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

Comparing libeio/eio.c (file contents):
Revision 1.30 by root, Wed Jun 3 12:24:49 2009 UTC vs.
Revision 1.157 by root, Thu Nov 29 21:48:44 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) 2007,2008,2009,2010,2011,2012,2013,2016,2017,2018 Marc Alexander Lehmann <libeio@schmorp.de>
5 * All rights reserved. 5 * All rights reserved.
6 * 6 *
7 * Redistribution and use in source and binary forms, with or without modifica- 7 * Redistribution and use in source and binary forms, with or without modifica-
8 * tion, are permitted provided that the following conditions are met: 8 * tion, are permitted provided that the following conditions are met:
9 * 9 *
10 * 1. Redistributions of source code must retain the above copyright notice, 10 * 1. Redistributions of source code must retain the above copyright notice,
11 * this list of conditions and the following disclaimer. 11 * this list of conditions and the following disclaimer.
12 * 12 *
13 * 2. Redistributions in binary form must reproduce the above copyright 13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the 14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution. 15 * documentation and/or other materials provided with the distribution.
16 * 16 *
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR IMPLIED 17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR IMPLIED
18 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MER- 18 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MER-
19 * CHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO 19 * CHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO
20 * EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPE- 20 * EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPE-
21 * CIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, 21 * CIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
… …
35 * and other provisions required by the GPL. If you do not delete the 35 * and other provisions required by the GPL. If you do not delete the
36 * provisions above, a recipient may use your version of this file under 36 * provisions above, a recipient may use your version of this file under
37 * either the BSD or the GPL. 37 * either the BSD or the GPL.
38 */ 38 */
39 39
40#ifndef _WIN32
41# include "config.h"
42#endif
43
40#include "eio.h" 44#include "eio.h"
41
42#ifdef EIO_STACKSIZE
43# define XTHREAD_STACKSIZE EIO_STACKSIZE
44#endif
45#include "xthread.h" 45#include "ecb.h"
46 46
47#include <errno.h> 47#include <errno.h>
48#include <stddef.h> 48#include <stddef.h>
49#include <stdlib.h> 49#include <stdlib.h>
50#include <string.h> 50#include <string.h>
… …
53#include <sys/stat.h> 53#include <sys/stat.h>
54#include <limits.h> 54#include <limits.h>
55#include <fcntl.h> 55#include <fcntl.h>
56#include <assert.h> 56#include <assert.h>
57 57
58/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
59/* intptr_t only comes from stdint.h, says idiot openbsd coder */
60#if HAVE_STDINT_H
61# include <stdint.h>
62#endif
63
64#ifndef ECANCELED
65# define ECANCELED EDOM
66#endif
67#ifndef ELOOP
68# define ELOOP EDOM
69#endif
70
71#if !defined(ENOTSOCK) && defined(WSAENOTSOCK)
72# define ENOTSOCK WSAENOTSOCK
73#endif
74
75static void eio_destroy (eio_req *req);
76
58#ifndef EIO_FINISH 77#ifndef EIO_FINISH
59# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0 78# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
60#endif 79#endif
61 80
62#ifndef EIO_DESTROY 81#ifndef EIO_DESTROY
… …
65 84
66#ifndef EIO_FEED 85#ifndef EIO_FEED
67# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0) 86# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)
68#endif 87#endif
69 88
89#ifndef EIO_FD_TO_WIN32_HANDLE
90# define EIO_FD_TO_WIN32_HANDLE(fd) _get_osfhandle (fd)
91#endif
92#ifndef EIO_WIN32_HANDLE_TO_FD
93# define EIO_WIN32_HANDLE_TO_FD(handle) _open_osfhandle (handle, 0)
94#endif
95
96#define EIO_ERRNO(errval,retval) ((errno = errval), retval)
97
98#define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1)
99
70#ifdef _WIN32 100#ifdef _WIN32
71 101
72 /*doh*/ 102 #undef PAGESIZE
103 #define PAGESIZE 4096 /* GetSystemInfo? */
104
105 /* TODO: look at how perl does stat (non-sloppy), unlink (ro-files), utime, link */
106
107 #ifdef EIO_STRUCT_STATI64
108 /* look at perl's non-sloppy stat */
109 #define stat(path,buf) _stati64 (path,buf)
110 #define fstat(fd,buf) _fstati64 (fd,buf)
111 #endif
112 #define lstat(path,buf) stat (path,buf)
113 #define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
114 #define mkdir(path,mode) _mkdir (path)
115 #define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
116
117 #define chmod(path,mode) _chmod (path, mode)
118 #define dup(fd) _dup (fd)
119 #define dup2(fd1,fd2) _dup2 (fd1, fd2)
120 #define pipe(fds) _pipe (fds, 4096, O_BINARY)
121
122 #define fcntl(fd,cmd,arg) EIO_ENOSYS ()
123 #define ioctl(fd,cmd,arg) EIO_ENOSYS ()
124 #define fchmod(fd,mode) EIO_ENOSYS ()
125 #define chown(path,uid,gid) EIO_ENOSYS ()
126 #define fchown(fd,uid,gid) EIO_ENOSYS ()
127 #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
128 #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
129 #define mknod(path,mode,dev) EIO_ENOSYS ()
130 #define sync() EIO_ENOSYS ()
131 #define readlink(path,buf,s) EIO_ENOSYS ()
132 #define statvfs(path,buf) EIO_ENOSYS ()
133 #define fstatvfs(fd,buf) EIO_ENOSYS ()
134
135 #define pread(fd,buf,count,offset) eio__pread (fd, buf, count, offset)
136 #define pwrite(fd,buf,count,offset) eio__pwrite (fd, buf, count, offset)
137
138 #if __GNUC__
139 typedef long long eio_off_t; /* signed for compatibility to msvc */
140 #else
141 typedef __int64 eio_off_t; /* unsigned not supported by msvc */
142 #endif
143
144 static eio_ssize_t
145 eio__pread (int fd, void *buf, eio_ssize_t count, eio_off_t offset)
146 {
147 OVERLAPPED o = { 0 };
148 DWORD got;
149
150 o.Offset = offset;
151 o.OffsetHigh = offset >> 32;
152
153 return ReadFile ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd), buf, count, &got, &o)
154 ? got : -1;
155 }
156
157 static eio_ssize_t
158 eio__pwrite (int fd, void *buf, eio_ssize_t count, eio_off_t offset)
159 {
160 OVERLAPPED o = { 0 };
161 DWORD got;
162
163 o.Offset = offset;
164 o.OffsetHigh = offset >> 32;
165
166 return WriteFile ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd), buf, count, &got, &o)
167 ? got : -1;
168 }
169
170 /* rename() uses MoveFile, which fails to overwrite */
171 #define rename(old,neu) eio__rename (old, neu)
172
173 static int
174 eio__rename (const char *old, const char *neu)
175 {
176 if (MoveFileEx (old, neu, MOVEFILE_REPLACE_EXISTING))
177 return 0;
178
179 /* should steal _dosmaperr */
180 switch (GetLastError ())
181 {
182 case ERROR_FILE_NOT_FOUND:
183 case ERROR_PATH_NOT_FOUND:
184 case ERROR_INVALID_DRIVE:
185 case ERROR_NO_MORE_FILES:
186 case ERROR_BAD_NETPATH:
187 case ERROR_BAD_NET_NAME:
188 case ERROR_BAD_PATHNAME:
189 case ERROR_FILENAME_EXCED_RANGE:
190 errno = ENOENT;
191 break;
192
193 default:
194 errno = EACCES;
195 break;
196 }
197
198 return -1;
199 }
200
201 /* we could even stat and see if it exists */
202 static int
203 symlink (const char *old, const char *neu)
204 {
205 #if WINVER >= 0x0600
206 int flags;
207
208 /* This tries out all combinations of SYMBOLIC_LINK_FLAG_DIRECTORY
209 * and SYMBOLIC_LINK_FLAG_ALLOW_UNPRIVILEGED_CREATE,
210 * with directory first.
211 */
212 for (flags = 3; flags >= 0; --flags)
213 if (CreateSymbolicLink (neu, old, flags))
214 return 0;
215 #endif
216
217 return EIO_ERRNO (ENOENT, -1);
218 }
219
220 /* POSIX API only, causing trouble for win32 apps */
221 #define CreateHardLink(neu,old,flags) 0 /* not really creating hardlink, still using relative paths? */
222 #define CreateSymbolicLink(neu,old,flags) 0 /* vista+ only */
223
224 struct statvfs
225 {
226 int dummy;
227 };
228
229 #define DT_DIR EIO_DT_DIR
230 #define DT_REG EIO_DT_REG
231 #define D_NAME(entp) entp.cFileName
232 #define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
73 233
74#else 234#else
75 235
76# include "config.h" 236 #include <sys/ioctl.h>
77# include <sys/time.h> 237 #include <sys/time.h>
78# include <sys/select.h> 238 #include <sys/select.h>
79# include <unistd.h> 239 #include <unistd.h>
240 #include <signal.h>
241 #include <dirent.h>
242
243 #ifdef ANDROID
244 #include <sys/vfs.h>
245 #define statvfs statfs
246 #define fstatvfs fstatfs
247 #include <asm/page.h> /* supposedly limits.h does #define PAGESIZE PAGESIZE */
248 #else
249 #include <sys/statvfs.h>
250 #endif
251
252 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
253 #include <sys/mman.h>
254 #endif
255
256 #define D_NAME(entp) entp->d_name
257
258 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
259 #if __FreeBSD__ || __NetBSD__ || __OpenBSD__
260 #define _DIRENT_HAVE_D_TYPE /* sigh */
261 #define D_INO(de) (de)->d_fileno
262 #define D_NAMLEN(de) (de)->d_namlen
263 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
264 #define D_INO(de) (de)->d_ino
265 #endif
266
267 #ifdef _D_EXACT_NAMLEN
268 #undef D_NAMLEN
269 #define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
270 #endif
271
272 #ifdef _DIRENT_HAVE_D_TYPE
273 #define D_TYPE(de) (de)->d_type
274 #endif
275
276 #ifndef EIO_STRUCT_DIRENT
277 #define EIO_STRUCT_DIRENT struct dirent
278 #endif
279
280#endif
281
282#if HAVE_UTIMES
80# include <utime.h> 283# include <utime.h>
81# include <signal.h>
82# include <dirent.h>
83
84# ifndef EIO_STRUCT_DIRENT
85# define EIO_STRUCT_DIRENT struct dirent
86# endif 284#endif
87 285
286#if HAVE_SYS_SYSCALL_H
287# include <sys/syscall.h>
88#endif 288#endif
89 289
90#if HAVE_SENDFILE 290#if HAVE_SENDFILE
91# if __linux 291# if __linux
92# include <sys/sendfile.h> 292# include <sys/sendfile.h>
93# elif __freebsd 293# elif __FreeBSD__ || defined __APPLE__
94# include <sys/socket.h> 294# include <sys/socket.h>
95# include <sys/uio.h> 295# include <sys/uio.h>
96# elif __hpux 296# elif __hpux
97# include <sys/socket.h> 297# include <sys/socket.h>
98# elif __solaris /* not yet */ 298# elif __solaris
99# include <sys/sendfile.h> 299# include <sys/sendfile.h>
100# else 300# else
101# error sendfile support requested but not available 301# error sendfile support requested but not available
102# endif 302# endif
103#endif 303#endif
104 304
105/* number of seconds after which an idle threads exit */ 305#if HAVE_RENAMEAT2
106#define IDLE_TIMEOUT 10 306# include <sys/syscall.h>
307# include <linux/fs.h>
308#endif
309
310#ifndef D_TYPE
311# define D_TYPE(de) 0
312#endif
313#ifndef D_INO
314# define D_INO(de) 0
315#endif
316#ifndef D_NAMLEN
317# define D_NAMLEN(entp) strlen (D_NAME (entp))
318#endif
107 319
108/* used for struct dirent, AIX doesn't provide it */ 320/* used for struct dirent, AIX doesn't provide it */
109#ifndef NAME_MAX 321#ifndef NAME_MAX
110# define NAME_MAX 4096 322# define NAME_MAX 4096
111#endif 323#endif
112 324
325/* used for readlink etc. */
326#ifndef PATH_MAX
327# define PATH_MAX 0
328#endif
