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

Comparing libeio/eio.c (file contents):
Revision 1.27 by root, Wed Oct 22 18:15:36 2008 UTC vs.
Revision 1.148 by root, Tue Feb 20 06:05:19 2018 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) 2007,2008,2009,2010,2011,2012,2013,2016,2017 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#include "xthread.h" 45#include "ecb.h"
42 46
43#include <errno.h> 47#include <errno.h>
44#include <stddef.h> 48#include <stddef.h>
45#include <stdlib.h> 49#include <stdlib.h>
46#include <string.h> 50#include <string.h>
… …
49#include <sys/stat.h> 53#include <sys/stat.h>
50#include <limits.h> 54#include <limits.h>
51#include <fcntl.h> 55#include <fcntl.h>
52#include <assert.h> 56#include <assert.h>
53 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
54#ifndef EIO_FINISH 77#ifndef EIO_FINISH
55# 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
56#endif 79#endif
57 80
58#ifndef EIO_DESTROY 81#ifndef EIO_DESTROY
… …
61 84
62#ifndef EIO_FEED 85#ifndef EIO_FEED
63# 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)
64#endif 87#endif
65 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
66#ifdef _WIN32 100#ifdef _WIN32
67 101
68 /*doh*/ 102 #undef PAGESIZE
103 #define PAGESIZE 4096 /* GetSystemInfo? */
104
105 #define O_CLOEXEC 0
106
107 /* TODO: look at how perl does stat (non-sloppy), unlink (ro-files), utime, link */
108
109 #ifdef EIO_STRUCT_STATI64
110 /* look at perl's non-sloppy stat */
111 #define stat(path,buf) _stati64 (path,buf)
112 #define fstat(fd,buf) _fstati64 (fd,buf)
113 #endif
114 #define lstat(path,buf) stat (path,buf)
115 #define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
116 #define mkdir(path,mode) _mkdir (path)
117 #define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
118
119 #define chmod(path,mode) _chmod (path, mode)
120 #define dup(fd) _dup (fd)
121 #define dup2(fd1,fd2) _dup2 (fd1, fd2)
122 #define pipe(fds) _pipe (fds, 4096, O_BINARY)
123
124 #define fcntl(fd,cmd,arg) EIO_ENOSYS ()
125 #define ioctl(fd,cmd,arg) EIO_ENOSYS ()
126 #define fchmod(fd,mode) EIO_ENOSYS ()
127 #define chown(path,uid,gid) EIO_ENOSYS ()
128 #define fchown(fd,uid,gid) EIO_ENOSYS ()
129 #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
130 #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
131 #define mknod(path,mode,dev) EIO_ENOSYS ()
132 #define sync() EIO_ENOSYS ()
133 #define readlink(path,buf,s) EIO_ENOSYS ()
134 #define statvfs(path,buf) EIO_ENOSYS ()
135 #define fstatvfs(fd,buf) EIO_ENOSYS ()
136
137 #define pread(fd,buf,count,offset) eio__pread (fd, buf, count, offset)
138 #define pwrite(fd,buf,count,offset) eio__pwrite (fd, buf, count, offset)
139
140 #if __GNUC__
141 typedef long long eio_off_t; /* signed for compatibility to msvc */
142 #else
143 typedef __int64 eio_off_t; /* unsigned not supported by msvc */
144 #endif
145
146 static eio_ssize_t
147 eio__pread (int fd, void *buf, eio_ssize_t count, eio_off_t offset)
148 {
149 OVERLAPPED o = { 0 };
150 DWORD got;
151
152 o.Offset = offset;
153 o.OffsetHigh = offset >> 32;
154
155 return ReadFile ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd), buf, count, &got, &o)
156 ? got : -1;
157 }
158
159 static eio_ssize_t
160 eio__pwrite (int fd, void *buf, eio_ssize_t count, eio_off_t offset)
161 {
162 OVERLAPPED o = { 0 };
163 DWORD got;
164
165 o.Offset = offset;
166 o.OffsetHigh = offset >> 32;
167
168 return WriteFile ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd), buf, count, &got, &o)
169 ? got : -1;
170 }
171
172 /* rename() uses MoveFile, which fails to overwrite */
173 #define rename(old,neu) eio__rename (old, neu)
174
175 static int
176 eio__rename (const char *old, const char *neu)
177 {
178 if (MoveFileEx (old, neu, MOVEFILE_REPLACE_EXISTING))
179 return 0;
180
181 /* should steal _dosmaperr */
182 switch (GetLastError ())
183 {
184 case ERROR_FILE_NOT_FOUND:
185 case ERROR_PATH_NOT_FOUND:
186 case ERROR_INVALID_DRIVE:
187 case ERROR_NO_MORE_FILES:
188 case ERROR_BAD_NETPATH:
189 case ERROR_BAD_NET_NAME:
190 case ERROR_BAD_PATHNAME:
191 case ERROR_FILENAME_EXCED_RANGE:
192 errno = ENOENT;
193 break;
194
195 default:
196 errno = EACCES;
197 break;
198 }
199
200 return -1;
201 }
202
203 /* we could even stat and see if it exists */
204 static int
205 symlink (const char *old, const char *neu)
206 {
207 #if WINVER >= 0x0600
208 int flags;
209
210 /* This tries out all combinations of SYMBOLIC_LINK_FLAG_DIRECTORY
211 * and SYMBOLIC_LINK_FLAG_ALLOW_UNPRIVILEGED_CREATE,
212 * with directory first.
213 */
214 for (flags = 3; flags >= 0; --flags)
215 if (CreateSymbolicLink (neu, old, flags))
216 return 0;
217 #endif
218
219 return EIO_ERRNO (ENOENT, -1);
220 }
221
222 /* POSIX API only, causing trouble for win32 apps */
223 #define CreateHardLink(neu,old,flags) 0 /* not really creating hardlink, still using relative paths? */
224 #define CreateSymbolicLink(neu,old,flags) 0 /* vista+ only */
225
226 struct statvfs
227 {
228 int dummy;
229 };
230
231 #define DT_DIR EIO_DT_DIR
232 #define DT_REG EIO_DT_REG
233 #define D_NAME(entp) entp.cFileName
234 #define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
69 235
70#else 236#else
71 237
72# include "config.h" 238 #include <sys/ioctl.h>
73# include <sys/time.h> 239 #include <sys/time.h>
74# include <sys/select.h> 240 #include <sys/select.h>
75# include <unistd.h> 241 #include <unistd.h>
242 #include <signal.h>
243 #include <dirent.h>
244
245 #ifdef ANDROID
246 #include <sys/vfs.h>
247 #define statvfs statfs
248 #define fstatvfs fstatfs
249 #include <asm/page.h> /* supposedly limits.h does #define PAGESIZE PAGESIZE */
250 #else
251 #include <sys/statvfs.h>
252 #endif
253
254 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
255 #include <sys/mman.h>
256 #endif
257
258 #define D_NAME(entp) entp->d_name
259
260 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
261 #if __FreeBSD__ || __NetBSD__ || __OpenBSD__
262 #define _DIRENT_HAVE_D_TYPE /* sigh */
263 #define D_INO(de) (de)->d_fileno
264 #define D_NAMLEN(de) (de)->d_namlen
265 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
266 #define D_INO(de) (de)->d_ino
267 #endif
268
269 #ifdef _D_EXACT_NAMLEN
270 #undef D_NAMLEN
271 #define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
272 #endif
273
274 #ifdef _DIRENT_HAVE_D_TYPE
275 #define D_TYPE(de) (de)->d_type
276 #endif
277
278 #ifndef EIO_STRUCT_DIRENT
279 #define EIO_STRUCT_DIRENT struct dirent
280 #endif
281
282#endif
283
284#if HAVE_UTIMES
76# include <utime.h> 285# include <utime.h>
77# include <signal.h>
78# include <dirent.h>
79
80# ifndef EIO_STRUCT_DIRENT
81# define EIO_STRUCT_DIRENT struct dirent
82# endif 286#endif
83 287
288#if HAVE_SYS_SYSCALL_H
289# include <sys/syscall.h>
84#endif 290#endif
85 291
86#if HAVE_SENDFILE 292#if HAVE_SENDFILE
87# if __linux 293# if __linux
88# include <sys/sendfile.h> 294# include <sys/sendfile.h>
89# elif __freebsd 295# elif __FreeBSD__ || defined __APPLE__
90# include <sys/socket.h> 296# include <sys/socket.h>
91# include <sys/uio.h> 297# include <sys/uio.h>
92# elif __hpux 298# elif __hpux
93# include <sys/socket.h> 299# include <sys/socket.h>
94# elif __solaris /* not yet */ 300# elif __solaris
95# include <sys/sendfile.h> 301# include <sys/sendfile.h>
96# else 302# else
97# error sendfile support requested but not available 303# error sendfile support requested but not available
98# endif 304# endif
99#endif 305#endif
100 306
101/* number of seconds after which an idle threads exit */ 307#if HAVE_RENAMEAT2
102#define IDLE_TIMEOUT 10 308# include <sys/syscall.h>
309# include <linux/fs.h>
310#endif
311
312#ifndef D_TYPE
313# define D_TYPE(de) 0
314#endif
315#ifndef D_INO
316# define D_INO(de) 0
317#endif
318#ifndef D_NAMLEN
319# define D_NAMLEN(entp) strlen (D_NAME (entp))
320#endif
103 321
104/* used for struct dirent, AIX doesn't provide it */ 322/* used for struct dirent, AIX doesn't provide it */
105#ifndef NAME_MAX 323#ifndef NAME_MAX
106# define NAME_MAX 4096 324# define NAME_MAX 4096
107#endif 325#endif
108 326
327/* used for readlink etc. */
328#ifndef PATH_MAX
329# define PATH_MAX 0
330#endif
331
332#ifndef EIO_PATH_MIN
333# define EIO_PATH_MIN 8160
334#endif
335
336#define EIO_PATH_MAX (PATH_MAX <= EIO_PATH_MIN ? EIO_PATH_MIN : PATH_MAX)
337
109/* buffer size for various temporary buffers */ 338/* buffer size for various temporary buffers */
110#define EIO_BUFSIZE 65536 339#define EIO_BUFSIZE 65536
111 340
112#define dBUF \ 341#define dBUF \
113 char *eio_buf; \
114 ETP_WORKER_LOCK (self); \
115 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 342 char *eio_buf = malloc (EIO_BUFSIZE); \
116 ETP_WORKER_UNLOCK (self); \
117 errno = ENOMEM; \ 343 errno = ENOMEM; \
118 if (!eio_buf) \ 344 if (!eio_buf) \
119 return -1; 345 return -1
120 346
121#define EIO_TICKS ((1000000 + 1023) >> 10) 347#define FUBd \
348 free (eio_buf)
122 349
123/*****************************************************************************/ 350/*****************************************************************************/
124 351
125#if __GNUC__ >= 3 352struct etp_tmpbuf;
126# define expect(expr,value) __builtin_expect ((expr),(value)) 353
354#if _POSIX_VERSION >= 200809L
355 #define HAVE_AT 1
356 #define WD2FD(wd) ((wd) ? (wd)->fd : AT_FDCWD)
357 #ifndef O_SEARCH
358 #define O_SEARCH O_RDONLY
359 #endif
127#else 360#else
128# define expect(expr,value) (expr) 361 #define HAVE_AT 0
362 static const char *wd_expand (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path);
129#endif 363#endif
130 364
131#define expect_false(expr) expect ((expr) != 0, 0) 365struct eio_pwd
132#define expect_true(expr) expect ((expr) != 0, 1) 366{
367#if HAVE_AT
368 int fd;
369#endif
370 int len;
371 char str[1]; /* actually, a 0-terminated canonical path */
372};
133 373
134/*****************************************************************************/ 374/*****************************************************************************/
135 375
136#define ETP_PRI_MIN EIO_PRI_MIN 376#define ETP_PRI_MIN EIO_PRI_MIN
137#define ETP_PRI_MAX EIO_PRI_MAX 377#define ETP_PRI_MAX EIO_PRI_MAX
138 378
379#define ETP_TYPE_QUIT -1
380#define ETP_TYPE_GROUP EIO_GROUP
381
382static void eio_nop_callback (void) { }
383static void (*eio_want_poll_cb)(void) = eio_nop_callback;
384static void (*eio_done_poll_cb)(void) = eio_nop_callback;
385
386#define ETP_WANT_POLL(pool) eio_want_poll_cb ()
387#define ETP_DONE_POLL(pool) eio_done_poll_cb ()
388
139struct etp_worker; 389struct etp_worker;
140
141#define ETP_REQ eio_req 390#define ETP_REQ eio_req
142#define ETP_DESTROY(req) eio_destroy (req) 391#define ETP_DESTROY(req) eio_destroy (req)
