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

Comparing libeio/eio.c (file contents):
Revision 1.49 by root, Sat Jan 2 14:24:32 2010 UTC vs.
Revision 1.146 by root, Sat Jan 6 01:04:43 2018 UTC

1/* 1/*
2 * libeio implementation 2 * libeio implementation
3 * 3 *
4 * Copyright (c) 2007,2008,2009,2010 Marc Alexander Lehmann <libeio@schmorp.de> 4 * Copyright (c) 2007,2008,2009,2010,2011,2012,2013,2016,2017 Marc Alexander Lehmann <libeio@schmorp.de>
5 * All rights reserved. 5 * All rights reserved.
6 * 6 *
7 * Redistribution and use in source and binary forms, with or without modifica- 7 * Redistribution and use in source and binary forms, with or without modifica-
8 * tion, are permitted provided that the following conditions are met: 8 * tion, are permitted provided that the following conditions are met:
9 * 9 *
10 * 1. Redistributions of source code must retain the above copyright notice, 10 * 1. Redistributions of source code must retain the above copyright notice,
11 * this list of conditions and the following disclaimer. 11 * this list of conditions and the following disclaimer.
12 * 12 *
13 * 2. Redistributions in binary form must reproduce the above copyright 13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the 14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution. 15 * documentation and/or other materials provided with the distribution.
16 * 16 *
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR IMPLIED 17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR IMPLIED
18 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MER- 18 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MER-
19 * CHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO 19 * CHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO
20 * EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPE- 20 * EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPE-
21 * CIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, 21 * CIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
… …
35 * and other provisions required by the GPL. If you do not delete the 35 * and other provisions required by the GPL. If you do not delete the
36 * provisions above, a recipient may use your version of this file under 36 * provisions above, a recipient may use your version of this file under
37 * either the BSD or the GPL. 37 * either the BSD or the GPL.
38 */ 38 */
39 39
40#ifndef _WIN32
41# include "config.h"
42#endif
43
40#include "eio.h" 44#include "eio.h"
41
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)
233
73#else 234#else
74 235
75# include "config.h" 236 #include <sys/ioctl.h>
76# include <sys/time.h> 237 #include <sys/time.h>
77# include <sys/select.h> 238 #include <sys/select.h>
78# include <sys/mman.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/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
85# if __freebsd || defined __NetBSD__ || defined __OpenBSD__
86# define _DIRENT_HAVE_D_TYPE /* sigh */
87# define D_INO(de) (de)->d_fileno
88# define D_NAMLEN(de) (de)->d_namlen
89# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
90# define D_INO(de) (de)->d_ino
91# endif 284#endif
92 285
93#ifdef _D_EXACT_NAMLEN 286#if HAVE_SYS_SYSCALL_H
94# undef D_NAMLEN 287# include <sys/syscall.h>
95# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
96#endif
97
98# ifdef _DIRENT_HAVE_D_TYPE
99# define D_TYPE(de) (de)->d_type
100# endif
101
102# ifndef EIO_STRUCT_DIRENT
103# define EIO_STRUCT_DIRENT struct dirent
104# endif
105
106#endif 288#endif
107 289
108#if HAVE_SENDFILE 290#if HAVE_SENDFILE
109# if __linux 291# if __linux
110# include <sys/sendfile.h> 292# include <sys/sendfile.h>
111# elif __freebsd || defined __APPLE__ 293# elif __FreeBSD__ || defined __APPLE__
112# include <sys/socket.h> 294# include <sys/socket.h>
113# include <sys/uio.h> 295# include <sys/uio.h>
114# elif __hpux 296# elif __hpux
