_page_align (&mem, &len);
921
922 {
923 intptr_t addr = (intptr_t)mem;
924 intptr_t end = addr + len;
925 intptr_t page = eio_pagesize ();
926
927 if (addr < end)
928 if (flags & EIO_MT_MODIFY) /* modify */
929 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
930 else
931 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
932 }
933
934 return 0;
935}
936
937/*****************************************************************************/
938/* requests implemented outside eio_execute, because they are so large */
939
940static void
941eio__lseek (eio_req *req)
942{
943 /* this usually gets optimised away completely, or your compiler sucks, */
944 /* or the whence constants really are not 0, 1, 2 */
945 int whence = req->int2 == EIO_SEEK_SET ? SEEK_SET
946 : req->int2 == EIO_SEEK_CUR ? SEEK_CUR
947 : req->int2 == EIO_SEEK_END ? SEEK_END
948 : req->int2;
949
950 req->offs = lseek (req->int1, req->offs, whence);
951 req->result = req->offs == (off_t)-1 ? -1 : 0;
952}
953
954/* result will always end up in tmpbuf, there is always space for adding a 0-byte */
955static int
956eio__realpath (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
957{
958 char *res;
959 const char *rel = path;
960 char *tmp1, *tmp2;
961#if SYMLOOP_MAX > 32
962 int symlinks = SYMLOOP_MAX;
963#else
964 int symlinks = 32;
965#endif
966
967 errno = EINVAL;
968 if (!rel)
969 return -1;
970
971 errno = ENOENT;
972 if (!*rel)
973 return -1;
974
975 res = etp_tmpbuf_get (tmpbuf, PATH_MAX * 3);
976#ifdef _WIN32
977 if (_access (rel, 4) != 0)
978 return -1;
979
980 symlinks = GetFullPathName (rel, PATH_MAX * 3, res, 0);
981
982 errno = ENAMETOOLONG;
983 if (symlinks >= PATH_MAX * 3)
984 return -1;
985
986 errno = EIO;
987 if (symlinks <= 0)
988 return -1;
989
990 return symlinks;
991
992#else
993 tmp1 = res + PATH_MAX;
994 tmp2 = tmp1 + PATH_MAX;
995
996#if 0 /* disabled, the musl way to do things is just too racy */
997#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
998 /* on linux we may be able to ask the kernel */
999 {
1000 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1001
1002 if (fd >= 0)
1003 {
1004 sprintf (tmp1, "/proc/self/fd/%d", fd);
1005 req->result = readlink (tmp1, res, PATH_MAX);
1006 /* here we should probably stat the open file and the disk file, to make sure they still match */
1007 close (fd);
1008
1009 if (req->result > 0)
1010 goto done;
1011 }
1012 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1013 return -1;
1014 }
1015#endif
1016#endif
1017
1018 if (*rel != '/')
1019 {
1020 int len;
1021
1022 errno = ENOENT;
1023 if (wd == EIO_INVALID_WD)
1024 return -1;
1025
1026 if (wd == EIO_CWD)
1027 {
1028 if (!getcwd (res, PATH_MAX))
1029 return -1;
1030
1031 len = strlen (res);
1032 }
1033 else
1034 memcpy (res, wd->str, len = wd->len);
1035
1036 if (res [1]) /* only use if not / */
1037 res += len;
1038 }
1039
1040 while (*rel)
1041 {
1042 eio_ssize_t len, linklen;
1043 const char *beg = rel;
1044
1045 while (*rel && *rel != '/')
1046 ++rel;
1047
1048 len = rel - beg;
1049
1050 if (!len) /* skip slashes */
1051 {
1052 ++rel;
1053 continue;
1054 }
1055
1056 if (beg [0] == '.')
1057 {
1058 if (len == 1)
1059 continue; /* . - nop */
1060
1061 if (beg [1] == '.' && len == 2)
1062 {
1063 /* .. - back up one component, if possible */
1064
1065 while (res != tmpbuf->ptr)
1066 if (*--res == '/')
1067 break;
1068
1069 continue;
1070 }
1071 }
1072
1073 errno = ENAMETOOLONG;
1074 if (res + 1 + len + 1 >= tmp1)
1075 return -1;
1076
1077 /* copy one component */
1078 *res = '/';
1079 memcpy (res + 1, beg, len);
1080
1081 /* zero-terminate, for readlink */
1082 res [len + 1] = 0;
1083
1084 /* now check if it's a symlink */
1085 linklen = readlink (tmpbuf->ptr, tmp1, PATH_MAX);
1086
1087 if (linklen < 0)
1088 {
1089 if (errno != EINVAL)
1090 return -1;
1091
1092 /* it's a normal directory. hopefully */
1093 res += len + 1;
1094 }
1095 else
1096 {
1097 /* yay, it was a symlink - build new path in tmp2 */
1098 int rellen = strlen (rel);
1099
1100 errno = ENAMETOOLONG;
1101 if (linklen + 1 + rellen >= PATH_MAX)
1102 return -1;
1103
1104 errno = ELOOP;
1105 if (!--symlinks)
1106 return -1;
1107
1108 if (*tmp1 == '/')
1109 res = tmpbuf->ptr; /* symlink resolves to an absolute path */
1110
1111 /* we need to be careful, as rel might point into tmp2 already */
1112 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1113 tmp2 [linklen] = '/';
1114 memcpy (tmp2, tmp1, linklen);
1115
1116 rel = tmp2;
1117 }
1118 }
1119
1120 /* special case for the lone root path */
1121 if (res == tmpbuf->ptr)
1122 *res++ = '/';
1123
1124 return res - (char *)tmpbuf->ptr;
1125#endif
1126}
1127
1128static signed char
1129eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1130{
1131 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1132 : a->inode < b->inode ? -1
1133 : a->inode > b->inode ? 1
1134 : 0;
1135}
1136
1137#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1138
1139#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1140#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1141
1142static void
1143eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1144{
1145 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1146 unsigned char *bit = bits;
1147
