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 1.161 by root, Fri Feb 16 21:20:52 2024 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) 2007,2008,2009,2010,2011,2012,2013,2016,2017,2018,2019 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#if defined(O_PATH)
351# define EIO_O_PATH O_PATH
352#elif defined(O_SEARCH)
353# define EIO_O_PATH O_SEARCH
354#else
355# define EIO_O_PATH 0
356#endif
357
358#define EIO_PATH_MAX (PATH_MAX <= EIO_PATH_MIN ? EIO_PATH_MIN : PATH_MAX)
359
113/* buffer size for various temporary buffers */ 360/* buffer size for various temporary buffers */
114#define EIO_BUFSIZE 65536 361#define EIO_BUFSIZE 65536
115 362
116#define dBUF \ 363#define dBUF \
117 char *eio_buf; \
118 ETP_WORKER_LOCK (self); \
119 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 364 char *eio_buf = malloc (EIO_BUFSIZE); \
120 ETP_WORKER_UNLOCK (self); \
121 errno = ENOMEM; \ 365 errno = ENOMEM; \
122 if (!eio_buf) \ 366 if (!eio_buf) \
123 return -1; 367 return -1
124 368
125#define EIO_TICKS ((1000000 + 1023) >> 10) 369#define FUBd \
370 free (eio_buf)
126 371
127/*****************************************************************************/ 372/*****************************************************************************/
128 373
129#if __GNUC__ >= 3 374struct etp_tmpbuf;
130# define expect(expr,value) __builtin_expect ((expr),(value)) 375
376#if _POSIX_VERSION >= 200809L
377 #define HAVE_AT 1
378 #define WD2FD(wd) ((wd) ? (wd)->fd : AT_FDCWD)
131#else 379#else
132# define expect(expr,value) (expr) 380 #define HAVE_AT 0
381 static const char *wd_expand (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path);
133#endif 382#endif
134 383
135#define expect_false(expr) expect ((expr) != 0, 0) 384struct eio_pwd
136#define expect_true(expr) expect ((expr) != 0, 1) 385{
386#if HAVE_AT
387 int fd;
388#endif
389 int len;
390 char str[1]; /* actually, a 0-terminated canonical path */
391};
137 392
138/*****************************************************************************/ 393/*****************************************************************************/
139 394
140#define ETP_PRI_MIN EIO_PRI_MIN 395#define ETP_PRI_MIN EIO_PRI_MIN
141#define ETP_PRI_MAX EIO_PRI_MAX 396#define ETP_PRI_MAX EIO_PRI_MAX
142 397
398#define ETP_TYPE_QUIT -1
399#define ETP_TYPE_GROUP EIO_GROUP
400
401static void eio_nop_callback (void) { }
402static void (*eio_want_poll_cb)(void) = eio_nop_callback;
403static void (*eio_done_poll_cb)(void) = eio_nop_callback;
404
405#define ETP_WANT_POLL(pool) eio_want_poll_cb ()
406#define ETP_DONE_POLL(pool) eio_done_poll_cb ()
407
143struct etp_worker; 408struct etp_worker;
144
145#define ETP_REQ eio_req 409#define ETP_REQ eio_req
146#define ETP_DESTROY(req) eio_destroy (req) 410#define ETP_DESTROY(req) eio_destroy (req)
147static int eio_finish (eio_req *req); 411static int eio_finish (eio_req *req);
148#define ETP_FINISH(req) eio_finish (req) 412#define ETP_FINISH(req) eio_finish (req)
149static void eio_execute (struct etp_worker *self, eio_req *req); 413static void eio_execute (struct etp_worker *self, eio_req *req);
150#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) 414#define ETP_EXECUTE(wrk,req) eio_execute (wrk, req)
151 415
152#define ETP_WORKER_CLEAR(req) \ 416#include "etp.c"
153 if (wrk->dbuf) \ 417
154 { \ 418static struct etp_pool eio_pool;
155 free (wrk->dbuf); \ 419#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;
167 420
168/*****************************************************************************/ 421/*****************************************************************************/
169 422
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 423static 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) 424grp_try_feed (eio_req *grp)
610{ 425{
611 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 426 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
612 { 427 {
613 grp->flags &= ~EIO_FLAG_GROUPADD; 428 grp->flags &= ~ETP_FLAG_GROUPADD;
614 429
615 EIO_FEED (grp); 430 EIO_FEED (grp);
616 431
617 /* stop if no progress has been made */ 432 /* stop if no progress has been made */
618 if (!(grp->flags & EIO_FLAG_GROUPADD)) 433 if (!(grp->flags & ETP_FLAG_GROUPADD))
619 { 434 {
620 grp->feed = 0; 435 grp->feed = 0;
621 break; 436 break;
622 } 437 }
623 } 438 }
624} 439}
625 440
441static int
626static int grp_dec (eio_req *grp) 442grp_dec (eio_req *grp)
627{ 443{
628 --grp->size; 444 --grp->size;
629 445
630 /* call feeder, if applicable */ 446 /* call feeder, if applicable */
631 grp_try_feed (grp); 447 grp_try_feed (grp);
632 448
633 /* finish, if done */ 449 /* finish, if done */
634 if (!grp->size && grp->int1) 450 if (!grp->size && grp->flags & ETP_FLAG_DELAYED)
635 return eio_finish (grp); 451 return eio_finish (grp);
636 else 452 else
637 return 0; 453 return 0;
638} 454}
639 455
456static void
640void eio_destroy (eio_req *req) 457eio_destroy (eio_req *req)
641{ 458{
642 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); 459 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
643 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); 460 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
644 461
645 EIO_DESTROY (req); 462 EIO_DESTROY (req);
646} 463}
647 464
465static int
648static int eio_finish (eio_req *req) 466eio_finish (eio_req *req)
649{ 467{
650 int res = EIO_FINISH (req); 468 int res = EIO_FINISH (req);
651 469
652 if (req->grp) 470 if (req->grp)
653 { 471 {
… …
661 if (grp->grp_first == req) 479 if (grp->grp_first == req)
662 grp->grp_first = req->grp_next; 480 grp->grp_first = req->grp_next;
663 481
664 res2 = grp_dec (grp); 482 res2 = grp_dec (grp);
665 483
666 if (!res && res2) 484 if (!res)
667 res = res2; 485 res = res2;
668 } 486 }
669 487
670 eio_destroy (req); 488 eio_destroy (req);
671 489
672 return res; 490 return res;
673} 491}
674 492
493void
675void eio_grp_cancel (eio_req *grp) 494eio_grp_cancel (eio_req *grp)
676{ 495{
677 for (grp = grp->grp_first; grp; grp = grp->grp_next) 496 etp_grp_cancel (EIO_POOL, grp);
678 eio_cancel (grp);
679} 497}
680 498
499void
681void eio_cancel (eio_req *req) 500eio_cancel (eio_req *req)
682{ 501{
683 etp_cancel (req); 502 etp_cancel (EIO_POOL, req);
684} 503}
685 504
505void
686void eio_submit (eio_req *req) 506eio_submit (eio_req *req)
687{ 507{
688 etp_submit (req); 508 etp_submit (EIO_POOL, req);
689} 509}
690 510
691unsigned int eio_nreqs (void) 511unsigned int
512eio_nreqs (void)
692{ 513{
693 return etp_nreqs (); 514 return etp_nreqs (EIO_POOL);
694} 515}
695 516
696unsigned int eio_nready (void) 517unsigned int
518eio_nready (void)
697{ 519{
698 return etp_nready (); 520 return etp_nready (EIO_POOL);
699} 521}
700 522