115# include <sys/socket.h> 297# include <sys/socket.h>
116# elif __solaris 298# elif __solaris
… …
118# else 300# else
119# error sendfile support requested but not available 301# error sendfile support requested but not available
120# endif 302# endif
121#endif 303#endif
122 304
305#if HAVE_RENAMEAT2
306# include <sys/syscall.h>
307# include <linux/fs.h>
308#endif
309
123#ifndef D_TYPE 310#ifndef D_TYPE
124# define D_TYPE(de) 0 311# define D_TYPE(de) 0
125#endif 312#endif
126#ifndef D_INO 313#ifndef D_INO
127# define D_INO(de) 0 314# define D_INO(de) 0
128#endif 315#endif
129#ifndef D_NAMLEN 316#ifndef D_NAMLEN
130# define D_NAMLEN(de) strlen ((de)->d_name) 317# define D_NAMLEN(entp) strlen (D_NAME (entp))
131#endif 318#endif
132
133/* number of seconds after which an idle threads exit */
134#define IDLE_TIMEOUT 10
135 319
136/* used for struct dirent, AIX doesn't provide it */ 320/* used for struct dirent, AIX doesn't provide it */
137#ifndef NAME_MAX 321#ifndef NAME_MAX
138# define NAME_MAX 4096 322# define NAME_MAX 4096
139#endif 323#endif
140 324
325/* used for readlink etc. */
326#ifndef PATH_MAX
327# define PATH_MAX 0
328#endif
329
330#ifndef EIO_PATH_MIN
331# define EIO_PATH_MIN 8160
332#endif
333
334#define EIO_PATH_MAX (PATH_MAX <= EIO_PATH_MIN ? EIO_PATH_MIN : PATH_MAX)
335
141/* buffer size for various temporary buffers */ 336/* buffer size for various temporary buffers */
142#define EIO_BUFSIZE 65536 337#define EIO_BUFSIZE 65536
143 338
144#define dBUF \ 339#define dBUF \
145 char *eio_buf; \
146 ETP_WORKER_LOCK (self); \
147 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 340 char *eio_buf = malloc (EIO_BUFSIZE); \
148 ETP_WORKER_UNLOCK (self); \
149 errno = ENOMEM; \ 341 errno = ENOMEM; \
150 if (!eio_buf) \ 342 if (!eio_buf) \
151 return -1; 343 return -1
152 344
153#define EIO_TICKS ((1000000 + 1023) >> 10) 345#define FUBd \
346 free (eio_buf)
154 347
155/*****************************************************************************/ 348/*****************************************************************************/
156 349
157#if __GNUC__ >= 3 350struct etp_tmpbuf;
158# define expect(expr,value) __builtin_expect ((expr),(value)) 351
352#if _POSIX_VERSION >= 200809L
353 #define HAVE_AT 1
354 #define WD2FD(wd) ((wd) ? (wd)->fd : AT_FDCWD)
355 #ifndef O_SEARCH
356 #define O_SEARCH O_RDONLY
357 #endif
159#else 358#else
160# define expect(expr,value) (expr) 359 #define HAVE_AT 0
360 static const char *wd_expand (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path);
161#endif 361#endif
162 362
163#define expect_false(expr) expect ((expr) != 0, 0) 363struct eio_pwd
164#define expect_true(expr) expect ((expr) != 0, 1) 364{
365#if HAVE_AT
366 int fd;
367#endif
368 int len;
369 char str[1]; /* actually, a 0-terminated canonical path */
370};
165 371
166/*****************************************************************************/ 372/*****************************************************************************/
167 373
168#define ETP_PRI_MIN EIO_PRI_MIN 374#define ETP_PRI_MIN EIO_PRI_MIN
169#define ETP_PRI_MAX EIO_PRI_MAX 375#define ETP_PRI_MAX EIO_PRI_MAX
170 376
377#define ETP_TYPE_QUIT -1
378#define ETP_TYPE_GROUP EIO_GROUP
379
380static void eio_nop_callback (void) { }
381static void (*eio_want_poll_cb)(void) = eio_nop_callback;
382static void (*eio_done_poll_cb)(void) = eio_nop_callback;
383
384#define ETP_WANT_POLL(pool) eio_want_poll_cb ()
385#define ETP_DONE_POLL(pool) eio_done_poll_cb ()
386
171struct etp_worker; 387struct etp_worker;
172
173#define ETP_REQ eio_req 388#define ETP_REQ eio_req
174#define ETP_DESTROY(req) eio_destroy (req) 389#define ETP_DESTROY(req) eio_destroy (req)
175static int eio_finish (eio_req *req); 390static int eio_finish (eio_req *req);
176#define ETP_FINISH(req) eio_finish (req) 391#define ETP_FINISH(req) eio_finish (req)
