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 1.145 by root, Fri Jun 23 03:10:19 2017 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 int flags;
207
208 /* This tries out all combinations of SYMBOLIC_LINK_FLAG_DIRECTORY
209 * and SYMBOLIC_LINK_FLAG_ALLOW_UNPRIVILEGED_CREATE,
210 * with directory first.
211 */
212 for (flags = 3; flags >= 0; --flags)
213 if (CreateSymbolicLink (neu, old, flags))
214 return 0;
215 #endif
216
217 return EIO_ERRNO (ENOENT, -1);
218 }
219
220 /* POSIX API only, causing trouble for win32 apps */
221 #define CreateHardLink(neu,old,flags) 0 /* not really creating hardlink, still using relative paths? */
222 #define CreateSymbolicLink(neu,old,flags) 0 /* vista+ only */
223
224 struct statvfs
225 {
226 int dummy;
227 };
228
229 #define DT_DIR EIO_DT_DIR
230 #define DT_REG EIO_DT_REG
231 #define D_NAME(entp) entp.cFileName
232 #define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
73 233
74#else 234#else
75 235
76# include "config.h" 236 #include <sys/ioctl.h>
77# include <sys/time.h> 237 #include <sys/time.h>
78# include <sys/select.h> 238 #include <sys/select.h>
79# include <unistd.h> 239 #include <unistd.h>
240 #include <signal.h>
241 #include <dirent.h>
242
243 #ifdef ANDROID
244 #include <sys/vfs.h>
245 #define statvfs statfs
246 #define fstatvfs fstatfs
247 #include <asm/page.h> /* supposedly limits.h does #define PAGESIZE PAGESIZE */
248 #else
249 #include <sys/statvfs.h>
250 #endif
251
252 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
253 #include <sys/mman.h>
254 #endif
255
256 #define D_NAME(entp) entp->d_name
257
258 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
259 #if __FreeBSD__ || __NetBSD__ || __OpenBSD__
260 #define _DIRENT_HAVE_D_TYPE /* sigh */
261 #define D_INO(de) (de)->d_fileno
262 #define D_NAMLEN(de) (de)->d_namlen
263 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
264 #define D_INO(de) (de)->d_ino
265 #endif
266
267 #ifdef _D_EXACT_NAMLEN
268 #undef D_NAMLEN
269 #define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
270 #endif
271
272 #ifdef _DIRENT_HAVE_D_TYPE
273 #define D_TYPE(de) (de)->d_type
274 #endif
275
276 #ifndef EIO_STRUCT_DIRENT
277 #define EIO_STRUCT_DIRENT struct dirent
278 #endif
279
280#endif
281
282#if HAVE_UTIMES
80# include <utime.h> 283# include <utime.h>
81# include <signal.h>
82# include <dirent.h>
83
84# ifndef EIO_STRUCT_DIRENT
85# define EIO_STRUCT_DIRENT struct dirent
86# endif 284#endif
87 285
286#if HAVE_SYS_SYSCALL_H
287# include <sys/syscall.h>
88#endif 288#endif
89 289
90#if HAVE_SENDFILE 290#if HAVE_SENDFILE
91# if __linux 291# if __linux
92# include <sys/sendfile.h> 292# include <sys/sendfile.h>
93# elif __freebsd 293# elif __FreeBSD__ || defined __APPLE__
94# include <sys/socket.h> 294# include <sys/socket.h>
95# include <sys/uio.h> 295# include <sys/uio.h>
96# elif __hpux 296# elif __hpux
97# include <sys/socket.h> 297# include <sys/socket.h>
98# elif __solaris /* not yet */ 298# elif __solaris
99# include <sys/sendfile.h> 299# include <sys/sendfile.h>
100# else 300# else
101# error sendfile support requested but not available 301# error sendfile support requested but not available
102# endif 302# endif
103#endif 303#endif
104 304
105/* number of seconds after which an idle threads exit */ 305#if HAVE_RENAMEAT2
106#define IDLE_TIMEOUT 10 306# include <sys/syscall.h>
307# include <linux/fs.h>
308#endif
309
310#ifndef D_TYPE
311# define D_TYPE(de) 0
312#endif
313#ifndef D_INO
314# define D_INO(de) 0
315#endif
316#ifndef D_NAMLEN
317# define D_NAMLEN(entp) strlen (D_NAME (entp))
318#endif
107 319
108/* used for struct dirent, AIX doesn't provide it */ 320/* used for struct dirent, AIX doesn't provide it */
109#ifndef NAME_MAX 321#ifndef NAME_MAX
110# define NAME_MAX 4096 322# define NAME_MAX 4096
111#endif 323#endif
112 324
325/* used for readlink etc. */
326#ifndef PATH_MAX
327# define PATH_MAX 0
328#endif
329
330#ifndef EIO_PATH_MIN
331# define EIO_PATH_MIN 8160
332#endif
333
334#define EIO_PATH_MAX (PATH_MAX <= EIO_PATH_MIN ? EIO_PATH_MIN : PATH_MAX)
335
113/* buffer size for various temporary buffers */ 336/* buffer size for various temporary buffers */
114#define EIO_BUFSIZE 65536 337#define EIO_BUFSIZE 65536
115 338
116#define dBUF \ 339#define dBUF \
117 char *eio_buf; \
118 ETP_WORKER_LOCK (self); \
119 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 340 char *eio_buf = malloc (EIO_BUFSIZE); \
120 ETP_WORKER_UNLOCK (self); \
121 errno = ENOMEM; \ 341 errno = ENOMEM; \
122 if (!eio_buf) \ 342 if (!eio_buf) \
123 return -1; 343 return -1
124 344
125#define EIO_TICKS ((1000000 + 1023) >> 10) 345#define FUBd \
346 free (eio_buf)
126 347
127/*****************************************************************************/ 348/*****************************************************************************/
128 349
129#if __GNUC__ >= 3 350struct etp_tmpbuf;
130# define expect(expr,value) __builtin_expect ((expr),(value)) 351
352#if _POSIX_VERSION >= 200809L
353 #define HAVE_AT 1
354 #define WD2FD(wd) ((wd) ? (wd)->fd : AT_FDCWD)
355 #ifndef O_SEARCH
356 #define O_SEARCH O_RDONLY
357 #endif
131#else 358#else
132# define expect(expr,value) (expr) 359 #define HAVE_AT 0
360 static const char *wd_expand (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path);
133#endif 361#endif
134 362
135#define expect_false(expr) expect ((expr) != 0, 0) 363struct eio_pwd
136#define expect_true(expr) expect ((expr) != 0, 1) 364{
365#if HAVE_AT
366 int fd;
367#endif
368 int len;
369 char str[1]; /* actually, a 0-terminated canonical path */
370};
137 371
138/*****************************************************************************/ 372/*****************************************************************************/
139 373
140#define ETP_PRI_MIN EIO_PRI_MIN 374#define ETP_PRI_MIN EIO_PRI_MIN
141#define ETP_PRI_MAX EIO_PRI_MAX 375#define ETP_PRI_MAX EIO_PRI_MAX
142 376
377#define ETP_TYPE_QUIT -1
378#define ETP_TYPE_GROUP EIO_GROUP
379
380static void eio_nop_callback (void) { }
381static void (*eio_want_poll_cb)(void) = eio_nop_callback;
382static void (*eio_done_poll_cb)(void) = eio_nop_callback;
383
384#define ETP_WANT_POLL(pool) eio_want_poll_cb ()
385#define ETP_DONE_POLL(pool) eio_done_poll_cb ()
386
143struct etp_worker; 387struct etp_worker;
144
145#define ETP_REQ eio_req 388#define ETP_REQ eio_req
146#define ETP_DESTROY(req) eio_destroy (req) 389#define ETP_DESTROY(req) eio_destroy (req)
147static int eio_finish (eio_req *req); 390static int eio_finish (eio_req *req);
148#define ETP_FINISH(req) eio_finish (req) 391#define ETP_FINISH(req) eio_finish (req)
149static void eio_execute (struct etp_worker *self, eio_req *req); 392static void eio_execute (struct etp_worker *self, eio_req *req);
150#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) 393#define ETP_EXECUTE(wrk,req) eio_execute (wrk, req)
151 394
152#define ETP_WORKER_CLEAR(req) \ 395#include "etp.c"
153 if (wrk->dbuf) \ 396
154 { \ 397static struct etp_pool eio_pool;
155 free (wrk->dbuf); \ 398#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;
167 399
168/*****************************************************************************/ 400/*****************************************************************************/
169 401
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 402static 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) 403grp_try_feed (eio_req *grp)
610{ 404{
611 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 405 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
612 { 406 {
613 grp->flags &= ~EIO_FLAG_GROUPADD; 407 grp->flags &= ~ETP_FLAG_GROUPADD;
614 408
615 EIO_FEED (grp); 409 EIO_FEED (grp);
616 410
617 /* stop if no progress has been made */ 411 /* stop if no progress has been made */
618 if (!(grp->flags & EIO_FLAG_GROUPADD)) 412 if (!(grp->flags & ETP_FLAG_GROUPADD))
619 { 413 {
620 grp->feed = 0; 414 grp->feed = 0;
621 break; 415 break;
622 } 416 }
623 } 417 }
624} 418}
625 419
420static int
626static int grp_dec (eio_req *grp) 421grp_dec (eio_req *grp)
627{ 422{
628 --grp->size; 423 --grp->size;
629 424
630 /* call feeder, if applicable */ 425 /* call feeder, if applicable */
631 grp_try_feed (grp); 426 grp_try_feed (grp);
632 427
633 /* finish, if done */ 428 /* finish, if done */
634 if (!grp->size && grp->int1) 429 if (!grp->size && grp->flags & ETP_FLAG_DELAYED)