329
330#ifndef O_CLOEXEC
331 #define O_CLOEXEC 0
332#endif
333
334#ifndef O_NONBLOCK
335 #define O_NONBLOCK 0
336#endif
337
338#ifndef O_SEARCH
339 #define O_SEARCH O_RDONLY
340#endif
341
342#ifndef O_DIRECTORY
343 #define O_DIRECTORY 0
344#endif
345
346#ifndef EIO_PATH_MIN
347# define EIO_PATH_MIN 8160
348#endif
349
350#define EIO_PATH_MAX (PATH_MAX <= EIO_PATH_MIN ? EIO_PATH_MIN : PATH_MAX)
351
113/* buffer size for various temporary buffers */ 352/* buffer size for various temporary buffers */
114#define EIO_BUFSIZE 65536 353#define EIO_BUFSIZE 65536
115 354
116#define dBUF \ 355#define dBUF \
117 char *eio_buf; \
118 ETP_WORKER_LOCK (self); \
119 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 356 char *eio_buf = malloc (EIO_BUFSIZE); \
120 ETP_WORKER_UNLOCK (self); \
121 errno = ENOMEM; \ 357 errno = ENOMEM; \
122 if (!eio_buf) \ 358 if (!eio_buf) \
123 return -1; 359 return -1
124 360
125#define EIO_TICKS ((1000000 + 1023) >> 10) 361#define FUBd \
362 free (eio_buf)
126 363
127/*****************************************************************************/ 364/*****************************************************************************/
128 365
129#if __GNUC__ >= 3 366struct etp_tmpbuf;
130# define expect(expr,value) __builtin_expect ((expr),(value)) 367
368#if _POSIX_VERSION >= 200809L
369 #define HAVE_AT 1
370 #define WD2FD(wd) ((wd) ? (wd)->fd : AT_FDCWD)
131#else 371#else
132# define expect(expr,value) (expr) 372 #define HAVE_AT 0
373 static const char *wd_expand (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path);
133#endif 374#endif
134 375
135#define expect_false(expr) expect ((expr) != 0, 0) 376struct eio_pwd
136#define expect_true(expr) expect ((expr) != 0, 1) 377{
378#if HAVE_AT
379 int fd;
380#endif
381 int len;
382 char str[1]; /* actually, a 0-terminated canonical path */
383};
137 384
138/*****************************************************************************/ 385/*****************************************************************************/
139 386
140#define ETP_PRI_MIN EIO_PRI_MIN 387#define ETP_PRI_MIN EIO_PRI_MIN
141#define ETP_PRI_MAX EIO_PRI_MAX 388#define ETP_PRI_MAX EIO_PRI_MAX
142 389
390#define ETP_TYPE_QUIT -1
391#define ETP_TYPE_GROUP EIO_GROUP
392
393static void eio_nop_callback (void) { }
394static void (*eio_want_poll_cb)(void) = eio_nop_callback;
395static void (*eio_done_poll_cb)(void) = eio_nop_callback;
396
397#define ETP_WANT_POLL(pool) eio_want_poll_cb ()
398#define ETP_DONE_POLL(pool) eio_done_poll_cb ()
399
143struct etp_worker; 400struct etp_worker;
144
145#define ETP_REQ eio_req 401#define ETP_REQ eio_req
146#define ETP_DESTROY(req) eio_destroy (req) 402#define ETP_DESTROY(req) eio_destroy (req)
147static int eio_finish (eio_req *req); 403static int eio_finish (eio_req *req);
148#define ETP_FINISH(req) eio_finish (req) 404#define ETP_FINISH(req) eio_finish (req)
149static void eio_execute (struct etp_worker *self, eio_req *req); 405static void eio_execute (struct etp_worker *self, eio_req *req);
150#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) 406#define ETP_EXECUTE(wrk,req) eio_execute (wrk, req)
151 407
152#define ETP_WORKER_CLEAR(req) \ 408#include "etp.c"
153 if (wrk->dbuf) \ 409
154 { \ 410static struct etp_pool eio_pool;
155 free (wrk->dbuf); \ 411#define EIO_POOL (&eio_pool)
156 wrk->dbuf = 0; \
157 } \
158 \
159 if (wrk->dirp) \
160 { \
161 closedir (wrk->dirp); \
162 wrk->dirp = 0; \
163 }
164#define ETP_WORKER_COMMON \
165 void *dbuf; \
166 DIR *dirp;
167 412
168/*****************************************************************************/ 413/*****************************************************************************/
169 414
170#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
171
172/* calculcate time difference in ~1/EIO_TICKS of a second */
173static int tvdiff (struct timeval *tv1, struct timeval *tv2)
174{
175 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
176 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
177}
178
179static unsigned int started, idle, wanted = 4;
180
181static void (*want_poll_cb) (void);
182static void (*done_poll_cb) (void);
183
184static unsigned int max_poll_time; /* reslock */
185static unsigned int max_poll_reqs; /* reslock */
186
187static volatile unsigned int nreqs; /* reqlock */
188static volatile unsigned int nready; /* reqlock */
189static volatile unsigned int npending; /* reqlock */
190static volatile unsigned int max_idle = 4;
191
192static mutex_t wrklock = X_MUTEX_INIT;
193static mutex_t reslock = X_MUTEX_INIT;
194static mutex_t reqlock = X_MUTEX_INIT;
195static cond_t reqwait = X_COND_INIT;
196
197#if !HAVE_PREADWRITE
198/*
199 * make our pread/pwrite emulation safe against themselves, but not against
200 * normal read/write by using a mutex. slows down execution a lot,
201 * but that's your problem, not mine.
202 */
203static mutex_t preadwritelock = X_MUTEX_INIT;
204#endif
205
206typedef struct etp_worker
207{
208 /* locked by wrklock */
209 struct etp_worker *prev, *next;
210
211 thread_t tid;
212
213 /* locked by reslock, reqlock or wrklock */
214 ETP_REQ *req; /* currently processed request */
215
216 ETP_WORKER_COMMON
217} etp_worker;
218
219static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */
220
221#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
222#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
223
224/* worker threads management */
225
226static void etp_worker_clear (etp_worker *wrk)
227{
228 ETP_WORKER_CLEAR (wrk);
229}
230
231static void etp_worker_free (etp_worker *wrk)
232{
233 wrk->next->prev = wrk->prev;
234 wrk->prev->next = wrk->next;
235
236 free (wrk);
237}
238
239static unsigned int etp_nreqs (void)
240{
241 int retval;
242 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
243 retval = nreqs;
244 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
245 return retval;
246}
247
248static unsigned int etp_nready (void)
249{
250 unsigned int retval;
251
252 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
253 retval = nready;
254 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
255
256 return retval;
257}
258
259static unsigned int etp_npending (void)
260{
261 unsigned int retval;
262
263 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
264 retval = npending;
265 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
266
267 return retval;
268}
269
270static unsigned int etp_nthreads (void)
271{
272 unsigned int retval;
273
274 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
275 retval = started;
276 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
277
278 return retval;
279}
280
281/*
282 * a somewhat faster data structure might be nice, but
283 * with 8 priorities this actually needs <20 insns
284 * per shift, the most expensive operation.
285 */
286typedef struct {
287 ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */
288 int size;
289} etp_reqq;
290
291static etp_reqq req_queue;
292static etp_reqq res_queue;
293
294static int reqq_push (etp_reqq *q, ETP_REQ *req)
295{
296 int pri = req->pri;
297 req->next = 0;
298
299 if (q->qe[pri])
300 {
301 q->qe[pri]->next = req;
302 q->qe[pri] = req;
303 }
304 else
305 q->qe[pri] = q->qs[pri] = req;
306
307 return q->size++;
308}
309
310static ETP_REQ *reqq_shift (etp_reqq *q)
311{
312 int pri;
313
314 if (!q->size)
315 return 0;
316
317 --q->size;
318
319 for (pri = ETP_NUM_PRI; pri--; )
320 {
321 eio_req *req = q->qs[pri];
322
323 if (req)
324 {
325 if (!(q->qs[pri] = (eio_req *)req->next))
326 q->qe[pri] = 0;
327
328 return req;
329 }
330 }
331
332 abort ();
333}
334
335static void etp_atfork_prepare (void)
336{
337 X_LOCK (wrklock);
338 X_LOCK (reqlock);
339 X_LOCK (reslock);
340#if !HAVE_PREADWRITE
341 X_LOCK (preadwritelock);
342#endif
343}
344
345static void etp_atfork_parent (void)
346{
347#if !HAVE_PREADWRITE
348 X_UNLOCK (preadwritelock);
349#endif
350 X_UNLOCK (reslock);
351 X_UNLOCK (reqlock);
352 X_UNLOCK (wrklock);
353}
354
355static void etp_atfork_child (void)
356{
357 ETP_REQ *prv;
358
359 while ((prv = reqq_shift (&req_queue)))
360 ETP_DESTROY (prv);
361
362 while ((prv = reqq_shift (&res_queue)))
363 ETP_DESTROY (prv);
364
365 while (wrk_first.next != &wrk_first)
366 {
367 etp_worker *wrk = wrk_first.next;
368
369 if (wrk->req)
370 ETP_DESTROY (wrk->req);
371
372 etp_worker_clear (wrk);
373 etp_worker_free (wrk);
374 }
375
376 started = 0;
377 idle = 0;
378 nreqs = 0;
379 nready = 0;
380 npending = 0;
381
382 etp_atfork_parent ();
383}
384
385static void 415static void
386etp_once_init (void)
387{
388 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
389}
390
391static int
392etp_init (void (*want_poll)(void), void (*done_poll)(void))
393{
394 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
395
396 pthread_once (&doinit, etp_once_init);
397
398 want_poll_cb = want_poll;
399 done_poll_cb = done_poll;
400
401 return 0;
402}
403
404X_THREAD_PROC (etp_proc);
405
406static void etp_start_thread (void)
407{
408 etp_worker *wrk = calloc (1, sizeof (etp_worker));
409
410 /*TODO*/
411 assert (("unable to allocate worker thread data", wrk));
412
413 X_LOCK (wrklock);
414
415 if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
416 {
417 wrk->prev = &wrk_first;
418 wrk->next = wrk_first.next;
419 wrk_first.next->prev = wrk;
420 wrk_first.next = wrk;
421 ++started;
422 }
423 else
424 free (wrk);
425
426 X_UNLOCK (wrklock);
427}
428
429static void etp_maybe_start_thread (void)
430{
431 if (expect_true (etp_nthreads () >= wanted))
432 return;
433
434 /* todo: maybe use idle here, but might be less exact */
435 if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
436 return;
437
438 etp_start_thread ();
439}
440
441static void etp_end_thread (void)
442{
443 eio_req *req = calloc (1, sizeof (eio_req));
444
445 req->type = -1;
446 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
447
448 X_LOCK (reqlock);
449 reqq_push (&req_queue, req);
450 X_COND_SIGNAL (reqwait);
451 X_UNLOCK (reqlock);
452
453 X_LOCK (wrklock);
454 --started;
455 X_UNLOCK (wrklock);
456}
457
458static int etp_poll (void)
459{
460 unsigned int maxreqs;
461 unsigned int maxtime;
462 struct timeval tv_start, tv_now;
463
464 X_LOCK (reslock);
465 maxreqs = max_poll_reqs;
466 maxtime = max_poll_time;
467 X_UNLOCK (reslock);
468
469 if (maxtime)
470 gettimeofday (&tv_start, 0);
471
472 for (;;)
473 {
474 ETP_REQ *req;
475
476 etp_maybe_start_thread ();
477
478 X_LOCK (reslock);
479 req = reqq_shift (&res_queue);
480
481 if (req)
482 {
483 --npending;
484
485 if (!res_queue.size && done_poll_cb)
486 done_poll_cb ();
487 }
488
489 X_UNLOCK (reslock);
490
491 if (!req)
492 return 0;
493
494 X_LOCK (reqlock);
495 --nreqs;
496 X_UNLOCK (reqlock);
497
498 if (expect_false (req->type == EIO_GROUP && req->size))
499 {
500 req->int1 = 1; /* mark request as delayed */
501 continue;
502 }
503 else
504 {
505 int res = ETP_FINISH (req);
506 if (expect_false (res))
507 return res;
508 }
509
510 if (expect_false (maxreqs && !--maxreqs))