701unsigned int eio_npending (void) 523unsigned int
524eio_npending (void)
702{ 525{
703 return etp_npending (); 526 return etp_npending (EIO_POOL);
704} 527}
705 528
706unsigned int eio_nthreads (void) 529unsigned int ecb_cold
530eio_nthreads (void)
707{ 531{
708 return etp_nthreads (); 532 return etp_nthreads (EIO_POOL);
709} 533}
710 534
535void ecb_cold
711void eio_set_max_poll_time (double nseconds) 536eio_set_max_poll_time (double nseconds)
712{ 537{
713 etp_set_max_poll_time (nseconds); 538 etp_set_max_poll_time (EIO_POOL, nseconds);
714} 539}
715 540
541void ecb_cold
716void eio_set_max_poll_reqs (unsigned int maxreqs) 542eio_set_max_poll_reqs (unsigned int maxreqs)
717{ 543{
718 etp_set_max_poll_reqs (maxreqs); 544 etp_set_max_poll_reqs (EIO_POOL, maxreqs);
719} 545}
720 546
547void ecb_cold
721void eio_set_max_idle (unsigned int nthreads) 548eio_set_max_idle (unsigned int nthreads)
722{ 549{
723 etp_set_max_idle (nthreads); 550 etp_set_max_idle (EIO_POOL, nthreads);
724} 551}
725 552
553void ecb_cold
554eio_set_idle_timeout (unsigned int seconds)
555{
556 etp_set_idle_timeout (EIO_POOL, seconds);
557}
558
559void ecb_cold
726void eio_set_min_parallel (unsigned int nthreads) 560eio_set_min_parallel (unsigned int nthreads)
727{ 561{
728 etp_set_min_parallel (nthreads); 562 etp_set_min_parallel (EIO_POOL, nthreads);
729} 563}
730 564
565void ecb_cold
731void eio_set_max_parallel (unsigned int nthreads) 566eio_set_max_parallel (unsigned int nthreads)
732{ 567{
733 etp_set_max_parallel (nthreads); 568 etp_set_max_parallel (EIO_POOL, nthreads);
734} 569}
735 570
736int eio_poll (void) 571int eio_poll (void)
737{ 572{
738 return etp_poll (); 573 return etp_poll (EIO_POOL);
739} 574}
740 575
741/*****************************************************************************/ 576/*****************************************************************************/
742/* work around various missing functions */ 577/* work around various missing functions */
743 578
744#if !HAVE_PREADWRITE 579#if HAVE_POSIX_CLOSE && !__linux
745# undef pread 580# define eio__close(fd) posix_close (fd, 0)
746# undef pwrite 581#else
747# define pread eio__pread 582# 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 583#endif
782 584
585/* close() without disturbing errno */
586static void
587silent_close (int fd)
588{
589 int saved_errno = errno;
590 eio__close (fd);
591 errno = saved_errno;
592}
593
783#ifndef HAVE_FUTIMES 594#ifndef HAVE_UTIMES
784 595
785# undef utimes 596# undef utimes
786# undef futimes
787# define utimes(path,times) eio__utimes (path, times) 597# define utimes(path,times) eio__utimes (path, times)
788# define futimes(fd,times) eio__futimes (fd, times)
789 598
790static int 599static int
791eio__utimes (const char *filename, const struct timeval times[2]) 600eio__utimes (const char *filename, const struct timeval times[2])
792{ 601{
793 if (times) 602 if (times)
… …
801 } 610 }
802 else 611 else
803 return utime (filename, 0); 612 return utime (filename, 0);
804} 613}
805 614
615#endif
616
617#ifndef HAVE_FUTIMES
618
619# undef futimes
620# define futimes(fd,times) eio__futimes (fd, times)
621
622static int
806static int eio__futimes (int fd, const struct timeval tv[2]) 623eio__futimes (int fd, const struct timeval tv[2])
807{ 624{
808 errno = ENOSYS; 625 errno = ENOSYS;
809 return -1; 626 return -1;
810} 627}
811 628
… …
814#if !HAVE_FDATASYNC 631#if !HAVE_FDATASYNC
815# undef fdatasync 632# undef fdatasync
816# define fdatasync(fd) fsync (fd) 633# define fdatasync(fd) fsync (fd)
817#endif 634#endif
818 635
636static int
637eio__syncfs (int fd)
638{
639 int res;
640
641#if HAVE_SYS_SYNCFS
642 res = (int)syscall (__NR_syncfs, (int)(fd));
643#else
644 res = EIO_ENOSYS ();
645#endif
646
647 if (res < 0 && errno == ENOSYS && fd >= 0)
648 sync ();
649
650 return res;
651}
652
819/* sync_file_range always needs emulation */ 653/* sync_file_range always needs emulation */
820int 654static int
821eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags) 655eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
822{ 656{
823#if HAVE_SYNC_FILE_RANGE 657#if HAVE_SYNC_FILE_RANGE
824 int res; 658 int res;
825 659
… …
838 if (!res || errno != ENOSYS) 672 if (!res || errno != ENOSYS)
839 return res; 673 return res;
840#endif 674#endif
841 675
842 /* even though we could play tricks with the flags, it's better to always 676 /* 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 */ 677 * call fdatasync, as that matches the expectation of its users best */
844 return fdatasync (fd); 678 return fdatasync (fd);
679}
680
681static int
682eio__fallocate (int fd, int mode, off_t offset, size_t len)
683{
684#if HAVE_LINUX_FALLOCATE
685 return fallocate (fd, mode, offset, len);
686#else
687 return EIO_ENOSYS ();
688#endif
845} 689}
846 690
847#if !HAVE_READAHEAD 691#if !HAVE_READAHEAD
848# undef readahead 692# undef readahead
849# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 693# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
850 694
851static ssize_t 695static eio_ssize_t
852eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) 696eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
853{ 697{
854 size_t todo = count; 698 size_t todo = count;
855 dBUF; 699 dBUF;
856 700
… …
861 pread (fd, eio_buf, len, offset); 705 pread (fd, eio_buf, len, offset);
862 offset += len; 706 offset += len;
863 todo -= len; 707 todo -= len;
864 } 708 }
865 709
866 errno = 0; 710 FUBd;
711
712 /* linux's readahead basically only fails for EBADF or EINVAL (not mmappable) */
713 /* but not for e.g. EIO or eof, so we also never fail */
867 return count; 714 return 0;
868} 715}
869 716
870#endif 717#endif
871 718
872/* sendfile always needs emulation */ 719/* sendfile always needs emulation */
873static ssize_t 720static eio_ssize_t
874eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) 721eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
875{ 722{
723 eio_ssize_t written = 0;
876 ssize_t res; 724 eio_ssize_t res;
877 725
878 if (!count) 726 if (!count)
879 return 0; 727 return 0;
880 728
729 for (;;)
730 {
731#ifdef __APPLE__
732# undef HAVE_SENDFILE /* broken, as everything on os x */
733#endif
881#if HAVE_SENDFILE 734#if HAVE_SENDFILE
882# if __linux 735# if __linux
736 off_t soffset = offset;
883 res = sendfile (ofd, ifd, &offset, count); 737 res = sendfile (ofd, ifd, &soffset, count);
884 738
885# elif __freebsd 739# elif __FreeBSD__
886 /* 740 /*
887 * Of course, the freebsd sendfile is a dire hack with no thoughts 741 * Of course, the freebsd sendfile is a dire hack with no thoughts
888 * wasted on making it similar to other I/O functions. 742 * wasted on making it similar to other I/O functions.