635 return eio_finish (grp); 430 return eio_finish (grp);
636 else 431 else
637 return 0; 432 return 0;
638} 433}
639 434
435static void
640void eio_destroy (eio_req *req) 436eio_destroy (eio_req *req)
641{ 437{
642 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); 438 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
643 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); 439 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
644 440
645 EIO_DESTROY (req); 441 EIO_DESTROY (req);
646} 442}
647 443
444static int
648static int eio_finish (eio_req *req) 445eio_finish (eio_req *req)
649{ 446{
650 int res = EIO_FINISH (req); 447 int res = EIO_FINISH (req);
651 448
652 if (req->grp) 449 if (req->grp)
653 { 450 {
… …
661 if (grp->grp_first == req) 458 if (grp->grp_first == req)
662 grp->grp_first = req->grp_next; 459 grp->grp_first = req->grp_next;
663 460
664 res2 = grp_dec (grp); 461 res2 = grp_dec (grp);
665 462
666 if (!res && res2) 463 if (!res)
667 res = res2; 464 res = res2;
668 } 465 }
669 466
670 eio_destroy (req); 467 eio_destroy (req);
671 468
672 return res; 469 return res;
673} 470}
674 471
472void
675void eio_grp_cancel (eio_req *grp) 473eio_grp_cancel (eio_req *grp)
676{ 474{
677 for (grp = grp->grp_first; grp; grp = grp->grp_next) 475 etp_grp_cancel (EIO_POOL, grp);
678 eio_cancel (grp);
679} 476}
680 477
478void
681void eio_cancel (eio_req *req) 479eio_cancel (eio_req *req)
682{ 480{
683 etp_cancel (req); 481 etp_cancel (EIO_POOL, req);
684} 482}
685 483
484void
686void eio_submit (eio_req *req) 485eio_submit (eio_req *req)
687{ 486{
688 etp_submit (req); 487 etp_submit (EIO_POOL, req);
689} 488}
690 489
691unsigned int eio_nreqs (void) 490unsigned int
491eio_nreqs (void)
692{ 492{
693 return etp_nreqs (); 493 return etp_nreqs (EIO_POOL);
694} 494}
695 495
696unsigned int eio_nready (void) 496unsigned int
497eio_nready (void)
697{ 498{
698 return etp_nready (); 499 return etp_nready (EIO_POOL);
699} 500}
700 501
701unsigned int eio_npending (void) 502unsigned int
503eio_npending (void)
702{ 504{
703 return etp_npending (); 505 return etp_npending (EIO_POOL);
704} 506}
705 507
706unsigned int eio_nthreads (void) 508unsigned int ecb_cold
509eio_nthreads (void)
707{ 510{
708 return etp_nthreads (); 511 return etp_nthreads (EIO_POOL);
709} 512}
710 513
514void ecb_cold
711void eio_set_max_poll_time (double nseconds) 515eio_set_max_poll_time (double nseconds)
712{ 516{
713 etp_set_max_poll_time (nseconds); 517 etp_set_max_poll_time (EIO_POOL, nseconds);
714} 518}
715 519
520void ecb_cold
716void eio_set_max_poll_reqs (unsigned int maxreqs) 521eio_set_max_poll_reqs (unsigned int maxreqs)
717{ 522{
718 etp_set_max_poll_reqs (maxreqs); 523 etp_set_max_poll_reqs (EIO_POOL, maxreqs);
719} 524}
720 525
526void ecb_cold
721void eio_set_max_idle (unsigned int nthreads) 527eio_set_max_idle (unsigned int nthreads)
722{ 528{
723 etp_set_max_idle (nthreads); 529 etp_set_max_idle (EIO_POOL, nthreads);
724} 530}
725 531
532void ecb_cold
533eio_set_idle_timeout (unsigned int seconds)
534{
535 etp_set_idle_timeout (EIO_POOL, seconds);
536}
537
538void ecb_cold
726void eio_set_min_parallel (unsigned int nthreads) 539eio_set_min_parallel (unsigned int nthreads)
727{ 540{
728 etp_set_min_parallel (nthreads); 541 etp_set_min_parallel (EIO_POOL, nthreads);
729} 542}
730 543
544void ecb_cold
731void eio_set_max_parallel (unsigned int nthreads) 545eio_set_max_parallel (unsigned int nthreads)
732{ 546{
733 etp_set_max_parallel (nthreads); 547 etp_set_max_parallel (EIO_POOL, nthreads);
734} 548}
735 549
736int eio_poll (void) 550int eio_poll (void)
737{ 551{
738 return etp_poll (); 552 return etp_poll (EIO_POOL);
739} 553}
740 554
741/*****************************************************************************/ 555/*****************************************************************************/
742/* work around various missing functions */ 556/* work around various missing functions */
743 557
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 558#ifndef HAVE_UTIMES
784 559
785# undef utimes 560# undef utimes
786# undef futimes
787# define utimes(path,times) eio__utimes (path, times) 561# define utimes(path,times) eio__utimes (path, times)
788# define futimes(fd,times) eio__futimes (fd, times)
789 562
790static int 563static int
791eio__utimes (const char *filename, const struct timeval times[2]) 564eio__utimes (const char *filename, const struct timeval times[2])
792{ 565{
793 if (times) 566 if (times)
… …
801 } 574 }
802 else 575 else
803 return utime (filename, 0); 576 return utime (filename, 0);
804} 577}
805 578
579#endif
580
581#ifndef HAVE_FUTIMES
582
583# undef futimes
584# define futimes(fd,times) eio__futimes (fd, times)
585
586static int
806static int eio__futimes (int fd, const struct timeval tv[2]) 587eio__futimes (int fd, const struct timeval tv[2])
807{ 588{
808 errno = ENOSYS; 589 errno = ENOSYS;
809 return -1; 590 return -1;
810} 591}
811 592
… …
814#if !HAVE_FDATASYNC 595#if !HAVE_FDATASYNC
815# undef fdatasync 596# undef fdatasync
816# define fdatasync(fd) fsync (fd) 597# define fdatasync(fd) fsync (fd)
817#endif 598#endif
818 599
600static int
601eio__syncfs (int fd)
602{
603 int res;
604
605#if HAVE_SYS_SYNCFS
606 res = (int)syscall (__NR_syncfs, (int)(fd));
607#else
608 res = EIO_ENOSYS ();
609#endif
610
611 if (res < 0 && errno == ENOSYS && fd >= 0)
612 sync ();
613
614 return res;
615}
616
819/* sync_file_range always needs emulation */ 617/* sync_file_range always needs emulation */
820int 618static int
821eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags) 619eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
822{ 620{
823#if HAVE_SYNC_FILE_RANGE 621#if HAVE_SYNC_FILE_RANGE
824 int res; 622 int res;
825 623
… …
838 if (!res || errno != ENOSYS) 636 if (!res || errno != ENOSYS)
839 return res; 637 return res;
840#endif 638#endif
841 639
842 /* even though we could play tricks with the flags, it's better to always 640 /* 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 */ 641 * call fdatasync, as that matches the expectation of its users best */
844 return fdatasync (fd); 642 return fdatasync (fd);
643}
644
645static int
646eio__fallocate (int fd, int mode, off_t offset, size_t len)
647{
648#if HAVE_LINUX_FALLOCATE
649 return fallocate (fd, mode, offset, len);
650#else
651 return EIO_ENOSYS ();
652#endif
845} 653}
846 654
847#if !HAVE_READAHEAD 655#if !HAVE_READAHEAD
848# undef readahead 656# undef readahead
849# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 657# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
850 658
851static ssize_t 659static eio_ssize_t
852eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) 660eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
853{ 661{
854 size_t todo = count; 662 size_t todo = count;
855 dBUF; 663 dBUF;
856 664
… …
861 pread (fd, eio_buf, len, offset); 669 pread (fd, eio_buf, len, offset);
862 offset += len; 670 offset += len;
863 todo -= len; 671 todo -= len;
864 } 672 }
865 673
866 errno = 0; 674 FUBd;
675
676 /* linux's readahead basically only fails for EBADF or EINVAL (not mmappable) */
677 /* but not for e.g. EIO or eof, so we also never fail */
867 return count; 678 return 0;
868} 679}
869 680
870#endif 681#endif
871 682
872/* sendfile always needs emulation */ 683/* sendfile always needs emulation */
873static ssize_t 684static eio_ssize_t
874eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) 685eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
875{ 686{
687 eio_ssize_t written = 0;
876 ssize_t res; 688 eio_ssize_t res;
877 689
878 if (!count) 690 if (!count)
879 return 0; 691 return 0;
880 692
693 for (;;)
694 {
695#ifdef __APPLE__
696# undef HAVE_SENDFILE /* broken, as everything on os x */
697#endif
881#if HAVE_SENDFILE 698#if HAVE_SENDFILE
882# if __linux 699# if __linux
700 off_t soffset = offset;
883 res = sendfile (ofd, ifd, &offset, count); 701 res = sendfile (ofd, ifd, &soffset, count);
884 702
885# elif __freebsd 703# elif __FreeBSD__
886 /* 704 /*
887 * Of course, the freebsd sendfile is a dire hack with no thoughts 705 * Of course, the freebsd sendfile is a dire hack with no thoughts
888 * wasted on making it similar to other I/O functions. 706 * wasted on making it similar to other I/O functions.