177static void eio_execute (struct etp_worker *self, eio_req *req); 392static void eio_execute (struct etp_worker *self, eio_req *req);
178#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) 393#define ETP_EXECUTE(wrk,req) eio_execute (wrk, req)
179 394
180#define ETP_WORKER_CLEAR(req) \ 395#include "etp.c"
181 if (wrk->dbuf) \
182 { \
183 free (wrk->dbuf); \
184 wrk->dbuf = 0; \
185 } \
186 \
187 if (wrk->dirp) \
188 { \
189 closedir (wrk->dirp); \
190 wrk->dirp = 0; \
191 }
192 396
193#define ETP_WORKER_COMMON \ 397static struct etp_pool eio_pool;
194 void *dbuf; \ 398#define EIO_POOL (&eio_pool)
195 DIR *dirp;
196 399
197/*****************************************************************************/ 400/*****************************************************************************/
198 401
199#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
200
201/* calculate time difference in ~1/EIO_TICKS of a second */
202static int tvdiff (struct timeval *tv1, struct timeval *tv2)
203{
204 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
205 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
206}
207
208static unsigned int started, idle, wanted = 4;
209
210static void (*want_poll_cb) (void);
211static void (*done_poll_cb) (void);
212
213static unsigned int max_poll_time; /* reslock */
214static unsigned int max_poll_reqs; /* reslock */
215
216static volatile unsigned int nreqs; /* reqlock */
217static volatile unsigned int nready; /* reqlock */
218static volatile unsigned int npending; /* reqlock */
219static volatile unsigned int max_idle = 4;
220
221static mutex_t wrklock = X_MUTEX_INIT;
222static mutex_t reslock = X_MUTEX_INIT;
223static mutex_t reqlock = X_MUTEX_INIT;
224static cond_t reqwait = X_COND_INIT;
225
226#if !HAVE_PREADWRITE
227/*
228 * make our pread/pwrite emulation safe against themselves, but not against
229 * normal read/write by using a mutex. slows down execution a lot,
230 * but that's your problem, not mine.
231 */
232static mutex_t preadwritelock = X_MUTEX_INIT;
233#endif
234
235typedef struct etp_worker
236{
237 /* locked by wrklock */
238 struct etp_worker *prev, *next;
239
240 thread_t tid;
241
242 /* locked by reslock, reqlock or wrklock */
243 ETP_REQ *req; /* currently processed request */
244
245 ETP_WORKER_COMMON
246} etp_worker;
247
248static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */
249
250#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
251#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
252
253/* worker threads management */
254
255static void etp_worker_clear (etp_worker *wrk)
256{
257 ETP_WORKER_CLEAR (wrk);
258}
259
260static void etp_worker_free (etp_worker *wrk)
261{
262 wrk->next->prev = wrk->prev;
263 wrk->prev->next = wrk->next;
264
265 free (wrk);
266}
267
268static unsigned int etp_nreqs (void)
269{
270 int retval;
271 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
272 retval = nreqs;
273 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
274 return retval;
275}
276
277static unsigned int etp_nready (void)
278{
279 unsigned int retval;
280
281 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
282 retval = nready;
283 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
284
285 return retval;
286}
287
288static unsigned int etp_npending (void)
289{
290 unsigned int retval;
291
292 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
293 retval = npending;
294 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
295
296 return retval;
297}
298
299static unsigned int etp_nthreads (void)
300{
301 unsigned int retval;
302
303 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
304 retval = started;
305 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
306
307 return retval;
308}
309
310/*
311 * a somewhat faster data structure might be nice, but
312 * with 8 priorities this actually needs <20 insns
313 * per shift, the most expensive operation.