889 */ 743 */
890 {
891 off_t sbytes; 744 off_t sbytes;
892 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 745 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
893 746
894 if (res < 0 && sbytes) 747 #if 0 /* according to the manpage, this is correct, but broken behaviour */
895 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 748 /* freebsd' sendfile will return 0 on success */
749 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
750 /* not on e.g. EIO or EPIPE - sounds broken */
751 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
896 res = sbytes; 752 res = sbytes;
897 } 753 #endif
754
755 /* according to source inspection, this is correct, and useful behaviour */
756 if (sbytes)
757 res = sbytes;
758
759# elif defined __APPLE__
760 off_t sbytes = count;
761 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
762
763 /* according to the manpage, sbytes is always valid */
764 if (sbytes)
765 res = sbytes;
898 766
899# elif __hpux 767# elif __hpux
900 res = sendfile (ofd, ifd, offset, count, 0, 0); 768 res = sendfile (ofd, ifd, offset, count, 0, 0);
901 769
902# elif __solaris 770# elif __solaris
903 {
904 struct sendfilevec vec; 771 struct sendfilevec vec;
905 size_t sbytes; 772 size_t sbytes;
906 773
907 vec.sfv_fd = ifd; 774 vec.sfv_fd = ifd;
908 vec.sfv_flag = 0; 775 vec.sfv_flag = 0;
909 vec.sfv_off = offset; 776 vec.sfv_off = offset;
910 vec.sfv_len = count; 777 vec.sfv_len = count;
911 778
912 res = sendfilev (ofd, &vec, 1, &sbytes); 779 res = sendfilev (ofd, &vec, 1, &sbytes);
913 780
914 if (res < 0 && sbytes) 781 if (res < 0 && sbytes)
915 res = sbytes; 782 res = sbytes;
916 }
917 783
918# endif 784# endif
785
786#elif defined (_WIN32) && 0
787 /* does not work, just for documentation of what would need to be done */
788 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
789 /* libeio guarantees that the file offset does not change, and windows */
790 /* has no way to get an independent handle to the same file description */
791 HANDLE h = TO_SOCKET (ifd);
792 SetFilePointer (h, offset, 0, FILE_BEGIN);
793 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
794
919#else 795#else
920 res = -1; 796 res = EIO_ENOSYS ();
921 errno = ENOSYS;
922#endif 797#endif
923 798
799 /* we assume sendfile can copy at least 128mb in one go */
800 if (res <= 128 * 1024 * 1024)
801 {
802 if (res > 0)
803 written += res;
804
805 if (written)
806 return written;
807
808 break;
809 }
810 else
811 {
812 /* if we requested more, then probably the kernel was lazy */
813 written += res;
814 offset += res;
815 count -= res;
816
817 if (!count)
818 return written;
819 }
820 }
821
924 if (res < 0 822 if (res < 0
925 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 823 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
824 /* BSDs */
825#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
826 || errno == ENOTSUP
827#endif
828#ifdef EOPNOTSUPP /* windows */
829 || errno == EOPNOTSUPP /* BSDs */
830#endif
926#if __solaris 831#if __solaris
927 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 832 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
928#endif 833#endif
929 ) 834 )
930 ) 835 )
… …
934 839
935 res = 0; 840 res = 0;
936 841
937 while (count) 842 while (count)
938 { 843 {
939 ssize_t cnt; 844 eio_ssize_t cnt;
940 845
941 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 846 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
942 847
943 if (cnt <= 0) 848 if (cnt <= 0)
944 { 849 {
… …
956 861
957 offset += cnt; 862 offset += cnt;
958 res += cnt; 863 res += cnt;
959 count -= cnt; 864 count -= cnt;
960 } 865 }
866
867 FUBd;
961 } 868 }
962 869
963 return res; 870 return res;
871}
872
873#ifdef PAGESIZE
874# define eio_pagesize() PAGESIZE
875#else
876static intptr_t
877eio_pagesize (void)
878{
879 static intptr_t page;
880
881 if (!page)
882 page = sysconf (_SC_PAGESIZE);
883
884 return page;
885}
886#endif
887
888static void
889eio_page_align (void **addr, size_t *length)
890{
891 intptr_t mask = eio_pagesize () - 1;
892
893 /* round down addr */
894 intptr_t adj = mask & (intptr_t)*addr;
895
896 *addr = (void *)((intptr_t)*addr - adj);
897 *length += adj;
898
899 /* round up length */
900 *length = (*length + mask) & ~mask;
901}
902
903#if !_POSIX_MEMLOCK
904# define eio__mlockall(a) EIO_ENOSYS ()
905#else
906
907static int
908eio__mlockall (int flags)
909{
910 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
911 extern int mallopt (int, int);
912 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
913 #endif
914
915 #ifndef MCL_ONFAULT
916 if (flags & EIO_MCL_ONFAULT)
917 return EIO_ERRNO (EINVAL, -1);
918 #define MCL_ONFAULT 4
919 #endif
920
921 if (EIO_MCL_CURRENT != MCL_CURRENT
922 || EIO_MCL_FUTURE != MCL_FUTURE
923 || EIO_MCL_ONFAULT != MCL_ONFAULT)
924 {
925 flags = 0
926 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
927 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0)
928 | (flags & EIO_MCL_ONFAULT ? MCL_ONFAULT : 0)
929 ;
930 }
931
932 return mlockall (flags);
933}
934#endif
935
936#if !_POSIX_MEMLOCK_RANGE
937# define eio__mlock(a,b) EIO_ENOSYS ()
938#else
939
940static int
941eio__mlock (void *addr, size_t length)
942{
943 eio_page_align (&addr, &length);
944
945 return mlock (addr, length);
946}
947
948#endif
949
950#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
951# define eio__msync(a,b,c) EIO_ENOSYS ()
952#else
953
954static int
955eio__msync (void *mem, size_t len, int flags)
956{
957 eio_page_align (&mem, &len);
958
959 if (EIO_MS_ASYNC != MS_SYNC
960 || EIO_MS_INVALIDATE != MS_INVALIDATE
961 || EIO_MS_SYNC != MS_SYNC)
962 {
963 flags = 0
964 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
965 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
966 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
967 }
968
969 return msync (mem, len, flags);
970}
971
972#endif
973
974static int
975eio__mtouch (eio_req *req)
976{
977 void *mem = req->ptr2;
978 size_t len = req->size;
979 int flags = req->int1;
980
981 eio_page_align (&mem, &len);
982
983 {
984 intptr_t addr = (intptr_t)mem;
985 intptr_t end = addr + len;
986 intptr_t page = eio_pagesize ();
987
988 if (addr < end)
989 if (flags & EIO_MT_MODIFY) /* modify */
990 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < end && !EIO_CANCELLED (req));
991 else
992 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < end && !EIO_CANCELLED (req));
993 }
994
995 return 0;
996}
997
998/*****************************************************************************/
999/* requests implemented outside eio_execute, because they are so large */
1000
1001static void
1002eio__lseek (eio_req *req)
1003{
1004 /* this usually gets optimised away completely, or your compiler sucks, */
1005 /* or the whence constants really are not 0, 1, 2 */
1006 int whence = req->int2 == EIO_SEEK_SET ? SEEK_SET
1007 : req->int2 == EIO_SEEK_CUR ? SEEK_CUR
1008 : req->int2 == EIO_SEEK_END ? SEEK_END
1009 : req->int2;
1010
1011 req->offs = lseek (req->int1, req->offs, whence);
1012 req->result = req->offs == (off_t)-1 ? -1 : 0;
1013}
1014
1015/* result will always end up in tmpbuf, there is always space for adding a 0-byte */
1016static int
1017eio__realpath (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
1018{
1019 char *res;
1020 const char *rel = path;
1021 char *tmp1, *tmp2;
1022#if SYMLOOP_MAX > 32
1023 int symlinks = SYMLOOP_MAX;
1024#else
1025 int symlinks = 32;
1026#endif
1027
1028 errno = EINVAL;
1029 if (!rel)
1030 return -1;
1031
1032 errno = ENOENT;
1033 if (!*rel)
1034 return -1;
1035
1036 res = etp_tmpbuf_get (tmpbuf, EIO_PATH_MAX * 3);
1037#ifdef _WIN32
1038 if (_access (rel, 4) != 0)
1039 return -1;
1040
1041 symlinks = GetFullPathName (rel, EIO_PATH_MAX * 3, res, 0);
1042
1043 errno = ENAMETOOLONG;
1044 if (symlinks >= EIO_PATH_MAX * 3)
1045 return -1;
1046
1047 errno = EIO;
1048 if (symlinks <= 0)
1049 return -1;
1050
1051 return symlinks;
1052
1053#else
1054 tmp1 = res + EIO_PATH_MAX;
1055 tmp2 = tmp1 + EIO_PATH_MAX;
1056
1057#if 0 /* disabled, the musl way to do things is just too racy */
1058#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1059 /* on linux we may be able to ask the kernel */
1060 {
1061 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1062
1063 if (fd >= 0)
1064 {
1065 sprintf (tmp1, "/proc/self/fd/%d", fd);
1066 req->result = readlink (tmp1, res, EIO_PATH_MAX);
1067 /* here we should probably stat the open file and the disk file, to make sure they still match */
1068 eio__close (fd);
1069
1070 if (req->result > 0)
1071 goto done;
1072 }
1073 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1074 return -1;
1075 }
1076#endif
1077#endif
1078
1079 if (*rel != '/')
1080 {
1081 int len;
1082
1083 errno = ENOENT;
1084 if (wd == EIO_INVALID_WD)
1085 return -1;
1086
1087 if (wd == EIO_CWD)
1088 {
1089 if (!getcwd (res, EIO_PATH_MAX))
1090 return -1;
1091
1092 len = strlen (res);
1093 }
1094 else
1095 memcpy (res, wd->str, len = wd->len);
1096
1097 if (res [1]) /* only use if not / */
1098 res += len;
1099 }
1100
1101 while (*rel)
1102 {
1103 eio_ssize_t len, linklen;
1104 const char *beg = rel;
1105
1106 while (*rel && *rel != '/')
1107 ++rel;
1108
1109 len = rel - beg;
1110
1111 if (!len) /* skip slashes */
1112 {
1113 ++rel;
1114 continue;
1115 }
1116
1117 if (beg [0] == '.')