511 break;
512
513 if (maxtime)
514 {
515 gettimeofday (&tv_now, 0);
516
517 if (tvdiff (&tv_start, &tv_now) >= maxtime)
518 break;
519 }
520 }
521
522 errno = EAGAIN;
523 return -1;
524}
525
526static void etp_cancel (ETP_REQ *req)
527{
528 X_LOCK (wrklock);
529 req->flags |= EIO_FLAG_CANCELLED;
530 X_UNLOCK (wrklock);
531
532 eio_grp_cancel (req);
533}
534
535static void etp_submit (ETP_REQ *req)
536{
537 req->pri -= ETP_PRI_MIN;
538
539 if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
540 if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
541
542 if (expect_false (req->type == EIO_GROUP))
543 {
544 /* I hope this is worth it :/ */
545 X_LOCK (reqlock);
546 ++nreqs;
547 X_UNLOCK (reqlock);
548
549 X_LOCK (reslock);
550
551 ++npending;
552
553 if (!reqq_push (&res_queue, req) && want_poll_cb)
554 want_poll_cb ();
555
556 X_UNLOCK (reslock);
557 }
558 else
559 {
560 X_LOCK (reqlock);
561 ++nreqs;
562 ++nready;
563 reqq_push (&req_queue, req);
564 X_COND_SIGNAL (reqwait);
565 X_UNLOCK (reqlock);
566
567 etp_maybe_start_thread ();
568 }
569}
570
571static void etp_set_max_poll_time (double nseconds)
572{
573 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
574 max_poll_time = nseconds;
575 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
576}
577
578static void etp_set_max_poll_reqs (unsigned int maxreqs)
579{
580 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
581 max_poll_reqs = maxreqs;
582 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
583}
584
585static void etp_set_max_idle (unsigned int nthreads)
586{
587 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
588 max_idle = nthreads <= 0 ? 1 : nthreads;
589 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
590}
591
592static void etp_set_min_parallel (unsigned int nthreads)
593{
594 if (wanted < nthreads)
595 wanted = nthreads;
596}
597
598static void etp_set_max_parallel (unsigned int nthreads)
599{
600 if (wanted > nthreads)
601 wanted = nthreads;
602
603 while (started > wanted)
604 etp_end_thread ();
605}
606
607/*****************************************************************************/
608
609static void grp_try_feed (eio_req *grp) 416grp_try_feed (eio_req *grp)
610{ 417{
611 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 418 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
612 { 419 {
613 grp->flags &= ~EIO_FLAG_GROUPADD; 420 grp->flags &= ~ETP_FLAG_GROUPADD;
614 421
615 EIO_FEED (grp); 422 EIO_FEED (grp);
616 423
617 /* stop if no progress has been made */ 424 /* stop if no progress has been made */
618 if (!(grp->flags & EIO_FLAG_GROUPADD)) 425 if (!(grp->flags & ETP_FLAG_GROUPADD))
619 { 426 {
620 grp->feed = 0; 427 grp->feed = 0;
621 break; 428 break;
622 } 429 }
623 } 430 }
624} 431}
625 432
433static int
626static int grp_dec (eio_req *grp) 434grp_dec (eio_req *grp)
627{ 435{
628 --grp->size; 436 --grp->size;
629 437
630 /* call feeder, if applicable */ 438 /* call feeder, if applicable */
631 grp_try_feed (grp); 439 grp_try_feed (grp);
632 440
633 /* finish, if done */ 441 /* finish, if done */
634 if (!grp->size && grp->int1) 442 if (!grp->size && grp->flags & ETP_FLAG_DELAYED)
635 return eio_finish (grp); 443 return eio_finish (grp);
636 else 444 else
637 return 0; 445 return 0;
638} 446}
639 447
448static void
640void eio_destroy (eio_req *req) 449eio_destroy (eio_req *req)
641{ 450{
642 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); 451 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
643 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); 452 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
644 453
645 EIO_DESTROY (req); 454 EIO_DESTROY (req);
646} 455}
647 456
457static int
648static int eio_finish (eio_req *req) 458eio_finish (eio_req *req)
649{ 459{
650 int res = EIO_FINISH (req); 460 int res = EIO_FINISH (req);
651 461
652 if (req->grp) 462 if (req->grp)
653 { 463 {
… …
661 if (grp->grp_first == req) 471 if (grp->grp_first == req)
662 grp->grp_first = req->grp_next; 472 grp->grp_first = req->grp_next;
663 473
664 res2 = grp_dec (grp); 474 res2 = grp_dec (grp);
665 475
666 if (!res && res2) 476 if (!res)
667 res = res2; 477 res = res2;
668 } 478 }
669 479
670 eio_destroy (req); 480 eio_destroy (req);
671 481
672 return res; 482 return res;
673} 483}
674 484
485void
675void eio_grp_cancel (eio_req *grp) 486eio_grp_cancel (eio_req *grp)
676{ 487{
677 for (grp = grp->grp_first; grp; grp = grp->grp_next) 488 etp_grp_cancel (EIO_POOL, grp);
678 eio_cancel (grp);
679} 489}
680 490
491void
681void eio_cancel (eio_req *req) 492eio_cancel (eio_req *req)
682{ 493{
683 etp_cancel (req); 494 etp_cancel (EIO_POOL, req);
684} 495}
685 496
497void
686void eio_submit (eio_req *req) 498eio_submit (eio_req *req)
687{ 499{
688 etp_submit (req); 500 etp_submit (EIO_POOL, req);
689} 501}
690 502
691unsigned int eio_nreqs (void) 503unsigned int
504eio_nreqs (void)
692{ 505{
693 return etp_nreqs (); 506 return etp_nreqs (EIO_POOL);
694} 507}
695 508
696unsigned int eio_nready (void) 509unsigned int
510eio_nready (void)
697{ 511{
698 return etp_nready (); 512 return etp_nready (EIO_POOL);
699} 513}
700 514
701unsigned int eio_npending (void) 515unsigned int
516eio_npending (void)
702{ 517{
703 return etp_npending (); 518 return etp_npending (EIO_POOL);
704} 519}
705 520
706unsigned int eio_nthreads (void) 521unsigned int ecb_cold
522eio_nthreads (void)
707{ 523{
708 return etp_nthreads (); 524 return etp_nthreads (EIO_POOL);
709} 525}
710 526
527void ecb_cold
711void eio_set_max_poll_time (double nseconds) 528eio_set_max_poll_time (double nseconds)
712{ 529{
713 etp_set_max_poll_time (nseconds); 530 etp_set_max_poll_time (EIO_POOL, nseconds);
714} 531}
715 532
533void ecb_cold
716void eio_set_max_poll_reqs (unsigned int maxreqs) 534eio_set_max_poll_reqs (unsigned int maxreqs)
717{ 535{
718 etp_set_max_poll_reqs (maxreqs); 536 etp_set_max_poll_reqs (EIO_POOL, maxreqs);
719} 537}
720 538
539void ecb_cold
721void eio_set_max_idle (unsigned int nthreads) 540eio_set_max_idle (unsigned int nthreads)
722{ 541{
723 etp_set_max_idle (nthreads); 542 etp_set_max_idle (EIO_POOL, nthreads);
724} 543}
725 544
545void ecb_cold
546eio_set_idle_timeout (unsigned int seconds)
547{
548 etp_set_idle_timeout (EIO_POOL, seconds);
549}
550
551void ecb_cold
726void eio_set_min_parallel (unsigned int nthreads) 552eio_set_min_parallel (unsigned int nthreads)
727{ 553{
728 etp_set_min_parallel (nthreads); 554 etp_set_min_parallel (EIO_POOL, nthreads);
729} 555}
730 556
557void ecb_cold
731void eio_set_max_parallel (unsigned int nthreads) 558eio_set_max_parallel (unsigned int nthreads)
732{ 559{
733 etp_set_max_parallel (nthreads); 560 etp_set_max_parallel (EIO_POOL, nthreads);
734} 561}
735 562
736int eio_poll (void) 563int eio_poll (void)
737{ 564{
738 return etp_poll (); 565 return etp_poll (EIO_POOL);
739} 566}
740 567
741/*****************************************************************************/ 568/*****************************************************************************/
742/* work around various missing functions */ 569/* work around various missing functions */
743 570
744#if !HAVE_PREADWRITE 571#if HAVE_POSIX_CLOSE && !__linux
745# undef pread 572# define eio__close(fd) posix_close (fd, 0)
746# undef pwrite 573#else
747# define pread eio__pread 574# define eio__close(fd) close (fd)
748# define pwrite eio__pwrite
749
750static ssize_t
751eio__pread (int fd, void *buf, size_t count, off_t offset)
752{
753 ssize_t res;
754 off_t ooffset;
755
756 X_LOCK (preadwritelock);
757 ooffset = lseek (fd, 0, SEEK_CUR);
758 lseek (fd, offset, SEEK_SET);
759 res = read (fd, buf, count);
760 lseek (fd, ooffset, SEEK_SET);
761 X_UNLOCK (preadwritelock);
762
763 return res;
764}
765
766static ssize_t
767eio__pwrite (int fd, void *buf, size_t count, off_t offset)
768{
769 ssize_t res;
770 off_t ooffset;
771
772 X_LOCK (preadwritelock);
773 ooffset = lseek (fd, 0, SEEK_CUR);
774 lseek (fd, offset, SEEK_SET);
775 res = write (fd, buf, count);
776 lseek (fd, ooffset, SEEK_SET);
777 X_UNLOCK (preadwritelock);
778
779 return res;
780}
781#endif 575#endif
782 576
577/* close() without disturbing errno */
578static void
579silent_close (int fd)
580{
581 int saved_errno = errno;
582 eio__close (fd);
583 errno = saved_errno;
584}
585
783#ifndef HAVE_FUTIMES 586#ifndef HAVE_UTIMES
784 587
785# undef utimes 588# undef utimes
786# undef futimes
787# define utimes(path,times) eio__utimes (path, times) 589# define utimes(path,times) eio__utimes (path, times)
788# define futimes(fd,times) eio__futimes (fd, times)
789 590
790static int 591static int
791eio__utimes (const char *filename, const struct timeval times[2]) 592eio__utimes (const char *filename, const struct timeval times[2])
792{ 593{
793 if (times) 594 if (times)
… …
801 } 602 }
802 else 603 else
803 return utime (filename, 0); 604 return utime (filename, 0);
804} 605}
805 606
607#endif
608
609#ifndef HAVE_FUTIMES
610
611# undef futimes
612# define futimes(fd,times) eio__futimes (fd, times)
613
614static int
806static int eio__futimes (int fd, const struct timeval tv[2]) 615eio__futimes (int fd, const struct timeval tv[2])
807{ 616{
808 errno = ENOSYS; 617 errno = ENOSYS;
809 return -1; 618 return -1;
810} 619}
811 620
… …
814#if !HAVE_FDATASYNC 623#if !HAVE_FDATASYNC
815# undef fdatasync 624# undef fdatasync
816# define fdatasync(fd) fsync (fd) 625# define fdatasync(fd) fsync (fd)
817#endif 626#endif
818 627
628static int
629eio__syncfs (int fd)
630{
631 int res;
632
633#if HAVE_SYS_SYNCFS
634 res = (int)syscall (__NR_syncfs, (int)(fd));
635#else
636 res = EIO_ENOSYS ();
637#endif
638
639 if (res < 0 && errno == ENOSYS && fd >= 0)
640 sync ();
641
642 return res;
643}
644
819/* sync_file_range always needs emulation */ 645/* sync_file_range always needs emulation */
820int 646static int
821eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags) 647eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
822{ 648{
823#if HAVE_SYNC_FILE_RANGE 649#if HAVE_SYNC_FILE_RANGE
824 int res; 650 int res;
825 651
… …
838 if (!res || errno != ENOSYS) 664 if (!res || errno != ENOSYS)
839 return res; 665 return res;
840#endif 666#endif
841 667
842 /* even though we could play tricks with the flags, it's better to always 668 /* even though we could play tricks with the flags, it's better to always
843 * call fdatasync, as thta matches the expectation of it's users best */ 669 * call fdatasync, as that matches the expectation of its users best */
844 return fdatasync (fd); 670 return fdatasync (fd);
671}
672
673static int
674eio__fallocate (int fd, int mode, off_t offset, size_t len)
675{
676#if HAVE_LINUX_FALLOCATE
677 return fallocate (fd, mode, offset, len);
678#else
679 return EIO_ENOSYS ();
680#endif
845} 681}
846 682
847#if !HAVE_READAHEAD 683#if !HAVE_READAHEAD
848# undef readahead 684# undef readahead
849# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 685# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
850 686