143static int eio_finish (eio_req *req); 392static int eio_finish (eio_req *req);
144#define ETP_FINISH(req) eio_finish (req) 393#define ETP_FINISH(req) eio_finish (req)
145static void eio_execute (struct etp_worker *self, eio_req *req); 394static void eio_execute (struct etp_worker *self, eio_req *req);
146#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) 395#define ETP_EXECUTE(wrk,req) eio_execute (wrk, req)
147 396
148#define ETP_WORKER_CLEAR(req) \ 397#include "etp.c"
149 if (wrk->dbuf) \ 398
150 { \ 399static struct etp_pool eio_pool;
151 free (wrk->dbuf); \ 400#define EIO_POOL (&eio_pool)
152 wrk->dbuf = 0; \
153 } \
154 \
155 if (wrk->dirp) \
156 { \
157 closedir (wrk->dirp); \
158 wrk->dirp = 0; \
159 }
160#define ETP_WORKER_COMMON \
161 void *dbuf; \
162 DIR *dirp;
163 401
164/*****************************************************************************/ 402/*****************************************************************************/
165 403
166#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
167
168/* calculcate time difference in ~1/EIO_TICKS of a second */
169static int tvdiff (struct timeval *tv1, struct timeval *tv2)
170{
171 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
172 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
173}
174
175static unsigned int started, idle, wanted = 4;
176
177static void (*want_poll_cb) (void);
178static void (*done_poll_cb) (void);
179
180static unsigned int max_poll_time; /* reslock */
181static unsigned int max_poll_reqs; /* reslock */
182
183static volatile unsigned int nreqs; /* reqlock */
184static volatile unsigned int nready; /* reqlock */
185static volatile unsigned int npending; /* reqlock */
186static volatile unsigned int max_idle = 4;
187
188static mutex_t wrklock = X_MUTEX_INIT;
189static mutex_t reslock = X_MUTEX_INIT;
190static mutex_t reqlock = X_MUTEX_INIT;
191static cond_t reqwait = X_COND_INIT;
192
193#if !HAVE_PREADWRITE
194/*
195 * make our pread/pwrite emulation safe against themselves, but not against
196 * normal read/write by using a mutex. slows down execution a lot,
197 * but that's your problem, not mine.
198 */
199static mutex_t preadwritelock = X_MUTEX_INIT;
200#endif
201
202typedef struct etp_worker
203{
204 /* locked by wrklock */
205 struct etp_worker *prev, *next;
206
207 thread_t tid;
208
209 /* locked by reslock, reqlock or wrklock */
210 ETP_REQ *req; /* currently processed request */
211
212 ETP_WORKER_COMMON
213} etp_worker;
214
215static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */
216
217#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
218#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
219
220/* worker threads management */
221
222static void etp_worker_clear (etp_worker *wrk)
223{
224 ETP_WORKER_CLEAR (wrk);
225}
226
227static void etp_worker_free (etp_worker *wrk)
228{
229 wrk->next->prev = wrk->prev;
230 wrk->prev->next = wrk->next;
231
232 free (wrk);
233}
234
235static unsigned int etp_nreqs (void)
236{
237 int retval;
238 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
239 retval = nreqs;
240 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
241 return retval;
242}
243
244static unsigned int etp_nready (void)
245{
246 unsigned int retval;
247
248 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
249 retval = nready;
250 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
251
252 return retval;
253}
254
255static unsigned int etp_npending (void)
256{
257 unsigned int retval;
258
259 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
260 retval = npending;
261 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
262
263 return retval;
264}
265
266static unsigned int etp_nthreads (void)
267{
268 unsigned int retval;
269
270 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
271 retval = started;
272 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
273
274 return retval;
275}
276
277/*
278 * a somewhat faster data structure might be nice, but
279 * with 8 priorities this actually needs <20 insns
280 * per shift, the most expensive operation.
281 */
282typedef struct {
283 ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */
284 int size;
285} etp_reqq;
286
287static etp_reqq req_queue;
288static etp_reqq res_queue;
289
290static int reqq_push (etp_reqq *q, ETP_REQ *req)
291{
292 int pri = req->pri;
293 req->next = 0;
294
295 if (q->qe[pri])
296 {
297 q->qe[pri]->next = req;
298 q->qe[pri] = req;
299 }
300 else
301 q->qe[pri] = q->qs[pri] = req;
302
303 return q->size++;
304}
305
306static ETP_REQ *reqq_shift (etp_reqq *q)
307{
308 int pri;
309
310 if (!q->size)
311 return 0;
312
313 --q->size;
314
315 for (pri = ETP_NUM_PRI; pri--; )
316 {
317 eio_req *req = q->qs[pri];
318
319 if (req)
320 {
321 if (!(q->qs[pri] = (eio_req *)req->next))
322 q->qe[pri] = 0;
323
324 return req;
325 }
326 }
327
328 abort ();
329}
330
331static void etp_atfork_prepare (void)
332{
333 X_LOCK (wrklock);
334 X_LOCK (reqlock);
335 X_LOCK (reslock);
336#if !HAVE_PREADWRITE
337 X_LOCK (preadwritelock);
338#endif
339}
340
341static void etp_atfork_parent (void)
342{
343#if !HAVE_PREADWRITE
344 X_UNLOCK (preadwritelock);
345#endif
346 X_UNLOCK (reslock);
347 X_UNLOCK (reqlock);
348 X_UNLOCK (wrklock);
349}
350
351static void etp_atfork_child (void)
352{
353 ETP_REQ *prv;
354
355 while ((prv = reqq_shift (&req_queue)))
356 ETP_DESTROY (prv);
357
358 while ((prv = reqq_shift (&res_queue)))
359 ETP_DESTROY (prv);
360
361 while (wrk_first.next != &wrk_first)
362 {
363 etp_worker *wrk = wrk_first.next;
364
365 if (wrk->req)
366 ETP_DESTROY (wrk->req);
367
368 etp_worker_clear (wrk);
369 etp_worker_free (wrk);
370 }
371
372 started = 0;
373 idle = 0;
374 nreqs = 0;
375 nready = 0;
376 npending = 0;
377
378 etp_atfork_parent ();
379}
380
381static void 404static void
382etp_once_init (void)
383{
384 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
385}
386
387static int
388etp_init (void (*want_poll)(void), void (*done_poll)(void))
389{
390 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
391
392 pthread_once (&doinit, etp_once_init);
393
394 want_poll_cb = want_poll;
395 done_poll_cb = done_poll;
396
397 return 0;
398}
399
400X_THREAD_PROC (etp_proc);
401
402static void etp_start_thread (void)
403{
404 etp_worker *wrk = calloc (1, sizeof (etp_worker));
405
406 /*TODO*/
407 assert (("unable to allocate worker thread data", wrk));
408
409 X_LOCK (wrklock);
410
411 if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
412 {
413 wrk->prev = &wrk_first;
414 wrk->next = wrk_first.next;
415 wrk_first.next->prev = wrk;
416 wrk_first.next = wrk;
417 ++started;
418 }
419 else
420 free (wrk);
421
422 X_UNLOCK (wrklock);
423}
424
425static void etp_maybe_start_thread (void)
426{
427 if (expect_true (etp_nthreads () >= wanted))
428 return;
429
430 /* todo: maybe use idle here, but might be less exact */
431 if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
432 return;
433
434 etp_start_thread ();
435}
436
437static void etp_end_thread (void)
438{
439 eio_req *req = calloc (1, sizeof (eio_req));
440
441 req->type = -1;
442 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
443
444 X_LOCK (reqlock);
445 reqq_push (&req_queue, req);
446 X_COND_SIGNAL (reqwait);
447 X_UNLOCK (reqlock);
448
449 X_LOCK (wrklock);
450 --started;
451 X_UNLOCK (wrklock);
452}
453
454static int etp_poll (void)
455{
456 unsigned int maxreqs;
457 unsigned int maxtime;
458 struct timeval tv_start, tv_now;
459
460 X_LOCK (reslock);
461 maxreqs = max_poll_reqs;
462 maxtime = max_poll_time;
463 X_UNLOCK (reslock);
464
465 if (maxtime)
466 gettimeofday (&tv_start, 0);
467
468 for (;;)
469 {
470 ETP_REQ *req;
471
472 etp_maybe_start_thread ();
473
474 X_LOCK (reslock);
475 req = reqq_shift (&res_queue);
476
477 if (req)
478 {
479 --npending;
480
481 if (!res_queue.size && done_poll_cb)
482 done_poll_cb ();
483 }
484
485 X_UNLOCK (reslock);
486
487 if (!req)
488 return 0;
489
490 X_LOCK (reqlock);
491 --nreqs;
492 X_UNLOCK (reqlock);
493
494 if (expect_false (req->type == EIO_GROUP && req->size))
495 {
496 req->int1 = 1; /* mark request as delayed */
497 continue;
498 }
499 else
500 {
501 int res = ETP_FINISH (req);
502 if (expect_false (res))
503 return res;
504 }
505
506 if (expect_false (maxreqs && !--maxreqs))
507 break;
508
509 if (maxtime)
510 {
511 gettimeofday (&tv_now, 0);
512
513 if (tvdiff (&tv_start, &tv_now) >= maxtime)
514 break;
515 }
516 }
517
518 errno = EAGAIN;
519 return -1;
520}
521
522static void etp_cancel (ETP_REQ *req)
523{
524 X_LOCK (wrklock);
525 req->flags |= EIO_FLAG_CANCELLED;
526 X_UNLOCK (wrklock);
527
528 eio_grp_cancel (req);
529}
530
531static void etp_submit (ETP_REQ *req)
532{
533 req->pri -= ETP_PRI_MIN;
534
535 if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
536 if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
537
538 if (expect_false (req->type == EIO_GROUP))
539 {
540 /* I hope this is worth it :/ */
541 X_LOCK (reqlock);
542 ++nreqs;
543 X_UNLOCK (reqlock);
544
545 X_LOCK (reslock);
546
547 ++npending;
548
549 if (!reqq_push (&res_queue, req) && want_poll_cb)
550 want_poll_cb ();
551
552 X_UNLOCK (reslock);
553 }
554 else
555 {
556 X_LOCK (reqlock);
557 ++nreqs;
558 ++nready;
559 reqq_push (&req_queue, req);
560 X_COND_SIGNAL (reqwait);
561 X_UNLOCK (reqlock);
562
563 etp_maybe_start_thread ();
564 }
565}
566
567static void etp_set_max_poll_time (double nseconds)
568{
569 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
570 max_poll_time = nseconds;
571 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
572}
573
574static void etp_set_max_poll_reqs (unsigned int maxreqs)
575{
576 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
577 max_poll_reqs = maxreqs;
578 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
579}
580
581static void etp_set_max_idle (unsigned int nthreads)
582{
583 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
584 max_idle = nthreads <= 0 ? 1 : nthreads;
585 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
586}
587
588static void etp_set_min_parallel (unsigned int nthreads)
589{
590 if (wanted < nthreads)
591 wanted = nthreads;
592}
593
594static void etp_set_max_parallel (unsigned int nthreads)
595{
596 if (wanted > nthreads)
597 wanted = nthreads;
598
599 while (started > wanted)
600 etp_end_thread ();
601}
602
603/*****************************************************************************/
604
605static void grp_try_feed (eio_req *grp) 405grp_try_feed (eio_req *grp)
606{ 406{
607 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 407 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