314 */
315typedef struct {
316 ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */
317 int size;
318} etp_reqq;
319
320static etp_reqq req_queue;
321static etp_reqq res_queue;
322
323static int reqq_push (etp_reqq *q, ETP_REQ *req)
324{
325 int pri = req->pri;
326 req->next = 0;
327
328 if (q->qe[pri])
329 {
330 q->qe[pri]->next = req;
331 q->qe[pri] = req;
332 }
333 else
334 q->qe[pri] = q->qs[pri] = req;
335
336 return q->size++;
337}
338
339static ETP_REQ *reqq_shift (etp_reqq *q)
340{
341 int pri;
342
343 if (!q->size)
344 return 0;
345
346 --q->size;
347
348 for (pri = ETP_NUM_PRI; pri--; )
349 {
350 eio_req *req = q->qs[pri];
351
352 if (req)
353 {
354 if (!(q->qs[pri] = (eio_req *)req->next))
355 q->qe[pri] = 0;
356
357 return req;
358 }
359 }
360
361 abort ();
362}
363
364static void etp_atfork_prepare (void)
365{
366 X_LOCK (wrklock);
367 X_LOCK (reqlock);
368 X_LOCK (reslock);
369#if !HAVE_PREADWRITE
370 X_LOCK (preadwritelock);
371#endif
372}
373
374static void etp_atfork_parent (void)
375{
376#if !HAVE_PREADWRITE
377 X_UNLOCK (preadwritelock);
378#endif
379 X_UNLOCK (reslock);
380 X_UNLOCK (reqlock);
381 X_UNLOCK (wrklock);
382}
383
384static void etp_atfork_child (void)
385{
386 ETP_REQ *prv;
387
388 while ((prv = reqq_shift (&req_queue)))
389 ETP_DESTROY (prv);
390
391 while ((prv = reqq_shift (&res_queue)))
392 ETP_DESTROY (prv);
393
394 while (wrk_first.next != &wrk_first)
395 {
396 etp_worker *wrk = wrk_first.next;
397
398 if (wrk->req)
399 ETP_DESTROY (wrk->req);
400
401 etp_worker_clear (wrk);
402 etp_worker_free (wrk);
403 }
404
405 started = 0;
406 idle = 0;
407 nreqs = 0;
408 nready = 0;
409 npending = 0;
410
411 etp_atfork_parent ();
412}
413
414static void 402static void
415etp_once_init (void)
416{
417 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
418}
419
420static int
421etp_init (void (*want_poll)(void), void (*done_poll)(void))
422{
423 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
424
425 pthread_once (&doinit, etp_once_init);
426
427 want_poll_cb = want_poll;
428 done_poll_cb = done_poll;
429
430 return 0;
431}
432
433X_THREAD_PROC (etp_proc);
434
435static void etp_start_thread (void)
436{
437 etp_worker *wrk = calloc (1, sizeof (etp_worker));
438
439 /*TODO*/
440 assert (("unable to allocate worker thread data", wrk));
441
442 X_LOCK (wrklock);
443
444 if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
445 {
446 wrk->prev = &wrk_first;
447 wrk->next = wrk_first.next;
448 wrk_first.next->prev = wrk;
449 wrk_first.next = wrk;
450 ++started;
451 }
452 else
453 free (wrk);
454
455 X_UNLOCK (wrklock);
456}
457
458static void etp_maybe_start_thread (void)
459{
460 if (expect_true (etp_nthreads () >= wanted))
461 return;
462
463 /* todo: maybe use idle here, but might be less exact */
464 if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
465 return;
466
467 etp_start_thread ();
468}
469
470static void etp_end_thread (void)
471{
472 eio_req *req = calloc (1, sizeof (eio_req));
473
474 req->type = -1;
475 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
476
477 X_LOCK (reqlock);
478 reqq_push (&req_queue, req);
479 X_COND_SIGNAL (reqwait);
480 X_UNLOCK (reqlock);
481
482 X_LOCK (wrklock);
483 --started;
484 X_UNLOCK (wrklock);
485}
486
487static int etp_poll (void)
488{
489 unsigned int maxreqs;
490 unsigned int maxtime;
491 struct timeval tv_start, tv_now;
492
493 X_LOCK (reslock);
494 maxreqs = max_poll_reqs;
495 maxtime = max_poll_time;
496 X_UNLOCK (reslock);
497
498 if (maxtime)
499 gettimeofday (&tv_start, 0);
500
501 for (;;)
502 {
503 ETP_REQ *req;
504
505 etp_maybe_start_thread ();
506
507 X_LOCK (reslock);
508 req = reqq_shift (&res_queue);
509
510 if (req)
511 {
512 --npending;
513
514 if (!res_queue.size && done_poll_cb)
515 done_poll_cb ();
516 }
517
518 X_UNLOCK (reslock);
519