1148 assert (CHAR_BIT == 8);
1149 assert (sizeof (eio_dirent) * 8 < 256);
1150 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1151 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1152
1153 if (size <= EIO_SORT_FAST)
1154 return;
1155
1156 /* first prepare an array of bits to test in our radix sort */
1157 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1158 /* inode_bits must contain all inodes ORed together */
1159 /* which is used to skip bits that are 0 everywhere, which is very common */
1160 {
1161 eio_ino_t endianness;
1162 int i, j;
1163
1164 /* we store the byte offset of byte n into byte n of "endianness" */
1165 for (i = 0; i < sizeof (eio_ino_t); ++i)
1166 ((unsigned char *)&endianness)[i] = i;
1167
1168 *bit++ = 0;
1169
1170 for (i = 0; i < sizeof (eio_ino_t); ++i)
1171 {
1172 /* shifting off the byte offsets out of "endianness" */
1173 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1174 endianness >>= 8;
1175
1176 for (j = 0; j < 8; ++j)
1177 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1178 *bit++ = offs + j;
1179 }
1180
1181 for (j = 0; j < 8; ++j)
1182 if (score_bits & (1 << j))
1183 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1184 }
1185
1186 /* now actually do the sorting (a variant of MSD radix sort) */
1187 {
1188 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1189 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1190 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1191 int stk_idx = 0;
1192
1193 base_stk [stk_idx] = dents;
1194 end_stk [stk_idx] = dents + size;
1195 bit_stk [stk_idx] = bit - 1;
1196
1197 do
1198 {
1199 base = base_stk [stk_idx];
1200 end = end_stk [stk_idx];
1201 bit = bit_stk [stk_idx];
1202
1203 for (;;)
1204 {
1205 unsigned char O = *bit >> 3;
1206 unsigned char M = 1 << (*bit & 7);
1207
1208 eio_dirent *a = base;
1209 eio_dirent *b = end;
1210
1211 if (b - a < EIO_SORT_CUTOFF)
1212 break;
1213
1214 /* now bit-partition the array on the bit */
1215 /* this ugly asymmetric loop seems to perform much better than typical */
1216 /* partition algos found in the literature */
1217 do
1218 if (!(((unsigned char *)a)[O] & M))
1219 ++a;
1220 else if (!(((unsigned char *)--b)[O] & M))
1221 {
1222 eio_dirent tmp = *a; *a = *b; *b = tmp;
1223 ++a;
1224 }
1225 while (b > a);
1226
1227 /* next bit, or stop, if no bits left in this path */
1228 if (!*--bit)
1229 break;
1230
1231 base_stk [stk_idx] = a;
1232 end_stk [stk_idx] = end;
1233 bit_stk [stk_idx] = bit;
1234 ++stk_idx;
1235
1236 end = a;
1237 }
1238 }
1239 while (stk_idx--);
1240 }
1241}
1242
1243static void
1244eio_dent_insertion_sort (eio_dirent *dents, int size)
1245{
1246 /* first move the smallest element to the front, to act as a sentinel */
1247 {
1248 int i;
1249 eio_dirent *min = dents;
1250
1251 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1252 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1253 if (EIO_DENT_CMP (dents [i], <, *min))
1254 min = &dents [i];
1255
1256 /* swap elements 0 and j (minimum) */
1257 {
1258 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1259 }
1260 }
1261
1262 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1263 {
1264 eio_dirent *i, *j;
1265
1266 for (i = dents + 1; i < dents + size; ++i)
1267 {
1268 eio_dirent value = *i;
1269
1270 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1271 j [1] = j [0];
1272
1273 j [1] = value;
1274 }
1275 }
1276}
1277
1278static void
1279eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1280{
1281 if (size <= 1)
1282 return; /* our insertion sort relies on size > 0 */
1283
1284 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1285 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1286 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1287
1288 /* use an insertion sort at the end, or for small arrays, */
1289 /* as insertion sort is more efficient for small partitions */
1290 eio_dent_insertion_sort (dents, size);
964} 1291}
965 1292
966/* read a full directory */ 1293/* read a full directory */
967static void 1294static void
968eio__scandir (eio_req *req, etp_worker *self) 1295eio__scandir (eio_req *req, etp_worker *self)
969{ 1296{
1297 char *name, *names;
1298 int namesalloc = 4096 - sizeof (void *) * 4;
1299 int namesoffs = 0;
1300 int flags = req->int1;
1301 eio_dirent *dents = 0;
1302 int dentalloc = 128;
1303 int dentoffs = 0;
1304 eio_ino_t inode_bits = 0;
1305#ifdef _WIN32
1306 HANDLE dirp;
1307 WIN32_FIND_DATA entp;
1308#else
970 DIR *dirp; 1309 DIR *dirp;
971 EIO_STRUCT_DIRENT *entp; 1310 EIO_STRUCT_DIRENT *entp;
972 char *name, *names; 1311#endif
973 int memlen = 4096;
974 int memofs = 0;
975 int res = 0;
976 1312
977 X_LOCK (wrklock); 1313 req->result = -1;
978 /* the corresponding closedir is in ETP_WORKER_CLEAR */
979 self->dirp = dirp = opendir (req->ptr1);
980 req->flags |= EIO_FLAG_PTR2_FREE;
981 req->ptr2 = names = malloc (memlen);
982 X_UNLOCK (wrklock);
983 1314
984 if (dirp && names) 1315 if (!(flags & EIO_READDIR_DENTS))
985 for (;;) 1316 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1317
1318#ifdef _WIN32
1319 {
1320 int len = strlen ((const char *)req->ptr1);
1321 char *path = malloc (MAX_PATH);
1322 const char *fmt;
1323 const char *reqpath = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1324