1118 {
1119 if (len == 1)
1120 continue; /* . - nop */
1121
1122 if (beg [1] == '.' && len == 2)
1123 {
1124 /* .. - back up one component, if possible */
1125
1126 while (res != tmpbuf->ptr)
1127 if (*--res == '/')
1128 break;
1129
1130 continue;
1131 }
1132 }
1133
1134 errno = ENAMETOOLONG;
1135 if (res + 1 + len + 1 >= tmp1)
1136 return -1;
1137
1138 /* copy one component */
1139 *res = '/';
1140 memcpy (res + 1, beg, len);
1141
1142 /* zero-terminate, for readlink */
1143 res [len + 1] = 0;
1144
1145 /* now check if it's a symlink */
1146 linklen = readlink (tmpbuf->ptr, tmp1, EIO_PATH_MAX);
1147
1148 if (linklen < 0)
1149 {
1150 if (errno != EINVAL)
1151 return -1;
1152
1153 /* it's a normal directory. hopefully */
1154 res += len + 1;
1155 }
1156 else
1157 {
1158 /* yay, it was a symlink - build new path in tmp2 */
1159 int rellen = strlen (rel);
1160
1161 errno = ENAMETOOLONG;
1162 if (linklen + 1 + rellen >= EIO_PATH_MAX) /* also catch linklen >= EIO_PATH_MAX */
1163 return -1;
1164
1165 errno = ELOOP;
1166 if (!--symlinks)
1167 return -1;
1168
1169 if (*tmp1 == '/')
1170 res = tmpbuf->ptr; /* symlink resolves to an absolute path */
1171
1172 /* we need to be careful, as rel might point into tmp2 already */
1173 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1174 tmp2 [linklen] = '/';
1175 memcpy (tmp2, tmp1, linklen);
1176
1177 rel = tmp2;
1178 }
1179 }
1180
1181 /* special case for the lone root path */
1182 if (res == tmpbuf->ptr)
1183 *res++ = '/';
1184
1185 return res - (char *)tmpbuf->ptr;
1186#endif
1187}
1188
1189static signed char
1190eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1191{
1192 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1193 : a->inode < b->inode ? -1
1194 : a->inode > b->inode ? 1
1195 : 0;
1196}
1197
1198#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1199
1200#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1201#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1202
1203static void
1204eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1205{
1206 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1207 unsigned char *bit = bits;
1208
1209 assert (CHAR_BIT == 8);
1210 assert (sizeof (eio_dirent) * 8 < 256);
1211 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1212 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1213
1214 if (size <= EIO_SORT_FAST)
1215 return;
1216
1217 /* first prepare an array of bits to test in our radix sort */
1218 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1219 /* inode_bits must contain all inodes ORed together */
1220 /* which is used to skip bits that are 0 everywhere, which is very common */
1221 {
1222 eio_ino_t endianness;
1223 int i, j;
1224
1225 /* we store the byte offset of byte n into byte n of "endianness" */
1226 for (i = 0; i < sizeof (eio_ino_t); ++i)
1227 ((unsigned char *)&endianness)[i] = i;
1228
1229 *bit++ = 0;
1230
1231 for (i = 0; i < sizeof (eio_ino_t); ++i)
1232 {
1233 /* shifting off the byte offsets out of "endianness" */
1234 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1235 endianness >>= 8;
1236
1237 for (j = 0; j < 8; ++j)
1238 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1239 *bit++ = offs + j;
1240 }
1241
1242 for (j = 0; j < 8; ++j)
1243 if (score_bits & (1 << j))
1244 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1245 }
1246
1247 /* now actually do the sorting (a variant of MSD radix sort) */
1248 {
1249 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1250 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1251 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1252 int stk_idx = 0;
1253
1254 base_stk [stk_idx] = dents;
1255 end_stk [stk_idx] = dents + size;
1256 bit_stk [stk_idx] = bit - 1;
1257
1258 do
1259 {
1260 base = base_stk [stk_idx];
1261 end = end_stk [stk_idx];
1262 bit = bit_stk [stk_idx];
1263
1264 for (;;)
1265 {
1266 unsigned char O = *bit >> 3;
1267 unsigned char M = 1 << (*bit & 7);
1268
1269 eio_dirent *a = base;
1270 eio_dirent *b = end;
1271
1272 if (b - a < EIO_SORT_CUTOFF)
1273 break;
1274
1275 /* now bit-partition the array on the bit */
1276 /* this ugly asymmetric loop seems to perform much better than typical */
1277 /* partition algos found in the literature */
1278 do
1279 if (!(((unsigned char *)a)[O] & M))
1280 ++a;
1281 else if (!(((unsigned char *)--b)[O] & M))
1282 {
1283 eio_dirent tmp = *a; *a = *b; *b = tmp;
1284 ++a;
1285 }
1286 while (b > a);
1287
1288 /* next bit, or stop, if no bits left in this path */
1289 if (!*--bit)
1290 break;
1291
1292 base_stk [stk_idx] = a;
1293 end_stk [stk_idx] = end;
1294 bit_stk [stk_idx] = bit;
1295 ++stk_idx;
1296
1297 end = a;
1298 }
1299 }
1300 while (stk_idx--);
1301 }
1302}
1303
1304static void
1305eio_dent_insertion_sort (eio_dirent *dents, int size)
1306{
1307 /* first move the smallest element to the front, to act as a sentinel */
1308 {
1309 int i;
1310 eio_dirent *min = dents;
1311
1312 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1313 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1314 if (EIO_DENT_CMP (dents [i], <, *min))
1315 min = &dents [i];
1316
1317 /* swap elements 0 and j (minimum) */
1318 {
1319 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1320 }
1321 }
1322
1323 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1324 {
1325 eio_dirent *i, *j;
1326
1327 for (i = dents + 1; i < dents + size; ++i)
1328 {
1329 eio_dirent value = *i;
1330
1331 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1332 j [1] = j [0];
1333
1334 j [1] = value;
1335 }
1336 }
1337}
1338
1339static void
1340eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1341{
1342 if (size <= 1)
1343 return; /* our insertion sort relies on size > 0 */
1344
1345 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1346 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1347 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1348
1349 /* use an insertion sort at the end, or for small arrays, */
1350 /* as insertion sort is more efficient for small partitions */
1351 eio_dent_insertion_sort (dents, size);
964} 1352}
965 1353
966/* read a full directory */ 1354/* read a full directory */
967static void 1355static void
968eio__scandir (eio_req *req, etp_worker *self) 1356eio__scandir (eio_req *req, etp_worker *self)
969{ 1357{
1358 char *name, *names;
1359 int namesalloc = 4096 - sizeof (void *) * 4;
1360 int namesoffs = 0;
1361 int flags = req->int1;
1362 eio_dirent *dents = 0;
1363 int dentalloc = 128;
1364 int dentoffs = 0;
1365 eio_ino_t inode_bits = 0;
1366#ifdef _WIN32
1367 HANDLE dirp;
1368 WIN32_FIND_DATA entp;
1369#else
970 DIR *dirp; 1370 DIR *dirp;
971 EIO_STRUCT_DIRENT *entp; 1371 EIO_STRUCT_DIRENT *entp;
972 char *name, *names; 1372#endif
973 int memlen = 4096;
974 int memofs = 0;
975 int res = 0;
976 1373
977 X_LOCK (wrklock); 1374 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 1375
984 if (dirp && names) 1376 if (!(flags & EIO_READDIR_DENTS))
985 for (;;) 1377 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1378