520 if (!req)
521 return 0;
522
523 X_LOCK (reqlock);
524 --nreqs;
525 X_UNLOCK (reqlock);
526
527 if (expect_false (req->type == EIO_GROUP && req->size))
528 {
529 req->int1 = 1; /* mark request as delayed */
530 continue;
531 }
532 else
533 {
534 int res = ETP_FINISH (req);
535 if (expect_false (res))
536 return res;
537 }
538
539 if (expect_false (maxreqs && !--maxreqs))
540 break;
541
542 if (maxtime)
543 {
544 gettimeofday (&tv_now, 0);
545
546 if (tvdiff (&tv_start, &tv_now) >= maxtime)
547 break;
548 }
549 }
550
551 errno = EAGAIN;
552 return -1;
553}
554
555static void etp_cancel (ETP_REQ *req)
556{
557 X_LOCK (wrklock);
558 req->flags |= EIO_FLAG_CANCELLED;
559 X_UNLOCK (wrklock);
560
561 eio_grp_cancel (req);
562}
563
564static void etp_submit (ETP_REQ *req)
565{
566 req->pri -= ETP_PRI_MIN;
567
568 if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
569 if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
570
571 if (expect_false (req->type == EIO_GROUP))
572 {
573 /* I hope this is worth it :/ */
574 X_LOCK (reqlock);
575 ++nreqs;
576 X_UNLOCK (reqlock);
577
578 X_LOCK (reslock);
579
580 ++npending;
581
582 if (!reqq_push (&res_queue, req) && want_poll_cb)
583 want_poll_cb ();
584
585 X_UNLOCK (reslock);
586 }
587 else
588 {
589 X_LOCK (reqlock);
590 ++nreqs;
591 ++nready;
592 reqq_push (&req_queue, req);
593 X_COND_SIGNAL (reqwait);
594 X_UNLOCK (reqlock);
595
596 etp_maybe_start_thread ();
597 }
598}
599
600static void etp_set_max_poll_time (double nseconds)
601{
602 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
603 max_poll_time = nseconds * EIO_TICKS;
604 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
605}
606
607static void etp_set_max_poll_reqs (unsigned int maxreqs)
608{
609 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
610 max_poll_reqs = maxreqs;
611 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
612}
613
614static void etp_set_max_idle (unsigned int nthreads)
615{
616 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
617 max_idle = nthreads <= 0 ? 1 : nthreads;
618 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
619}
620
621static void etp_set_min_parallel (unsigned int nthreads)
622{
623 if (wanted < nthreads)
624 wanted = nthreads;
625}
626
627static void etp_set_max_parallel (unsigned int nthreads)
628{
629 if (wanted > nthreads)
630 wanted = nthreads;
631
632 while (started > wanted)
633 etp_end_thread ();
634}
635
636/*****************************************************************************/
637
638static void grp_try_feed (eio_req *grp) 403grp_try_feed (eio_req *grp)
639{ 404{
640 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 405 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
641 { 406 {
642 grp->flags &= ~EIO_FLAG_GROUPADD; 407 grp->flags &= ~ETP_FLAG_GROUPADD;
643 408
644 EIO_FEED (grp); 409 EIO_FEED (grp);
645 410
646 /* stop if no progress has been made */ 411 /* stop if no progress has been made */
647 if (!(grp->flags & EIO_FLAG_GROUPADD)) 412 if (!(grp->flags & ETP_FLAG_GROUPADD))
648 { 413 {
649 grp->feed = 0; 414 grp->feed = 0;
650 break; 415 break;
651 } 416 }
652 } 417 }
653} 418}
654 419
420static int
655static int grp_dec (eio_req *grp) 421grp_dec (eio_req *grp)
656{ 422{
657 --grp->size; 423 --grp->size;
658 424
659 /* call feeder, if applicable */ 425 /* call feeder, if applicable */
660 grp_try_feed (grp); 426 grp_try_feed (grp);
661 427
662 /* finish, if done */ 428 /* finish, if done */
663 if (!grp->size && grp->int1) 429 if (!grp->size && grp->flags & ETP_FLAG_DELAYED)
664 return eio_finish (grp); 430 return eio_finish (grp);
665 else 431 else
666 return 0; 432 return 0;
667} 433}
668 434
435static void
669void eio_destroy (eio_req *req) 436eio_destroy (eio_req *req)
670{ 437{