1325 if (!len)
1326 fmt = "./*";
1327 else if (reqpath[len - 1] == '/' || reqpath[len - 1] == '\\')
1328 fmt = "%s*";
1329 else
1330 fmt = "%s/*";
1331
1332 _snprintf (path, MAX_PATH, fmt, reqpath);
1333 dirp = FindFirstFile (path, &entp);
1334 free (path);
1335
1336 if (dirp == INVALID_HANDLE_VALUE)
1337 {
1338 /* should steal _dosmaperr */
1339 switch (GetLastError ())
1340 {
1341 case ERROR_FILE_NOT_FOUND:
1342 req->result = 0;
1343 break;
1344
1345 case ERROR_INVALID_NAME:
1346 case ERROR_PATH_NOT_FOUND:
1347 case ERROR_NO_MORE_FILES:
1348 errno = ENOENT;
1349 break;
1350
1351 case ERROR_NOT_ENOUGH_MEMORY:
1352 errno = ENOMEM;
1353 break;
1354
1355 default:
1356 errno = EINVAL;
1357 break;
1358 }
1359
1360 return;
1361 }
1362 }
1363#else
1364 #if HAVE_AT
1365 if (req->wd)
986 { 1366 {
1367 int fd = openat (WD2FD (req->wd), req->ptr1, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1368
1369 if (fd < 0)
1370 return;
1371
1372 dirp = fdopendir (fd);
1373
1374 if (!dirp)
1375 close (fd);
1376 }
1377 else
1378 dirp = opendir (req->ptr1);
1379 #else
1380 dirp = opendir (wd_expand (&self->tmpbuf, req->wd, req->ptr1));
1381 #endif
1382
1383 if (!dirp)
1384 return;
1385#endif
1386
1387 if (req->flags & EIO_FLAG_PTR1_FREE)
1388 free (req->ptr1);
1389
1390 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1391 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1392 req->ptr2 = names = malloc (namesalloc);
1393
1394 if (!names || (flags && !dents))
1395 return;
1396
1397 for (;;)
1398 {
1399 int done;
1400
1401#ifdef _WIN32
1402 done = !dirp;
1403#else
987 errno = 0; 1404 errno = 0;
988 entp = readdir (dirp); 1405 entp = readdir (dirp);
1406 done = !entp;
1407#endif
989 1408
1409 if (done)
1410 {
1411#ifndef _WIN32
1412 int old_errno = errno;
1413 closedir (dirp);
1414 errno = old_errno;
1415
990 if (!entp) 1416 if (errno)
1417 break;
1418#endif
1419
1420 /* sort etc. */
1421 req->int1 = flags;
1422 req->result = dentoffs;
1423
1424 if (flags & EIO_READDIR_STAT_ORDER)
1425 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1426 else if (flags & EIO_READDIR_DIRS_FIRST)
1427 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1428 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1429 else
1430 {
1431 /* in this case, all is known, and we just put dirs first and sort them */
1432 eio_dirent *oth = dents + dentoffs;
1433 eio_dirent *dir = dents;
1434
1435 /* now partition dirs to the front, and non-dirs to the back */
1436 /* by walking from both sides and swapping if necessary */
1437 while (oth > dir)
1438 {
1439 if (dir->type == EIO_DT_DIR)
1440 ++dir;
1441 else if ((--oth)->type == EIO_DT_DIR)
1442 {
1443 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1444
1445 ++dir;
1446 }
1447 }
1448
1449 /* now sort the dirs only (dirs all have the same score) */
1450 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1451 }
1452
991 break; 1453 break;
1454 }
992 1455
993 name = entp->d_name; 1456 /* now add the entry to our list(s) */
1457 name = D_NAME (entp);
994 1458
1459 /* skip . and .. entries */
995 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1460 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
996 { 1461 {
997 int len = strlen (name) + 1; 1462 int len = D_NAMLEN (entp) + 1;
998 1463
999 res++; 1464 while (ecb_expect_false (namesoffs + len > namesalloc))
1000
1001 while (memofs + len > memlen)
1002 { 1465 {
1003 memlen *= 2; 1466 namesalloc *= 2;
1004 X_LOCK (wrklock);
1005 req->ptr2 = names = realloc (names, memlen); 1467 req->ptr2 = names = realloc (names, namesalloc);
1006 X_UNLOCK (wrklock);
1007 1468
1008 if (!names) 1469 if (!names)
1470 break;
1471 }
1472
1473 memcpy (names + namesoffs, name, len);
1474
1475 if (dents)
1476 {
1477 struct eio_dirent *ent;
1478
1479 if (ecb_expect_false (dentoffs == dentalloc))
1480 {
1481 dentalloc *= 2;
1482 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1483
1484 if (!dents)
1009 break; 1485 break;
1010 } 1486 }
1011 1487
1012 memcpy (names + memofs, name, len); 1488 ent = dents + dentoffs;
1013 memofs += len; 1489
1490 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1491 ent->namelen = len - 1;
1492 ent->inode = D_INO (entp);
1493
1494 inode_bits |= ent->inode;
1495
1496 switch (D_TYPE (entp))
1497 {
1498 default:
1499 ent->type = EIO_DT_UNKNOWN;
1500 flags |= EIO_READDIR_FOUND_UNKNOWN;
1501 break;
1502
1503 #ifdef DT_FIFO
1504 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1505 #endif
1506 #ifdef DT_CHR
1507 case DT_CHR: ent->type = EIO_DT_CHR; break;
1508 #endif
1509 #ifdef DT_MPC
1510 case DT_MPC: ent->type = EIO_DT_MPC; break;
1511 #endif
1512 #ifdef DT_DIR
1513 case DT_DIR: ent->type = EIO_DT_DIR; break;
1514 #endif
1515 #ifdef DT_NAM
1516 case DT_NAM: ent->type = EIO_DT_NAM; break;
1517 #endif
1518 #ifdef DT_BLK
1519 case DT_BLK: ent->type = EIO_DT_BLK; break;
1520 #endif
1521 #ifdef DT_MPB
1522 case DT_MPB: ent->type = EIO_DT_MPB; break;
1523 #endif
1524 #ifdef DT_REG
1525 case DT_REG: ent->type = EIO_DT_REG; break;
1526 #endif
1527 #ifdef DT_NWK
1528 case DT_NWK: ent->type = EIO_DT_NWK; break;
1529 #endif
1530 #ifdef DT_CMP
1531 case DT_CMP: ent->type = EIO_DT_CMP; break;
1532 #endif
1533 #ifdef DT_LNK
1534 case DT_LNK: ent->type = EIO_DT_LNK; break;