889 */ 707 */
890 {
891 off_t sbytes; 708 off_t sbytes;
892 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 709 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
893 710
894 if (res < 0 && sbytes) 711 #if 0 /* according to the manpage, this is correct, but broken behaviour */
895 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 712 /* freebsd' sendfile will return 0 on success */
713 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
714 /* not on e.g. EIO or EPIPE - sounds broken */
715 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
896 res = sbytes; 716 res = sbytes;
897 } 717 #endif
718
719 /* according to source inspection, this is correct, and useful behaviour */
720 if (sbytes)
721 res = sbytes;
722
723# elif defined __APPLE__
724 off_t sbytes = count;
725 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
726
727 /* according to the manpage, sbytes is always valid */
728 if (sbytes)
729 res = sbytes;
898 730
899# elif __hpux 731# elif __hpux
900 res = sendfile (ofd, ifd, offset, count, 0, 0); 732 res = sendfile (ofd, ifd, offset, count, 0, 0);
901 733
902# elif __solaris 734# elif __solaris
903 {
904 struct sendfilevec vec; 735 struct sendfilevec vec;
905 size_t sbytes; 736 size_t sbytes;
906 737
907 vec.sfv_fd = ifd; 738 vec.sfv_fd = ifd;
908 vec.sfv_flag = 0; 739 vec.sfv_flag = 0;
909 vec.sfv_off = offset; 740 vec.sfv_off = offset;
910 vec.sfv_len = count; 741 vec.sfv_len = count;
911 742
912 res = sendfilev (ofd, &vec, 1, &sbytes); 743 res = sendfilev (ofd, &vec, 1, &sbytes);
913 744
914 if (res < 0 && sbytes) 745 if (res < 0 && sbytes)
915 res = sbytes; 746 res = sbytes;
916 }
917 747
918# endif 748# endif
749
750#elif defined (_WIN32) && 0
751 /* does not work, just for documentation of what would need to be done */
752 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
753 /* libeio guarantees that the file offset does not change, and windows */
754 /* has no way to get an independent handle to the same file description */
755 HANDLE h = TO_SOCKET (ifd);
756 SetFilePointer (h, offset, 0, FILE_BEGIN);
757 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
758
919#else 759#else
920 res = -1; 760 res = EIO_ENOSYS ();
921 errno = ENOSYS;
922#endif 761#endif
923 762
763 /* we assume sendfile can copy at least 128mb in one go */
764 if (res <= 128 * 1024 * 1024)
765 {
766 if (res > 0)
767 written += res;
768
769 if (written)
770 return written;
771
772 break;
773 }
774 else
775 {
776 /* if we requested more, then probably the kernel was lazy */
777 written += res;
778 offset += res;
779 count -= res;
780
781 if (!count)
782 return written;
783 }
784 }
785
924 if (res < 0 786 if (res < 0
925 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 787 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
788 /* BSDs */
789#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
790 || errno == ENOTSUP
791#endif
792#ifdef EOPNOTSUPP /* windows */
793 || errno == EOPNOTSUPP /* BSDs */
794#endif
926#if __solaris 795#if __solaris
927 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 796 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
928#endif 797#endif
929 ) 798 )
930 ) 799 )
… …
934 803
935 res = 0; 804 res = 0;
936 805
937 while (count) 806 while (count)
938 { 807 {
939 ssize_t cnt; 808 eio_ssize_t cnt;
940 809
941 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 810 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
942 811
943 if (cnt <= 0) 812 if (cnt <= 0)
944 { 813 {
… …
956 825
957 offset += cnt; 826 offset += cnt;
958 res += cnt; 827 res += cnt;
959 count -= cnt; 828 count -= cnt;
960 } 829 }
830
831 FUBd;
961 } 832 }
962 833
963 return res; 834 return res;
835}
836
837#ifdef PAGESIZE
838# define eio_pagesize() PAGESIZE
839#else
840static intptr_t
841eio_pagesize (void)
842{
843 static intptr_t page;
844
845 if (!page)
846 page = sysconf (_SC_PAGESIZE);
847
848 return page;
849}
850#endif
851
852static void
853eio_page_align (void **addr, size_t *length)
854{
855 intptr_t mask = eio_pagesize () - 1;
856
857 /* round down addr */
858 intptr_t adj = mask & (intptr_t)*addr;
859
860 *addr = (void *)((intptr_t)*addr - adj);
861 *length += adj;
862
863 /* round up length */
864 *length = (*length + mask) & ~mask;
865}
866
867#if !_POSIX_MEMLOCK
868# define eio__mlockall(a) EIO_ENOSYS ()
869#else
870
871static int
872eio__mlockall (int flags)
873{
874 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
875 extern int mallopt (int, int);
876 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
877 #endif
878
879 if (EIO_MCL_CURRENT != MCL_CURRENT
880 || EIO_MCL_FUTURE != MCL_FUTURE)
881 {
882 flags = 0
883 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
884 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
885 }
886
887 return mlockall (flags);
888}
889#endif
890
891#if !_POSIX_MEMLOCK_RANGE
892# define eio__mlock(a,b) EIO_ENOSYS ()
893#else
894
895static int
896eio__mlock (void *addr, size_t length)
897{
898 eio_page_align (&addr, &length);
899
900 return mlock (addr, length);
901}
902
903#endif
904
905#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
906# define eio__msync(a,b,c) EIO_ENOSYS ()
907#else
908
909static int
910eio__msync (void *mem, size_t len, int flags)
911{
912 eio_page_align (&mem, &len);
913
914 if (EIO_MS_ASYNC != MS_SYNC
915 || EIO_MS_INVALIDATE != MS_INVALIDATE
916 || EIO_MS_SYNC != MS_SYNC)
917 {
918 flags = 0
919 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
920 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
921 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
922 }
923
924 return msync (mem, len, flags);
925}
926
927#endif
928
929static int
930eio__mtouch (eio_req *req)
931{
932 void *mem = req->ptr2;
933 size_t len = req->size;
934 int flags = req->int1;
935
936 eio_page_align (&mem, &len);
937
938 {
939 intptr_t addr = (intptr_t)mem;
940 intptr_t end = addr + len;
941 intptr_t page = eio_pagesize ();
942
943 if (addr < end)
944 if (flags & EIO_MT_MODIFY) /* modify */
945 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
946 else
947 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
948 }
949
950 return 0;
951}
952
953/*****************************************************************************/
954/* requests implemented outside eio_execute, because they are so large */
955
956static void
957eio__lseek (eio_req *req)
958{
959 /* this usually gets optimised away completely, or your compiler sucks, */
960 /* or the whence constants really are not 0, 1, 2 */
961 int whence = req->int2 == EIO_SEEK_SET ? SEEK_SET
962 : req->int2 == EIO_SEEK_CUR ? SEEK_CUR
963 : req->int2 == EIO_SEEK_END ? SEEK_END
964 : req->int2;
965
966 req->offs = lseek (req->int1, req->offs, whence);
967 req->result = req->offs == (off_t)-1 ? -1 : 0;
968}
969
970/* result will always end up in tmpbuf, there is always space for adding a 0-byte */
971static int
972eio__realpath (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
973{
974 char *res;
975 const char *rel = path;
976 char *tmp1, *tmp2;
977#if SYMLOOP_MAX > 32
978 int symlinks = SYMLOOP_MAX;
979#else
980 int symlinks = 32;