671 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); 438 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
672 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); 439 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
673 440
674 EIO_DESTROY (req); 441 EIO_DESTROY (req);
675} 442}
676 443
444static int
677static int eio_finish (eio_req *req) 445eio_finish (eio_req *req)
678{ 446{
679 int res = EIO_FINISH (req); 447 int res = EIO_FINISH (req);
680 448
681 if (req->grp) 449 if (req->grp)
682 { 450 {
… …
690 if (grp->grp_first == req) 458 if (grp->grp_first == req)
691 grp->grp_first = req->grp_next; 459 grp->grp_first = req->grp_next;
692 460
693 res2 = grp_dec (grp); 461 res2 = grp_dec (grp);
694 462
695 if (!res && res2) 463 if (!res)
696 res = res2; 464 res = res2;
697 } 465 }
698 466
699 eio_destroy (req); 467 eio_destroy (req);
700 468
701 return res; 469 return res;
702} 470}
703 471
472void
704void eio_grp_cancel (eio_req *grp) 473eio_grp_cancel (eio_req *grp)
705{ 474{
706 for (grp = grp->grp_first; grp; grp = grp->grp_next) 475 etp_grp_cancel (EIO_POOL, grp);
707 eio_cancel (grp);
708} 476}
709 477
478void
710void eio_cancel (eio_req *req) 479eio_cancel (eio_req *req)
711{ 480{
712 etp_cancel (req); 481 etp_cancel (EIO_POOL, req);
713} 482}
714 483
484void
715void eio_submit (eio_req *req) 485eio_submit (eio_req *req)
716{ 486{
717 etp_submit (req); 487 etp_submit (EIO_POOL, req);
718} 488}
719 489
720unsigned int eio_nreqs (void) 490unsigned int
491eio_nreqs (void)
721{ 492{
722 return etp_nreqs (); 493 return etp_nreqs (EIO_POOL);
723} 494}
724 495
725unsigned int eio_nready (void) 496unsigned int
497eio_nready (void)
726{ 498{
727 return etp_nready (); 499 return etp_nready (EIO_POOL);
728} 500}
729 501
730unsigned int eio_npending (void) 502unsigned int
503eio_npending (void)
731{ 504{
732 return etp_npending (); 505 return etp_npending (EIO_POOL);
733} 506}
734 507
735unsigned int eio_nthreads (void) 508unsigned int ecb_cold
509eio_nthreads (void)
736{ 510{
737 return etp_nthreads (); 511 return etp_nthreads (EIO_POOL);
738} 512}
739 513
514void ecb_cold
740void eio_set_max_poll_time (double nseconds) 515eio_set_max_poll_time (double nseconds)
741{ 516{
742 etp_set_max_poll_time (nseconds); 517 etp_set_max_poll_time (EIO_POOL, nseconds);
743} 518}
744 519
520void ecb_cold
745void eio_set_max_poll_reqs (unsigned int maxreqs) 521eio_set_max_poll_reqs (unsigned int maxreqs)
746{ 522{
747 etp_set_max_poll_reqs (maxreqs); 523 etp_set_max_poll_reqs (EIO_POOL, maxreqs);
748} 524}
749 525
526void ecb_cold
750void eio_set_max_idle (unsigned int nthreads) 527eio_set_max_idle (unsigned int nthreads)
751{ 528{
752 etp_set_max_idle (nthreads); 529 etp_set_max_idle (EIO_POOL, nthreads);
753} 530}
754 531
532void ecb_cold
533eio_set_idle_timeout (unsigned int seconds)
534{
535 etp_set_idle_timeout (EIO_POOL, seconds);
536}
537
538void ecb_cold
755void eio_set_min_parallel (unsigned int nthreads) 539eio_set_min_parallel (unsigned int nthreads)
756{ 540{
757 etp_set_min_parallel (nthreads); 541 etp_set_min_parallel (EIO_POOL, nthreads);
758} 542}
759 543
544void ecb_cold
760void eio_set_max_parallel (unsigned int nthreads) 545eio_set_max_parallel (unsigned int nthreads)
761{ 546{
762 etp_set_max_parallel (nthreads); 547 etp_set_max_parallel (EIO_POOL, nthreads);
763} 548}
764 549
765int eio_poll (void) 550int eio_poll (void)
766{ 551{
767 return etp_poll (); 552 return etp_poll (EIO_POOL);
768} 553}
769 554
770/*****************************************************************************/ 555/*****************************************************************************/
771/* work around various missing functions */ 556/* work around various missing functions */
772 557
773#if !HAVE_PREADWRITE 558#if HAVE_POSIX_CLOSE && !__linux
774# undef pread 559# define eio__close(fd) posix_close (fd, 0)