1535 #endif
1536 #ifdef DT_SOCK
1537 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1538 #endif
1539 #ifdef DT_DOOR
1540 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1541 #endif
1542 #ifdef DT_WHT
1543 case DT_WHT: ent->type = EIO_DT_WHT; break;
1544 #endif
1545 }
1546
1547 ent->score = 7;
1548
1549 if (flags & EIO_READDIR_DIRS_FIRST)
1550 {
1551 if (ent->type == EIO_DT_UNKNOWN)
1552 {
1553 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1554 ent->score = 1;
1555 else if (!strchr (name, '.')) /* absence of dots indicate likely dirs */
1556 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1557 }
1558 else if (ent->type == EIO_DT_DIR)
1559 ent->score = 0;
1560 }
1014 } 1561 }
1562
1563 namesoffs += len;
1564 ++dentoffs;
1015 } 1565 }
1016 1566
1017 if (errno) 1567 if (EIO_CANCELLED (req))
1018 res = -1; 1568 {
1569 errno = ECANCELED;
1570 break;
1571 }
1572
1573#ifdef _WIN32
1574 if (!FindNextFile (dirp, &entp))
1575 {
1576 FindClose (dirp);
1577 dirp = 0;
1578 }
1579#endif
1580 }
1581}
1582
1583/*****************************************************************************/
1584/* working directory stuff */
1585/* various deficiencies in the posix 2008 api force us to */
1586/* keep the absolute path in string form at all times */
1587/* fuck yeah. */
1588
1589#if !HAVE_AT
1590
1591/* a bit like realpath, but usually faster because it doesn'T have to return */
1592/* an absolute or canonical path */
1593static const char *
1594wd_expand (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
1595{
1596 if (!wd || *path == '/')
1597 return path;
1598
1599 if (path [0] == '.' && !path [1])
1600 return wd->str;
1601
1019 1602 {
1020 req->result = res; 1603 int l1 = wd->len;
1021} 1604 int l2 = strlen (path);
1022 1605
1023#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) 1606 char *res = etp_tmpbuf_get (tmpbuf, l1 + l2 + 2);
1024# undef msync
1025# define msync(a,b,c) ((errno = ENOSYS), -1)
1026#endif
1027 1607
1028int 1608 memcpy (res, wd->str, l1);
1029eio__mtouch (void *mem, size_t len, int flags) 1609 res [l1] = '/';
1030{ 1610 memcpy (res + l1 + 1, path, l2 + 1);
1031 intptr_t addr = (intptr_t)mem;
1032 intptr_t end = addr + len;
1033#ifdef PAGESIZE
1034 const intptr_t page = PAGESIZE;
1035#else
1036 static intptr_t page;
1037 1611
1038 if (!page) 1612 return res;
1039 page = sysconf (_SC_PAGESIZE); 1613 }
1040#endif 1614}
1041 1615
1042 addr &= ~(page - 1); /* assume page size is always a power of two */ 1616#endif
1043 1617
1044 if (addr < end) 1618static eio_wd
1045 if (flags) /* modify */ 1619eio__wd_open_sync (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
1046 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len); 1620{
1047 else 1621 int fd;
1048 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len); 1622 eio_wd res;
1623 int len = eio__realpath (tmpbuf, wd, path);
1049 1624
1625 if (len < 0)
1626 return EIO_INVALID_WD;
1627
1628#if HAVE_AT
1629 fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1630
1631 if (fd < 0)
1632 return EIO_INVALID_WD;
1633#endif
1634
1635 res = malloc (sizeof (*res) + len); /* one extra 0-byte */
1636
1637#if HAVE_AT
1638 res->fd = fd;
1639#endif
1640
1641 res->len = len;
1642 memcpy (res->str, tmpbuf->ptr, len);
1643 res->str [len] = 0;
1644
1645 return res;
1646}
1647
1648eio_wd
1649eio_wd_open_sync (eio_wd wd, const char *path)
1650{
1651 struct etp_tmpbuf tmpbuf = { };
1652 wd = eio__wd_open_sync (&tmpbuf, wd, path);
1653 free (tmpbuf.ptr);
1654
1050 return 0; 1655 return wd;
1051} 1656}
1657
1658void
1659eio_wd_close_sync (eio_wd wd)
1660{
1661 if (wd != EIO_INVALID_WD && wd != EIO_CWD)
1662 {
1663 #if HAVE_AT
1664 close (wd->fd);
1665 #endif
1666 free (wd);
1667 }
1668}
1669
1670#if HAVE_AT
1671
1672/* they forgot these */
1673
1674static int
1675eio__truncateat (int dirfd, const char *path, off_t length)
1676{
1677 int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
1678 int res;
1679
1680 if (fd < 0)
1681 return fd;
1682
1683 res = ftruncate (fd, length);
1684 close (fd);
1685 return res;
1686}
1687
1688static int
1689eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
1690{
1691 int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
1692 int res;
1693
1694 if (fd < 0)
1695 return fd;
1696
1697 res = fstatvfs (fd, buf);
1698 close (fd);
1699 return res;
1700
1701}
1702
1703#endif
1052 1704
1053/*****************************************************************************/ 1705/*****************************************************************************/
1054 1706
1055#define ALLOC(len) \ 1707#define ALLOC(len) \
1056 if (!req->ptr2) \ 1708 if (!req->ptr2) \
1057 { \ 1709 { \
1058 X_LOCK (wrklock); \ 1710 X_LOCK (EIO_POOL->wrklock); \
1059 req->flags |= EIO_FLAG_PTR2_FREE; \ 1711 req->flags |= EIO_FLAG_PTR2_FREE; \
1060 X_UNLOCK (wrklock); \ 1712 X_UNLOCK (EIO_POOL->wrklock); \
1061 req->ptr2 = malloc (len); \ 1713 req->ptr2 = malloc (len); \
1062 if (!req->ptr2) \ 1714 if (!req->ptr2) \
1063 { \ 1715 { \
1064 errno = ENOMEM; \ 1716 errno = ENOMEM; \
1065 req->result = -1; \ 1717 req->result = -1; \
1066 break; \ 1718 break; \
1067 } \ 1719 } \
1068 } 1720 }
1069 1721
1070X_THREAD_PROC (etp_proc)
1071{
1072 ETP_REQ *req;
1073 struct timespec ts;