851static ssize_t 687static eio_ssize_t
852eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) 688eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
853{ 689{
854 size_t todo = count; 690 size_t todo = count;
855 dBUF; 691 dBUF;
856 692
… …
861 pread (fd, eio_buf, len, offset); 697 pread (fd, eio_buf, len, offset);
862 offset += len; 698 offset += len;
863 todo -= len; 699 todo -= len;
864 } 700 }
865 701
866 errno = 0; 702 FUBd;
703
704 /* linux's readahead basically only fails for EBADF or EINVAL (not mmappable) */
705 /* but not for e.g. EIO or eof, so we also never fail */
867 return count; 706 return 0;
868} 707}
869 708
870#endif 709#endif
871 710
872/* sendfile always needs emulation */ 711/* sendfile always needs emulation */
873static ssize_t 712static eio_ssize_t
874eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) 713eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
875{ 714{
715 eio_ssize_t written = 0;
876 ssize_t res; 716 eio_ssize_t res;
877 717
878 if (!count) 718 if (!count)
879 return 0; 719 return 0;
880 720
721 for (;;)
722 {
723#ifdef __APPLE__
724# undef HAVE_SENDFILE /* broken, as everything on os x */
725#endif
881#if HAVE_SENDFILE 726#if HAVE_SENDFILE
882# if __linux 727# if __linux
728 off_t soffset = offset;
883 res = sendfile (ofd, ifd, &offset, count); 729 res = sendfile (ofd, ifd, &soffset, count);
884 730
885# elif __freebsd 731# elif __FreeBSD__
886 /* 732 /*
887 * Of course, the freebsd sendfile is a dire hack with no thoughts 733 * Of course, the freebsd sendfile is a dire hack with no thoughts
888 * wasted on making it similar to other I/O functions. 734 * wasted on making it similar to other I/O functions.
889 */ 735 */
890 {
891 off_t sbytes; 736 off_t sbytes;
892 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 737 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
893 738
894 if (res < 0 && sbytes) 739 #if 0 /* according to the manpage, this is correct, but broken behaviour */
895 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 740 /* freebsd' sendfile will return 0 on success */
741 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
742 /* not on e.g. EIO or EPIPE - sounds broken */
743 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
896 res = sbytes; 744 res = sbytes;
897 } 745 #endif
746
747 /* according to source inspection, this is correct, and useful behaviour */
748 if (sbytes)
749 res = sbytes;
750
751# elif defined __APPLE__
752 off_t sbytes = count;
753 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
754
755 /* according to the manpage, sbytes is always valid */
756 if (sbytes)
757 res = sbytes;
898 758
899# elif __hpux 759# elif __hpux
900 res = sendfile (ofd, ifd, offset, count, 0, 0); 760 res = sendfile (ofd, ifd, offset, count, 0, 0);
901 761
902# elif __solaris 762# elif __solaris
903 {
904 struct sendfilevec vec; 763 struct sendfilevec vec;
905 size_t sbytes; 764 size_t sbytes;
906 765
907 vec.sfv_fd = ifd; 766 vec.sfv_fd = ifd;
908 vec.sfv_flag = 0; 767 vec.sfv_flag = 0;
909 vec.sfv_off = offset; 768 vec.sfv_off = offset;
910 vec.sfv_len = count; 769 vec.sfv_len = count;
911 770
912 res = sendfilev (ofd, &vec, 1, &sbytes); 771 res = sendfilev (ofd, &vec, 1, &sbytes);
913 772
914 if (res < 0 && sbytes) 773 if (res < 0 && sbytes)
915 res = sbytes; 774 res = sbytes;
916 }
917 775
918# endif 776# endif
777
778#elif defined (_WIN32) && 0
779 /* does not work, just for documentation of what would need to be done */
780 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
781 /* libeio guarantees that the file offset does not change, and windows */
782 /* has no way to get an independent handle to the same file description */
783 HANDLE h = TO_SOCKET (ifd);
784 SetFilePointer (h, offset, 0, FILE_BEGIN);
785 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
786
919#else 787#else
920 res = -1; 788 res = EIO_ENOSYS ();
921 errno = ENOSYS;
922#endif 789#endif
923 790
791 /* we assume sendfile can copy at least 128mb in one go */
792 if (res <= 128 * 1024 * 1024)
793 {
794 if (res > 0)
795 written += res;
796
797 if (written)
798 return written;
799
800 break;
801 }
802 else
803 {
804 /* if we requested more, then probably the kernel was lazy */
805 written += res;
806 offset += res;
807 count -= res;
808
809 if (!count)
810 return written;
811 }
812 }
813
924 if (res < 0 814 if (res < 0
925 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 815 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
816 /* BSDs */
817#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
818 || errno == ENOTSUP
819#endif
820#ifdef EOPNOTSUPP /* windows */
821 || errno == EOPNOTSUPP /* BSDs */
822#endif
926#if __solaris 823#if __solaris
927 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 824 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
928#endif 825#endif
929 ) 826 )
930 ) 827 )
… …
934 831
935 res = 0; 832 res = 0;
936 833
937 while (count) 834 while (count)
938 { 835 {
939 ssize_t cnt; 836 eio_ssize_t cnt;
940 837
941 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 838 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
942 839
943 if (cnt <= 0) 840 if (cnt <= 0)
944 { 841 {
… …
956 853
957 offset += cnt; 854 offset += cnt;
958 res += cnt; 855 res += cnt;
959 count -= cnt; 856 count -= cnt;
960 } 857 }
858
859 FUBd;
961 } 860 }
962 861
963 return res; 862 return res;
863}
864
865#ifdef PAGESIZE
866# define eio_pagesize() PAGESIZE
867#else
868static intptr_t
869eio_pagesize (void)
870{
871 static intptr_t page;
872
873 if (!page)
874 page = sysconf (_SC_PAGESIZE);
875
876 return page;
877}
878#endif
879
880static void
881eio_page_align (void **addr, size_t *length)
882{
883 intptr_t mask = eio_pagesize () - 1;
884
885 /* round down addr */
886 intptr_t adj = mask & (intptr_t)*addr;
887
888 *addr = (void *)((intptr_t)*addr - adj);
889 *length += adj;
890
891 /* round up length */
892 *length = (*length + mask) & ~mask;
893}
894
895#if !_POSIX_MEMLOCK
896# define eio__mlockall(a) EIO_ENOSYS ()
897#else
898
899static int
900eio__mlockall (int flags)
901{
902 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
903 extern int mallopt (int, int);
904 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
905 #endif
906
907 #ifndef MCL_ONFAULT
908 if (flags & EIO_MCL_ONFAULT)
909 return EIO_ERRNO (EINVAL, -1);
910 #define MCL_ONFAULT 4
911 #endif
912
913 if (EIO_MCL_CURRENT != MCL_CURRENT
914 || EIO_MCL_FUTURE != MCL_FUTURE
915 || EIO_MCL_ONFAULT != MCL_ONFAULT)
916 {
917 flags = 0
918 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
919 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0)
920 | (flags & EIO_MCL_ONFAULT ? MCL_ONFAULT : 0)
921 ;
922 }
923
924 return mlockall (flags);
925}
926#endif
927
928#if !_POSIX_MEMLOCK_RANGE
929# define eio__mlock(a,b) EIO_ENOSYS ()
930#else
931
932static int
933eio__mlock (void *addr, size_t length)
934{
935 eio_page_align (&addr, &length);
936
937 return mlock (addr, length);
938}
939
940#endif
941
942#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
943# define eio__msync(a,b,c) EIO_ENOSYS ()
944#else
945
946static int
947eio__msync (void *mem, size_t len, int flags)
948{
949 eio_page_align (&mem, &len);
950
951 if (EIO_MS_ASYNC != MS_SYNC
952 || EIO_MS_INVALIDATE != MS_INVALIDATE
953 || EIO_MS_SYNC != MS_SYNC)
954 {
955 flags = 0
956 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
957 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
958 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
959 }
960
961 return msync (mem, len, flags);
962}
963
964#endif
965
966static int
967eio__mtouch (eio_req *req)
968{
969 void *mem = req->ptr2;
970 size_t len = req->size;
971 int flags = req->int1;
972
973 eio_page_align (&mem, &len);
974
975 {
976 intptr_t addr = (intptr_t)mem;
977 intptr_t end = addr + len;
978 intptr_t page = eio_pagesize ();
979
980 if (addr < end)
981 if (flags & EIO_MT_MODIFY) /* modify */
982 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < end && !EIO_CANCELLED (req));
983 else
984 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < end && !EIO_CANCELLED (req));
985 }
986
987 return 0;
988}
989
990/*****************************************************************************/
991/* requests implemented outside eio_execute, because they are so large */
992
993static void
994eio__lseek (eio_req *req)
995{
996 /* this usually gets optimised away completely, or your compiler sucks, */
997 /* or the whence constants really are not 0, 1, 2 */
998 int whence = req->int2 == EIO_SEEK_SET ? SEEK_SET
999 : req->int2 == EIO_SEEK_CUR ? SEEK_CUR
1000 : req->int2 == EIO_SEEK_END ? SEEK_END
1001 : req->int2;
1002
1003 req->offs = lseek (req->int1, req->offs, whence);
1004 req->result = req->offs == (off_t)-1 ? -1 : 0;
1005}
1006
1007/* result will always end up in tmpbuf, there is always space for adding a 0-byte */
1008static int
1009eio__realpath (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
1010{
1011 char *res;
1012 const char *rel = path;
1013 char *tmp1, *tmp2;
1014#if SYMLOOP_MAX > 32
1015 int symlinks = SYMLOOP_MAX;
1016#else
1017 int symlinks = 32;
1018#endif
1019
1020 errno = EINVAL;
1021 if (!rel)
1022 return -1;
1023
1024 errno = ENOENT;
1025 if (!*rel)
1026 return -1;
1027
1028 res = etp_tmpbuf_get (tmpbuf, EIO_PATH_MAX * 3);
1029#ifdef _WIN32
1030 if (_access (rel, 4) != 0)
1031 return -1;
1032
1033 symlinks = GetFullPathName (rel, EIO_PATH_MAX * 3, res, 0);
1034
1035 errno = ENAMETOOLONG;
1036 if (symlinks >= EIO_PATH_MAX * 3)
1037 return -1;
1038
1039 errno = EIO;
1040 if (symlinks <= 0)
1041 return -1;
1042
1043 return symlinks;
1044
1045#else
1046 tmp1 = res + EIO_PATH_MAX;
1047 tmp2 = tmp1 + EIO_PATH_MAX;
1048
1049#if 0 /* disabled, the musl way to do things is just too racy */
1050#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1051 /* on linux we may be able to ask the kernel */
1052 {
1053 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1054
1055 if (fd >= 0)
1056 {
1057 sprintf (tmp1, "/proc/self/fd/%d", fd);
1058 req->result = readlink (tmp1, res, EIO_PATH_MAX);
1059 /* here we should probably stat the open file and the disk file, to make sure they still match */
1060 eio__close (fd);
1061
1062 if (req->result > 0)
1063 goto done;
1064 }
1065 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1066 return -1;
1067 }
1068#endif
1069#endif
1070
1071 if (*rel != '/')
1072 {
1073 int len;
1074
1075 errno = ENOENT;
1076 if (wd == EIO_INVALID_WD)
1077 return -1;
1078
1079 if (wd == EIO_CWD)
1080 {
1081 if (!getcwd (res, EIO_PATH_MAX))
1082 return -1;
1083
1084 len = strlen (res);
1085 }
1086 else
1087 memcpy (res, wd->str, len = wd->len);
1088
1089 if (res [1]) /* only use if not / */
1090 res += len;
1091 }
1092
1093 while (*rel)
1094 {
1095 eio_ssize_t len, linklen;
1096 const char *beg = rel;
1097
1098 while (*rel && *rel != '/')
1099 ++rel;
1100
1101 len = rel - beg;
1102
1103 if (!len) /* skip slashes */
1104 {
1105 ++rel;
1106 continue;
1107 }
1108
1109 if (beg [0] == '.')