608 { 408 {
609 grp->flags &= ~EIO_FLAG_GROUPADD; 409 grp->flags &= ~ETP_FLAG_GROUPADD;
610 410
611 EIO_FEED (grp); 411 EIO_FEED (grp);
612 412
613 /* stop if no progress has been made */ 413 /* stop if no progress has been made */
614 if (!(grp->flags & EIO_FLAG_GROUPADD)) 414 if (!(grp->flags & ETP_FLAG_GROUPADD))
615 { 415 {
616 grp->feed = 0; 416 grp->feed = 0;
617 break; 417 break;
618 } 418 }
619 } 419 }
620} 420}
621 421
422static int
622static int grp_dec (eio_req *grp) 423grp_dec (eio_req *grp)
623{ 424{
624 --grp->size; 425 --grp->size;
625 426
626 /* call feeder, if applicable */ 427 /* call feeder, if applicable */
627 grp_try_feed (grp); 428 grp_try_feed (grp);
628 429
629 /* finish, if done */ 430 /* finish, if done */
630 if (!grp->size && grp->int1) 431 if (!grp->size && grp->flags & ETP_FLAG_DELAYED)
631 return eio_finish (grp); 432 return eio_finish (grp);
632 else 433 else
633 return 0; 434 return 0;
634} 435}
635 436
437static void
636void eio_destroy (eio_req *req) 438eio_destroy (eio_req *req)
637{ 439{
638 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); 440 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
639 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); 441 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
640 442
641 EIO_DESTROY (req); 443 EIO_DESTROY (req);
642} 444}
643 445
446static int
644static int eio_finish (eio_req *req) 447eio_finish (eio_req *req)
645{ 448{
646 int res = EIO_FINISH (req); 449 int res = EIO_FINISH (req);
647 450
648 if (req->grp) 451 if (req->grp)
649 { 452 {
… …
657 if (grp->grp_first == req) 460 if (grp->grp_first == req)
658 grp->grp_first = req->grp_next; 461 grp->grp_first = req->grp_next;
659 462
660 res2 = grp_dec (grp); 463 res2 = grp_dec (grp);
661 464
662 if (!res && res2) 465 if (!res)
663 res = res2; 466 res = res2;
664 } 467 }
665 468
666 eio_destroy (req); 469 eio_destroy (req);
667 470
668 return res; 471 return res;
669} 472}
670 473
474void
671void eio_grp_cancel (eio_req *grp) 475eio_grp_cancel (eio_req *grp)
672{ 476{
673 for (grp = grp->grp_first; grp; grp = grp->grp_next) 477 etp_grp_cancel (EIO_POOL, grp);
674 eio_cancel (grp);
675} 478}
676 479
480void
677void eio_cancel (eio_req *req) 481eio_cancel (eio_req *req)
678{ 482{
679 etp_cancel (req); 483 etp_cancel (EIO_POOL, req);
680} 484}
681 485
486void
682void eio_submit (eio_req *req) 487eio_submit (eio_req *req)
683{ 488{
684 etp_submit (req); 489 etp_submit (EIO_POOL, req);
685} 490}
686 491
687unsigned int eio_nreqs (void) 492unsigned int
493eio_nreqs (void)
688{ 494{
689 return etp_nreqs (); 495 return etp_nreqs (EIO_POOL);
690} 496}
691 497
692unsigned int eio_nready (void) 498unsigned int
499eio_nready (void)
693{ 500{
694 return etp_nready (); 501 return etp_nready (EIO_POOL);
695} 502}
696 503
697unsigned int eio_npending (void) 504unsigned int
505eio_npending (void)
698{ 506{
699 return etp_npending (); 507 return etp_npending (EIO_POOL);
700} 508}
701 509
702unsigned int eio_nthreads (void) 510unsigned int ecb_cold
511eio_nthreads (void)
703{ 512{
704 return etp_nthreads (); 513 return etp_nthreads (EIO_POOL);
705} 514}
706 515
516void ecb_cold
707void eio_set_max_poll_time (double nseconds) 517eio_set_max_poll_time (double nseconds)
708{ 518{
709 etp_set_max_poll_time (nseconds); 519 etp_set_max_poll_time (EIO_POOL, nseconds);
710} 520}
711 521
522void ecb_cold
712void eio_set_max_poll_reqs (unsigned int maxreqs) 523eio_set_max_poll_reqs (unsigned int maxreqs)
713{ 524{
714 etp_set_max_poll_reqs (maxreqs); 525 etp_set_max_poll_reqs (EIO_POOL, maxreqs);
715} 526}
716 527
528void ecb_cold
717void eio_set_max_idle (unsigned int nthreads) 529eio_set_max_idle (unsigned int nthreads)
718{ 530{
719 etp_set_max_idle (nthreads); 531 etp_set_max_idle (EIO_POOL, nthreads);
720} 532}
721 533
534void ecb_cold
535eio_set_idle_timeout (unsigned int seconds)
536{
537 etp_set_idle_timeout (EIO_POOL, seconds);
538}
539
540void ecb_cold
722void eio_set_min_parallel (unsigned int nthreads) 541eio_set_min_parallel (unsigned int nthreads)
723{ 542{
724 etp_set_min_parallel (nthreads); 543 etp_set_min_parallel (EIO_POOL, nthreads);
725} 544}
726 545
546void ecb_cold
727void eio_set_max_parallel (unsigned int nthreads) 547eio_set_max_parallel (unsigned int nthreads)
728{ 548{
729 etp_set_max_parallel (nthreads); 549 etp_set_max_parallel (EIO_POOL, nthreads);
730} 550}
731 551
732int eio_poll (void) 552int eio_poll (void)
733{ 553{
734 return etp_poll (); 554 return etp_poll (EIO_POOL);
735} 555}
736 556
737/*****************************************************************************/ 557/*****************************************************************************/
738/* work around various missing functions */ 558/* work around various missing functions */
739 559
740#if !HAVE_PREADWRITE 560#if HAVE_POSIX_CLOSE && !__linux
741# undef pread 561# define eio__close(fd) posix_close (fd, 0)
742# undef pwrite 562#else
743# define pread eio__pread 563# define eio__close(fd) close (fd)
744# define pwrite eio__pwrite
745
746static ssize_t
747eio__pread (int fd, void *buf, size_t count, off_t offset)
748{
749 ssize_t res;
750 off_t ooffset;
751
752 X_LOCK (preadwritelock);
753 ooffset = lseek (fd, 0, SEEK_CUR);
754 lseek (fd, offset, SEEK_SET);
755 res = read (fd, buf, count);
756 lseek (fd, ooffset, SEEK_SET);
757 X_UNLOCK (preadwritelock);
758
759 return res;
760}
761
762static ssize_t
763eio__pwrite (int fd, void *buf, size_t count, off_t offset)
764{
765 ssize_t res;
766 off_t ooffset;
767
768 X_LOCK (preadwritelock);
769 ooffset = lseek (fd, 0, SEEK_CUR);
770 lseek (fd, offset, SEEK_SET);
771 res = write (fd, buf, count);
772 lseek (fd, offset, SEEK_SET);
773 X_UNLOCK (preadwritelock);
774
775 return res;
776}
777#endif 564#endif
778 565
566/* close() without disturbing errno */
567static void
568silent_close (int fd)
569{
570 int saved_errno = errno;
571 eio__close (fd);
572 errno = saved_errno;
573}
574
779#ifndef HAVE_FUTIMES 575#ifndef HAVE_UTIMES
780 576
781# undef utimes 577# undef utimes
782# undef futimes
783# define utimes(path,times) eio__utimes (path, times) 578# define utimes(path,times) eio__utimes (path, times)
784# define futimes(fd,times) eio__futimes (fd, times)
785 579
786static int 580static int
787eio__utimes (const char *filename, const struct timeval times[2]) 581eio__utimes (const char *filename, const struct timeval times[2])
788{ 582{
789 if (times) 583 if (times)
… …
797 } 591 }
798 else 592 else
799 return utime (filename, 0); 593 return utime (filename, 0);
800} 594}
801 595
596#endif
597
598#ifndef HAVE_FUTIMES
599
600# undef futimes
601# define futimes(fd,times) eio__futimes (fd, times)
602
603static int
802static int eio__futimes (int fd, const struct timeval tv[2]) 604eio__futimes (int fd, const struct timeval tv[2])
803{ 605{
804 errno = ENOSYS; 606 errno = ENOSYS;
805 return -1; 607 return -1;
806} 608}
807 609
… …
810#if !HAVE_FDATASYNC 612#if !HAVE_FDATASYNC
811# undef fdatasync 613# undef fdatasync
812# define fdatasync(fd) fsync (fd) 614# define fdatasync(fd) fsync (fd)
813#endif 615#endif
814 616
617static int
618eio__syncfs (int fd)
619{
620 int res;
621
622#if HAVE_SYS_SYNCFS
623 res = (int)syscall (__NR_syncfs, (int)(fd));
624#else
625 res = EIO_ENOSYS ();
626#endif
627
628 if (res < 0 && errno == ENOSYS && fd >= 0)
629 sync ();
630
631 return res;
632}
633
815/* sync_file_range always needs emulation */ 634/* sync_file_range always needs emulation */
816int 635static int
817eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags) 636eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
818{ 637{
819#if HAVE_SYNC_FILE_RANGE 638#if HAVE_SYNC_FILE_RANGE
820 int res; 639 int res;
821 640
… …
829 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0); 648 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
830 } 649 }
831 650
832 res = sync_file_range (fd, offset, nbytes, flags); 651 res = sync_file_range (fd, offset, nbytes, flags);
833 652
834 if (res != ENOSYS) 653 if (!res || errno != ENOSYS)
835 return res; 654 return res;
836#endif 655#endif
837 656
838 /* even though we could play tricks with the flags, it's better to always 657 /* even though we could play tricks with the flags, it's better to always
839 * call fdatasync, as thta matches the expectation of it's users best */ 658 * call fdatasync, as that matches the expectation of its users best */
840 return fdatasync (fd); 659 return fdatasync (fd);
660}
661
662static int
663eio__fallocate (int fd, int mode, off_t offset, size_t len)
664{
665#if HAVE_LINUX_FALLOCATE
666 return fallocate (fd, mode, offset, len);
667#else
668 return EIO_ENOSYS ();
669#endif
841} 670}
842 671
843#if !HAVE_READAHEAD 672#if !HAVE_READAHEAD
844# undef readahead 673# undef readahead
845# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 674# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
846 675
847static ssize_t 676static eio_ssize_t
848eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) 677eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
849{ 678{
850 size_t todo = count; 679 size_t todo = count;
851 dBUF; 680 dBUF;
852 681
… …
857 pread (fd, eio_buf, len, offset); 686 pread (fd, eio_buf, len, offset);
858 offset += len; 687 offset += len;
859 todo -= len; 688 todo -= len;
860 } 689 }
861 690
862 errno = 0; 691 FUBd;
692
693 /* linux's readahead basically only fails for EBADF or EINVAL (not mmappable) */
694 /* but not for e.g. EIO or eof, so we also never fail */
863 return count; 695 return 0;
864} 696}
865 697
866#endif 698#endif
867 699
868/* sendfile always needs emulation */ 700/* sendfile always needs emulation */
869static ssize_t 701static eio_ssize_t
870eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) 702eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
871{ 703{
704 eio_ssize_t written = 0;
872 ssize_t res; 705 eio_ssize_t res;
873 706
874 if (!count) 707 if (!count)
875 return 0; 708 return 0;
876 709
710 for (;;)
711 {
712#ifdef __APPLE__
713# undef HAVE_SENDFILE /* broken, as everything on os x */
714#endif
877#if HAVE_SENDFILE 715#if HAVE_SENDFILE
878# if __linux 716# if __linux
717 off_t soffset = offset;
879 res = sendfile (ofd, ifd, &offset, count); 718 res = sendfile (ofd, ifd, &soffset, count);
880 719
881# elif __freebsd 720# elif __FreeBSD__
882 /* 721 /*
883 * Of course, the freebsd sendfile is a dire hack with no thoughts 722 * Of course, the freebsd sendfile is a dire hack with no thoughts
884 * wasted on making it similar to other I/O functions. 723 * wasted on making it similar to other I/O functions.