775# undef pwrite 560#else
776# define pread eio__pread 561# define eio__close(fd) close (fd)
777# define pwrite eio__pwrite
778
779static ssize_t
780eio__pread (int fd, void *buf, size_t count, off_t offset)
781{
782 ssize_t res;
783 off_t ooffset;
784
785 X_LOCK (preadwritelock);
786 ooffset = lseek (fd, 0, SEEK_CUR);
787 lseek (fd, offset, SEEK_SET);
788 res = read (fd, buf, count);
789 lseek (fd, ooffset, SEEK_SET);
790 X_UNLOCK (preadwritelock);
791
792 return res;
793}
794
795static ssize_t
796eio__pwrite (int fd, void *buf, size_t count, off_t offset)
797{
798 ssize_t res;
799 off_t ooffset;
800
801 X_LOCK (preadwritelock);
802 ooffset = lseek (fd, 0, SEEK_CUR);
803 lseek (fd, offset, SEEK_SET);
804 res = write (fd, buf, count);
805 lseek (fd, ooffset, SEEK_SET);
806 X_UNLOCK (preadwritelock);
807
808 return res;
809}
810#endif 562#endif
811 563
564/* close() without disturbing errno */
565static void
566silent_close (int fd)
567{
568 int saved_errno = errno;
569 eio__close (fd);
570 errno = saved_errno;
571}
572
812#ifndef HAVE_FUTIMES 573#ifndef HAVE_UTIMES
813 574
814# undef utimes 575# undef utimes
815# undef futimes
816# define utimes(path,times) eio__utimes (path, times) 576# define utimes(path,times) eio__utimes (path, times)
817# define futimes(fd,times) eio__futimes (fd, times)
818 577
819static int 578static int
820eio__utimes (const char *filename, const struct timeval times[2]) 579eio__utimes (const char *filename, const struct timeval times[2])
821{ 580{
822 if (times) 581 if (times)
… …
830 } 589 }
831 else 590 else
832 return utime (filename, 0); 591 return utime (filename, 0);
833} 592}
834 593
594#endif
595
596#ifndef HAVE_FUTIMES
597
598# undef futimes
599# define futimes(fd,times) eio__futimes (fd, times)
600
601static int
835static int eio__futimes (int fd, const struct timeval tv[2]) 602eio__futimes (int fd, const struct timeval tv[2])
836{ 603{
837 errno = ENOSYS; 604 errno = ENOSYS;
838 return -1; 605 return -1;
839} 606}
840 607
… …
843#if !HAVE_FDATASYNC 610#if !HAVE_FDATASYNC
844# undef fdatasync 611# undef fdatasync
845# define fdatasync(fd) fsync (fd) 612# define fdatasync(fd) fsync (fd)
846#endif 613#endif
847 614
615static int
616eio__syncfs (int fd)
617{
618 int res;
619
620#if HAVE_SYS_SYNCFS
621 res = (int)syscall (__NR_syncfs, (int)(fd));
622#else
623 res = EIO_ENOSYS ();
624#endif
625
626 if (res < 0 && errno == ENOSYS && fd >= 0)
627 sync ();
628
629 return res;
630}
631
848/* sync_file_range always needs emulation */ 632/* sync_file_range always needs emulation */
849int 633static int
850eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags) 634eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
851{ 635{
852#if HAVE_SYNC_FILE_RANGE 636#if HAVE_SYNC_FILE_RANGE
853 int res; 637 int res;
854 638
… …
871 /* even though we could play tricks with the flags, it's better to always 655 /* even though we could play tricks with the flags, it's better to always
872 * call fdatasync, as that matches the expectation of its users best */ 656 * call fdatasync, as that matches the expectation of its users best */
873 return fdatasync (fd); 657 return fdatasync (fd);
874} 658}
875 659
660static int
661eio__fallocate (int fd, int mode, off_t offset, size_t len)
662{
663#if HAVE_LINUX_FALLOCATE
664 return fallocate (fd, mode, offset, len);
665#else
666 return EIO_ENOSYS ();
667#endif
668}
669
876#if !HAVE_READAHEAD 670#if !HAVE_READAHEAD
877# undef readahead 671# undef readahead
878# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 672# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
879 673
880static ssize_t 674static eio_ssize_t
881eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) 675eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
882{ 676{
883 size_t todo = count; 677 size_t todo = count;