1110 {
1111 if (len == 1)
1112 continue; /* . - nop */
1113
1114 if (beg [1] == '.' && len == 2)
1115 {
1116 /* .. - back up one component, if possible */
1117
1118 while (res != tmpbuf->ptr)
1119 if (*--res == '/')
1120 break;
1121
1122 continue;
1123 }
1124 }
1125
1126 errno = ENAMETOOLONG;
1127 if (res + 1 + len + 1 >= tmp1)
1128 return -1;
1129
1130 /* copy one component */
1131 *res = '/';
1132 memcpy (res + 1, beg, len);
1133
1134 /* zero-terminate, for readlink */
1135 res [len + 1] = 0;
1136
1137 /* now check if it's a symlink */
1138 linklen = readlink (tmpbuf->ptr, tmp1, EIO_PATH_MAX);
1139
1140 if (linklen < 0)
1141 {
1142 if (errno != EINVAL)
1143 return -1;
1144
1145 /* it's a normal directory. hopefully */
1146 res += len + 1;
1147 }
1148 else
1149 {
1150 /* yay, it was a symlink - build new path in tmp2 */
1151 int rellen = strlen (rel);
1152
1153 errno = ENAMETOOLONG;
1154 if (linklen + 1 + rellen >= EIO_PATH_MAX) /* also catch linklen >= EIO_PATH_MAX */
1155 return -1;
1156
1157 errno = ELOOP;
1158 if (!--symlinks)
1159 return -1;
1160
1161 if (*tmp1 == '/')
1162 res = tmpbuf->ptr; /* symlink resolves to an absolute path */
1163
1164 /* we need to be careful, as rel might point into tmp2 already */
1165 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1166 tmp2 [linklen] = '/';
1167 memcpy (tmp2, tmp1, linklen);
1168
1169 rel = tmp2;
1170 }
1171 }
1172
1173 /* special case for the lone root path */
1174 if (res == tmpbuf->ptr)
1175 *res++ = '/';
1176
1177 return res - (char *)tmpbuf->ptr;
1178#endif
1179}
1180
1181static signed char
1182eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1183{
1184 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1185 : a->inode < b->inode ? -1
1186 : a->inode > b->inode ? 1
1187 : 0;
1188}
1189
1190#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1191
1192#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1193#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1194
1195static void
1196eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1197{
1198 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1199 unsigned char *bit = bits;
1200
1201 assert (CHAR_BIT == 8);
1202 assert (sizeof (eio_dirent) * 8 < 256);
1203 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1204 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1205
1206 if (size <= EIO_SORT_FAST)
1207 return;
1208
1209 /* first prepare an array of bits to test in our radix sort */
1210 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1211 /* inode_bits must contain all inodes ORed together */
1212 /* which is used to skip bits that are 0 everywhere, which is very common */
1213 {
1214 eio_ino_t endianness;
1215 int i, j;
1216
1217 /* we store the byte offset of byte n into byte n of "endianness" */
1218 for (i = 0; i < sizeof (eio_ino_t); ++i)
1219 ((unsigned char *)&endianness)[i] = i;
1220
1221 *bit++ = 0;
1222
1223 for (i = 0; i < sizeof (eio_ino_t); ++i)
1224 {
1225 /* shifting off the byte offsets out of "endianness" */
1226 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1227 endianness >>= 8;
1228
1229 for (j = 0; j < 8; ++j)
1230 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1231 *bit++ = offs + j;
1232 }
1233
1234 for (j = 0; j < 8; ++j)
1235 if (score_bits & (1 << j))
1236 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1237 }
1238
1239 /* now actually do the sorting (a variant of MSD radix sort) */
1240 {
1241 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1242 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1243 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1244 int stk_idx = 0;
1245
1246 base_stk [stk_idx] = dents;
1247 end_stk [stk_idx] = dents + size;
1248 bit_stk [stk_idx] = bit - 1;
1249
1250 do
1251 {
1252 base = base_stk [stk_idx];
1253 end = end_stk [stk_idx];
1254 bit = bit_stk [stk_idx];
1255
1256 for (;;)
1257 {
1258 unsigned char O = *bit >> 3;
1259 unsigned char M = 1 << (*bit & 7);
1260
1261 eio_dirent *a = base;
1262 eio_dirent *b = end;
1263
1264 if (b - a < EIO_SORT_CUTOFF)
1265 break;
1266
1267 /* now bit-partition the array on the bit */
1268 /* this ugly asymmetric loop seems to perform much better than typical */
1269 /* partition algos found in the literature */
1270 do
1271 if (!(((unsigned char *)a)[O] & M))
1272 ++a;
1273 else if (!(((unsigned char *)--b)[O] & M))
1274 {
1275 eio_dirent tmp = *a; *a = *b; *b = tmp;
1276 ++a;
1277 }
1278 while (b > a);
1279
1280 /* next bit, or stop, if no bits left in this path */
1281 if (!*--bit)
1282 break;
1283
1284 base_stk [stk_idx] = a;
1285 end_stk [stk_idx] = end;
1286 bit_stk [stk_idx] = bit;
1287 ++stk_idx;
1288
1289 end = a;
1290 }
1291 }
1292 while (stk_idx--);
1293 }
1294}
1295
1296static void
1297eio_dent_insertion_sort (eio_dirent *dents, int size)
1298{
1299 /* first move the smallest element to the front, to act as a sentinel */
1300 {
1301 int i;
1302 eio_dirent *min = dents;
1303
1304 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1305 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1306 if (EIO_DENT_CMP (dents [i], <, *min))
1307 min = &dents [i];
1308
1309 /* swap elements 0 and j (minimum) */
1310 {
1311 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1312 }
1313 }
1314
1315 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1316 {
1317 eio_dirent *i, *j;
1318
1319 for (i = dents + 1; i < dents + size; ++i)
1320 {
1321 eio_dirent value = *i;
1322
1323 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1324 j [1] = j [0];
1325
1326 j [1] = value;
1327 }
1328 }
1329}
1330
1331static void
1332eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1333{
1334 if (size <= 1)
1335 return; /* our insertion sort relies on size > 0 */
1336
1337 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1338 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1339 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1340
1341 /* use an insertion sort at the end, or for small arrays, */
1342 /* as insertion sort is more efficient for small partitions */
1343 eio_dent_insertion_sort (dents, size);
964} 1344}
965 1345
966/* read a full directory */ 1346/* read a full directory */
967static void 1347static void
968eio__scandir (eio_req *req, etp_worker *self) 1348eio__scandir (eio_req *req, etp_worker *self)
969{ 1349{
1350 char *name, *names;
1351 int namesalloc = 4096 - sizeof (void *) * 4;
1352 int namesoffs = 0;
1353 int flags = req->int1;
1354 eio_dirent *dents = 0;
1355 int dentalloc = 128;
1356 int dentoffs = 0;
1357 eio_ino_t inode_bits = 0;
1358#ifdef _WIN32
1359 HANDLE dirp;
1360 WIN32_FIND_DATA entp;
1361#else
970 DIR *dirp; 1362 DIR *dirp;
971 EIO_STRUCT_DIRENT *entp; 1363 EIO_STRUCT_DIRENT *entp;
972 char *name, *names; 1364#endif
973 int memlen = 4096;
974 int memofs = 0;
975 int res = 0;
976 1365
977 X_LOCK (wrklock); 1366 req->result = -1;
978 /* the corresponding closedir is in ETP_WORKER_CLEAR */
979 self->dirp = dirp = opendir (req->ptr1);
980 req->flags |= EIO_FLAG_PTR2_FREE;
981 req->ptr2 = names = malloc (memlen);
982 X_UNLOCK (wrklock);
983 1367
984 if (dirp && names) 1368 if (!(flags & EIO_READDIR_DENTS))
985 for (;;) 1369 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1370
1371#ifdef _WIN32
1372 {
1373 int len = strlen ((const char *)req->ptr1);
1374 char *path = malloc (MAX_PATH);
1375 const char *fmt;
1376 const char *reqpath = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1377
1378 if (!len)
1379 fmt = "./*";
1380 else if (reqpath[len - 1] == '/' || reqpath[len - 1] == '\\')
1381 fmt = "%s*";
1382 else
1383 fmt = "%s/*";
1384
1385 _snprintf (path, MAX_PATH, fmt, reqpath);
1386 dirp = FindFirstFile (path, &entp);
1387 free (path);
1388
1389 if (dirp == INVALID_HANDLE_VALUE)
986 { 1390 {
1391 /* should steal _dosmaperr */
1392 switch (GetLastError ())
1393 {
1394 case ERROR_FILE_NOT_FOUND:
1395 req->result = 0;
1396 break;
1397
1398 case ERROR_INVALID_NAME:
1399 case ERROR_PATH_NOT_FOUND:
1400 case ERROR_NO_MORE_FILES:
1401 errno = ENOENT;
1402 break;
1403
1404 case ERROR_NOT_ENOUGH_MEMORY:
1405 errno = ENOMEM;
1406 break;
1407
1408 default:
1409 errno = EINVAL;
1410 break;
1411 }
1412
1413 return;
1414 }
1415 }
1416#else
1417 #if HAVE_AT
1418 {
1419 int fd = openat (WD2FD (req->wd), req->ptr1, O_CLOEXEC | O_SEARCH | O_DIRECTORY | O_NONBLOCK);
1420
1421 if (fd < 0)
1422 return;
1423
1424 dirp = fdopendir (fd);
1425
1426 if (!dirp)
1427 {
1428 silent_close (fd);
1429 return;
1430 }
1431 }
1432 #else
1433 dirp = opendir (wd_expand (&self->tmpbuf, req->wd, req->ptr1));
1434
1435 if (!dirp)
1436 return;
1437 #endif
1438#endif
1439
1440 if (req->flags & EIO_FLAG_PTR1_FREE)
1441 free (req->ptr1);
1442
1443 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1444 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1445 req->ptr2 = names = malloc (namesalloc);
1446
1447 if (!names || (flags && !dents))
1448 return;
1449
1450 for (;;)
1451 {
1452 int done;
1453
1454#ifdef _WIN32
1455 done = !dirp;
1456#else
987 errno = 0; 1457 errno = 0;
988 entp = readdir (dirp); 1458 entp = readdir (dirp);
1459 done = !entp;
1460#endif
989 1461
1462 if (done)
1463 {
1464#ifndef _WIN32
1465 int old_errno = errno;
1466 closedir (dirp);
1467 errno = old_errno;
1468
990 if (!entp) 1469 if (errno)
1470 break;
1471#endif
1472
1473 /* sort etc. */
1474 req->int1 = flags;
1475 req->result = dentoffs;
1476
1477 if (flags & EIO_READDIR_STAT_ORDER)
1478 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1479 else if (flags & EIO_READDIR_DIRS_FIRST)
1480 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1481 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1482 else
1483 {
1484 /* in this case, all is known, and we just put dirs first and sort them */
1485 eio_dirent *oth = dents + dentoffs;
1486 eio_dirent *dir = dents;
1487
1488 /* now partition dirs to the front, and non-dirs to the back */
1489 /* by walking from both sides and swapping if necessary */