885 */ 724 */
886 {
887 off_t sbytes; 725 off_t sbytes;
888 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 726 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
889 727
890 if (res < 0 && sbytes) 728 #if 0 /* according to the manpage, this is correct, but broken behaviour */
891 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 729 /* freebsd' sendfile will return 0 on success */
730 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
731 /* not on e.g. EIO or EPIPE - sounds broken */
732 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
892 res = sbytes; 733 res = sbytes;
893 } 734 #endif
735
736 /* according to source inspection, this is correct, and useful behaviour */
737 if (sbytes)
738 res = sbytes;
739
740# elif defined __APPLE__
741 off_t sbytes = count;
742 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
743
744 /* according to the manpage, sbytes is always valid */
745 if (sbytes)
746 res = sbytes;
894 747
895# elif __hpux 748# elif __hpux
896 res = sendfile (ofd, ifd, offset, count, 0, 0); 749 res = sendfile (ofd, ifd, offset, count, 0, 0);
897 750
898# elif __solaris 751# elif __solaris
899 {
900 struct sendfilevec vec; 752 struct sendfilevec vec;
901 size_t sbytes; 753 size_t sbytes;
902 754
903 vec.sfv_fd = ifd; 755 vec.sfv_fd = ifd;
904 vec.sfv_flag = 0; 756 vec.sfv_flag = 0;
905 vec.sfv_off = offset; 757 vec.sfv_off = offset;
906 vec.sfv_len = count; 758 vec.sfv_len = count;
907 759
908 res = sendfilev (ofd, &vec, 1, &sbytes); 760 res = sendfilev (ofd, &vec, 1, &sbytes);
909 761
910 if (res < 0 && sbytes) 762 if (res < 0 && sbytes)
911 res = sbytes; 763 res = sbytes;
912 }
913 764
914# endif 765# endif
766
767#elif defined (_WIN32) && 0
768 /* does not work, just for documentation of what would need to be done */
769 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
770 /* libeio guarantees that the file offset does not change, and windows */
771 /* has no way to get an independent handle to the same file description */
772 HANDLE h = TO_SOCKET (ifd);
773 SetFilePointer (h, offset, 0, FILE_BEGIN);
774 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
775
915#else 776#else
916 res = -1; 777 res = EIO_ENOSYS ();
917 errno = ENOSYS;
918#endif 778#endif
919 779
780 /* we assume sendfile can copy at least 128mb in one go */
781 if (res <= 128 * 1024 * 1024)
782 {
783 if (res > 0)
784 written += res;
785
786 if (written)
787 return written;
788
789 break;
790 }
791 else
792 {
793 /* if we requested more, then probably the kernel was lazy */
794 written += res;
795 offset += res;
796 count -= res;
797
798 if (!count)
799 return written;
800 }
801 }
802
920 if (res < 0 803 if (res < 0
921 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 804 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
805 /* BSDs */
806#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
807 || errno == ENOTSUP
808#endif
809#ifdef EOPNOTSUPP /* windows */
810 || errno == EOPNOTSUPP /* BSDs */
811#endif
922#if __solaris 812#if __solaris
923 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 813 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
924#endif 814#endif
925 ) 815 )
926 ) 816 )
… …
930 820
931 res = 0; 821 res = 0;
932 822
933 while (count) 823 while (count)
934 { 824 {
935 ssize_t cnt; 825 eio_ssize_t cnt;
936 826
937 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 827 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
938 828
939 if (cnt <= 0) 829 if (cnt <= 0)
940 { 830 {
… …
952 842
953 offset += cnt; 843 offset += cnt;
954 res += cnt; 844 res += cnt;
955 count -= cnt; 845 count -= cnt;
956 } 846 }
847
848 FUBd;
957 } 849 }
958 850
959 return res; 851 return res;
852}
853
854#ifdef PAGESIZE
855# define eio_pagesize() PAGESIZE
856#else
857static intptr_t
858eio_pagesize (void)
859{
860 static intptr_t page;
861
862 if (!page)
863 page = sysconf (_SC_PAGESIZE);
864
865 return page;
866}
867#endif
868
869static void
870eio_page_align (void **addr, size_t *length)
871{
872 intptr_t mask = eio_pagesize () - 1;
873
874 /* round down addr */
875 intptr_t adj = mask & (intptr_t)*addr;
876
877 *addr = (void *)((intptr_t)*addr - adj);
878 *length += adj;
879
880 /* round up length */
881 *length = (*length + mask) & ~mask;
882}
883
884#if !_POSIX_MEMLOCK
885# define eio__mlockall(a) EIO_ENOSYS ()
886#else
887
888static int
889eio__mlockall (int flags)
890{
891 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
892 extern int mallopt (int, int);
893 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
894 #endif
895
896 if (EIO_MCL_CURRENT != MCL_CURRENT
897 || EIO_MCL_FUTURE != MCL_FUTURE)
898 {
899 flags = 0
900 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
901 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
902 }
903
904 return mlockall (flags);
905}
906#endif
907
908#if !_POSIX_MEMLOCK_RANGE
909# define eio__mlock(a,b) EIO_ENOSYS ()
910#else
911
912static int
913eio__mlock (void *addr, size_t length)
914{
915 eio_page_align (&addr, &length);
916
917 return mlock (addr, length);
918}
919
920#endif
921
922#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
923# define eio__msync(a,b,c) EIO_ENOSYS ()
924#else
925
926static int
927eio__msync (void *mem, size_t len, int flags)
928{
929 eio_page_align (&mem, &len);
930
931 if (EIO_MS_ASYNC != MS_SYNC
932 || EIO_MS_INVALIDATE != MS_INVALIDATE
933 || EIO_MS_SYNC != MS_SYNC)
934 {
935 flags = 0
936 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
937 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
938 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
939 }
940
941 return msync (mem, len, flags);
942}
943
944#endif
945
946static int
947eio__mtouch (eio_req *req)
948{
949 void *mem = req->ptr2;
950 size_t len = req->size;
951 int flags = req->int1;
952
953 eio_page_align (&mem, &len);
954
955 {
956 intptr_t addr = (intptr_t)mem;
957 intptr_t end = addr + len;
958 intptr_t page = eio_pagesize ();
959
960 if (addr < end)
961 if (flags & EIO_MT_MODIFY) /* modify */
962 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
963 else
964 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
965 }
966
967 return 0;
968}
969
970/*****************************************************************************/
971/* requests implemented outside eio_execute, because they are so large */
972
973static void
974eio__lseek (eio_req *req)
975{
976 /* this usually gets optimised away completely, or your compiler sucks, */
977 /* or the whence constants really are not 0, 1, 2 */
978 int whence = req->int2 == EIO_SEEK_SET ? SEEK_SET
979 : req->int2 == EIO_SEEK_CUR ? SEEK_CUR
980 : req->int2 == EIO_SEEK_END ? SEEK_END
981 : req->int2;
982
983 req->offs = lseek (req->int1, req->offs, whence);
984 req->result = req->offs == (off_t)-1 ? -1 : 0;
985}
986
987/* result will always end up in tmpbuf, there is always space for adding a 0-byte */
988static int
989eio__realpath (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
990{
991 char *res;
992 const char *rel = path;
993 char *tmp1, *tmp2;
994#if SYMLOOP_MAX > 32
995 int symlinks = SYMLOOP_MAX;
996#else
997 int symlinks = 32;
998#endif
999
1000 errno = EINVAL;
1001 if (!rel)
1002 return -1;
1003
1004 errno = ENOENT;
1005 if (!*rel)
1006 return -1;
1007
1008 res = etp_tmpbuf_get (tmpbuf, EIO_PATH_MAX * 3);
1009#ifdef _WIN32
1010 if (_access (rel, 4) != 0)
1011 return -1;
1012
1013 symlinks = GetFullPathName (rel, EIO_PATH_MAX * 3, res, 0);
1014
1015 errno = ENAMETOOLONG;
1016 if (symlinks >= EIO_PATH_MAX * 3)
1017 return -1;
1018
1019 errno = EIO;
1020 if (symlinks <= 0)
1021 return -1;
1022
1023 return symlinks;
1024
1025#else
1026 tmp1 = res + EIO_PATH_MAX;
1027 tmp2 = tmp1 + EIO_PATH_MAX;
1028
1029#if 0 /* disabled, the musl way to do things is just too racy */
1030#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1031 /* on linux we may be able to ask the kernel */
1032 {
1033 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1034
1035 if (fd >= 0)
1036 {
1037 sprintf (tmp1, "/proc/self/fd/%d", fd);
1038 req->result = readlink (tmp1, res, EIO_PATH_MAX);
1039 /* here we should probably stat the open file and the disk file, to make sure they still match */
1040 eio__close (fd);
1041
1042 if (req->result > 0)
1043 goto done;
1044 }
1045 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1046 return -1;
1047 }
1048#endif
1049#endif
1050
1051 if (*rel != '/')
1052 {
1053 int len;
1054
1055 errno = ENOENT;
1056 if (wd == EIO_INVALID_WD)
1057 return -1;
1058
1059 if (wd == EIO_CWD)
1060 {
1061 if (!getcwd (res, EIO_PATH_MAX))
1062 return -1;
1063
1064 len = strlen (res);
1065 }
1066 else
1067 memcpy (res, wd->str, len = wd->len);
1068
1069 if (res [1]) /* only use if not / */
1070 res += len;
1071 }
1072
1073 while (*rel)
1074 {
1075 eio_ssize_t len, linklen;
1076 const char *beg = rel;
1077
1078 while (*rel && *rel != '/')
1079 ++rel;
1080
1081 len = rel - beg;
1082
1083 if (!len) /* skip slashes */
1084 {
1085 ++rel;
1086 continue;
1087 }
1088
1089 if (beg [0] == '.')