1379#ifdef _WIN32
1380 {
1381 int len = strlen ((const char *)req->ptr1);
1382 char *path = malloc (MAX_PATH);
1383 const char *fmt;
1384 const char *reqpath = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1385
1386 if (!len)
1387 fmt = "./*";
1388 else if (reqpath[len - 1] == '/' || reqpath[len - 1] == '\\')
1389 fmt = "%s*";
1390 else
1391 fmt = "%s/*";
1392
1393 _snprintf (path, MAX_PATH, fmt, reqpath);
1394 dirp = FindFirstFile (path, &entp);
1395 free (path);
1396
1397 if (dirp == INVALID_HANDLE_VALUE)
986 { 1398 {
1399 /* should steal _dosmaperr */
1400 switch (GetLastError ())
1401 {
1402 case ERROR_FILE_NOT_FOUND:
1403 req->result = 0;
1404 break;
1405
1406 case ERROR_INVALID_NAME:
1407 case ERROR_PATH_NOT_FOUND:
1408 case ERROR_NO_MORE_FILES:
1409 errno = ENOENT;
1410 break;
1411
1412 case ERROR_NOT_ENOUGH_MEMORY:
1413 errno = ENOMEM;
1414 break;
1415
1416 default:
1417 errno = EINVAL;
1418 break;
1419 }
1420
1421 return;
1422 }
1423 }
1424#else
1425 #if HAVE_AT
1426 {
1427 int fd = openat (WD2FD (req->wd), req->ptr1, O_CLOEXEC | O_RDONLY | O_DIRECTORY | O_NONBLOCK);
1428
1429 if (fd < 0)
1430 return;
1431
1432 dirp = fdopendir (fd);
1433
1434 if (!dirp)
1435 {
1436 silent_close (fd);
1437 return;
1438 }
1439 }
1440 #else
1441 dirp = opendir (wd_expand (&self->tmpbuf, req->wd, req->ptr1));
1442
1443 if (!dirp)
1444 return;
1445 #endif
1446#endif
1447
1448 if (req->flags & EIO_FLAG_PTR1_FREE)
1449 free (req->ptr1);
1450
1451 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1452 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1453 req->ptr2 = names = malloc (namesalloc);
1454
1455 if (!names || (flags && !dents))
1456 return;
1457
1458 for (;;)
1459 {
1460 int done;
1461
1462#ifdef _WIN32
1463 done = !dirp;
1464#else
987 errno = 0; 1465 errno = 0;
988 entp = readdir (dirp); 1466 entp = readdir (dirp);
1467 done = !entp;
1468#endif
989 1469
1470 if (done)
1471 {
1472#ifndef _WIN32
1473 int old_errno = errno;
1474 closedir (dirp);
1475 errno = old_errno;
1476
990 if (!entp) 1477 if (errno)
1478 break;
1479#endif
1480
1481 /* sort etc. */
1482 req->int1 = flags;
1483 req->result = dentoffs;
1484
1485 if (flags & EIO_READDIR_STAT_ORDER)
1486 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1487 else if (flags & EIO_READDIR_DIRS_FIRST)
1488 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1489 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1490 else
1491 {
1492 /* in this case, all is known, and we just put dirs first and sort them */
1493 eio_dirent *oth = dents + dentoffs;
1494 eio_dirent *dir = dents;
1495
1496 /* now partition dirs to the front, and non-dirs to the back */
1497 /* by walking from both sides and swapping if necessary */
1498 while (oth > dir)
1499 {
1500 if (dir->type == EIO_DT_DIR)
1501 ++dir;
1502 else if ((--oth)->type == EIO_DT_DIR)
1503 {
1504 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1505
1506 ++dir;
1507 }
1508 }
1509
1510 /* now sort the dirs only (dirs all have the same score) */
1511 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1512 }
1513
991 break; 1514 break;
1515 }
992 1516
993 name = entp->d_name; 1517 /* now add the entry to our list(s) */
1518 name = D_NAME (entp);
994 1519
1520 /* skip . and .. entries */
995 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1521 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
996 { 1522 {
997 int len = strlen (name) + 1; 1523 int len = D_NAMLEN (entp) + 1;
998 1524
999 res++; 1525 while (ecb_expect_false (namesoffs + len > namesalloc))
1000
1001 while (memofs + len > memlen)
1002 { 1526 {
1003 memlen *= 2; 1527 namesalloc *= 2;
1004 X_LOCK (wrklock);
1005 req->ptr2 = names = realloc (names, memlen); 1528 req->ptr2 = names = realloc (names, namesalloc);
1006 X_UNLOCK (wrklock);
1007 1529
1008 if (!names) 1530 if (!names)
1531 break;
1532 }
1533
1534 memcpy (names + namesoffs, name, len);
1535
1536 if (dents)
1537 {
1538 struct eio_dirent *ent;
1539
1540 if (ecb_expect_false (dentoffs == dentalloc))
1541 {
1542 dentalloc *= 2;
1543 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1544
1545 if (!dents)
1009 break; 1546 break;
1010 } 1547 }
1011 1548
1012 memcpy (names + memofs, name, len); 1549 ent = dents + dentoffs;
1013 memofs += len; 1550
1551 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1552 ent->namelen = len - 1;
1553 ent->inode = D_INO (entp);
1554
1555 inode_bits |= ent->inode;
1556
1557 switch (D_TYPE (entp))
1558 {
1559 default:
1560 ent->type = EIO_DT_UNKNOWN;
1561 flags |= EIO_READDIR_FOUND_UNKNOWN;
1562 break;
1563
1564 #ifdef DT_FIFO
1565 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1566 #endif
1567 #ifdef DT_CHR
1568 case DT_CHR: ent->type = EIO_DT_CHR; break;
1569 #endif
1570 #ifdef DT_MPC
1571 case DT_MPC: ent->type = EIO_DT_MPC; break;
1572 #endif
1573 #ifdef DT_DIR
1574 case DT_DIR: ent->type = EIO_DT_DIR; break;
1575 #endif
1576 #ifdef DT_NAM
1577 case DT_NAM: ent->type = EIO_DT_NAM; break;
1578 #endif
1579 #ifdef DT_BLK
1580 case DT_BLK: ent->type = EIO_DT_BLK; break;
1581 #endif
1582 #ifdef DT_MPB
1583 case DT_MPB: ent->type = EIO_DT_MPB; break;
1584 #endif
1585 #ifdef DT_REG
1586 case DT_REG: ent->type = EIO_DT_REG; break;
1587 #endif
1588 #ifdef DT_NWK
1589 case DT_NWK: ent->type = EIO_DT_NWK; break;
1590 #endif
1591 #ifdef DT_CMP
1592 case DT_CMP: ent->type = EIO_DT_CMP; break;
1593 #endif
1594 #ifdef DT_LNK
1595 case DT_LNK: ent->type = EIO_DT_LNK; break;
1596 #endif
1597 #ifdef DT_SOCK
1598 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1599 #endif
1600 #ifdef DT_DOOR
1601 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1602 #endif
1603 #ifdef DT_WHT
1604 case DT_WHT: ent->type = EIO_DT_WHT; break;
1605 #endif
1606 }
1607
1608 ent->score = 7;
1609
1610 if (flags & EIO_READDIR_DIRS_FIRST)
1611 {
1612 if (ent->type == EIO_DT_UNKNOWN)
1613 {
1614 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1615 ent->score = 1;
1616 else if (!strchr (name, '.')) /* absence of dots indicate likely dirs */
1617 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1618 }
1619 else if (ent->type == EIO_DT_DIR)
1620 ent->score = 0;
1621 }
1014 } 1622 }
1623
1624 namesoffs += len;
1625 ++dentoffs;
1015 } 1626 }
1016 1627
1017 if (errno) 1628 if (EIO_CANCELLED (req))
1018 res = -1; 1629 {
1630 errno = ECANCELED;
1631 break;
1632 }
1633
1634#ifdef _WIN32
1635 if (!FindNextFile (dirp, &entp))
1636 {
1637 FindClose (dirp);
1638 dirp = 0;
1639 }
1640#endif
1641 }
1642}
1643
1644/*****************************************************************************/
1645/* working directory stuff */
1646/* various deficiencies in the posix 2008 api force us to */
1647/* keep the absolute path in string form at all times */
1648/* fuck yeah. */
1649
1650#if !HAVE_AT
1651
1652/* a bit like realpath, but usually faster because it doesn't have to return */
1653/* an absolute or canonical path */
1654static const char *
1655wd_expand (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
1656{
1657 if (!wd || *path == '/')