1490 while (oth > dir)
1491 {
1492 if (dir->type == EIO_DT_DIR)
1493 ++dir;
1494 else if ((--oth)->type == EIO_DT_DIR)
1495 {
1496 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1497
1498 ++dir;
1499 }
1500 }
1501
1502 /* now sort the dirs only (dirs all have the same score) */
1503 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1504 }
1505
991 break; 1506 break;
1507 }
992 1508
993 name = entp->d_name; 1509 /* now add the entry to our list(s) */
1510 name = D_NAME (entp);
994 1511
1512 /* skip . and .. entries */
995 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1513 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
996 { 1514 {
997 int len = strlen (name) + 1; 1515 int len = D_NAMLEN (entp) + 1;
998 1516
999 res++; 1517 while (ecb_expect_false (namesoffs + len > namesalloc))
1000
1001 while (memofs + len > memlen)
1002 { 1518 {
1003 memlen *= 2; 1519 namesalloc *= 2;
1004 X_LOCK (wrklock);
1005 req->ptr2 = names = realloc (names, memlen); 1520 req->ptr2 = names = realloc (names, namesalloc);
1006 X_UNLOCK (wrklock);
1007 1521
1008 if (!names) 1522 if (!names)
1523 break;
1524 }
1525
1526 memcpy (names + namesoffs, name, len);
1527
1528 if (dents)
1529 {
1530 struct eio_dirent *ent;
1531
1532 if (ecb_expect_false (dentoffs == dentalloc))
1533 {
1534 dentalloc *= 2;
1535 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1536
1537 if (!dents)
1009 break; 1538 break;
1010 } 1539 }
1011 1540
1012 memcpy (names + memofs, name, len); 1541 ent = dents + dentoffs;
1013 memofs += len; 1542
1543 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1544 ent->namelen = len - 1;
1545 ent->inode = D_INO (entp);
1546
1547 inode_bits |= ent->inode;
1548
1549 switch (D_TYPE (entp))
1550 {
1551 default:
1552 ent->type = EIO_DT_UNKNOWN;
1553 flags |= EIO_READDIR_FOUND_UNKNOWN;
1554 break;
1555
1556 #ifdef DT_FIFO
1557 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1558 #endif
1559 #ifdef DT_CHR
1560 case DT_CHR: ent->type = EIO_DT_CHR; break;
1561 #endif
1562 #ifdef DT_MPC
1563 case DT_MPC: ent->type = EIO_DT_MPC; break;
1564 #endif
1565 #ifdef DT_DIR
1566 case DT_DIR: ent->type = EIO_DT_DIR; break;
1567 #endif
1568 #ifdef DT_NAM
1569 case DT_NAM: ent->type = EIO_DT_NAM; break;
1570 #endif
1571 #ifdef DT_BLK
1572 case DT_BLK: ent->type = EIO_DT_BLK; break;
1573 #endif
1574 #ifdef DT_MPB
1575 case DT_MPB: ent->type = EIO_DT_MPB; break;
1576 #endif
1577 #ifdef DT_REG
1578 case DT_REG: ent->type = EIO_DT_REG; break;
1579 #endif
1580 #ifdef DT_NWK
1581 case DT_NWK: ent->type = EIO_DT_NWK; break;
1582 #endif
1583 #ifdef DT_CMP
1584 case DT_CMP: ent->type = EIO_DT_CMP; break;
1585 #endif
1586 #ifdef DT_LNK
1587 case DT_LNK: ent->type = EIO_DT_LNK; break;
1588 #endif
1589 #ifdef DT_SOCK
1590 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1591 #endif
1592 #ifdef DT_DOOR
1593 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1594 #endif
1595 #ifdef DT_WHT
1596 case DT_WHT: ent->type = EIO_DT_WHT; break;
1597 #endif
1598 }
1599
1600 ent->score = 7;
1601
1602 if (flags & EIO_READDIR_DIRS_FIRST)
1603 {
1604 if (ent->type == EIO_DT_UNKNOWN)
1605 {
1606 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1607 ent->score = 1;
1608 else if (!strchr (name, '.')) /* absence of dots indicate likely dirs */
1609 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1610 }
1611 else if (ent->type == EIO_DT_DIR)
1612 ent->score = 0;
1613 }
1014 } 1614 }
1615
1616 namesoffs += len;
1617 ++dentoffs;
1015 } 1618 }
1016 1619
1017 if (errno) 1620 if (EIO_CANCELLED (req))
1018 res = -1; 1621 {
1622 errno = ECANCELED;
1623 break;
1624 }
1625
1626#ifdef _WIN32
1627 if (!FindNextFile (dirp, &entp))
1628 {
1629 FindClose (dirp);
1630 dirp = 0;
1631 }
1632#endif
1633 }
1634}
1635
1636/*****************************************************************************/
1637/* working directory stuff */
1638/* various deficiencies in the posix 2008 api force us to */
1639/* keep the absolute path in string form at all times */
1640/* fuck yeah. */
1641
1642#if !HAVE_AT
1643
1644/* a bit like realpath, but usually faster because it doesn'T have to return */
1645/* an absolute or canonical path */
1646static const char *
1647wd_expand (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
1648{
1649 if (!wd || *path == '/')
1650 return path;
1651
1652 if (path [0] == '.' && !path [1])
1653 return wd->str;
1654
1019 1655 {
1020 req->result = res; 1656 int l1 = wd->len;
1021} 1657 int l2 = strlen (path);
1022 1658
1023#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) 1659 char *res = etp_tmpbuf_get (tmpbuf, l1 + l2 + 2);
1024# undef msync
1025# define msync(a,b,c) ((errno = ENOSYS), -1)
1026#endif
1027 1660
1028int 1661 memcpy (res, wd->str, l1);
1029eio__mtouch (void *mem, size_t len, int flags) 1662 res [l1] = '/';
1663 memcpy (res + l1 + 1, path, l2 + 1);
1664
1665 return res;
1666 }
1667}
1668
1669#endif
1670
1671static eio_wd
1672eio__wd_open_sync (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
1030{ 1673{
1031 intptr_t addr = (intptr_t)mem; 1674 int fd;
1032 intptr_t end = addr + len; 1675 eio_wd res;
1033#ifdef PAGESIZE 1676 int len = eio__realpath (tmpbuf, wd, path);
1034 const intptr_t page = PAGESIZE; 1677
1678 if (len < 0)
1679 return EIO_INVALID_WD;
1680
1681#if HAVE_AT
1682 fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY | O_NONBLOCK);
1683
1684 if (fd < 0)
1685 return EIO_INVALID_WD;
1686#endif
1687
1688 res = malloc (sizeof (*res) + len); /* one extra 0-byte */
1689
1690#if HAVE_AT
1691 res->fd = fd;
1692#endif
1693
1694 res->len = len;
1695 memcpy (res->str, tmpbuf->ptr, len);
1696 res->str [len] = 0;
1697
1698 return res;
1699}
1700
1701eio_wd
1702eio_wd_open_sync (eio_wd wd, const char *path)
1703{
1704 struct etp_tmpbuf tmpbuf = { 0 };
1705 wd = eio__wd_open_sync (&tmpbuf, wd, path);
1706 free (tmpbuf.ptr);
1707
1708 return wd;
1709}
1710
1711void
1712eio_wd_close_sync (eio_wd wd)
1713{
1714 if (wd != EIO_INVALID_WD && wd != EIO_CWD)
1715 {
1716 #if HAVE_AT
1717 eio__close (wd->fd);
1718 #endif
1719 free (wd);
1720 }
1721}
1722
1723#if HAVE_AT
1724
1725static int
1726eio__renameat2 (int olddirfd, const char *oldpath, int newdirfd, const char *newpath, unsigned int flags)
1727{
1728#if HAVE_RENAMEAT2
1729 return syscall (SYS_renameat2, olddirfd, oldpath, newdirfd, newpath, flags);
1035#else 1730#else
1036 static intptr_t page;
1037
1038 if (!page) 1731 if (flags)
1039 page = sysconf (_SC_PAGESIZE); 1732 return EIO_ENOSYS ();
1040#endif
1041 1733
1042 addr &= ~(page - 1); /* assume page size is always a power of two */ 1734 return renameat (olddirfd, oldpath, newdirfd, newpath);
1735#endif
1736}
1043 1737
1044 if (addr < end) 1738/* they forgot these */
1045 if (flags) /* modify */
1046 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
1047 else
1048 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
1049 1739
1740static int
1741eio__truncateat (int dirfd, const char *path, off_t length)
1742{
1743 int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC | O_NONBLOCK);
1744 int res;
1745
1746 if (fd < 0)
1747 return fd;
1748
1749 res = ftruncate (fd, length);
1750 silent_close (fd);
1050 return 0; 1751 return res;
1051} 1752}
1753
1754static int
1755eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
1756{
1757 int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC | O_NONBLOCK);
1758 int res;
1759
1760 if (fd < 0)
1761 return fd;
1762
1763 res = fstatvfs (fd, buf);
1764 silent_close (fd);
1765 return res;
1766}
1767
1768#endif
1052 1769
1053/*****************************************************************************/ 1770/*****************************************************************************/
1054 1771
1055#define ALLOC(len) \ 1772#define ALLOC(len) \
1056 if (!req->ptr2) \ 1773 if (!req->ptr2) \
1057 { \ 1774 { \
1058 X_LOCK (wrklock); \ 1775 X_LOCK (EIO_POOL->wrklock); \
1059 req->flags |= EIO_FLAG_PTR2_FREE; \ 1776 req->flags |= EIO_FLAG_PTR2_FREE; \
1060 X_UNLOCK (wrklock); \ 1777 X_UNLOCK (EIO_POOL->wrklock); \
1061 req->ptr2 = malloc (len); \ 1778 req->ptr2 = malloc (len); \
1062 if (!req->ptr2) \ 1779 if (!req->ptr2) \
1063 { \ 1780 { \
1064 errno = ENOMEM; \ 1781 errno = ENOMEM; \
1065 req->result = -1; \ 1782 req->result = -1; \
1066 break; \ 1783 goto alloc_fail; \
1067 } \ 1784 } \
1068 } 1785 }
1069 1786
1070X_THREAD_PROC (etp_proc) 1787/*****************************************************************************/
1071{
1072 ETP_REQ *req;
1073 struct timespec ts;
1074 etp_worker *self = (etp_worker *)thr_arg;
1075 1788
1076 /* try to distribute timeouts somewhat randomly */ 1789static void
1077 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 1790eio__slurp (int fd, eio_req *req)
1791{
1792 req->result = fd;
1078 1793
1079 for (;;) 1794 if (fd < 0)
1795 return;
1796
1797 if (req->offs < 0 || !req->size) /* do we need the size? */
1080 { 1798 {
1081 X_LOCK (reqlock); 1799 off_t size = lseek (fd, 0, SEEK_END);
1082 1800
1083 for (;;)
1084 {
1085 self->req = req = reqq_shift (&req_queue);
1086
1087 if (req)
1088 break;
1089
1090 ++idle;
1091
1092 ts.tv_sec = time (0) + IDLE_TIMEOUT;
1093 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
1094 {
1095 if (idle > max_idle)
1096 {
1097 --idle;
1098 X_UNLOCK (reqlock);
1099 X_LOCK (wrklock);
1100 --started;
1101 X_UNLOCK (wrklock);
1102 goto quit;
1103 }
1104
1105 /* we are allowed to idle, so do so without any timeout */
1106 X_COND_WAIT (reqwait, reqlock);
1107 }
1108
1109 --idle;
1110 }
1111
1112 --nready;
1113
1114 X_UNLOCK (reqlock);
1115
1116 if (req->type < 0) 1801 if (req->offs < 0)
1117 goto quit; 1802 req->offs += size;
1118 1803
1119 if (!EIO_CANCELLED (req)) 1804 if (!req->size)
1120 ETP_EXECUTE (self, req); 1805 req->size = size - req->offs;
1121
1122 X_LOCK (reslock);
1123
1124 ++npending;
1125
1126 if (!reqq_push (&res_queue, req) && want_poll_cb)
1127 want_poll_cb ();
1128
1129 self->req = 0;
1130 etp_worker_clear (self);
1131