1074 etp_worker *self = (etp_worker *)thr_arg;
1075
1076 /* try to distribute timeouts somewhat randomly */
1077 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1078
1079 for (;;)
1080 {
1081 X_LOCK (reqlock);
1082
1083 for (;;)
1084 {
1085 self->req = req = reqq_shift (&req_queue);
1086
1087 if (req)
1088 break;
1089
1090 ++idle;
1091
1092 ts.tv_sec = time (0) + IDLE_TIMEOUT;
1093 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
1094 {
1095 if (idle > max_idle)
1096 {
1097 --idle;
1098 X_UNLOCK (reqlock);
1099 X_LOCK (wrklock);
1100 --started;
1101 X_UNLOCK (wrklock);
1102 goto quit;
1103 }
1104
1105 /* we are allowed to idle, so do so without any timeout */
1106 X_COND_WAIT (reqwait, reqlock);
1107 }
1108
1109 --idle;
1110 }
1111
1112 --nready;
1113
1114 X_UNLOCK (reqlock);
1115
1116 if (req->type < 0)
1117 goto quit;
1118
1119 if (!EIO_CANCELLED (req))
1120 ETP_EXECUTE (self, req);
1121
1122 X_LOCK (reslock);
1123
1124 ++npending;
1125
1126 if (!reqq_push (&res_queue, req) && want_poll_cb)
1127 want_poll_cb ();
1128
1129 self->req = 0;
1130 etp_worker_clear (self);
1131
1132 X_UNLOCK (reslock);
1133 }
1134
1135quit:
1136 X_LOCK (wrklock);
1137 etp_worker_free (self);
1138 X_UNLOCK (wrklock);
1139
1140 return 0;
1141}
1142
1143/*****************************************************************************/ 1722/*****************************************************************************/
1144 1723
1724int ecb_cold
1145int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1725eio_init (void (*want_poll)(void), void (*done_poll)(void))
1146{ 1726{
1147 return etp_init (want_poll, done_poll); 1727 eio_want_poll_cb = want_poll;
1148} 1728 eio_done_poll_cb = done_poll;
1149 1729
1730 return etp_init (EIO_POOL, 0, 0, 0);
1731}
1732
1733ecb_inline void
1150static void eio_api_destroy (eio_req *req) 1734eio_api_destroy (eio_req *req)
1151{ 1735{
1152 free (req); 1736 free (req);
1153} 1737}
1154 1738
1155#define REQ(rtype) \ 1739#define REQ(rtype) \
1156 eio_req *req; \ 1740 eio_req *req; \
1157 \ 1741 \
1158 req = (eio_req *)calloc (1, sizeof *req); \ 1742 req = (eio_req *)calloc (1, sizeof *req); \
1159 if (!req) \ 1743 if (!req) \
1160 return 0; \ 1744 return 0; \
… …
1174 { \ 1758 { \
1175 eio_api_destroy (req); \ 1759 eio_api_destroy (req); \
1176 return 0; \ 1760 return 0; \
1177 } 1761 }
1178 1762
1763#define SINGLEDOT(ptr) (0[(char *)(ptr)] == '.' && !1[(char *)(ptr)])
1764
1765static void
1179static void eio_execute (etp_worker *self, eio_req *req) 1766eio_execute (etp_worker *self, eio_req *req)
1180{ 1767{
1181 errno = 0; 1768#if HAVE_AT
1769 int dirfd;
1770#else
1771 const char *path;
1772#endif
1773
1774 if (ecb_expect_false (EIO_CANCELLED (req)))
1775 {
1776 req->result = -1;
1777 req->errorno = ECANCELED;
1778 return;
1779 }
1780
1781 if (ecb_expect_false (req->wd == EIO_INVALID_WD))
1782 {
1783 req->result = -1;
1784 req->errorno = ENOENT;
1785 return;
1786 }
1787
1788 if (req->type >= EIO_OPEN)
1789 {
1790 #if HAVE_AT
1791 dirfd = WD2FD (req->wd);
1792 #else
1793 path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1794 #endif
1795 }
1182 1796
1183 switch (req->type) 1797 switch (req->type)
1184 { 1798 {
1799 case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
1800 req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
1801 break;
1802 case EIO_WD_CLOSE: req->result = 0;
1803 eio_wd_close_sync (req->wd); break;
1804
1805 case EIO_SEEK: eio__lseek (req); break;
1185 case EIO_READ: ALLOC (req->size); 1806 case EIO_READ: ALLOC (req->size);
1186 req->result = req->offs >= 0 1807 req->result = req->offs >= 0
1187 ? pread (req->int1, req->ptr2, req->size, req->offs) 1808 ? pread (req->int1, req->ptr2, req->size, req->offs)
1188 : read (req->int1, req->ptr2, req->size); break; 1809 : read (req->int1, req->ptr2, req->size); break;
1189 case EIO_WRITE: req->result = req->offs >= 0 1810 case EIO_WRITE: req->result = req->offs >= 0
1190 ? pwrite (req->int1, req->ptr2, req->size, req->offs) 1811 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1191 : write (req->int1, req->ptr2, req->size); break; 1812 : write (req->int1, req->ptr2, req->size); break;
1192 1813
1814 case EIO_FCNTL: req->result = fcntl (req->int1, (int) req->int2, req->ptr2); break;
1815 case EIO_IOCTL: req->result = ioctl (req->int1, (unsigned long)req->int2, req->ptr2); break;
1816
1193 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break; 1817 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1194 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break; 1818 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
1819
1820#if HAVE_AT
1195 1821
1196 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1822 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1197 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 1823 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break;
1198 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1824 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1199 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 1825 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, AT_SYMLINK_NOFOLLOW); break;
1200 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1201 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1202