1090 {
1091 if (len == 1)
1092 continue; /* . - nop */
1093
1094 if (beg [1] == '.' && len == 2)
1095 {
1096 /* .. - back up one component, if possible */
1097
1098 while (res != tmpbuf->ptr)
1099 if (*--res == '/')
1100 break;
1101
1102 continue;
1103 }
1104 }
1105
1106 errno = ENAMETOOLONG;
1107 if (res + 1 + len + 1 >= tmp1)
1108 return -1;
1109
1110 /* copy one component */
1111 *res = '/';
1112 memcpy (res + 1, beg, len);
1113
1114 /* zero-terminate, for readlink */
1115 res [len + 1] = 0;
1116
1117 /* now check if it's a symlink */
1118 linklen = readlink (tmpbuf->ptr, tmp1, EIO_PATH_MAX);
1119
1120 if (linklen < 0)
1121 {
1122 if (errno != EINVAL)
1123 return -1;
1124
1125 /* it's a normal directory. hopefully */
1126 res += len + 1;
1127 }
1128 else
1129 {
1130 /* yay, it was a symlink - build new path in tmp2 */
1131 int rellen = strlen (rel);
1132
1133 errno = ENAMETOOLONG;
1134 if (linklen + 1 + rellen >= EIO_PATH_MAX) /* also catch linklen >= EIO_PATH_MAX */
1135 return -1;
1136
1137 errno = ELOOP;
1138 if (!--symlinks)
1139 return -1;
1140
1141 if (*tmp1 == '/')
1142 res = tmpbuf->ptr; /* symlink resolves to an absolute path */
1143
1144 /* we need to be careful, as rel might point into tmp2 already */
1145 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1146 tmp2 [linklen] = '/';
1147 memcpy (tmp2, tmp1, linklen);
1148
1149 rel = tmp2;
1150 }
1151 }
1152
1153 /* special case for the lone root path */
1154 if (res == tmpbuf->ptr)
1155 *res++ = '/';
1156
1157 return res - (char *)tmpbuf->ptr;
1158#endif
1159}
1160
1161static signed char
1162eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1163{
1164 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1165 : a->inode < b->inode ? -1
1166 : a->inode > b->inode ? 1
1167 : 0;
1168}
1169
1170#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1171
1172#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1173#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1174
1175static void
1176eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1177{
1178 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1179 unsigned char *bit = bits;
1180
1181 assert (CHAR_BIT == 8);
1182 assert (sizeof (eio_dirent) * 8 < 256);
1183 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1184 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1185
1186 if (size <= EIO_SORT_FAST)
1187 return;
1188
1189 /* first prepare an array of bits to test in our radix sort */
1190 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1191 /* inode_bits must contain all inodes ORed together */
1192 /* which is used to skip bits that are 0 everywhere, which is very common */
1193 {
1194 eio_ino_t endianness;
1195 int i, j;
1196
1197 /* we store the byte offset of byte n into byte n of "endianness" */
1198 for (i = 0; i < sizeof (eio_ino_t); ++i)
1199 ((unsigned char *)&endianness)[i] = i;
1200
1201 *bit++ = 0;
1202
1203 for (i = 0; i < sizeof (eio_ino_t); ++i)
1204 {
1205 /* shifting off the byte offsets out of "endianness" */
1206 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1207 endianness >>= 8;
1208
1209 for (j = 0; j < 8; ++j)
1210 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1211 *bit++ = offs + j;
1212 }
1213
1214 for (j = 0; j < 8; ++j)
1215 if (score_bits & (1 << j))
1216 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1217 }
1218
1219 /* now actually do the sorting (a variant of MSD radix sort) */
1220 {
1221 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1222 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1223 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1224 int stk_idx = 0;
1225
1226 base_stk [stk_idx] = dents;
1227 end_stk [stk_idx] = dents + size;
1228 bit_stk [stk_idx] = bit - 1;
1229
1230 do
1231 {
1232 base = base_stk [stk_idx];
1233 end = end_stk [stk_idx];
1234 bit = bit_stk [stk_idx];
1235
1236 for (;;)
1237 {
1238 unsigned char O = *bit >> 3;
1239 unsigned char M = 1 << (*bit & 7);
1240
1241 eio_dirent *a = base;
1242 eio_dirent *b = end;
1243
1244 if (b - a < EIO_SORT_CUTOFF)
1245 break;
1246
1247 /* now bit-partition the array on the bit */
1248 /* this ugly asymmetric loop seems to perform much better than typical */
1249 /* partition algos found in the literature */
1250 do
1251 if (!(((unsigned char *)a)[O] & M))
1252 ++a;
1253 else if (!(((unsigned char *)--b)[O] & M))
1254 {
1255 eio_dirent tmp = *a; *a = *b; *b = tmp;
1256 ++a;
1257 }
1258 while (b > a);
1259
1260 /* next bit, or stop, if no bits left in this path */
1261 if (!*--bit)
1262 break;
1263
1264 base_stk [stk_idx] = a;
1265 end_stk [stk_idx] = end;
1266 bit_stk [stk_idx] = bit;
1267 ++stk_idx;
1268
1269 end = a;
1270 }
1271 }
1272 while (stk_idx--);
1273 }
1274}
1275
1276static void
1277eio_dent_insertion_sort (eio_dirent *dents, int size)
1278{
1279 /* first move the smallest element to the front, to act as a sentinel */
1280 {
1281 int i;
1282 eio_dirent *min = dents;
1283
1284 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1285 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1286 if (EIO_DENT_CMP (dents [i], <, *min))
1287 min = &dents [i];
1288
1289 /* swap elements 0 and j (minimum) */
1290 {
1291 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1292 }
1293 }
1294
1295 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1296 {
1297 eio_dirent *i, *j;
1298
1299 for (i = dents + 1; i < dents + size; ++i)
1300 {
1301 eio_dirent value = *i;
1302
1303 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1304 j [1] = j [0];
1305
1306 j [1] = value;
1307 }
1308 }
1309}
1310
1311static void
1312eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1313{
1314 if (size <= 1)
1315 return; /* our insertion sort relies on size > 0 */
1316
1317 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1318 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1319 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1320
1321 /* use an insertion sort at the end, or for small arrays, */
1322 /* as insertion sort is more efficient for small partitions */
1323 eio_dent_insertion_sort (dents, size);
960} 1324}
961 1325
962/* read a full directory */ 1326/* read a full directory */
963static void 1327static void
964eio__scandir (eio_req *req, etp_worker *self) 1328eio__scandir (eio_req *req, etp_worker *self)
965{ 1329{
1330 char *name, *names;
1331 int namesalloc = 4096 - sizeof (void *) * 4;
1332 int namesoffs = 0;
1333 int flags = req->int1;
1334 eio_dirent *dents = 0;
1335 int dentalloc = 128;
1336 int dentoffs = 0;
1337 eio_ino_t inode_bits = 0;
1338#ifdef _WIN32
1339 HANDLE dirp;
1340 WIN32_FIND_DATA entp;
1341#else
966 DIR *dirp; 1342 DIR *dirp;
967 EIO_STRUCT_DIRENT *entp; 1343 EIO_STRUCT_DIRENT *entp;
968 char *name, *names; 1344#endif
969 int memlen = 4096;
970 int memofs = 0;
971 int res = 0;
972 1345
973 X_LOCK (wrklock); 1346 req->result = -1;
974 /* the corresponding closedir is in ETP_WORKER_CLEAR */
975 self->dirp = dirp = opendir (req->ptr1);
976 req->flags |= EIO_FLAG_PTR2_FREE;
977 req->ptr2 = names = malloc (memlen);
978 X_UNLOCK (wrklock);
979 1347
980 if (dirp && names) 1348 if (!(flags & EIO_READDIR_DENTS))
981 for (;;) 1349 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1350
1351#ifdef _WIN32
1352 {
1353 int len = strlen ((const char *)req->ptr1);
1354 char *path = malloc (MAX_PATH);
1355 const char *fmt;
1356 const char *reqpath = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1357
1358 if (!len)
1359 fmt = "./*";
1360 else if (reqpath[len - 1] == '/' || reqpath[len - 1] == '\\')
1361 fmt = "%s*";
1362 else
1363 fmt = "%s/*";
1364
1365 _snprintf (path, MAX_PATH, fmt, reqpath);
1366 dirp = FindFirstFile (path, &entp);
1367 free (path);
1368
1369 if (dirp == INVALID_HANDLE_VALUE)
1370 {
1371 /* should steal _dosmaperr */
1372 switch (GetLastError ())
1373 {
1374 case ERROR_FILE_NOT_FOUND:
1375 req->result = 0;
1376 break;
1377
1378 case ERROR_INVALID_NAME:
1379 case ERROR_PATH_NOT_FOUND:
1380 case ERROR_NO_MORE_FILES:
1381 errno = ENOENT;
1382 break;
1383
1384 case ERROR_NOT_ENOUGH_MEMORY:
1385 errno = ENOMEM;
1386 break;
1387
1388 default:
1389 errno = EINVAL;
1390 break;
1391 }
1392
1393 return;
1394 }
1395 }
1396#else
1397 #if HAVE_AT
1398 if (req->wd)
982 { 1399 {
1400 int fd = openat (WD2FD (req->wd), req->ptr1, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1401
1402 if (fd < 0)
1403 return;
1404
1405 dirp = fdopendir (fd);
1406
1407 if (!dirp)
1408 silent_close (fd);
1409 }
1410 else
1411 dirp = opendir (req->ptr1);
1412 #else
1413 dirp = opendir (wd_expand (&self->tmpbuf, req->wd, req->ptr1));
1414 #endif
1415
1416 if (!dirp)
1417 return;
1418#endif
1419
1420 if (req->flags & EIO_FLAG_PTR1_FREE)
1421 free (req->ptr1);
1422
1423 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1424 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1425 req->ptr2 = names = malloc (namesalloc);
1426
1427 if (!names || (flags && !dents))
1428 return;
1429
1430 for (;;)
1431 {
1432 int done;
1433
1434#ifdef _WIN32
1435 done = !dirp;
1436#else
983 errno = 0; 1437 errno = 0;
984 entp = readdir (dirp); 1438 entp = readdir (dirp);
1439 done = !entp;
1440#endif
985 1441
1442 if (done)
1443 {
1444#ifndef _WIN32
1445 int old_errno = errno;
1446 closedir (dirp);
1447 errno = old_errno;
1448
986 if (!entp) 1449 if (errno)
1450 break;
1451#endif
1452
1453 /* sort etc. */
1454 req->int1 = flags;
1455 req->result = dentoffs;
1456
1457 if (flags & EIO_READDIR_STAT_ORDER)
1458 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1459 else if (flags & EIO_READDIR_DIRS_FIRST)
1460 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1461 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1462 else
1463 {
1464 /* in this case, all is known, and we just put dirs first and sort them */
1465 eio_dirent *oth = dents + dentoffs;
1466 eio_dirent *dir = dents;
1467
1468 /* now partition dirs to the front, and non-dirs to the back */