981#endif
982
983 errno = EINVAL;
984 if (!rel)
985 return -1;
986
987 errno = ENOENT;
988 if (!*rel)
989 return -1;
990
991 res = etp_tmpbuf_get (tmpbuf, EIO_PATH_MAX * 3);
992#ifdef _WIN32
993 if (_access (rel, 4) != 0)
994 return -1;
995
996 symlinks = GetFullPathName (rel, EIO_PATH_MAX * 3, res, 0);
997
998 errno = ENAMETOOLONG;
999 if (symlinks >= EIO_PATH_MAX * 3)
1000 return -1;
1001
1002 errno = EIO;
1003 if (symlinks <= 0)
1004 return -1;
1005
1006 return symlinks;
1007
1008#else
1009 tmp1 = res + EIO_PATH_MAX;
1010 tmp2 = tmp1 + EIO_PATH_MAX;
1011
1012#if 0 /* disabled, the musl way to do things is just too racy */
1013#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1014 /* on linux we may be able to ask the kernel */
1015 {
1016 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1017
1018 if (fd >= 0)
1019 {
1020 sprintf (tmp1, "/proc/self/fd/%d", fd);
1021 req->result = readlink (tmp1, res, EIO_PATH_MAX);
1022 /* here we should probably stat the open file and the disk file, to make sure they still match */
1023 close (fd);
1024
1025 if (req->result > 0)
1026 goto done;
1027 }
1028 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1029 return -1;
1030 }
1031#endif
1032#endif
1033
1034 if (*rel != '/')
1035 {
1036 int len;
1037
1038 errno = ENOENT;
1039 if (wd == EIO_INVALID_WD)
1040 return -1;
1041
1042 if (wd == EIO_CWD)
1043 {
1044 if (!getcwd (res, EIO_PATH_MAX))
1045 return -1;
1046
1047 len = strlen (res);
1048 }
1049 else
1050 memcpy (res, wd->str, len = wd->len);
1051
1052 if (res [1]) /* only use if not / */
1053 res += len;
1054 }
1055
1056 while (*rel)
1057 {
1058 eio_ssize_t len, linklen;
1059 const char *beg = rel;
1060
1061 while (*rel && *rel != '/')
1062 ++rel;
1063
1064 len = rel - beg;
1065
1066 if (!len) /* skip slashes */
1067 {
1068 ++rel;
1069 continue;
1070 }
1071
1072 if (beg [0] == '.')
1073 {
1074 if (len == 1)
1075 continue; /* . - nop */
1076
1077 if (beg [1] == '.' && len == 2)
1078 {
1079 /* .. - back up one component, if possible */
1080
1081 while (res != tmpbuf->ptr)
1082 if (*--res == '/')
1083 break;
1084
1085 continue;
1086 }
1087 }
1088
1089 errno = ENAMETOOLONG;
1090 if (res + 1 + len + 1 >= tmp1)
1091 return -1;
1092
1093 /* copy one component */
1094 *res = '/';
1095 memcpy (res + 1, beg, len);
1096
1097 /* zero-terminate, for readlink */
1098 res [len + 1] = 0;
1099
1100 /* now check if it's a symlink */
1101 linklen = readlink (tmpbuf->ptr, tmp1, EIO_PATH_MAX);
1102
1103 if (linklen < 0)
1104 {
1105 if (errno != EINVAL)
1106 return -1;
1107
1108 /* it's a normal directory. hopefully */
1109 res += len + 1;
1110 }
1111 else
1112 {
1113 /* yay, it was a symlink - build new path in tmp2 */
1114 int rellen = strlen (rel);
1115
1116 errno = ENAMETOOLONG;
1117 if (linklen + 1 + rellen >= EIO_PATH_MAX) /* also catch linklen >= EIO_PATH_MAX */
1118 return -1;
1119
1120 errno = ELOOP;
1121 if (!--symlinks)
1122 return -1;
1123
1124 if (*tmp1 == '/')
1125 res = tmpbuf->ptr; /* symlink resolves to an absolute path */
1126
1127 /* we need to be careful, as rel might point into tmp2 already */
1128 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1129 tmp2 [linklen] = '/';
1130 memcpy (tmp2, tmp1, linklen);
1131
1132 rel = tmp2;
1133 }
1134 }
1135
1136 /* special case for the lone root path */
1137 if (res == tmpbuf->ptr)
1138 *res++ = '/';
1139
1140 return res - (char *)tmpbuf->ptr;
1141#endif
1142}
1143
1144static signed char
1145eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1146{
1147 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1148 : a->inode < b->inode ? -1
1149 : a->inode > b->inode ? 1
1150 : 0;
1151}
1152
1153#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1154
1155#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1156#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1157
1158static void
1159eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1160{
1161 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1162 unsigned char *bit = bits;
1163
1164 assert (CHAR_BIT == 8);
1165 assert (sizeof (eio_dirent) * 8 < 256);
1166 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1167 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1168
1169 if (size <= EIO_SORT_FAST)
1170 return;
1171
1172 /* first prepare an array of bits to test in our radix sort */
1173 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1174 /* inode_bits must contain all inodes ORed together */
1175 /* which is used to skip bits that are 0 everywhere, which is very common */
1176 {
1177 eio_ino_t endianness;
1178 int i, j;
1179
1180 /* we store the byte offset of byte n into byte n of "endianness" */
1181 for (i = 0; i < sizeof (eio_ino_t); ++i)
1182 ((unsigned char *)&endianness)[i] = i;
1183
1184 *bit++ = 0;
1185
1186 for (i = 0; i < sizeof (eio_ino_t); ++i)
1187 {
1188 /* shifting off the byte offsets out of "endianness" */
1189 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1190 endianness >>= 8;
1191
1192 for (j = 0; j < 8; ++j)
1193 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1194 *bit++ = offs + j;
1195 }
1196
1197 for (j = 0; j < 8; ++j)
1198 if (score_bits & (1 << j))
1199 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1200 }
1201
1202 /* now actually do the sorting (a variant of MSD radix sort) */
1203 {
1204 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1205 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1206 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1207 int stk_idx = 0;
1208
1209 base_stk [stk_idx] = dents;
1210 end_stk [stk_idx] = dents + size;
1211 bit_stk [stk_idx] = bit - 1;
1212
1213 do
1214 {
1215 base = base_stk [stk_idx];
1216 end = end_stk [stk_idx];
1217 bit = bit_stk [stk_idx];
1218
1219 for (;;)
1220 {
1221 unsigned char O = *bit >> 3;
1222 unsigned char M = 1 << (*bit & 7);
1223
1224 eio_dirent *a = base;
1225 eio_dirent *b = end;
1226
1227 if (b - a < EIO_SORT_CUTOFF)
1228 break;
1229
1230 /* now bit-partition the array on the bit */
1231 /* this ugly asymmetric loop seems to perform much better than typical */
1232 /* partition algos found in the literature */
1233 do
1234 if (!(((unsigned char *)a)[O] & M))
1235 ++a;
1236 else if (!(((unsigned char *)--b)[O] & M))
1237 {
1238 eio_dirent tmp = *a; *a = *b; *b = tmp;
1239 ++a;
1240 }
1241 while (b > a);
1242
1243 /* next bit, or stop, if no bits left in this path */
1244 if (!*--bit)
1245 break;
1246
1247 base_stk [stk_idx] = a;
1248 end_stk [stk_idx] = end;
1249 bit_stk [stk_idx] = bit;
1250 ++stk_idx;
1251
1252 end = a;
1253 }
1254 }
1255 while (stk_idx--);
1256 }
1257}
1258
1259static void
1260eio_dent_insertion_sort (eio_dirent *dents, int size)
1261{
1262 /* first move the smallest element to the front, to act as a sentinel */
1263 {
1264 int i;
1265 eio_dirent *min = dents;
1266
1267 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1268 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1269 if (EIO_DENT_CMP (dents [i], <, *min))
1270 min = &dents [i];
1271
1272 /* swap elements 0 and j (minimum) */
1273 {
1274 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1275 }