1203 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; 1826 case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break;
1204 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1205 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break; 1827 case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break;
1206 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1207 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break; 1828 case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break;
1208 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1209
1210 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break; 1829 case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break;
1211 case EIO_CLOSE: req->result = close (req->int1); break; 1830
1212 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1213 case EIO_UNLINK: req->result = unlink (req->ptr1); break; 1831 case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
1214 case EIO_RMDIR: req->result = rmdir (req->ptr1); break; 1832 case EIO_RMDIR: /* complications arise because "." cannot be removed, so we might have to expand */
1833 req->result = req->wd && SINGLEDOT (req->ptr1)
1834 ? rmdir (req->wd->str)
1835 : unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break;
1215 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; 1836 case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); break;
1216 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; 1837 case EIO_RENAME: /* complications arise because "." cannot be renamed, so we might have to expand */
1838 req->result = req->wd && SINGLEDOT (req->ptr1)
1839 ? rename (req->wd->str, req->ptr2)
1840 : renameat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2); break;
1217 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; 1841 case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2, 0); break;
1218 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; 1842 case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
1219 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break; 1843 case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1220
1221 case EIO_READLINK: ALLOC (NAME_MAX); 1844 case EIO_READLINK: ALLOC (PATH_MAX);
1222 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; 1845 req->result = readlinkat (dirfd, req->ptr1, req->ptr2, PATH_MAX); break;
1223 1846 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1224 case EIO_SYNC: req->result = 0; sync (); break; 1847 req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1225 case EIO_FSYNC: req->result = fsync (req->int1); break;
1226 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1227 case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break;
1228 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;
1229 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1230
1231 case EIO_READDIR: eio__scandir (req, self); break;
1232
1233 case EIO_BUSY: 1848 case EIO_UTIME:
1234#ifdef _WIN32 1849 case EIO_FUTIME:
1235 Sleep (req->nv1 * 1000.); 1850 {
1851 struct timespec ts[2];
1852 struct timespec *times;
1853
1854 if (req->nv1 != -1. || req->nv2 != -1.)
1855 {
1856 ts[0].tv_sec = req->nv1;
1857 ts[0].tv_nsec = (req->nv1 - ts[0].tv_sec) * 1e9;
1858 ts[1].tv_sec = req->nv2;
1859 ts[1].tv_nsec = (req->nv2 - ts[1].tv_sec) * 1e9;
1860
1861 times = ts;
1862 }
1863 else
1864 times = 0;
1865
1866 req->result = req->type == EIO_FUTIME
1867 ? futimens (req->int1, times)
1868 : utimensat (dirfd, req->ptr1, times, 0);
1869 }
1870 break;
1871
1236#else 1872#else
1237 {
1238 struct timeval tv;
1239 1873
1240 tv.tv_sec = req->nv1; 1874 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1241 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; 1875 req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
1876 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1877 req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
1878 case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
1879 case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
1880 case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
1881 case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
1242 1882
1243 req->result = select (0, 0, 0, 0, &tv); 1883 case EIO_UNLINK: req->result = unlink (path ); break;
1244 } 1884 case EIO_RMDIR: req->result = rmdir (path ); break;
1245#endif 1885 case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
1246 break; 1886 case EIO_RENAME: req->result = rename (path , req->ptr2); break;
1887 case EIO_LINK: req->result = link (path , req->ptr2); break;
1888 case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
1889 case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
1890 case EIO_READLINK: ALLOC (PATH_MAX);
1891 req->result = readlink (path, req->ptr2, PATH_MAX); break;
1892 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1893 req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
1247 1894
1248 case EIO_UTIME: 1895 case EIO_UTIME:
1249 case EIO_FUTIME: 1896 case EIO_FUTIME:
1250 { 1897 {
1251 struct timeval tv[2]; 1898 struct timeval tv[2];
1252 struct timeval *times; 1899 struct timeval *times;
1253 1900
1254 if (req->nv1 != -1. || req->nv2 != -1.) 1901 if (req->nv1 != -1. || req->nv2 != -1.)