1469 /* by walking from both sides and swapping if necessary */
1470 while (oth > dir)
1471 {
1472 if (dir->type == EIO_DT_DIR)
1473 ++dir;
1474 else if ((--oth)->type == EIO_DT_DIR)
1475 {
1476 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1477
1478 ++dir;
1479 }
1480 }
1481
1482 /* now sort the dirs only (dirs all have the same score) */
1483 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1484 }
1485
987 break; 1486 break;
1487 }
988 1488
989 name = entp->d_name; 1489 /* now add the entry to our list(s) */
1490 name = D_NAME (entp);
990 1491
1492 /* skip . and .. entries */
991 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1493 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
992 { 1494 {
993 int len = strlen (name) + 1; 1495 int len = D_NAMLEN (entp) + 1;
994 1496
995 res++; 1497 while (ecb_expect_false (namesoffs + len > namesalloc))
996
997 while (memofs + len > memlen)
998 { 1498 {
999 memlen *= 2; 1499 namesalloc *= 2;
1000 X_LOCK (wrklock);
1001 req->ptr2 = names = realloc (names, memlen); 1500 req->ptr2 = names = realloc (names, namesalloc);
1002 X_UNLOCK (wrklock);
1003 1501
1004 if (!names) 1502 if (!names)
1503 break;
1504 }
1505
1506 memcpy (names + namesoffs, name, len);
1507
1508 if (dents)
1509 {
1510 struct eio_dirent *ent;
1511
1512 if (ecb_expect_false (dentoffs == dentalloc))
1513 {
1514 dentalloc *= 2;
1515 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1516
1517 if (!dents)
1005 break; 1518 break;
1006 } 1519 }
1007 1520
1008 memcpy (names + memofs, name, len); 1521 ent = dents + dentoffs;
1009 memofs += len; 1522
1523 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1524 ent->namelen = len - 1;
1525 ent->inode = D_INO (entp);
1526
1527 inode_bits |= ent->inode;
1528
1529 switch (D_TYPE (entp))
1530 {
1531 default:
1532 ent->type = EIO_DT_UNKNOWN;
1533 flags |= EIO_READDIR_FOUND_UNKNOWN;
1534 break;
1535
1536 #ifdef DT_FIFO
1537 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1538 #endif
1539 #ifdef DT_CHR
1540 case DT_CHR: ent->type = EIO_DT_CHR; break;
1541 #endif
1542 #ifdef DT_MPC
1543 case DT_MPC: ent->type = EIO_DT_MPC; break;
1544 #endif
1545 #ifdef DT_DIR
1546 case DT_DIR: ent->type = EIO_DT_DIR; break;
1547 #endif
1548 #ifdef DT_NAM
1549 case DT_NAM: ent->type = EIO_DT_NAM; break;
1550 #endif
1551 #ifdef DT_BLK
1552 case DT_BLK: ent->type = EIO_DT_BLK; break;
1553 #endif
1554 #ifdef DT_MPB
1555 case DT_MPB: ent->type = EIO_DT_MPB; break;
1556 #endif
1557 #ifdef DT_REG
1558 case DT_REG: ent->type = EIO_DT_REG; break;
1559 #endif
1560 #ifdef DT_NWK
1561 case DT_NWK: ent->type = EIO_DT_NWK; break;
1562 #endif
1563 #ifdef DT_CMP
1564 case DT_CMP: ent->type = EIO_DT_CMP; break;
1565 #endif
1566 #ifdef DT_LNK
1567 case DT_LNK: ent->type = EIO_DT_LNK; break;
1568 #endif
1569 #ifdef DT_SOCK
1570 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1571 #endif
1572 #ifdef DT_DOOR
1573 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1574 #endif
1575 #ifdef DT_WHT
1576 case DT_WHT: ent->type = EIO_DT_WHT; break;
1577 #endif
1578 }
1579
1580 ent->score = 7;
1581
1582 if (flags & EIO_READDIR_DIRS_FIRST)
1583 {
1584 if (ent->type == EIO_DT_UNKNOWN)
1585 {
1586 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1587 ent->score = 1;
1588 else if (!strchr (name, '.')) /* absence of dots indicate likely dirs */
1589 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1590 }
1591 else if (ent->type == EIO_DT_DIR)
1592 ent->score = 0;
1593 }
1010 } 1594 }
1595
1596 namesoffs += len;
1597 ++dentoffs;
1011 } 1598 }
1012 1599
1013 if (errno) 1600 if (EIO_CANCELLED (req))
1014 res = -1; 1601 {
1602 errno = ECANCELED;
1603 break;
1604 }
1605
1606#ifdef _WIN32
1607 if (!FindNextFile (dirp, &entp))
1608 {
1609 FindClose (dirp);
1610 dirp = 0;
1611 }
1612#endif
1613 }
1614}
1615
1616/*****************************************************************************/
1617/* working directory stuff */
1618/* various deficiencies in the posix 2008 api force us to */
1619/* keep the absolute path in string form at all times */
1620/* fuck yeah. */
1621
1622#if !HAVE_AT
1623
1624/* a bit like realpath, but usually faster because it doesn'T have to return */
1625/* an absolute or canonical path */
1626static const char *
1627wd_expand (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
1628{
1629 if (!wd || *path == '/')
1630 return path;
1631
1632 if (path [0] == '.' && !path [1])
1633 return wd->str;
1634
1015 1635 {
1016 req->result = res; 1636 int l1 = wd->len;
1017} 1637 int l2 = strlen (path);
1018 1638
1019#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) 1639 char *res = etp_tmpbuf_get (tmpbuf, l1 + l2 + 2);
1020# undef msync
1021# define msync(a,b,c) ENOSYS
1022#endif
1023 1640
1024int 1641 memcpy (res, wd->str, l1);
1025eio__mtouch (void *mem, size_t len, int flags) 1642 res [l1] = '/';
1643 memcpy (res + l1 + 1, path, l2 + 1);
1644
1645 return res;
1646 }
1647}
1648
1649#endif
1650
1651static eio_wd
1652eio__wd_open_sync (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
1026{ 1653{
1027 intptr_t addr = (intptr_t)mem; 1654 int fd;
1028 intptr_t end = addr + len; 1655 eio_wd res;
1029#ifdef PAGESIZE 1656 int len = eio__realpath (tmpbuf, wd, path);
1030 const intptr_t page = PAGESIZE; 1657
1658 if (len < 0)
1659 return EIO_INVALID_WD;
1660
1661#if HAVE_AT
1662 fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1663
1664 if (fd < 0)
1665 return EIO_INVALID_WD;
1666#endif
1667
1668 res = malloc (sizeof (*res) + len); /* one extra 0-byte */
1669
1670#if HAVE_AT
1671 res->fd = fd;
1672#endif
1673
1674 res->len = len;
1675 memcpy (res->str, tmpbuf->ptr, len);
1676 res->str [len] = 0;
1677
1678 return res;
1679}
1680
1681eio_wd
1682eio_wd_open_sync (eio_wd wd, const char *path)
1683{
1684 struct etp_tmpbuf tmpbuf = { };
1685 wd = eio__wd_open_sync (&tmpbuf, wd, path);
1686 free (tmpbuf.ptr);
1687
1688 return wd;
1689}
1690
1691void
1692eio_wd_close_sync (eio_wd wd)
1693{
1694 if (wd != EIO_INVALID_WD && wd != EIO_CWD)
1695 {
1696 #if HAVE_AT
1697 eio__close (wd->fd);
1698 #endif
1699 free (wd);
1700 }
1701}
1702
1703#if HAVE_AT
1704
1705static int
1706eio__renameat2 (int olddirfd, const char *oldpath, int newdirfd, const char *newpath, unsigned int flags)
1707{
1708#if HAVE_RENAMEAT2
1709 return syscall (SYS_renameat2, olddirfd, oldpath, newdirfd, newpath, flags);
1031#else 1710#else
1032 static intptr_t page;
1033
1034 if (!page) 1711 if (flags)
1035 page = sysconf (_SC_PAGESIZE); 1712 return EIO_ENOSYS ();
1036#endif
1037 1713
1038 addr &= ~(page - 1); /* assume page size is always a power of two */ 1714 return renameat (olddirfd, oldpath, newdirfd, newpath);
1715#endif
1716}
1039 1717
1040 if (addr < end) 1718/* they forgot these */
1041 if (flags) /* modify */
1042 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
1043 else
1044 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
1045 1719
1720static int
1721eio__truncateat (int dirfd, const char *path, off_t length)
1722{
1723 int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
1724 int res;
1725
1726 if (fd < 0)
1727 return fd;
1728
1729 res = ftruncate (fd, length);
1730 silent_close (fd);
1046 return 0; 1731 return res;
1047} 1732}
1733
1734static int
1735eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
1736{
1737 int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
1738 int res;
1739
1740 if (fd < 0)
1741 return fd;
1742
1743 res = fstatvfs (fd, buf);
1744 silent_close (fd);
1745 return res;
1746}
1747
1748#endif
1048 1749
1049/*****************************************************************************/ 1750/*****************************************************************************/
1050 1751
1051#define ALLOC(len) \ 1752#define ALLOC(len) \
1052 if (!req->ptr2) \ 1753 if (!req->ptr2) \
1053 { \ 1754 { \
1054 X_LOCK (wrklock); \ 1755 X_LOCK (EIO_POOL->wrklock); \
1055 req->flags |= EIO_FLAG_PTR2_FREE; \ 1756 req->flags |= EIO_FLAG_PTR2_FREE; \
1056 X_UNLOCK (wrklock); \ 1757 X_UNLOCK (EIO_POOL->wrklock); \
1057 req->ptr2 = malloc (len); \ 1758 req->ptr2 = malloc (len); \
1058 if (!req->ptr2) \ 1759 if (!req->ptr2) \
1059 { \ 1760 { \
1060 errno = ENOMEM; \ 1761 errno = ENOMEM; \
1061 req->result = -1; \ 1762 req->result = -1; \
1062 break; \ 1763 goto alloc_fail; \
1063 } \ 1764 } \
1064 } 1765 }
1065 1766
1066X_THREAD_PROC (etp_proc) 1767/*****************************************************************************/
1067{
1068 ETP_REQ *req;
1069 struct timespec ts;
1070 etp_worker *self = (etp_worker *)thr_arg;
1071 1768
1072 /* try to distribute timeouts somewhat randomly */ 1769static void
1073 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 1770eio__slurp (int fd, eio_req *req)
1771{
1772 req->result = fd;
1074 1773
1075 for (;;) 1774 if (fd < 0)
1775 return;
1776
1777 if (req->offs < 0 || !req->size) /* do we need the size? */
1076 { 1778 {
1077 X_LOCK (reqlock); 1779 off_t size = lseek (fd, 0, SEEK_END);
1078 1780
1079 for (;;)
1080 {
1081 self->req = req = reqq_shift (&req_queue);
1082
1083 if (req)
1084 break;
1085
1086 ++idle;
1087
1088 ts.tv_sec = time (0) + IDLE_TIMEOUT;
1089 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
1090 {
1091 if (idle > max_idle)
1092 {
1093 --idle;
1094 X_UNLOCK (reqlock);
1095 X_LOCK (wrklock);
1096 --started;
1097 X_UNLOCK (wrklock);
1098 goto quit;
1099 }
1100
1101 /* we are allowed to idle, so do so without any timeout */
1102 X_COND_WAIT (reqwait, reqlock);
1103 }
1104
1105 --idle;
1106 }
1107
1108 --nready;
1109
1110 X_UNLOCK (reqlock);
1111
1112 if (req->type < 0) 1781 if (req->offs < 0)
1113 goto quit; 1782 req->offs += size;
1114 1783
1115 if (!EIO_CANCELLED (req)) 1784 if (!req->size)
1116 ETP_EXECUTE (self, req); 1785 req->size = size - req->offs;
1117
1118 X_LOCK (reslock);
1119
1120 ++npending;
1121
1122 if (!reqq_push (&res_queue, req) && want_poll_cb)
1123 want_poll_cb ();
1124
1125 self->req = 0;
1126 etp_worker_clear (self);
1127
1128 X_UNLOCK (reslock);