1276 }
1277
1278 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1279 {
1280 eio_dirent *i, *j;
1281
1282 for (i = dents + 1; i < dents + size; ++i)
1283 {
1284 eio_dirent value = *i;
1285
1286 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1287 j [1] = j [0];
1288
1289 j [1] = value;
1290 }
1291 }
1292}
1293
1294static void
1295eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1296{
1297 if (size <= 1)
1298 return; /* our insertion sort relies on size > 0 */
1299
1300 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1301 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1302 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1303
1304 /* use an insertion sort at the end, or for small arrays, */
1305 /* as insertion sort is more efficient for small partitions */
1306 eio_dent_insertion_sort (dents, size);
964} 1307}
965 1308
966/* read a full directory */ 1309/* read a full directory */
967static void 1310static void
968eio__scandir (eio_req *req, etp_worker *self) 1311eio__scandir (eio_req *req, etp_worker *self)
969{ 1312{
1313 char *name, *names;
1314 int namesalloc = 4096 - sizeof (void *) * 4;
1315 int namesoffs = 0;
1316 int flags = req->int1;
1317 eio_dirent *dents = 0;
1318 int dentalloc = 128;
1319 int dentoffs = 0;
1320 eio_ino_t inode_bits = 0;
1321#ifdef _WIN32
1322 HANDLE dirp;
1323 WIN32_FIND_DATA entp;
1324#else
970 DIR *dirp; 1325 DIR *dirp;
971 EIO_STRUCT_DIRENT *entp; 1326 EIO_STRUCT_DIRENT *entp;
972 char *name, *names; 1327#endif
973 int memlen = 4096;
974 int memofs = 0;
975 int res = 0;
976 1328
977 X_LOCK (wrklock); 1329 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 1330
984 if (dirp && names) 1331 if (!(flags & EIO_READDIR_DENTS))
985 for (;;) 1332 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1333
1334#ifdef _WIN32
1335 {
1336 int len = strlen ((const char *)req->ptr1);
1337 char *path = malloc (MAX_PATH);
1338 const char *fmt;
1339 const char *reqpath = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1340
1341 if (!len)
1342 fmt = "./*";
1343 else if (reqpath[len - 1] == '/' || reqpath[len - 1] == '\\')
1344 fmt = "%s*";
1345 else
1346 fmt = "%s/*";
1347
1348 _snprintf (path, MAX_PATH, fmt, reqpath);
1349 dirp = FindFirstFile (path, &entp);
1350 free (path);
1351
1352 if (dirp == INVALID_HANDLE_VALUE)
1353 {
1354 /* should steal _dosmaperr */
1355 switch (GetLastError ())
1356 {
1357 case ERROR_FILE_NOT_FOUND:
1358 req->result = 0;
1359 break;
1360
1361 case ERROR_INVALID_NAME:
1362 case ERROR_PATH_NOT_FOUND:
1363 case ERROR_NO_MORE_FILES:
1364 errno = ENOENT;
1365 break;
1366
1367 case ERROR_NOT_ENOUGH_MEMORY:
1368 errno = ENOMEM;
1369 break;
1370
1371 default:
1372 errno = EINVAL;
1373 break;
1374 }
1375
1376 return;
1377 }
1378 }
1379#else
1380 #if HAVE_AT
1381 if (req->wd)
986 { 1382 {
1383 int fd = openat (WD2FD (req->wd), req->ptr1, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1384
1385 if (fd < 0)
1386 return;
1387
1388 dirp = fdopendir (fd);
1389
1390 if (!dirp)
1391 close (fd);
1392 }
1393 else
1394 dirp = opendir (req->ptr1);
1395 #else
1396 dirp = opendir (wd_expand (&self->tmpbuf, req->wd, req->ptr1));
1397 #endif
1398
1399 if (!dirp)
1400 return;
1401#endif
1402
1403 if (req->flags & EIO_FLAG_PTR1_FREE)
1404 free (req->ptr1);
1405
1406 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1407 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1408 req->ptr2 = names = malloc (namesalloc);
1409
1410 if (!names || (flags && !dents))
1411 return;
1412
1413 for (;;)
1414 {
1415 int done;
1416
1417#ifdef _WIN32
1418 done = !dirp;
1419#else
987 errno = 0; 1420 errno = 0;
988 entp = readdir (dirp); 1421 entp = readdir (dirp);
1422 done = !entp;
1423#endif
989 1424
1425 if (done)
1426 {
1427#ifndef _WIN32
1428 int old_errno = errno;
1429 closedir (dirp);
1430 errno = old_errno;
1431
990 if (!entp) 1432 if (errno)
1433 break;
1434#endif
1435
1436 /* sort etc. */
1437 req->int1 = flags;
1438 req->result = dentoffs;
1439
1440 if (flags & EIO_READDIR_STAT_ORDER)
1441 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1442 else if (flags & EIO_READDIR_DIRS_FIRST)
1443 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1444 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1445 else
1446 {
1447 /* in this case, all is known, and we just put dirs first and sort them */
1448 eio_dirent *oth = dents + dentoffs;
1449 eio_dirent *dir = dents;
1450
1451 /* now partition dirs to the front, and non-dirs to the back */
1452 /* by walking from both sides and swapping if necessary */
1453 while (oth > dir)
1454 {
1455 if (dir->type == EIO_DT_DIR)
1456 ++dir;
1457 else if ((--oth)->type == EIO_DT_DIR)
1458 {
1459 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1460
1461 ++dir;
1462 }
1463 }
1464
1465 /* now sort the dirs only (dirs all have the same score) */
1466 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1467 }
1468
991 break; 1469 break;
1470 }
992 1471
993 name = entp->d_name; 1472 /* now add the entry to our list(s) */
1473 name = D_NAME (entp);
994 1474
1475 /* skip . and .. entries */
995 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1476 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
996 { 1477 {
997 int len = strlen (name) + 1; 1478 int len = D_NAMLEN (entp) + 1;
998 1479
999 res++; 1480 while (ecb_expect_false (namesoffs + len > namesalloc))
1000
1001 while (memofs + len > memlen)
1002 { 1481 {
1003 memlen *= 2; 1482 namesalloc *= 2;
1004 X_LOCK (wrklock);
1005 req->ptr2 = names = realloc (names, memlen); 1483 req->ptr2 = names = realloc (names, namesalloc);
1006 X_UNLOCK (wrklock);
1007 1484
1008 if (!names) 1485 if (!names)
1486 break;
1487 }
1488
1489 memcpy (names + namesoffs, name, len);
1490
1491 if (dents)
1492 {
1493 struct eio_dirent *ent;
1494
1495 if (ecb_expect_false (dentoffs == dentalloc))
1496 {
1497 dentalloc *= 2;
1498 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1499
1500 if (!dents)
1009 break; 1501 break;
1010 } 1502 }
1011 1503
1012 memcpy (names + memofs, name, len); 1504 ent = dents + dentoffs;
1013 memofs += len; 1505
1506 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1507 ent->namelen = len - 1;
1508 ent->inode = D_INO (entp);
1509
1510 inode_bits |= ent->inode;
1511
1512 switch (D_TYPE (entp))
1513 {
1514 default:
1515 ent->type = EIO_DT_UNKNOWN;
1516 flags |= EIO_READDIR_FOUND_UNKNOWN;
1517 break;
1518
1519 #ifdef DT_FIFO
1520 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1521 #endif
1522 #ifdef DT_CHR
1523 case DT_CHR: ent->type = EIO_DT_CHR; break;
1524 #endif
1525 #ifdef DT_MPC
1526 case DT_MPC: ent->type = EIO_DT_MPC; break;
1527 #endif
1528 #ifdef DT_DIR
1529 case DT_DIR: ent->type = EIO_DT_DIR; break;
1530 #endif
1531 #ifdef DT_NAM
1532 case DT_NAM: ent->type = EIO_DT_NAM; break;
1533 #endif
1534 #ifdef DT_BLK
1535 case DT_BLK: ent->type = EIO_DT_BLK; break;
1536 #endif
1537 #ifdef DT_MPB
1538 case DT_MPB: ent->type = EIO_DT_MPB; break;
1539 #endif
1540 #ifdef DT_REG
1541 case DT_REG: ent->type = EIO_DT_REG; break;
1542 #endif
1543 #ifdef DT_NWK
1544 case DT_NWK: ent->type = EIO_DT_NWK; break;