1658 return path;
1659
1660 if (path [0] == '.' && !path [1])
1661 return wd->str;
1662
1019 1663 {
1020 req->result = res; 1664 int l1 = wd->len;
1021} 1665 int l2 = strlen (path);
1022 1666
1023#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) 1667 char *res = etp_tmpbuf_get (tmpbuf, l1 + l2 + 2);
1024# undef msync 1668
1025# define msync(a,b,c) ((errno = ENOSYS), -1) 1669 memcpy (res, wd->str, l1);
1670 res [l1] = '/';
1671 memcpy (res + l1 + 1, path, l2 + 1);
1672
1673 return res;
1674 }
1675}
1676
1677#endif
1678
1679static eio_wd
1680eio__wd_open_sync (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
1681{
1682 int fd;
1683 eio_wd res;
1684 int len = eio__realpath (tmpbuf, wd, path);
1685
1686 if (len < 0)
1687 return EIO_INVALID_WD;
1688
1689#if HAVE_AT
1690 fd = openat (WD2FD (wd), path, EIO_O_PATH | O_DIRECTORY | O_NONBLOCK | O_CLOEXEC);
1691
1692 /* 0 is a valid fd, but we use it for EIO_CWD, so in the very unlikely */
1693 /* case of fd 0 being available (almost certainly an a pplication bug) */
1694 /* make sure we use another fd value */
1695 #if EIO_CWD
1696 error EIO_CWD must be 0
1026#endif 1697 #endif
1698 if (ecb_expect_false (fd == 0))
1699 {
1700 int fd2 = fcntl (fd, F_DUPFD_CLOEXEC ? F_DUPFD_CLOEXEC : F_DUPFD);
1701 fcntl (fd2, F_SETFD, FD_CLOEXEC);
1702 eio__close (fd);
1703 fd = fd2;
1704 }
1027 1705
1028int 1706 if (fd < 0)
1029eio__mtouch (void *mem, size_t len, int flags) 1707 return EIO_INVALID_WD;
1708#endif
1709
1710 res = malloc (sizeof (*res) + len); /* one extra 0-byte */
1711
1712#if HAVE_AT
1713 res->fd = fd;
1714#endif
1715
1716 res->len = len;
1717 memcpy (res->str, tmpbuf->ptr, len);
1718 res->str [len] = 0;
1719
1720 return res;
1721}
1722
1723eio_wd
1724eio_wd_open_sync (eio_wd wd, const char *path)
1030{ 1725{
1031 intptr_t addr = (intptr_t)mem; 1726 struct etp_tmpbuf tmpbuf = { 0 };
1032 intptr_t end = addr + len; 1727 wd = eio__wd_open_sync (&tmpbuf, wd, path);
1033#ifdef PAGESIZE 1728 free (tmpbuf.ptr);
1034 const intptr_t page = PAGESIZE; 1729
1730 return wd;
1731}
1732
1733void
1734eio_wd_close_sync (eio_wd wd)
1735{
1736 if (wd != EIO_INVALID_WD && wd != EIO_CWD)
1737 {
1738 #if HAVE_AT
1739 eio__close (wd->fd);
1740 #endif
1741 free (wd);
1742 }
1743}
1744
1745#if HAVE_AT
1746
1747static int
1748eio__renameat2 (int olddirfd, const char *oldpath, int newdirfd, const char *newpath, unsigned int flags)
1749{
1750#if HAVE_RENAMEAT2
1751 return syscall (SYS_renameat2, olddirfd, oldpath, newdirfd, newpath, flags);
1035#else 1752#else
1036 static intptr_t page;
1037
1038 if (!page) 1753 if (flags)
1039 page = sysconf (_SC_PAGESIZE); 1754 return EIO_ENOSYS ();
1040#endif
1041 1755
1042 addr &= ~(page - 1); /* assume page size is always a power of two */ 1756 return renameat (olddirfd, oldpath, newdirfd, newpath);
1757#endif
1758}
1043 1759
1044 if (addr < end) 1760/* 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 1761
1762static int
1763eio__truncateat (int dirfd, const char *path, off_t length)
1764{
1765 int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC | O_NONBLOCK);
1766 int res;
1767
1768 if (fd < 0)
1769 return fd;
1770
1771 res = ftruncate (fd, length);
1772 silent_close (fd);
1050 return 0; 1773 return res;
1051} 1774}
1775
1776static int
1777eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
1778{
1779 int fd = openat (dirfd, path, EIO_O_PATH | O_CLOEXEC | O_NONBLOCK);
1780 int res;
1781
1782 if (fd < 0)
1783 return fd;
1784
1785 res = fstatvfs (fd, buf);
1786 silent_close (fd);
1787 return res;
1788}
1789
1790#endif
1052 1791
1053/*****************************************************************************/ 1792/*****************************************************************************/
1054 1793
1055#define ALLOC(len) \ 1794#define ALLOC(len) \
1056 if (!req->ptr2) \ 1795 if (!req->ptr2) \
1057 { \ 1796 { \
1058 X_LOCK (wrklock); \ 1797 X_LOCK (EIO_POOL->wrklock); \
1059 req->flags |= EIO_FLAG_PTR2_FREE; \ 1798 req->flags |= EIO_FLAG_PTR2_FREE; \
1060 X_UNLOCK (wrklock); \ 1799 X_UNLOCK (EIO_POOL->wrklock); \
1061 req->ptr2 = malloc (len); \ 1800 req->ptr2 = malloc (len); \
1062 if (!req->ptr2) \ 1801 if (!req->ptr2) \
1063 { \ 1802 { \
1064 errno = ENOMEM; \ 1803 errno = ENOMEM; \
1065 req->result = -1; \ 1804 req->result = -1; \
1066 break; \ 1805 goto alloc_fail; \
1067 } \ 1806 } \
1068 } 1807 }
1069 1808
1070X_THREAD_PROC (etp_proc) 1809/*****************************************************************************/
1071{
1072 ETP_REQ *req;
1073 struct timespec ts;
1074 etp_worker *self = (etp_worker *)thr_arg;
1075 1810
1076 /* try to distribute timeouts somewhat randomly */ 1811static void
1077 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 1812eio__slurp (int fd, eio_req *req)
1813{
1814 req->result = fd;
1078 1815
1079 for (;;) 1816 if (fd < 0)
1817 return;
1818
1819 if (req->offs < 0 || !req->size) /* do we need the size? */
1080 { 1820 {
1081 X_LOCK (reqlock); 1821 off_t size = lseek (fd, 0, SEEK_END);
1082 1822
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) 1823 if (req->offs < 0)
1117 goto quit; 1824 req->offs += size;
1118 1825
1119 if (!EIO_CANCELLED (req)) 1826 if (!req->size)
1120 ETP_EXECUTE (self, req); 1827 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 } 1828 }
1134 1829
1135quit: 1830 ALLOC (req->size);
1136 X_LOCK (wrklock); 1831 req->result = pread (fd, req->ptr2, req->size, req->offs);
1137 etp_worker_free (self);
1138 X_UNLOCK (wrklock);
1139 1832
1140 return 0; 1833 silent_close (fd);
1141}
1142 1834
1143/*****************************************************************************/ 1835alloc_fail:
1836 ;
1837}
1144 1838
1839int ecb_cold
1145int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1840eio_init (void (*want_poll)(void), void (*done_poll)(void))
1146{ 1841{
1147 return etp_init (want_poll, done_poll); 1842 eio_want_poll_cb = want_poll;
1148} 1843 eio_done_poll_cb = done_poll;
1149 1844
1845 return etp_init (EIO_POOL, 0, 0, 0);
1846}
1847
1848ecb_inline void
1150static void eio_api_destroy (eio_req *req) 1849eio_api_destroy (eio_req *req)
1151{ 1850{
1152 free (req); 1851 free (req);
1153} 1852}
1154 1853
1155#define REQ(rtype) \ 1854#define REQ(rtype) \
1156 eio_req *req; \ 1855 eio_req *req; \
1157 \ 1856 \
1158 req = (eio_req *)calloc (1, sizeof *req); \ 1857 req = (eio_req *)calloc (1, sizeof *req); \
1159 if (!req) \ 1858 if (!req) \
1160 return 0; \ 1859 return 0; \
… …
1174 { \ 1873 { \
1175 eio_api_destroy (req); \ 1874 eio_api_destroy (req); \
1176 return 0; \ 1875 return 0; \
1177 } 1876 }
1178 1877
1878#define SINGLEDOT(ptr) (0[(char *)(ptr)] == '.' && !1[(char *)(ptr)])
1879
1880static void
1179static void eio_execute (etp_worker *self, eio_req *req) 1881eio_execute (etp_worker *self, eio_req *req)
1180{ 1882{
1181 errno = 0; 1883#if HAVE_AT
1884 int dirfd;
1885#else
1886 const char *path;
1887#endif
1888
1889 if (ecb_expect_false (EIO_CANCELLED (req)))
1890 {
1891 req->result = -1;
1892 req->errorno = ECANCELED;
1893 return;