1132 X_UNLOCK (reslock);
1133 } 1806 }
1134 1807
1135quit: 1808 ALLOC (req->size);
1136 X_LOCK (wrklock); 1809 req->result = pread (fd, req->ptr2, req->size, req->offs);
1137 etp_worker_free (self);
1138 X_UNLOCK (wrklock);
1139 1810
1140 return 0; 1811 silent_close (fd);
1141}
1142 1812
1143/*****************************************************************************/ 1813alloc_fail:
1814 ;
1815}
1144 1816
1817int ecb_cold
1145int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1818eio_init (void (*want_poll)(void), void (*done_poll)(void))
1146{ 1819{
1147 return etp_init (want_poll, done_poll); 1820 eio_want_poll_cb = want_poll;
1148} 1821 eio_done_poll_cb = done_poll;
1149 1822
1823 return etp_init (EIO_POOL, 0, 0, 0);
1824}
1825
1826ecb_inline void
1150static void eio_api_destroy (eio_req *req) 1827eio_api_destroy (eio_req *req)
1151{ 1828{
1152 free (req); 1829 free (req);
1153} 1830}
1154 1831
1155#define REQ(rtype) \ 1832#define REQ(rtype) \
1156 eio_req *req; \ 1833 eio_req *req; \
1157 \ 1834 \
1158 req = (eio_req *)calloc (1, sizeof *req); \ 1835 req = (eio_req *)calloc (1, sizeof *req); \
1159 if (!req) \ 1836 if (!req) \
1160 return 0; \ 1837 return 0; \
… …
1174 { \ 1851 { \
1175 eio_api_destroy (req); \ 1852 eio_api_destroy (req); \
1176 return 0; \ 1853 return 0; \
1177 } 1854 }
1178 1855
1856#define SINGLEDOT(ptr) (0[(char *)(ptr)] == '.' && !1[(char *)(ptr)])
1857
1858static void
1179static void eio_execute (etp_worker *self, eio_req *req) 1859eio_execute (etp_worker *self, eio_req *req)
1180{ 1860{
1181 errno = 0; 1861#if HAVE_AT
1862 int dirfd;
1863#else
1864 const char *path;
1865#endif
1866
1867 if (ecb_expect_false (EIO_CANCELLED (req)))
1868 {
1869 req->result = -1;
1870 req->errorno = ECANCELED;
1871 return;
1872 }
1873
1874 if (ecb_expect_false (req->wd == EIO_INVALID_WD))
1875 {
1876 req->result = -1;
1877 req->errorno = ENOENT;
1878 return;
1879 }
1880
1881 if (req->type >= EIO_OPEN)
1882 {
1883 #if HAVE_AT
1884 dirfd = WD2FD (req->wd);
1885 #else
1886 path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1887 #endif
1888 }
1182 1889
1183 switch (req->type) 1890 switch (req->type)
1184 { 1891 {
1892 case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
1893 req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
1894 break;
1895 case EIO_WD_CLOSE: req->result = 0;
1896 eio_wd_close_sync (req->wd); break;
1897
1898 case EIO_SEEK: eio__lseek (req); break;
1185 case EIO_READ: ALLOC (req->size); 1899 case EIO_READ: ALLOC (req->size);
1186 req->result = req->offs >= 0 1900 req->result = req->offs >= 0
1187 ? pread (req->int1, req->ptr2, req->size, req->offs) 1901 ? pread (req->int1, req->ptr2, req->size, req->offs)
1188 : read (req->int1, req->ptr2, req->size); break; 1902 : read (req->int1, req->ptr2, req->size); break;
1189 case EIO_WRITE: req->result = req->offs >= 0 1903 case EIO_WRITE: req->result = req->offs >= 0
1190 ? pwrite (req->int1, req->ptr2, req->size, req->offs) 1904 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1191 : write (req->int1, req->ptr2, req->size); break; 1905 : write (req->int1, req->ptr2, req->size); break;
1192 1906
1907 case EIO_FCNTL: req->result = fcntl (req->int1, (int) req->int2, req->ptr2); break;
1908 case EIO_IOCTL: req->result = ioctl (req->int1, (unsigned long)req->int2, req->ptr2); break;
1909
1193 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break; 1910 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1194 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break; 1911 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
1912
1913#if HAVE_AT
1195 1914
1196 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1915 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1197 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 1916 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break;
1198 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1917 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1199 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 1918 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, AT_SYMLINK_NOFOLLOW); break;
1200 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1201 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1202
1203 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; 1919 case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break;
1204 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1205 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break; 1920 case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break;
1206 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1207 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break; 1921 case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break;
1208 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1209
1210 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break; 1922 case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break;
1211 case EIO_CLOSE: req->result = close (req->int1); break; 1923 case EIO_SLURP: eio__slurp ( openat (dirfd, req->ptr1, O_RDONLY | O_CLOEXEC), req); break;
1212 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break; 1924
1213 case EIO_UNLINK: req->result = unlink (req->ptr1); break; 1925 case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
1214 case EIO_RMDIR: req->result = rmdir (req->ptr1); break; 1926 case EIO_RMDIR: /* complications arise because "." cannot be removed, so we might have to expand */
1927 req->result = req->wd && SINGLEDOT (req->ptr1)
1928 ? rmdir (req->wd->str)
1929 : unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break;
1215 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; 1930 case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); break;
1216 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; 1931 case EIO_RENAME: req->result = eio__renameat2 (
1932 dirfd,
1933 /* complications arise because "." cannot be renamed, so we might have to expand */
1934 req->wd && SINGLEDOT (req->ptr1) ? req->wd->str : req->ptr1,
1935 WD2FD ((eio_wd)req->int3),
1936 req->ptr2,
1937 req->int2
1938 );
1939 break;
1217 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; 1940 case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2, 0); break;
1218 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; 1941 case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
1219 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break; 1942 case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1220 1943 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1944 req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1221 case EIO_READLINK: ALLOC (NAME_MAX); 1945 case EIO_READLINK: ALLOC (EIO_PATH_MAX);
1222 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; 1946 req->result = readlinkat (dirfd, req->ptr1, req->ptr2, EIO_PATH_MAX);
1223 1947 if (req->result == EIO_PATH_MAX)
1224 case EIO_SYNC: req->result = 0; sync (); break; 1948 {
1225 case EIO_FSYNC: req->result = fsync (req->int1); break; 1949 req->result = -1;
1226 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 1950 errno = ENAMETOOLONG;
1227 case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break; 1951 }
1228 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break; 1952 break;
1229 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1230
1231 case EIO_READDIR: eio__scandir (req, self); break;
1232
1233 case EIO_BUSY: 1953 case EIO_UTIME:
1234#ifdef _WIN32 1954 case EIO_FUTIME:
1235 Sleep (req->nv1 * 1000.); 1955 {
1956 struct timespec ts[2];
1957 struct timespec *times;
1958
1959 if (req->nv1 != -1. || req->nv2 != -1.)
1960 {
1961 ts[0].tv_sec = req->nv1;
1962 ts[0].tv_nsec = (req->nv1 - ts[0].tv_sec) * 1e9;
1963 ts[1].tv_sec = req->nv2;
1964 ts[1].tv_nsec = (req->nv2 - ts[1].tv_sec) * 1e9;
1965
1966 times = ts;
1967 }
1968 else
1969 times = 0;
1970
1971 req->result = req->type == EIO_FUTIME
1972 ? futimens (req->int1, times)
1973 : utimensat (dirfd, req->ptr1, times, 0);
1974 }
1975 break;
1976
1236#else 1977#else
1237 {
1238 struct timeval tv;
1239 1978
1240 tv.tv_sec = req->nv1; 1979 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1241 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; 1980 req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
1981 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1982 req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
1983 case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
1984 case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
1985 case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
1986 case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
1987 case EIO_SLURP: eio__slurp ( open (path , O_RDONLY | O_CLOEXEC), req); break;
1242 1988
1243 req->result = select (0, 0, 0, 0, &tv); 1989 case EIO_UNLINK: req->result = unlink (path ); break;
1244 } 1990 case EIO_RMDIR: req->result = rmdir (path ); break;
1245#endif 1991 case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
1246 break; 1992 case EIO_RENAME: req->result = req->int2 ? EIO_ENOSYS () : rename (path, req->ptr2); break;
1993 case EIO_LINK: req->result = link (path , req->ptr2); break;
1994 case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
1995 case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
1996 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1997 req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
1998 case EIO_READLINK: ALLOC (EIO_PATH_MAX);
1999 req->result = readlink (path, req->ptr2, EIO_PATH_MAX);
2000 if (req->result == EIO_PATH_MAX)
2001 {
2002 req->result = -1;
2003 errno = ENAMETOOLONG;
2004 }
2005 break;
1247 2006
1248 case EIO_UTIME: 2007 case EIO_UTIME:
1249 case EIO_FUTIME: 2008 case EIO_FUTIME:
1250 { 2009 {
1251 struct timeval tv[2]; 2010 struct timeval tv[2];
1252 struct timeval *times; 2011 struct timeval *times;
1253 2012
1254 if (req->nv1 != -1. || req->nv2 != -1.) 2013 if (req->nv1 != -1. || req->nv2 != -1.)