1545 #endif
1546 #ifdef DT_CMP
1547 case DT_CMP: ent->type = EIO_DT_CMP; break;
1548 #endif
1549 #ifdef DT_LNK
1550 case DT_LNK: ent->type = EIO_DT_LNK; break;
1551 #endif
1552 #ifdef DT_SOCK
1553 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1554 #endif
1555 #ifdef DT_DOOR
1556 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1557 #endif
1558 #ifdef DT_WHT
1559 case DT_WHT: ent->type = EIO_DT_WHT; break;
1560 #endif
1561 }
1562
1563 ent->score = 7;
1564
1565 if (flags & EIO_READDIR_DIRS_FIRST)
1566 {
1567 if (ent->type == EIO_DT_UNKNOWN)
1568 {
1569 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1570 ent->score = 1;
1571 else if (!strchr (name, '.')) /* absence of dots indicate likely dirs */
1572 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1573 }
1574 else if (ent->type == EIO_DT_DIR)
1575 ent->score = 0;
1576 }
1014 } 1577 }
1578
1579 namesoffs += len;
1580 ++dentoffs;
1015 } 1581 }
1016 1582
1017 if (errno) 1583 if (EIO_CANCELLED (req))
1018 res = -1; 1584 {
1585 errno = ECANCELED;
1586 break;
1587 }
1588
1589#ifdef _WIN32
1590 if (!FindNextFile (dirp, &entp))
1591 {
1592 FindClose (dirp);
1593 dirp = 0;
1594 }
1595#endif
1596 }
1597}
1598
1599/*****************************************************************************/
1600/* working directory stuff */
1601/* various deficiencies in the posix 2008 api force us to */
1602/* keep the absolute path in string form at all times */
1603/* fuck yeah. */
1604
1605#if !HAVE_AT
1606
1607/* a bit like realpath, but usually faster because it doesn'T have to return */
1608/* an absolute or canonical path */
1609static const char *
1610wd_expand (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
1611{
1612 if (!wd || *path == '/')
1613 return path;
1614
1615 if (path [0] == '.' && !path [1])
1616 return wd->str;
1617
1019 1618 {
1020 req->result = res; 1619 int l1 = wd->len;
1021} 1620 int l2 = strlen (path);
1022 1621
1023#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) 1622 char *res = etp_tmpbuf_get (tmpbuf, l1 + l2 + 2);
1024# undef msync
1025# define msync(a,b,c) ((errno = ENOSYS), -1)
1026#endif
1027 1623
1028int 1624 memcpy (res, wd->str, l1);
1029eio__mtouch (void *mem, size_t len, int flags) 1625 res [l1] = '/';
1626 memcpy (res + l1 + 1, path, l2 + 1);
1627
1628 return res;
1629 }
1630}
1631
1632#endif
1633
1634static eio_wd
1635eio__wd_open_sync (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
1030{ 1636{
1031 intptr_t addr = (intptr_t)mem; 1637 int fd;
1032 intptr_t end = addr + len; 1638 eio_wd res;
1033#ifdef PAGESIZE 1639 int len = eio__realpath (tmpbuf, wd, path);
1034 const intptr_t page = PAGESIZE; 1640
1641 if (len < 0)
1642 return EIO_INVALID_WD;
1643
1644#if HAVE_AT
1645 fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1646
1647 if (fd < 0)
1648 return EIO_INVALID_WD;
1649#endif
1650
1651 res = malloc (sizeof (*res) + len); /* one extra 0-byte */
1652
1653#if HAVE_AT
1654 res->fd = fd;
1655#endif
1656
1657 res->len = len;
1658 memcpy (res->str, tmpbuf->ptr, len);
1659 res->str [len] = 0;
1660
1661 return res;
1662}
1663
1664eio_wd
1665eio_wd_open_sync (eio_wd wd, const char *path)
1666{
1667 struct etp_tmpbuf tmpbuf = { };
1668 wd = eio__wd_open_sync (&tmpbuf, wd, path);
1669 free (tmpbuf.ptr);
1670
1671 return wd;
1672}
1673
1674void
1675eio_wd_close_sync (eio_wd wd)
1676{
1677 if (wd != EIO_INVALID_WD && wd != EIO_CWD)
1678 {
1679 #if HAVE_AT
1680 close (wd->fd);
1681 #endif
1682 free (wd);
1683 }
1684}
1685
1686#if HAVE_AT
1687
1688static int
1689eio__renameat2 (int olddirfd, const char *oldpath, int newdirfd, const char *newpath, unsigned int flags)
1690{
1691#if HAVE_RENAMEAT2
1692 return syscall (SYS_renameat2, olddirfd, oldpath, newdirfd, newpath, flags);
1035#else 1693#else
1036 static intptr_t page;
1037
1038 if (!page) 1694 if (flags)
1039 page = sysconf (_SC_PAGESIZE); 1695 return EIO_ENOSYS ();
1040#endif
1041 1696
1042 addr &= ~(page - 1); /* assume page size is always a power of two */ 1697 return renameat (olddirfd, oldpath, newdirfd, newpath);
1698#endif
1699}
1043 1700
1044 if (addr < end) 1701/* they forgot these */
1045 if (flags) /* modify */
1046 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
1047 else
1048 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
1049 1702
1703static int
1704eio__truncateat (int dirfd, const char *path, off_t length)
1705{
1706 int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
1707 int res;
1708
1709 if (fd < 0)
1710 return fd;
1711
1712 res = ftruncate (fd, length);
1713 close (fd);
1050 return 0; 1714 return res;
1051} 1715}
1716
1717static int
1718eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
1719{
1720 int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
1721 int res;
1722
1723 if (fd < 0)
1724 return fd;
1725
1726 res = fstatvfs (fd, buf);
1727 close (fd);
1728 return res;
1729
1730}
1731
1732#endif
1052 1733
1053/*****************************************************************************/ 1734/*****************************************************************************/
1054 1735
1055#define ALLOC(len) \ 1736#define ALLOC(len) \
1056 if (!req->ptr2) \ 1737 if (!req->ptr2) \
1057 { \ 1738 { \
1058 X_LOCK (wrklock); \ 1739 X_LOCK (EIO_POOL->wrklock); \
1059 req->flags |= EIO_FLAG_PTR2_FREE; \ 1740 req->flags |= EIO_FLAG_PTR2_FREE; \
1060 X_UNLOCK (wrklock); \ 1741 X_UNLOCK (EIO_POOL->wrklock); \
1061 req->ptr2 = malloc (len); \ 1742 req->ptr2 = malloc (len); \
1062 if (!req->ptr2) \ 1743 if (!req->ptr2) \
1063 { \ 1744 { \
1064 errno = ENOMEM; \ 1745 errno = ENOMEM; \
1065 req->result = -1; \ 1746 req->result = -1; \
1066 break; \ 1747 break; \
1067 } \ 1748 } \
1068 } 1749 }
1069 1750
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/*****************************************************************************/ 1751/*****************************************************************************/
1144 1752
1753int ecb_cold
1145int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1754eio_init (void (*want_poll)(void), void (*done_poll)(void))
1146{ 1755{
1147 return etp_init (want_poll, done_poll); 1756 eio_want_poll_cb = want_poll;
1148} 1757 eio_done_poll_cb = done_poll;
1149 1758
1759 return etp_init (EIO_POOL, 0, 0, 0);
1760}
1761
1762ecb_inline void
1150static void eio_api_destroy (eio_req *req) 1763eio_api_destroy (eio_req *req)
1151{ 1764{
1152 free (req); 1765 free (req);
1153} 1766}
1154 1767
1155#define REQ(rtype) \ 1768#define REQ(rtype) \
1156 eio_req *req; \ 1769 eio_req *req; \
1157 \ 1770 \
1158 req = (eio_req *)calloc (1, sizeof *req); \ 1771 req = (eio_req *)calloc (1, sizeof *req); \
1159 if (!req) \ 1772 if (!req) \
1160 return 0; \ 1773 return 0; \
… …
1174 { \ 1787 { \
1175 eio_api_destroy (req); \ 1788 eio_api_destroy (req); \
1176 return 0; \ 1789 return 0; \
1177 } 1790 }
1178 1791
1792#define SINGLEDOT(ptr) (0[(char *)(ptr)] == '.' && !1[(char *)(ptr)])
1793
1794static void
1179static void eio_execute (etp_worker *self, eio_req *req) 1795eio_execute (etp_worker *self, eio_req *req)
1180{ 1796{
1181 errno = 0; 1797#if HAVE_AT
1798 int dirfd;