1255 { 1902 {
1256 tv[0].tv_sec = req->nv1; 1903 tv[0].tv_sec = req->nv1;
1257 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.; 1904 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
1258 tv[1].tv_sec = req->nv2; 1905 tv[1].tv_sec = req->nv2;
1259 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.; 1906 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
1260 1907
1261 times = tv; 1908 times = tv;
1262 } 1909 }
1263 else 1910 else
1264 times = 0; 1911 times = 0;
1265
1266 1912
1267 req->result = req->type == EIO_FUTIME 1913 req->result = req->type == EIO_FUTIME
1268 ? futimes (req->int1, times) 1914 ? futimes (req->int1, times)
1269 : utimes (req->ptr1, times); 1915 : utimes (req->ptr1, times);
1270 } 1916 }
1271 break; 1917 break;
1272 1918
1919#endif
1920
1921 case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
1922 {
1923 ALLOC (req->result);
1924 memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
1925 }
1926 break;
1927
1928 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1929 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1930
1931 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1932 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1933
1934 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1935 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1936 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1937
1938 case EIO_CLOSE: req->result = close (req->int1); break;
1939 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1940 case EIO_SYNC: req->result = 0; sync (); break;
1941 case EIO_FSYNC: req->result = fsync (req->int1); break;
1942 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1943 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
1944 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1945 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1946 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1947 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1948 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1949 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
1950
1951 case EIO_READDIR: eio__scandir (req, self); break;
1952
1953 case EIO_BUSY:
1954#ifdef _WIN32
1955 Sleep (req->nv1 * 1e3);
1956#else
1957 {
1958 struct timeval tv;
1959
1960 tv.tv_sec = req->nv1;
1961 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1962
1963 req->result = select (0, 0, 0, 0, &tv);
1964 }
1965#endif
1966 break;
1967
1968#if 0
1273 case EIO_GROUP: 1969 case EIO_GROUP:
1274 abort (); /* handled in eio_request */ 1970 abort (); /* handled in eio_request */
1971#endif
1275 1972
1276 case EIO_NOP: 1973 case EIO_NOP:
1277 req->result = 0; 1974 req->result = 0;
1278 break; 1975 break;
1279 1976
1280 case EIO_CUSTOM: 1977 case EIO_CUSTOM:
1281 ((void (*)(eio_req *))req->feed) (req); 1978 req->feed (req);
1282 break; 1979 break;
1283 1980
1284 default: 1981 default:
1285 req->result = -1; 1982 req->result = EIO_ENOSYS ();
1286 break; 1983 break;
1287 } 1984 }
1288 1985
1289 req->errorno = errno; 1986 req->errorno = errno;
1290} 1987}
1291 1988
1292#ifndef EIO_NO_WRAPPERS 1989#ifndef EIO_NO_WRAPPERS
1293 1990
1991eio_req *eio_wd_open (const char *path, int pri, eio_cb cb, void *data)
1992{
1993 REQ (EIO_WD_OPEN); PATH; SEND;
1994}
1995
1996eio_req *eio_wd_close (eio_wd wd, int pri, eio_cb cb, void *data)
1997{
1998 REQ (EIO_WD_CLOSE); req->wd = wd; SEND;
1999}
2000
1294eio_req *eio_nop (int pri, eio_cb cb, void *data) 2001eio_req *eio_nop (int pri, eio_cb cb, void *data)
1295{ 2002{
1296 REQ (EIO_NOP); SEND; 2003 REQ (EIO_NOP); SEND;
1297} 2004}
1298 2005
… …
1314eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2021eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1315{ 2022{
1316 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2023 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1317} 2024}
1318 2025
2026eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2027{
2028 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2029}
2030
2031eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
2032{
2033 REQ (EIO_SYNCFS); req->int1 = fd; SEND;
2034}
2035
2036eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2037{
2038 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2039}
2040
1319eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2041eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1320{ 2042{
1321 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2043 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1322} 2044}
1323 2045
1324eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) 2046eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
1325{ 2047{
1326 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; 2048 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
1327} 2049}
1328 2050
1329eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) 2051eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
1330{ 2052{
1331 REQ (EIO_FDATASYNC); req->int1 = fd; SEND; 2053 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2054}
2055
2056eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2057{
2058 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
1332} 2059}
1333 2060
1334eio_req *eio_close (int fd, int pri, eio_cb cb, void *data) 2061eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
1335{ 2062{
1336 REQ (EIO_CLOSE); req->int1 = fd; SEND; 2063 REQ (EIO_CLOSE); req->int1 = fd; SEND;
… …
1339eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data) 2066eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
1340{ 2067{
1341 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND; 2068 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
1342} 2069}
1343 2070
2071eio_req *eio_seek (int fd, off_t offset, int whence, int pri, eio_cb cb, void *data)
2072{
2073 REQ (EIO_SEEK); req->int1 = fd; req->offs = offset; req->int2 = whence; SEND;
2074}
2075
1344eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data) 2076eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1345{ 2077{
1346 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND; 2078 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1347} 2079}
1348 2080
1349eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data) 2081eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1350{ 2082{
1351 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND; 2083 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1352} 2084}
1353 2085
2086eio_req *eio_fcntl (int fd, int cmd, void *arg, int pri, eio_cb cb, void *data)
2087{
2088 REQ (EIO_IOCTL); req->int1 = fd; req->int2 = cmd; req->ptr2 = arg; SEND;
2089}
2090
2091eio_req *eio_ioctl (int fd, unsigned long request, void *buf, int pri, eio_cb cb, void *data)
2092{
2093 REQ (EIO_IOCTL); req->int1 = fd; req->int2 = request; req->ptr2 = buf; SEND;
2094}
2095
1354eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) 2096eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1355{ 2097{