1255 { 2014 {
1256 tv[0].tv_sec = req->nv1; 2015 tv[0].tv_sec = req->nv1;
1257 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.; 2016 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
1258 tv[1].tv_sec = req->nv2; 2017 tv[1].tv_sec = req->nv2;
1259 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.; 2018 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
1260 2019
1261 times = tv; 2020 times = tv;
1262 } 2021 }
1263 else 2022 else
1264 times = 0; 2023 times = 0;
1265
1266 2024
1267 req->result = req->type == EIO_FUTIME 2025 req->result = req->type == EIO_FUTIME
1268 ? futimes (req->int1, times) 2026 ? futimes (req->int1, times)
1269 : utimes (req->ptr1, times); 2027 : utimes (req->ptr1, times);
1270 } 2028 }
1271 break; 2029 break;
1272 2030
2031#endif
2032
2033 case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
2034 {
2035 ALLOC (req->result);
2036 memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
2037 }
2038 break;
2039
2040 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2041 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
2042
2043 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2044 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2045
2046 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
2047 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
2048 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
2049
2050 case EIO_CLOSE: req->result = eio__close (req->int1); break;
2051 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
2052 case EIO_SYNC: req->result = 0; sync (); break;
2053 case EIO_FSYNC: req->result = fsync (req->int1); break;
2054 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
2055 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
2056 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
2057 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
2058 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
2059 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
2060 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
2061 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
2062
2063 case EIO_READDIR: eio__scandir (req, self); break;
2064
2065 case EIO_BUSY:
2066#ifdef _WIN32
2067 Sleep (req->nv1 * 1e3);
2068#else
2069 {
2070 struct timeval tv;
2071
2072 tv.tv_sec = req->nv1;
2073 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
2074
2075 req->result = select (0, 0, 0, 0, &tv);
2076 }
2077#endif
2078 break;
2079
2080#if 0
1273 case EIO_GROUP: 2081 case EIO_GROUP:
1274 abort (); /* handled in eio_request */ 2082 abort (); /* handled in eio_request */
2083#endif
1275 2084
1276 case EIO_NOP: 2085 case EIO_NOP:
1277 req->result = 0; 2086 req->result = 0;
1278 break; 2087 break;
1279 2088
1280 case EIO_CUSTOM: 2089 case EIO_CUSTOM:
1281 ((void (*)(eio_req *))req->feed) (req); 2090 req->feed (req);
1282 break; 2091 break;
1283 2092
1284 default: 2093 default:
1285 req->result = -1; 2094 req->result = EIO_ENOSYS ();
1286 break; 2095 break;
1287 } 2096 }
1288 2097
2098alloc_fail:
1289 req->errorno = errno; 2099 req->errorno = errno;
1290} 2100}
1291 2101
1292#ifndef EIO_NO_WRAPPERS 2102#ifndef EIO_NO_WRAPPERS
1293 2103
2104eio_req *eio_wd_open (const char *path, int pri, eio_cb cb, void *data)
2105{
2106 REQ (EIO_WD_OPEN); PATH; SEND;
2107}
2108
2109eio_req *eio_wd_close (eio_wd wd, int pri, eio_cb cb, void *data)
2110{
2111 REQ (EIO_WD_CLOSE); req->wd = wd; SEND;
2112}
2113
1294eio_req *eio_nop (int pri, eio_cb cb, void *data) 2114eio_req *eio_nop (int pri, eio_cb cb, void *data)
1295{ 2115{
1296 REQ (EIO_NOP); SEND; 2116 REQ (EIO_NOP); SEND;
1297} 2117}
1298 2118
… …
1314eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2134eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1315{ 2135{
1316 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2136 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1317} 2137}
1318 2138
2139eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2140{
2141 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2142}
2143
2144eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
2145{
2146 REQ (EIO_SYNCFS); req->int1 = fd; SEND;
2147}
2148
2149eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2150{
2151 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2152}
2153
1319eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2154eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1320{ 2155{
1321 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2156 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1322} 2157}
1323 2158
1324eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) 2159eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
1325{ 2160{
1326 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; 2161 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
1327} 2162}
1328 2163
1329eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) 2164eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
1330{ 2165{
1331 REQ (EIO_FDATASYNC); req->int1 = fd; SEND; 2166 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2167}
2168
2169eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2170{
2171 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
1332} 2172}
1333 2173
1334eio_req *eio_close (int fd, int pri, eio_cb cb, void *data) 2174eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
1335{ 2175{
1336 REQ (EIO_CLOSE); req->int1 = fd; SEND; 2176 REQ (EIO_CLOSE); req->int1 = fd; SEND;
… …
1339eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data) 2179eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
1340{ 2180{
1341 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND; 2181 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
1342} 2182}
1343 2183
2184eio_req *eio_seek (int fd, off_t offset, int whence, int pri, eio_cb cb, void *data)
2185{
2186 REQ (EIO_SEEK); req->int1 = fd; req->offs = offset; req->int2 = whence; SEND;
2187}
2188
1344eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data) 2189eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1345{ 2190{
1346 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND; 2191 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1347} 2192}
1348 2193
1349eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data) 2194eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1350{ 2195{
1351 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND; 2196 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1352} 2197}
1353 2198
2199eio_req *eio_fcntl (int fd, int cmd, void *arg, int pri, eio_cb cb, void *data)
2200{
2201 REQ (EIO_IOCTL); req->int1 = fd; req->int2 = cmd; req->ptr2 = arg; SEND;
2202}
2203
2204eio_req *eio_ioctl (int fd, unsigned long request, void *buf, int pri, eio_cb cb, void *data)
2205{
2206 REQ (EIO_IOCTL); req->int1 = fd; req->int2 = request; req->ptr2 = buf; SEND;
2207}
2208
1354eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) 2209eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1355{ 2210{
1356 REQ (EIO_FSTAT); req->int1 = fd; SEND; 2211 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1357} 2212}
1358 2213
2214eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
2215{
2216 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
2217}
2218
1359eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) 2219eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1360{ 2220{
1361 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; 2221 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1362} 2222}
1363 2223
… …
1369eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) 2229eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1370{ 2230{
1371 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2231 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1372} 2232}
1373 2233
1374eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2234eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1375{ 2235{
1376 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2236 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1377} 2237}
1378 2238
1379eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) 2239eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
… …
1399eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data) 2259eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1400{ 2260{
1401 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2261 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1402} 2262}
1403 2263
1404eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2264eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1405{ 2265{
1406 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2266 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1407} 2267}
1408 2268
1409eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) 2269eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
… …
1425eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data) 2285eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1426{ 2286{
1427 return eio__1path (EIO_READLINK, path, pri, cb, data); 2287 return eio__1path (EIO_READLINK, path, pri, cb, data);
1428} 2288}
1429 2289
2290eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2291{
2292 return eio__1path (EIO_REALPATH, path, pri, cb, data);
2293}
2294
1430eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) 2295eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1431{ 2296{
1432 return eio__1path (EIO_STAT, path, pri, cb, data); 2297 return eio__1path (EIO_STAT, path, pri, cb, data);
1433} 2298}
1434 2299
1435eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) 2300eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1436{ 2301{
1437 return eio__1path (EIO_LSTAT, path, pri, cb, data); 2302 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1438} 2303}
1439 2304
2305eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
2306{
2307 return eio__1path (EIO_STATVFS, path, pri, cb, data);
2308}
2309
1440eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) 2310eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1441{ 2311{
1442 return eio__1path (EIO_UNLINK, path, pri, cb, data); 2312 return eio__1path (EIO_UNLINK, path, pri, cb, data);
1443} 2313}
1444 2314
1445eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data) 2315eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1446{ 2316{
1447 return eio__1path (EIO_RMDIR, path, pri, cb, data); 2317 return eio__1path (EIO_RMDIR, path, pri, cb, data);
1448} 2318}
1449 2319
1450eio_req *eio_readdir (const char *path, int pri, eio_cb cb, void *data) 2320eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1451{ 2321{
1452 return eio__1path (EIO_READDIR, path, pri, cb, data); 2322 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1453} 2323}
1454 2324
1455eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) 2325eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1456{ 2326{
1457 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; 2327 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1458} 2328}
1459 2329
1460static eio_req * 2330static eio_req *
1461eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2331eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1462{ 2332{
… …
1486eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2356eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1487{ 2357{
1488 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2358 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1489} 2359}
1490 2360
2361eio_req *eio_slurp (const char *path, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
2362{
2363 REQ (EIO_SLURP); PATH; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
2364}
2365
1491eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) 2366eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
1492{ 2367{
1493 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND; 2368 REQ (EIO_CUSTOM); req->feed = execute; SEND;
1494} 2369}
1495 2370
1496#endif 2371#endif
1497 2372
1498eio_req *eio_grp (eio_cb cb, void *data) 2373eio_req *eio_grp (eio_cb cb, void *data)
… …
1507#undef SEND 2382#undef SEND
1508 2383
1509/*****************************************************************************/ 2384/*****************************************************************************/
1510/* grp functions */ 2385/* grp functions */
1511 2386
2387void
1512void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2388eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1513{ 2389{
1514 grp->int2 = limit; 2390 grp->int2 = limit;
1515 grp->feed = feed; 2391 grp->feed = feed;
1516 2392
1517 grp_try_feed (grp); 2393 grp_try_feed (grp);
1518} 2394}
1519 2395
2396void
1520void eio_grp_limit (eio_req *grp, int limit) 2397eio_grp_limit (eio_req *grp, int limit)
1521{ 2398{
1522 grp->int2 = limit; 2399 grp->int2 = limit;
1523 2400
1524 grp_try_feed (grp); 2401 grp_try_feed (grp);
1525} 2402}
1526 2403
2404void
1527void eio_grp_add (eio_req *grp, eio_req *req) 2405eio_grp_add (eio_req *grp, eio_req *req)
1528{ 2406{
1529 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2407 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1530 2408
1531 grp->flags |= EIO_FLAG_GROUPADD; 2409 grp->flags |= ETP_FLAG_GROUPADD;
1532 2410
1533 ++grp->size; 2411 ++grp->size;
1534 req->grp = grp; 2412 req->grp = grp;
1535 2413
1536 req->grp_prev = 0; 2414 req->grp_prev = 0;
… …
1543} 2421}
1544 2422
1545/*****************************************************************************/ 2423/*****************************************************************************/
1546/* misc garbage */ 2424/* misc garbage */
1547 2425
2426eio_ssize_t
1548ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2427eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1549{ 2428{
1550 etp_worker wrk;
1551
1552 wrk.dbuf = 0;
1553
1554 eio__sendfile (ofd, ifd, offset, count, &wrk); 2429 return eio__sendfile (ofd, ifd, offset, count);
1555
1556 if (wrk.dbuf)
1557 free (wrk.dbuf);
1558} 2430}
1559 2431
2432int eio_mlockall_sync (int flags)
2433{
2434 return eio__mlockall (flags);
2435}
2436

Diff Legend

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