1129 } 1786 }
1130 1787
1131quit: 1788 ALLOC (req->size);
1132 X_LOCK (wrklock); 1789 req->result = pread (fd, req->ptr2, req->size, req->offs);
1133 etp_worker_free (self);
1134 X_UNLOCK (wrklock);
1135 1790
1136 return 0; 1791 silent_close (fd);
1137}
1138 1792
1139/*****************************************************************************/ 1793alloc_fail:
1794 ;
1795}
1140 1796
1797int ecb_cold
1141int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1798eio_init (void (*want_poll)(void), void (*done_poll)(void))
1142{ 1799{
1143 return etp_init (want_poll, done_poll); 1800 eio_want_poll_cb = want_poll;
1144} 1801 eio_done_poll_cb = done_poll;
1145 1802
1803 return etp_init (EIO_POOL, 0, 0, 0);
1804}
1805
1806ecb_inline void
1146static void eio_api_destroy (eio_req *req) 1807eio_api_destroy (eio_req *req)
1147{ 1808{
1148 free (req); 1809 free (req);
1149} 1810}
1150 1811
1151#define REQ(rtype) \ 1812#define REQ(rtype) \
1152 eio_req *req; \ 1813 eio_req *req; \
1153 \ 1814 \
1154 req = (eio_req *)calloc (1, sizeof *req); \ 1815 req = (eio_req *)calloc (1, sizeof *req); \
1155 if (!req) \ 1816 if (!req) \
1156 return 0; \ 1817 return 0; \
… …
1170 { \ 1831 { \
1171 eio_api_destroy (req); \ 1832 eio_api_destroy (req); \
1172 return 0; \ 1833 return 0; \
1173 } 1834 }
1174 1835
1836#define SINGLEDOT(ptr) (0[(char *)(ptr)] == '.' && !1[(char *)(ptr)])
1837
1838static void
1175static void eio_execute (etp_worker *self, eio_req *req) 1839eio_execute (etp_worker *self, eio_req *req)
1176{ 1840{
1177 errno = 0; 1841#if HAVE_AT
1842 int dirfd;
1843#else
1844 const char *path;
1845#endif
1846
1847 if (ecb_expect_false (EIO_CANCELLED (req)))
1848 {
1849 req->result = -1;
1850 req->errorno = ECANCELED;
1851 return;
1852 }
1853
1854 if (ecb_expect_false (req->wd == EIO_INVALID_WD))
1855 {
1856 req->result = -1;
1857 req->errorno = ENOENT;
1858 return;
1859 }
1860
1861 if (req->type >= EIO_OPEN)
1862 {
1863 #if HAVE_AT
1864 dirfd = WD2FD (req->wd);
1865 #else
1866 path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1867 #endif
1868 }
1178 1869
1179 switch (req->type) 1870 switch (req->type)
1180 { 1871 {
1872 case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
1873 req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
1874 break;
1875 case EIO_WD_CLOSE: req->result = 0;
1876 eio_wd_close_sync (req->wd); break;
1877
1878 case EIO_SEEK: eio__lseek (req); break;
1181 case EIO_READ: ALLOC (req->size); 1879 case EIO_READ: ALLOC (req->size);
1182 req->result = req->offs >= 0 1880 req->result = req->offs >= 0
1183 ? pread (req->int1, req->ptr2, req->size, req->offs) 1881 ? pread (req->int1, req->ptr2, req->size, req->offs)
1184 : read (req->int1, req->ptr2, req->size); break; 1882 : read (req->int1, req->ptr2, req->size); break;
1185 case EIO_WRITE: req->result = req->offs >= 0 1883 case EIO_WRITE: req->result = req->offs >= 0
1186 ? pwrite (req->int1, req->ptr2, req->size, req->offs) 1884 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1187 : write (req->int1, req->ptr2, req->size); break; 1885 : write (req->int1, req->ptr2, req->size); break;
1188 1886
1887 case EIO_FCNTL: req->result = fcntl (req->int1, (int) req->int2, req->ptr2); break;
1888 case EIO_IOCTL: req->result = ioctl (req->int1, (unsigned long)req->int2, req->ptr2); break;
1889
1189 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break; 1890 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1190 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break; 1891 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
1892
1893#if HAVE_AT
1191 1894
1192 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1895 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1193 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 1896 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break;
1194 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1897 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1195 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 1898 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, AT_SYMLINK_NOFOLLOW); break;
1196 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1197 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1198
1199 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; 1899 case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break;
1200 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1201 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break; 1900 case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break;
1202 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1203 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break; 1901 case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break;
1204 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1205
1206 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break; 1902 case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break;
1207 case EIO_CLOSE: req->result = close (req->int1); break; 1903 case EIO_SLURP: eio__slurp ( openat (dirfd, req->ptr1, O_RDONLY | O_CLOEXEC), req); break;
1208 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break; 1904
1209 case EIO_UNLINK: req->result = unlink (req->ptr1); break; 1905 case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
1210 case EIO_RMDIR: req->result = rmdir (req->ptr1); break; 1906 case EIO_RMDIR: /* complications arise because "." cannot be removed, so we might have to expand */
1907 req->result = req->wd && SINGLEDOT (req->ptr1)
1908 ? rmdir (req->wd->str)
1909 : unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break;
1211 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; 1910 case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); break;
1212 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; 1911 case EIO_RENAME: req->result = eio__renameat2 (
1912 dirfd,
1913 /* complications arise because "." cannot be renamed, so we might have to expand */
1914 req->wd && SINGLEDOT (req->ptr1) ? req->wd->str : req->ptr1,
1915 WD2FD ((eio_wd)req->int3),
1916 req->ptr2,
1917 req->int2
1918 );
1919 break;
1213 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; 1920 case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2, 0); break;
1214 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; 1921 case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
1215 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break; 1922 case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1216 1923 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1924 req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1217 case EIO_READLINK: ALLOC (NAME_MAX); 1925 case EIO_READLINK: ALLOC (EIO_PATH_MAX);
1218 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; 1926 req->result = readlinkat (dirfd, req->ptr1, req->ptr2, EIO_PATH_MAX);
1219 1927 if (req->result == EIO_PATH_MAX)
1220 case EIO_SYNC: req->result = 0; sync (); break; 1928 {
1221 case EIO_FSYNC: req->result = fsync (req->int1); break; 1929 req->result = -1;
1222 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 1930 errno = ENAMETOOLONG;
1223 case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break; 1931 }
1224 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break; 1932 break;
1225 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1226
1227 case EIO_READDIR: eio__scandir (req, self); break;
1228
1229 case EIO_BUSY: 1933 case EIO_UTIME:
1230#ifdef _WIN32 1934 case EIO_FUTIME:
1231 Sleep (req->nv1 * 1000.); 1935 {
1936 struct timespec ts[2];
1937 struct timespec *times;
1938
1939 if (req->nv1 != -1. || req->nv2 != -1.)
1940 {
1941 ts[0].tv_sec = req->nv1;
1942 ts[0].tv_nsec = (req->nv1 - ts[0].tv_sec) * 1e9;
1943 ts[1].tv_sec = req->nv2;
1944 ts[1].tv_nsec = (req->nv2 - ts[1].tv_sec) * 1e9;
1945
1946 times = ts;
1947 }
1948 else
1949 times = 0;
1950
1951 req->result = req->type == EIO_FUTIME
1952 ? futimens (req->int1, times)
1953 : utimensat (dirfd, req->ptr1, times, 0);
1954 }
1955 break;
1956
1232#else 1957#else
1233 {
1234 struct timeval tv;
1235 1958
1236 tv.tv_sec = req->nv1; 1959 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1237 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; 1960 req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
1961 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1962 req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
1963 case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
1964 case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
1965 case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
1966 case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
1967 case EIO_SLURP: eio__slurp ( open (path , O_RDONLY | O_CLOEXEC), req); break;
1238 1968
1239 req->result = select (0, 0, 0, 0, &tv); 1969 case EIO_UNLINK: req->result = unlink (path ); break;
1240 } 1970 case EIO_RMDIR: req->result = rmdir (path ); break;
1241#endif 1971 case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
1242 break; 1972 case EIO_RENAME: req->result = req->int2 ? EIO_ENOSYS () : rename (path, req->ptr2); break;
1973 case EIO_LINK: req->result = link (path , req->ptr2); break;
1974 case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
1975 case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
1976 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1977 req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
1978 case EIO_READLINK: ALLOC (EIO_PATH_MAX);
1979 req->result = readlink (path, req->ptr2, EIO_PATH_MAX);
1980 if (req->result == EIO_PATH_MAX)
1981 {
1982 req->result = -1;
1983 errno = ENAMETOOLONG;
1984 }
1985 break;
1243 1986
1244 case EIO_UTIME: 1987 case EIO_UTIME:
1245 case EIO_FUTIME: 1988 case EIO_FUTIME:
1246 { 1989 {
1247 struct timeval tv[2]; 1990 struct timeval tv[2];
1248 struct timeval *times; 1991 struct timeval *times;
1249 1992
1250 if (req->nv1 != -1. || req->nv2 != -1.) 1993 if (req->nv1 != -1. || req->nv2 != -1.)