1356 REQ (EIO_FSTAT); req->int1 = fd; SEND; 2098 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1357} 2099}
1358 2100
2101eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
2102{
2103 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
2104}
2105
1359eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) 2106eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1360{ 2107{
1361 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; 2108 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1362} 2109}
1363 2110
… …
1369eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) 2116eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1370{ 2117{
1371 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2118 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1372} 2119}
1373 2120
1374eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2121eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1375{ 2122{
1376 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2123 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1377} 2124}
1378 2125
1379eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) 2126eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
… …
1399eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data) 2146eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1400{ 2147{
1401 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2148 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1402} 2149}
1403 2150
1404eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2151eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1405{ 2152{
1406 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2153 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1407} 2154}
1408 2155
1409eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) 2156eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
… …
1425eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data) 2172eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1426{ 2173{
1427 return eio__1path (EIO_READLINK, path, pri, cb, data); 2174 return eio__1path (EIO_READLINK, path, pri, cb, data);
1428} 2175}
1429 2176
2177eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2178{
2179 return eio__1path (EIO_REALPATH, path, pri, cb, data);
2180}
2181
1430eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) 2182eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1431{ 2183{
1432 return eio__1path (EIO_STAT, path, pri, cb, data); 2184 return eio__1path (EIO_STAT, path, pri, cb, data);
1433} 2185}
1434 2186
1435eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) 2187eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1436{ 2188{
1437 return eio__1path (EIO_LSTAT, path, pri, cb, data); 2189 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1438} 2190}
1439 2191
2192eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
2193{
2194 return eio__1path (EIO_STATVFS, path, pri, cb, data);
2195}
2196
1440eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) 2197eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1441{ 2198{
1442 return eio__1path (EIO_UNLINK, path, pri, cb, data); 2199 return eio__1path (EIO_UNLINK, path, pri, cb, data);
1443} 2200}
1444 2201
1445eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data) 2202eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1446{ 2203{
1447 return eio__1path (EIO_RMDIR, path, pri, cb, data); 2204 return eio__1path (EIO_RMDIR, path, pri, cb, data);
1448} 2205}
1449 2206
1450eio_req *eio_readdir (const char *path, int pri, eio_cb cb, void *data) 2207eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1451{ 2208{
1452 return eio__1path (EIO_READDIR, path, pri, cb, data); 2209 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1453} 2210}
1454 2211
1455eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) 2212eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1456{ 2213{
1457 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; 2214 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1458} 2215}
1459 2216
1460static eio_req * 2217static eio_req *
1461eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2218eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1462{ 2219{
… …
1486eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2243eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1487{ 2244{
1488 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2245 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1489} 2246}
1490 2247
1491eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) 2248eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
1492{ 2249{
1493 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND; 2250 REQ (EIO_CUSTOM); req->feed = execute; SEND;
1494} 2251}
1495 2252
1496#endif 2253#endif
1497 2254
1498eio_req *eio_grp (eio_cb cb, void *data) 2255eio_req *eio_grp (eio_cb cb, void *data)
… …
1507#undef SEND 2264#undef SEND
1508 2265
1509/*****************************************************************************/ 2266/*****************************************************************************/
1510/* grp functions */ 2267/* grp functions */
1511 2268
2269void
1512void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2270eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1513{ 2271{
1514 grp->int2 = limit; 2272 grp->int2 = limit;
1515 grp->feed = feed; 2273 grp->feed = feed;
1516 2274
1517 grp_try_feed (grp); 2275 grp_try_feed (grp);
1518} 2276}
1519 2277
2278void
1520void eio_grp_limit (eio_req *grp, int limit) 2279eio_grp_limit (eio_req *grp, int limit)
1521{ 2280{
1522 grp->int2 = limit; 2281 grp->int2 = limit;
1523 2282
1524 grp_try_feed (grp); 2283 grp_try_feed (grp);
1525} 2284}
1526 2285
2286void
1527void eio_grp_add (eio_req *grp, eio_req *req) 2287eio_grp_add (eio_req *grp, eio_req *req)
1528{ 2288{
1529 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2289 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1530 2290
1531 grp->flags |= EIO_FLAG_GROUPADD; 2291 grp->flags |= ETP_FLAG_GROUPADD;
1532 2292
1533 ++grp->size; 2293 ++grp->size;
1534 req->grp = grp; 2294 req->grp = grp;
1535 2295
1536 req->grp_prev = 0; 2296 req->grp_prev = 0;
… …
1543} 2303}
1544 2304
1545/*****************************************************************************/ 2305/*****************************************************************************/
1546/* misc garbage */ 2306/* misc garbage */
1547 2307
2308eio_ssize_t
1548ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2309eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1549{ 2310{
1550 etp_worker wrk;
1551
1552 wrk.dbuf = 0;
1553
1554 eio__sendfile (ofd, ifd, offset, count, &wrk); 2311 return eio__sendfile (ofd, ifd, offset, count);
1555
1556 if (wrk.dbuf)
1557 free (wrk.dbuf);
1558} 2312}
1559 2313

Diff Legend

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