1894 }
1895
1896 if (ecb_expect_false (req->wd == EIO_INVALID_WD))
1897 {
1898 req->result = -1;
1899 req->errorno = ENOENT;
1900 return;
1901 }
1902
1903 if (req->type >= EIO_OPEN)
1904 {
1905 #if HAVE_AT
1906 dirfd = WD2FD (req->wd);
1907 #else
1908 path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1909 #endif
1910 }
1182 1911
1183 switch (req->type) 1912 switch (req->type)
1184 { 1913 {
1914 case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
1915 req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
1916 break;
1917 case EIO_WD_CLOSE: req->result = 0;
1918 eio_wd_close_sync (req->wd); break;
1919
1920 case EIO_SEEK: eio__lseek (req); break;
1185 case EIO_READ: ALLOC (req->size); 1921 case EIO_READ: ALLOC (req->size);
1186 req->result = req->offs >= 0 1922 req->result = req->offs >= 0
1187 ? pread (req->int1, req->ptr2, req->size, req->offs) 1923 ? pread (req->int1, req->ptr2, req->size, req->offs)
1188 : read (req->int1, req->ptr2, req->size); break; 1924 : read (req->int1, req->ptr2, req->size); break;
1189 case EIO_WRITE: req->result = req->offs >= 0 1925 case EIO_WRITE: req->result = req->offs >= 0
1190 ? pwrite (req->int1, req->ptr2, req->size, req->offs) 1926 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1191 : write (req->int1, req->ptr2, req->size); break; 1927 : write (req->int1, req->ptr2, req->size); break;
1192 1928
1929 case EIO_FCNTL: req->result = fcntl (req->int1, (int) req->int2, req->ptr2); break;
1930 case EIO_IOCTL: req->result = ioctl (req->int1, (unsigned long)req->int2, req->ptr2); break;
1931
1193 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break; 1932 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; 1933 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
1934
1935#if HAVE_AT
1195 1936
1196 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1937 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1197 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 1938 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break;
1198 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1939 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1199 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 1940 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; 1941 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; 1942 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; 1943 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; 1944 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; 1945 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; 1946
1213 case EIO_UNLINK: req->result = unlink (req->ptr1); break; 1947 case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
1214 case EIO_RMDIR: req->result = rmdir (req->ptr1); break; 1948 case EIO_RMDIR: /* complications arise because "." cannot be removed, so we might have to expand */
1949 req->result = req->wd && SINGLEDOT (req->ptr1)
1950 ? rmdir (req->wd->str)
1951 : unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break;
1215 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; 1952 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; 1953 case EIO_RENAME: req->result = eio__renameat2 (
1954 dirfd,
1955 /* complications arise because "." cannot be renamed, so we might have to expand */
1956 req->wd && SINGLEDOT (req->ptr1) ? req->wd->str : req->ptr1,
1957 WD2FD ((eio_wd)req->int3),
1958 req->ptr2,
1959 req->int2
1960 );
1961 break;
1217 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; 1962 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; 1963 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; 1964 case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1220 1965 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1966 req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1221 case EIO_READLINK: ALLOC (NAME_MAX); 1967 case EIO_READLINK: ALLOC (EIO_PATH_MAX);
1222 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; 1968 req->result = readlinkat (dirfd, req->ptr1, req->ptr2, EIO_PATH_MAX);
1223 1969 if (req->result == EIO_PATH_MAX)
1224 case EIO_SYNC: req->result = 0; sync (); break; 1970 {
1225 case EIO_FSYNC: req->result = fsync (req->int1); break; 1971 req->result = -1;
1226 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 1972 errno = ENAMETOOLONG;
1227 case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break; 1973 }
1228 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break; 1974 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: 1975 case EIO_UTIME:
1234#ifdef _WIN32 1976 case EIO_FUTIME:
1235 Sleep (req->nv1 * 1000.); 1977 {
1978 struct timespec ts[2];
1979 struct timespec *times;
1980
1981 if (req->nv1 != -1. || req->nv2 != -1.)
1982 {
1983 ts[0].tv_sec = req->nv1;
1984 ts[0].tv_nsec = (req->nv1 - ts[0].tv_sec) * 1e9;
1985 ts[1].tv_sec = req->nv2;
1986 ts[1].tv_nsec = (req->nv2 - ts[1].tv_sec) * 1e9;
1987
1988 times = ts;
1989 }
1990 else
1991 times = 0;
1992
1993 req->result = req->type == EIO_FUTIME
1994 ? futimens (req->int1, times)
1995 : utimensat (dirfd, req->ptr1, times, 0);
1996 }
1997 break;
1998
1236#else 1999#else
1237 {
1238 struct timeval tv;
1239 2000
1240 tv.tv_sec = req->nv1; 2001 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1241 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; 2002 req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
2003 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2004 req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
2005 case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
2006 case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
2007 case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
2008 case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
2009 case EIO_SLURP: eio__slurp ( open (path , O_RDONLY | O_CLOEXEC), req); break;
1242 2010
1243 req->result = select (0, 0, 0, 0, &tv); 2011 case EIO_UNLINK: req->result = unlink (path ); break;
1244 } 2012 case EIO_RMDIR: req->result = rmdir (path ); break;
1245#endif 2013 case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
1246 break; 2014 case EIO_RENAME: req->result = req->int2 ? EIO_ENOSYS () : rename (path, req->ptr2); break;
2015 case EIO_LINK: req->result = link (path , req->ptr2); break;
2016 case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
2017 case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
2018 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2019 req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
2020 case EIO_READLINK: ALLOC (EIO_PATH_MAX);
2021 req->result = readlink (path, req->ptr2, EIO_PATH_MAX);
2022 if (req->result == EIO_PATH_MAX)
2023 {
2024 req->result = -1;
2025 errno = ENAMETOOLONG;
2026 }
2027 break;
1247 2028
1248 case EIO_UTIME: 2029 case EIO_UTIME:
1249 case EIO_FUTIME: 2030 case EIO_FUTIME:
1250 { 2031 {
1251 struct timeval tv[2]; 2032 struct timeval tv[2];
1252 struct timeval *times; 2033 struct timeval *times;
1253 2034
1254 if (req->nv1 != -1. || req->nv2 != -1.) 2035 if (req->nv1 != -1. || req->nv2 != -1.)