1799#else
1800 const char *path;
1801#endif
1802
1803 if (ecb_expect_false (EIO_CANCELLED (req)))
1804 {
1805 req->result = -1;
1806 req->errorno = ECANCELED;
1807 return;
1808 }
1809
1810 if (ecb_expect_false (req->wd == EIO_INVALID_WD))
1811 {
1812 req->result = -1;
1813 req->errorno = ENOENT;
1814 return;
1815 }
1816
1817 if (req->type >= EIO_OPEN)
1818 {
1819 #if HAVE_AT
1820 dirfd = WD2FD (req->wd);
1821 #else
1822 path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1823 #endif
1824 }
1182 1825
1183 switch (req->type) 1826 switch (req->type)
1184 { 1827 {
1828 case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
1829 req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
1830 break;
1831 case EIO_WD_CLOSE: req->result = 0;
1832 eio_wd_close_sync (req->wd); break;
1833
1834 case EIO_SEEK: eio__lseek (req); break;
1185 case EIO_READ: ALLOC (req->size); 1835 case EIO_READ: ALLOC (req->size);
1186 req->result = req->offs >= 0 1836 req->result = req->offs >= 0
1187 ? pread (req->int1, req->ptr2, req->size, req->offs) 1837 ? pread (req->int1, req->ptr2, req->size, req->offs)
1188 : read (req->int1, req->ptr2, req->size); break; 1838 : read (req->int1, req->ptr2, req->size); break;
1189 case EIO_WRITE: req->result = req->offs >= 0 1839 case EIO_WRITE: req->result = req->offs >= 0
1190 ? pwrite (req->int1, req->ptr2, req->size, req->offs) 1840 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1191 : write (req->int1, req->ptr2, req->size); break; 1841 : write (req->int1, req->ptr2, req->size); break;
1192 1842
1843 case EIO_FCNTL: req->result = fcntl (req->int1, (int) req->int2, req->ptr2); break;
1844 case EIO_IOCTL: req->result = ioctl (req->int1, (unsigned long)req->int2, req->ptr2); break;
1845
1193 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break; 1846 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; 1847 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
1848
1849#if HAVE_AT
1195 1850
1196 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1851 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1197 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 1852 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break;
1198 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1853 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1199 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 1854 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; 1855 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; 1856 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; 1857 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; 1858 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; 1859
1212 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1213 case EIO_UNLINK: req->result = unlink (req->ptr1); break; 1860 case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
1214 case EIO_RMDIR: req->result = rmdir (req->ptr1); break; 1861 case EIO_RMDIR: /* complications arise because "." cannot be removed, so we might have to expand */
1862 req->result = req->wd && SINGLEDOT (req->ptr1)
1863 ? rmdir (req->wd->str)
1864 : unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break;
1215 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; 1865 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; 1866 case EIO_RENAME: req->result = eio__renameat2 (
1867 dirfd,
1868 /* complications arise because "." cannot be renamed, so we might have to expand */
1869 req->wd && SINGLEDOT (req->ptr1) ? req->wd->str : req->ptr1,
1870 WD2FD ((eio_wd)req->int3),
1871 req->ptr2,
1872 req->int2
1873 );
1874 break;
1217 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; 1875 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; 1876 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; 1877 case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1220 1878 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1879 req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1221 case EIO_READLINK: ALLOC (NAME_MAX); 1880 case EIO_READLINK: ALLOC (EIO_PATH_MAX);
1222 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; 1881 req->result = readlinkat (dirfd, req->ptr1, req->ptr2, EIO_PATH_MAX);
1223 1882 if (req->result == EIO_PATH_MAX)
1224 case EIO_SYNC: req->result = 0; sync (); break; 1883 {
1225 case EIO_FSYNC: req->result = fsync (req->int1); break; 1884 req->result = -1;
1226 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 1885 errno = ENAMETOOLONG;
1227 case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break; 1886 }
1228 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break; 1887 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: 1888 case EIO_UTIME:
1234#ifdef _WIN32 1889 case EIO_FUTIME:
1235 Sleep (req->nv1 * 1000.); 1890 {
1891 struct timespec ts[2];
1892 struct timespec *times;
1893
1894 if (req->nv1 != -1. || req->nv2 != -1.)
1895 {
1896 ts[0].tv_sec = req->nv1;
1897 ts[0].tv_nsec = (req->nv1 - ts[0].tv_sec) * 1e9;
1898 ts[1].tv_sec = req->nv2;
1899 ts[1].tv_nsec = (req->nv2 - ts[1].tv_sec) * 1e9;
1900
1901 times = ts;
1902 }
1903 else
1904 times = 0;
1905
1906 req->result = req->type == EIO_FUTIME
1907 ? futimens (req->int1, times)
1908 : utimensat (dirfd, req->ptr1, times, 0);
1909 }
1910 break;
1911
1236#else 1912#else
1237 {
1238 struct timeval tv;
1239 1913
1240 tv.tv_sec = req->nv1; 1914 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1241 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; 1915 req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
1916 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1917 req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
1918 case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
1919 case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
1920 case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
1921 case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
1242 1922
1243 req->result = select (0, 0, 0, 0, &tv); 1923 case EIO_UNLINK: req->result = unlink (path ); break;
1244 } 1924 case EIO_RMDIR: req->result = rmdir (path ); break;
1245#endif 1925 case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
1246 break; 1926 case EIO_RENAME: req->result = req->int2 ? EIO_ENOSYS () : rename (path, req->ptr2); break;
1927 case EIO_LINK: req->result = link (path , req->ptr2); break;
1928 case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
1929 case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
1930 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1931 req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
1932 case EIO_READLINK: ALLOC (EIO_PATH_MAX);
1933 req->result = readlink (path, req->ptr2, EIO_PATH_MAX);
1934 if (req->result == EIO_PATH_MAX)
1935 {
1936 req->result = -1;
1937 errno = ENAMETOOLONG;
1938 }
1939 break;
1247 1940
1248 case EIO_UTIME: 1941 case EIO_UTIME:
1249 case EIO_FUTIME: 1942 case EIO_FUTIME:
1250 { 1943 {
1251 struct timeval tv[2]; 1944 struct timeval tv[2];
1252 struct timeval *times; 1945 struct timeval *times;
1253 1946
1254 if (req->nv1 != -1. || req->nv2 != -1.) 1947 if (req->nv1 != -1. || req->nv2 != -1.)