1251 { 1994 {
1252 tv[0].tv_sec = req->nv1; 1995 tv[0].tv_sec = req->nv1;
1253 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.; 1996 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
1254 tv[1].tv_sec = req->nv2; 1997 tv[1].tv_sec = req->nv2;
1255 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.; 1998 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
1256 1999
1257 times = tv; 2000 times = tv;
1258 } 2001 }
1259 else 2002 else
1260 times = 0; 2003 times = 0;
1261
1262 2004
1263 req->result = req->type == EIO_FUTIME 2005 req->result = req->type == EIO_FUTIME
1264 ? futimes (req->int1, times) 2006 ? futimes (req->int1, times)
1265 : utimes (req->ptr1, times); 2007 : utimes (req->ptr1, times);
1266 } 2008 }
1267 break; 2009 break;
1268 2010
2011#endif
2012
2013 case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
2014 {
2015 ALLOC (req->result);
2016 memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
2017 }
2018 break;
2019
2020 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2021 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
2022
2023 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2024 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2025
2026 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
2027 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
2028 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
2029
2030 case EIO_CLOSE: req->result = eio__close (req->int1); break;
2031 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
2032 case EIO_SYNC: req->result = 0; sync (); break;
2033 case EIO_FSYNC: req->result = fsync (req->int1); break;
2034 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
2035 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
2036 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
2037 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
2038 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
2039 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
2040 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
2041 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
2042
2043 case EIO_READDIR: eio__scandir (req, self); break;
2044
2045 case EIO_BUSY:
2046#ifdef _WIN32
2047 Sleep (req->nv1 * 1e3);
2048#else
2049 {
2050 struct timeval tv;
2051
2052 tv.tv_sec = req->nv1;
2053 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
2054
2055 req->result = select (0, 0, 0, 0, &tv);
2056 }
2057#endif
2058 break;
2059
2060#if 0
1269 case EIO_GROUP: 2061 case EIO_GROUP:
1270 abort (); /* handled in eio_request */ 2062 abort (); /* handled in eio_request */
2063#endif
1271 2064
1272 case EIO_NOP: 2065 case EIO_NOP:
1273 req->result = 0; 2066 req->result = 0;
1274 break; 2067 break;
1275 2068
1276 case EIO_CUSTOM: 2069 case EIO_CUSTOM:
1277 ((void (*)(eio_req *))req->feed) (req); 2070 req->feed (req);
1278 break; 2071 break;
1279 2072
1280 default: 2073 default:
1281 req->result = -1; 2074 req->result = EIO_ENOSYS ();
1282 break; 2075 break;
1283 } 2076 }
1284 2077
2078alloc_fail:
1285 req->errorno = errno; 2079 req->errorno = errno;
1286} 2080}
1287 2081
1288#ifndef EIO_NO_WRAPPERS 2082#ifndef EIO_NO_WRAPPERS
1289 2083
2084eio_req *eio_wd_open (const char *path, int pri, eio_cb cb, void *data)
2085{
2086 REQ (EIO_WD_OPEN); PATH; SEND;
2087}
2088
2089eio_req *eio_wd_close (eio_wd wd, int pri, eio_cb cb, void *data)
2090{
2091 REQ (EIO_WD_CLOSE); req->wd = wd; SEND;
2092}
2093
1290eio_req *eio_nop (int pri, eio_cb cb, void *data) 2094eio_req *eio_nop (int pri, eio_cb cb, void *data)
1291{ 2095{
1292 REQ (EIO_NOP); SEND; 2096 REQ (EIO_NOP); SEND;
1293} 2097}
1294 2098
… …
1310eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2114eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1311{ 2115{
1312 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2116 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1313} 2117}
1314 2118
2119eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2120{
2121 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2122}
2123
2124eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
2125{
2126 REQ (EIO_SYNCFS); req->int1 = fd; SEND;
2127}
2128
2129eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2130{
2131 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2132}
2133
1315eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2134eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1316{ 2135{
1317 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2136 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1318} 2137}
1319 2138
1320eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) 2139eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
1321{ 2140{
1322 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; 2141 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
1323} 2142}
1324 2143
1325eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) 2144eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
1326{ 2145{
1327 REQ (EIO_FDATASYNC); req->int1 = fd; SEND; 2146 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2147}
2148
2149eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2150{
2151 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
1328} 2152}
1329 2153
1330eio_req *eio_close (int fd, int pri, eio_cb cb, void *data) 2154eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
1331{ 2155{
1332 REQ (EIO_CLOSE); req->int1 = fd; SEND; 2156 REQ (EIO_CLOSE); req->int1 = fd; SEND;
… …
1335eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data) 2159eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
1336{ 2160{
1337 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND; 2161 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
1338} 2162}
1339 2163
2164eio_req *eio_seek (int fd, off_t offset, int whence, int pri, eio_cb cb, void *data)
2165{
2166 REQ (EIO_SEEK); req->int1 = fd; req->offs = offset; req->int2 = whence; SEND;
2167}
2168
1340eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data) 2169eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1341{ 2170{
1342 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND; 2171 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1343} 2172}
1344 2173
1345eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data) 2174eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1346{ 2175{
1347 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND; 2176 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1348} 2177}
1349 2178
2179eio_req *eio_fcntl (int fd, int cmd, void *arg, int pri, eio_cb cb, void *data)
2180{
2181 REQ (EIO_IOCTL); req->int1 = fd; req->int2 = cmd; req->ptr2 = arg; SEND;
2182}
2183
2184eio_req *eio_ioctl (int fd, unsigned long request, void *buf, int pri, eio_cb cb, void *data)
2185{
2186 REQ (EIO_IOCTL); req->int1 = fd; req->int2 = request; req->ptr2 = buf; SEND;
2187}
2188
1350eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) 2189eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1351{ 2190{
1352 REQ (EIO_FSTAT); req->int1 = fd; SEND; 2191 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1353} 2192}
1354 2193
2194eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
2195{
2196 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
2197}
2198
1355eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) 2199eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1356{ 2200{
1357 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; 2201 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1358} 2202}
1359 2203
… …
1365eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) 2209eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1366{ 2210{
1367 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2211 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1368} 2212}
1369 2213
1370eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2214eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1371{ 2215{
1372 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2216 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1373} 2217}
1374 2218
1375eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) 2219eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
… …
1395eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data) 2239eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1396{ 2240{
1397 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2241 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1398} 2242}
1399 2243
1400eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2244eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1401{ 2245{
1402 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2246 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1403} 2247}
1404 2248
1405eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) 2249eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
… …
1421eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data) 2265eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1422{ 2266{
1423 return eio__1path (EIO_READLINK, path, pri, cb, data); 2267 return eio__1path (EIO_READLINK, path, pri, cb, data);
1424} 2268}
1425 2269
2270eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2271{
2272 return eio__1path (EIO_REALPATH, path, pri, cb, data);
2273}
2274
1426eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) 2275eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1427{ 2276{
1428 return eio__1path (EIO_STAT, path, pri, cb, data); 2277 return eio__1path (EIO_STAT, path, pri, cb, data);
1429} 2278}
1430 2279
1431eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) 2280eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1432{ 2281{
1433 return eio__1path (EIO_LSTAT, path, pri, cb, data); 2282 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1434} 2283}
1435 2284
2285eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
2286{
2287 return eio__1path (EIO_STATVFS, path, pri, cb, data);
2288}
2289
1436eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) 2290eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1437{ 2291{
1438 return eio__1path (EIO_UNLINK, path, pri, cb, data); 2292 return eio__1path (EIO_UNLINK, path, pri, cb, data);
1439} 2293}
1440 2294
1441eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data) 2295eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1442{ 2296{
1443 return eio__1path (EIO_RMDIR, path, pri, cb, data); 2297 return eio__1path (EIO_RMDIR, path, pri, cb, data);
1444} 2298}
1445 2299
1446eio_req *eio_readdir (const char *path, int pri, eio_cb cb, void *data) 2300eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1447{ 2301{
1448 return eio__1path (EIO_READDIR, path, pri, cb, data); 2302 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1449} 2303}
1450 2304
1451eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) 2305eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1452{ 2306{
1453 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; 2307 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1454} 2308}
1455 2309
1456static eio_req * 2310static eio_req *
1457eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2311eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1458{ 2312{
… …
1482eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2336eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1483{ 2337{
1484 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2338 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1485} 2339}
1486 2340
2341eio_req *eio_slurp (const char *path, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
2342{
2343 REQ (EIO_SLURP); PATH; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
2344}
2345
1487eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) 2346eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
1488{ 2347{
1489 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND; 2348 REQ (EIO_CUSTOM); req->feed = execute; SEND;
1490} 2349}
1491 2350
1492#endif 2351#endif
1493 2352
1494eio_req *eio_grp (eio_cb cb, void *data) 2353eio_req *eio_grp (eio_cb cb, void *data)
… …
1503#undef SEND 2362#undef SEND
1504 2363
1505/*****************************************************************************/ 2364/*****************************************************************************/
1506/* grp functions */ 2365/* grp functions */
1507 2366
2367void
1508void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2368eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1509{ 2369{
1510 grp->int2 = limit; 2370 grp->int2 = limit;
1511 grp->feed = feed; 2371 grp->feed = feed;
1512 2372
1513 grp_try_feed (grp); 2373 grp_try_feed (grp);
1514} 2374}
1515 2375
2376void
1516void eio_grp_limit (eio_req *grp, int limit) 2377eio_grp_limit (eio_req *grp, int limit)
1517{ 2378{
1518 grp->int2 = limit; 2379 grp->int2 = limit;
1519 2380
1520 grp_try_feed (grp); 2381 grp_try_feed (grp);
1521} 2382}
1522 2383
2384void
1523void eio_grp_add (eio_req *grp, eio_req *req) 2385eio_grp_add (eio_req *grp, eio_req *req)
1524{ 2386{
1525 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2387 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1526 2388
1527 grp->flags |= EIO_FLAG_GROUPADD; 2389 grp->flags |= ETP_FLAG_GROUPADD;
1528 2390
1529 ++grp->size; 2391 ++grp->size;
1530 req->grp = grp; 2392 req->grp = grp;
1531 2393
1532 req->grp_prev = 0; 2394 req->grp_prev = 0;
… …
1539} 2401}
1540 2402
1541/*****************************************************************************/ 2403/*****************************************************************************/
1542/* misc garbage */ 2404/* misc garbage */
1543 2405
2406eio_ssize_t
1544ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2407eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1545{ 2408{
1546 etp_worker wrk;
1547
1548 wrk.dbuf = 0;
1549
1550 eio__sendfile (ofd, ifd, offset, count, &wrk); 2409 return eio__sendfile (ofd, ifd, offset, count);
1551
1552 if (wrk.dbuf)
1553 free (wrk.dbuf);
1554} 2410}
1555 2411

Diff Legend

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