1255 { 2036 {
1256 tv[0].tv_sec = req->nv1; 2037 tv[0].tv_sec = req->nv1;
1257 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.; 2038 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
1258 tv[1].tv_sec = req->nv2; 2039 tv[1].tv_sec = req->nv2;
1259 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.; 2040 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
1260 2041
1261 times = tv; 2042 times = tv;
1262 } 2043 }
1263 else 2044 else
1264 times = 0; 2045 times = 0;
1265
1266 2046
1267 req->result = req->type == EIO_FUTIME 2047 req->result = req->type == EIO_FUTIME
1268 ? futimes (req->int1, times) 2048 ? futimes (req->int1, times)
1269 : utimes (req->ptr1, times); 2049 : utimes (req->ptr1, times);
1270 } 2050 }
1271 break; 2051 break;
1272 2052
2053#endif
2054
2055 case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
2056 {
2057 ALLOC (req->result);
2058 memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
2059 }
2060 break;
2061
2062 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2063 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
2064
2065 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2066 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2067
2068 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
2069 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
2070 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
2071
2072 case EIO_CLOSE: req->result = eio__close (req->int1); break;
2073 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
2074 case EIO_SYNC: req->result = 0; sync (); break;
2075 case EIO_FSYNC: req->result = fsync (req->int1); break;
2076 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
2077 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
2078 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
2079 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
2080 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
2081 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
2082 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
2083 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
2084
2085 case EIO_READDIR: eio__scandir (req, self); break;
2086
2087 case EIO_BUSY:
2088#ifdef _WIN32
2089 Sleep (req->nv1 * 1e3);
2090#else
2091 {
2092 struct timeval tv;
2093
2094 tv.tv_sec = req->nv1;
2095 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
2096
2097 req->result = select (0, 0, 0, 0, &tv);
2098 }
2099#endif
2100 break;
2101
2102#if 0
1273 case EIO_GROUP: 2103 case EIO_GROUP:
1274 abort (); /* handled in eio_request */ 2104 abort (); /* handled in eio_request */
2105#endif
1275 2106
1276 case EIO_NOP: 2107 case EIO_NOP:
1277 req->result = 0; 2108 req->result = 0;
1278 break; 2109 break;
1279 2110
1280 case EIO_CUSTOM: 2111 case EIO_CUSTOM:
1281 ((void (*)(eio_req *))req->feed) (req); 2112 req->feed (req);
1282 break; 2113 break;
1283 2114
1284 default: 2115 default:
1285 req->result = -1; 2116 req->result = EIO_ENOSYS ();
1286 break; 2117 break;
1287 } 2118 }
1288 2119
2120alloc_fail:
1289 req->errorno = errno; 2121 req->errorno = errno;
1290} 2122}
1291 2123
1292#ifndef EIO_NO_WRAPPERS 2124#ifndef EIO_NO_WRAPPERS
1293 2125
2126eio_req *eio_wd_open (const char *path, int pri, eio_cb cb, void *data)
2127{
2128 REQ (EIO_WD_OPEN); PATH; SEND;
2129}
2130
2131eio_req *eio_wd_close (eio_wd wd, int pri, eio_cb cb, void *data)
2132{
2133 REQ (EIO_WD_CLOSE); req->wd = wd; SEND;
2134}
2135
1294eio_req *eio_nop (int pri, eio_cb cb, void *data) 2136eio_req *eio_nop (int pri, eio_cb cb, void *data)
1295{ 2137{
1296 REQ (EIO_NOP); SEND; 2138 REQ (EIO_NOP); SEND;
1297} 2139}
1298 2140
… …
1314eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2156eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1315{ 2157{
1316 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2158 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1317} 2159}
1318 2160
2161eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2162{
2163 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2164}
2165
2166eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
2167{
2168 REQ (EIO_SYNCFS); req->int1 = fd; SEND;
2169}
2170
2171eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2172{
2173 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2174}
2175
1319eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2176eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1320{ 2177{
1321 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2178 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1322} 2179}
1323 2180
1324eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) 2181eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
1325{ 2182{
1326 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; 2183 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
1327} 2184}
1328 2185
1329eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) 2186eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
1330{ 2187{
1331 REQ (EIO_FDATASYNC); req->int1 = fd; SEND; 2188 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2189}
2190
2191eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2192{
2193 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
1332} 2194}
1333 2195
1334eio_req *eio_close (int fd, int pri, eio_cb cb, void *data) 2196eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
1335{ 2197{
1336 REQ (EIO_CLOSE); req->int1 = fd; SEND; 2198 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) 2201eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
1340{ 2202{
1341 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND; 2203 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
1342} 2204}
1343 2205
2206eio_req *eio_seek (int fd, off_t offset, int whence, int pri, eio_cb cb, void *data)
2207{
2208 REQ (EIO_SEEK); req->int1 = fd; req->offs = offset; req->int2 = whence; SEND;
2209}
2210
1344eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data) 2211eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1345{ 2212{
1346 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND; 2213 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1347} 2214}
1348 2215
1349eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data) 2216eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1350{ 2217{
1351 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND; 2218 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1352} 2219}
1353 2220
2221eio_req *eio_fcntl (int fd, int cmd, void *arg, int pri, eio_cb cb, void *data)
2222{
2223 REQ (EIO_IOCTL); req->int1 = fd; req->int2 = cmd; req->ptr2 = arg; SEND;
2224}
2225
2226eio_req *eio_ioctl (int fd, unsigned long request, void *buf, int pri, eio_cb cb, void *data)
2227{
2228 REQ (EIO_IOCTL); req->int1 = fd; req->int2 = request; req->ptr2 = buf; SEND;
2229}
2230
1354eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) 2231eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1355{ 2232{
1356 REQ (EIO_FSTAT); req->int1 = fd; SEND; 2233 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1357} 2234}
1358 2235
2236eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
2237{
2238 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
2239}
2240
1359eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) 2241eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1360{ 2242{
1361 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; 2243 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1362} 2244}
1363 2245
… …
1369eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) 2251eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1370{ 2252{
1371 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2253 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1372} 2254}
1373 2255
1374eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2256eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1375{ 2257{
1376 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2258 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1377} 2259}
1378 2260
1379eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) 2261eio_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) 2281eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1400{ 2282{
1401 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2283 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1402} 2284}
1403 2285
1404eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2286eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1405{ 2287{
1406 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2288 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1407} 2289}
1408 2290
1409eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) 2291eio_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) 2307eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1426{ 2308{
1427 return eio__1path (EIO_READLINK, path, pri, cb, data); 2309 return eio__1path (EIO_READLINK, path, pri, cb, data);
1428} 2310}
1429 2311
2312eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2313{
2314 return eio__1path (EIO_REALPATH, path, pri, cb, data);
2315}
2316
1430eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) 2317eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1431{ 2318{
1432 return eio__1path (EIO_STAT, path, pri, cb, data); 2319 return eio__1path (EIO_STAT, path, pri, cb, data);
1433} 2320}
1434 2321
1435eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) 2322eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1436{ 2323{
1437 return eio__1path (EIO_LSTAT, path, pri, cb, data); 2324 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1438} 2325}
1439 2326
2327eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
2328{
2329 return eio__1path (EIO_STATVFS, path, pri, cb, data);
2330}
2331
1440eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) 2332eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1441{ 2333{
1442 return eio__1path (EIO_UNLINK, path, pri, cb, data); 2334 return eio__1path (EIO_UNLINK, path, pri, cb, data);
1443} 2335}
1444 2336
1445eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data) 2337eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1446{ 2338{
1447 return eio__1path (EIO_RMDIR, path, pri, cb, data); 2339 return eio__1path (EIO_RMDIR, path, pri, cb, data);
1448} 2340}
1449 2341
1450eio_req *eio_readdir (const char *path, int pri, eio_cb cb, void *data) 2342eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1451{ 2343{
1452 return eio__1path (EIO_READDIR, path, pri, cb, data); 2344 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1453} 2345}
1454 2346
1455eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) 2347eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1456{ 2348{
1457 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; 2349 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1458} 2350}
1459 2351
1460static eio_req * 2352static eio_req *
1461eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2353eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1462{ 2354{
… …
1486eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2378eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1487{ 2379{
1488 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2380 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1489} 2381}
1490 2382
2383eio_req *eio_slurp (const char *path, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
2384{
2385 REQ (EIO_SLURP); PATH; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
2386}
2387
1491eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) 2388eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
1492{ 2389{
1493 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND; 2390 REQ (EIO_CUSTOM); req->feed = execute; SEND;
1494} 2391}
1495 2392
1496#endif 2393#endif
1497 2394
1498eio_req *eio_grp (eio_cb cb, void *data) 2395eio_req *eio_grp (eio_cb cb, void *data)
… …
1507#undef SEND 2404#undef SEND
1508 2405
1509/*****************************************************************************/ 2406/*****************************************************************************/
1510/* grp functions */ 2407/* grp functions */
1511 2408
2409void
1512void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2410eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1513{ 2411{
1514 grp->int2 = limit; 2412 grp->int2 = limit;
1515 grp->feed = feed; 2413 grp->feed = feed;
1516 2414
1517 grp_try_feed (grp); 2415 grp_try_feed (grp);
1518} 2416}
1519 2417
2418void
1520void eio_grp_limit (eio_req *grp, int limit) 2419eio_grp_limit (eio_req *grp, int limit)
1521{ 2420{
1522 grp->int2 = limit; 2421 grp->int2 = limit;
1523 2422
1524 grp_try_feed (grp); 2423 grp_try_feed (grp);
1525} 2424}
1526 2425
2426void
1527void eio_grp_add (eio_req *grp, eio_req *req) 2427eio_grp_add (eio_req *grp, eio_req *req)
1528{ 2428{
1529 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2429 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1530 2430
1531 grp->flags |= EIO_FLAG_GROUPADD; 2431 grp->flags |= ETP_FLAG_GROUPADD;
1532 2432
1533 ++grp->size; 2433 ++grp->size;
1534 req->grp = grp; 2434 req->grp = grp;
1535 2435
1536 req->grp_prev = 0; 2436 req->grp_prev = 0;
… …
1543} 2443}
1544 2444
1545/*****************************************************************************/ 2445/*****************************************************************************/
1546/* misc garbage */ 2446/* misc garbage */
1547 2447
2448eio_ssize_t
1548ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2449eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1549{ 2450{
1550 etp_worker wrk;
1551
1552 wrk.dbuf = 0;
1553
1554 eio__sendfile (ofd, ifd, offset, count, &wrk); 2451 return eio__sendfile (ofd, ifd, offset, count);
1555
1556 if (wrk.dbuf)
1557 free (wrk.dbuf);
1558} 2452}
1559 2453
2454int eio_mlockall_sync (int flags)
2455{
2456 return eio__mlockall (flags);
2457}
2458

Diff Legend

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