1255 { 1948 {
1256 tv[0].tv_sec = req->nv1; 1949 tv[0].tv_sec = req->nv1;
1257 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.; 1950 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
1258 tv[1].tv_sec = req->nv2; 1951 tv[1].tv_sec = req->nv2;
1259 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.; 1952 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
1260 1953
1261 times = tv; 1954 times = tv;
1262 } 1955 }
1263 else 1956 else
1264 times = 0; 1957 times = 0;
1265
1266 1958
1267 req->result = req->type == EIO_FUTIME 1959 req->result = req->type == EIO_FUTIME
1268 ? futimes (req->int1, times) 1960 ? futimes (req->int1, times)
1269 : utimes (req->ptr1, times); 1961 : utimes (req->ptr1, times);
1270 } 1962 }
1271 break; 1963 break;
1272 1964
1965#endif
1966
1967 case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
1968 {
1969 ALLOC (req->result);
1970 memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
1971 }
1972 break;
1973
1974 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1975 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1976
1977 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1978 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1979
1980 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1981 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1982 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1983
1984 case EIO_CLOSE: req->result = close (req->int1); break;
1985 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1986 case EIO_SYNC: req->result = 0; sync (); break;
1987 case EIO_FSYNC: req->result = fsync (req->int1); break;
1988 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1989 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
1990 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1991 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1992 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1993 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1994 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1995 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
1996
1997 case EIO_READDIR: eio__scandir (req, self); break;
1998
1999 case EIO_BUSY:
2000#ifdef _WIN32
2001 Sleep (req->nv1 * 1e3);
2002#else
2003 {
2004 struct timeval tv;
2005
2006 tv.tv_sec = req->nv1;
2007 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
2008
2009 req->result = select (0, 0, 0, 0, &tv);
2010 }
2011#endif
2012 break;
2013
2014#if 0
1273 case EIO_GROUP: 2015 case EIO_GROUP:
1274 abort (); /* handled in eio_request */ 2016 abort (); /* handled in eio_request */
2017#endif
1275 2018
1276 case EIO_NOP: 2019 case EIO_NOP:
1277 req->result = 0; 2020 req->result = 0;
1278 break; 2021 break;
1279 2022
1280 case EIO_CUSTOM: 2023 case EIO_CUSTOM:
1281 ((void (*)(eio_req *))req->feed) (req); 2024 req->feed (req);
1282 break; 2025 break;
1283 2026
1284 default: 2027 default:
1285 req->result = -1; 2028 req->result = EIO_ENOSYS ();
1286 break; 2029 break;
1287 } 2030 }
1288 2031
1289 req->errorno = errno; 2032 req->errorno = errno;
1290} 2033}
1291 2034
1292#ifndef EIO_NO_WRAPPERS 2035#ifndef EIO_NO_WRAPPERS
1293 2036
2037eio_req *eio_wd_open (const char *path, int pri, eio_cb cb, void *data)
2038{
2039 REQ (EIO_WD_OPEN); PATH; SEND;
2040}
2041
2042eio_req *eio_wd_close (eio_wd wd, int pri, eio_cb cb, void *data)
2043{
2044 REQ (EIO_WD_CLOSE); req->wd = wd; SEND;
2045}
2046
1294eio_req *eio_nop (int pri, eio_cb cb, void *data) 2047eio_req *eio_nop (int pri, eio_cb cb, void *data)
1295{ 2048{
1296 REQ (EIO_NOP); SEND; 2049 REQ (EIO_NOP); SEND;
1297} 2050}
1298 2051
… …
1314eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2067eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1315{ 2068{
1316 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2069 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1317} 2070}
1318 2071
2072eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2073{
2074 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2075}
2076
2077eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
2078{
2079 REQ (EIO_SYNCFS); req->int1 = fd; SEND;
2080}
2081
2082eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2083{
2084 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2085}
2086
1319eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2087eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1320{ 2088{
1321 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2089 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1322} 2090}
1323 2091
1324eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) 2092eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
1325{ 2093{
1326 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; 2094 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
1327} 2095}
1328 2096
1329eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) 2097eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
1330{ 2098{
1331 REQ (EIO_FDATASYNC); req->int1 = fd; SEND; 2099 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2100}
2101
2102eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2103{
2104 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
1332} 2105}
1333 2106
1334eio_req *eio_close (int fd, int pri, eio_cb cb, void *data) 2107eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
1335{ 2108{
1336 REQ (EIO_CLOSE); req->int1 = fd; SEND; 2109 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) 2112eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
1340{ 2113{
1341 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND; 2114 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
1342} 2115}
1343 2116
2117eio_req *eio_seek (int fd, off_t offset, int whence, int pri, eio_cb cb, void *data)
2118{
2119 REQ (EIO_SEEK); req->int1 = fd; req->offs = offset; req->int2 = whence; SEND;
2120}
2121
1344eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data) 2122eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1345{ 2123{
1346 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND; 2124 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1347} 2125}
1348 2126
1349eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data) 2127eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1350{ 2128{
1351 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND; 2129 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1352} 2130}
1353 2131
2132eio_req *eio_fcntl (int fd, int cmd, void *arg, int pri, eio_cb cb, void *data)
2133{
2134 REQ (EIO_IOCTL); req->int1 = fd; req->int2 = cmd; req->ptr2 = arg; SEND;
2135}
2136
2137eio_req *eio_ioctl (int fd, unsigned long request, void *buf, int pri, eio_cb cb, void *data)
2138{
2139 REQ (EIO_IOCTL); req->int1 = fd; req->int2 = request; req->ptr2 = buf; SEND;
2140}
2141
1354eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) 2142eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1355{ 2143{
1356 REQ (EIO_FSTAT); req->int1 = fd; SEND; 2144 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1357} 2145}
1358 2146
2147eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
2148{
2149 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
2150}
2151
1359eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) 2152eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1360{ 2153{
1361 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; 2154 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1362} 2155}
1363 2156
… …
1369eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) 2162eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1370{ 2163{
1371 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2164 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1372} 2165}
1373 2166
1374eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2167eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1375{ 2168{
1376 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2169 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1377} 2170}
1378 2171
1379eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) 2172eio_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) 2192eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1400{ 2193{
1401 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2194 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1402} 2195}
1403 2196
1404eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2197eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1405{ 2198{
1406 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2199 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1407} 2200}
1408 2201
1409eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) 2202eio_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) 2218eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1426{ 2219{
1427 return eio__1path (EIO_READLINK, path, pri, cb, data); 2220 return eio__1path (EIO_READLINK, path, pri, cb, data);
1428} 2221}
1429 2222
2223eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2224{
2225 return eio__1path (EIO_REALPATH, path, pri, cb, data);
2226}
2227
1430eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) 2228eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1431{ 2229{
1432 return eio__1path (EIO_STAT, path, pri, cb, data); 2230 return eio__1path (EIO_STAT, path, pri, cb, data);
1433} 2231}
1434 2232
1435eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) 2233eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1436{ 2234{
1437 return eio__1path (EIO_LSTAT, path, pri, cb, data); 2235 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1438} 2236}
1439 2237
2238eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
2239{
2240 return eio__1path (EIO_STATVFS, path, pri, cb, data);
2241}
2242
1440eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) 2243eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1441{ 2244{
1442 return eio__1path (EIO_UNLINK, path, pri, cb, data); 2245 return eio__1path (EIO_UNLINK, path, pri, cb, data);
1443} 2246}
1444 2247
1445eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data) 2248eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1446{ 2249{
1447 return eio__1path (EIO_RMDIR, path, pri, cb, data); 2250 return eio__1path (EIO_RMDIR, path, pri, cb, data);
1448} 2251}
1449 2252
1450eio_req *eio_readdir (const char *path, int pri, eio_cb cb, void *data) 2253eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1451{ 2254{
1452 return eio__1path (EIO_READDIR, path, pri, cb, data); 2255 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1453} 2256}
1454 2257
1455eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) 2258eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1456{ 2259{
1457 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; 2260 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1458} 2261}
1459 2262
1460static eio_req * 2263static eio_req *
1461eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2264eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1462{ 2265{
… …
1486eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2289eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1487{ 2290{
1488 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2291 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1489} 2292}
1490 2293
1491eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) 2294eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
1492{ 2295{
1493 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND; 2296 REQ (EIO_CUSTOM); req->feed = execute; SEND;
1494} 2297}
1495 2298
1496#endif 2299#endif
1497 2300
1498eio_req *eio_grp (eio_cb cb, void *data) 2301eio_req *eio_grp (eio_cb cb, void *data)
… …
1507#undef SEND 2310#undef SEND
1508 2311
1509/*****************************************************************************/ 2312/*****************************************************************************/
1510/* grp functions */ 2313/* grp functions */
1511 2314
2315void
1512void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2316eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1513{ 2317{
1514 grp->int2 = limit; 2318 grp->int2 = limit;
1515 grp->feed = feed; 2319 grp->feed = feed;
1516 2320
1517 grp_try_feed (grp); 2321 grp_try_feed (grp);
1518} 2322}
1519 2323
2324void
1520void eio_grp_limit (eio_req *grp, int limit) 2325eio_grp_limit (eio_req *grp, int limit)
1521{ 2326{
1522 grp->int2 = limit; 2327 grp->int2 = limit;
1523 2328
1524 grp_try_feed (grp); 2329 grp_try_feed (grp);
1525} 2330}
1526 2331
2332void
1527void eio_grp_add (eio_req *grp, eio_req *req) 2333eio_grp_add (eio_req *grp, eio_req *req)
1528{ 2334{
1529 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2335 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1530 2336
1531 grp->flags |= EIO_FLAG_GROUPADD; 2337 grp->flags |= ETP_FLAG_GROUPADD;
1532 2338
1533 ++grp->size; 2339 ++grp->size;
1534 req->grp = grp; 2340 req->grp = grp;
1535 2341
1536 req->grp_prev = 0; 2342 req->grp_prev = 0;
… …
1543} 2349}
1544 2350
1545/*****************************************************************************/ 2351/*****************************************************************************/
1546/* misc garbage */ 2352/* misc garbage */
1547 2353
2354eio_ssize_t
1548ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2355eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1549{ 2356{
1550 etp_worker wrk;
1551
1552 wrk.dbuf = 0;
1553
1554 eio__sendfile (ofd, ifd, offset, count, &wrk); 2357 return eio__sendfile (ofd, ifd, offset, count);
1555
1556 if (wrk.dbuf)
1557 free (wrk.dbuf);
1558} 2358}
1559 2359

Diff Legend

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