884 dBUF; 678 dBUF;
885 679
… …
890 pread (fd, eio_buf, len, offset); 684 pread (fd, eio_buf, len, offset);
891 offset += len; 685 offset += len;
892 todo -= len; 686 todo -= len;
893 } 687 }
894 688
895 errno = 0; 689 FUBd;
690
691 /* linux's readahead basically only fails for EBADF or EINVAL (not mmappable) */
692 /* but not for e.g. EIO or eof, so we also never fail */
896 return count; 693 return 0;
897} 694}
898 695
899#endif 696#endif
900 697
901/* sendfile always needs emulation */ 698/* sendfile always needs emulation */
902static ssize_t 699static eio_ssize_t
903eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) 700eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
904{ 701{
702 eio_ssize_t written = 0;
905 ssize_t res; 703 eio_ssize_t res;
906 704
907 if (!count) 705 if (!count)
908 return 0; 706 return 0;
909 707
708 for (;;)
709 {
710#ifdef __APPLE__
711# undef HAVE_SENDFILE /* broken, as everything on os x */
712#endif
910#if HAVE_SENDFILE 713#if HAVE_SENDFILE
911# if __linux 714# if __linux
715 off_t soffset = offset;
912 res = sendfile (ofd, ifd, &offset, count); 716 res = sendfile (ofd, ifd, &soffset, count);
913 717
914# elif __freebsd 718# elif __FreeBSD__
915 /* 719 /*
916 * Of course, the freebsd sendfile is a dire hack with no thoughts 720 * Of course, the freebsd sendfile is a dire hack with no thoughts
917 * wasted on making it similar to other I/O functions. 721 * wasted on making it similar to other I/O functions.
918 */ 722 */
919 {
920 off_t sbytes; 723 off_t sbytes;
921 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 724 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
922 725
923 if (res < 0 && sbytes) 726 #if 0 /* according to the manpage, this is correct, but broken behaviour */
924 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 727 /* freebsd' sendfile will return 0 on success */
728 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
729 /* not on e.g. EIO or EPIPE - sounds broken */
730 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
925 res = sbytes; 731 res = sbytes;
926 } 732 #endif
927 733
734 /* according to source inspection, this is correct, and useful behaviour */
735 if (sbytes)
736 res = sbytes;
737
928# elif defined (__APPLE__) 738# elif defined __APPLE__
929
930 {
931 off_t sbytes = count; 739 off_t sbytes = count;
932 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0); 740 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
933 741
934 if (res < 0 && errno == EAGAIN && sbytes) 742 /* according to the manpage, sbytes is always valid */
743 if (sbytes)
935 res = sbytes; 744 res = sbytes;
936 }
937 745
938# elif __hpux 746# elif __hpux
939 res = sendfile (ofd, ifd, offset, count, 0, 0); 747 res = sendfile (ofd, ifd, offset, count, 0, 0);
940 748
941# elif __solaris 749# elif __solaris
942 {
943 struct sendfilevec vec; 750 struct sendfilevec vec;
944 size_t sbytes; 751 size_t sbytes;
945 752
946 vec.sfv_fd = ifd; 753 vec.sfv_fd = ifd;
947 vec.sfv_flag = 0; 754 vec.sfv_flag = 0;
948 vec.sfv_off = offset; 755 vec.sfv_off = offset;
949 vec.sfv_len = count; 756 vec.sfv_len = count;
950 757
951 res = sendfilev (ofd, &vec, 1, &sbytes); 758 res = sendfilev (ofd, &vec, 1, &sbytes);
952 759
953 if (res < 0 && sbytes) 760 if (res < 0 && sbytes)
954 res = sbytes; 761 res = sbytes;
955 }
956 762
957# endif 763# endif
958 764
959#elif defined (_WIN32) 765#elif defined (_WIN32) && 0
960
961 /* does not work, just for documentation of what would need to be done */ 766 /* does not work, just for documentation of what would need to be done */
962 { 767 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
768 /* libeio guarantees that the file offset does not change, and windows */
769 /* has no way to get an independent handle to the same file description */
963 HANDLE h = TO_SOCKET (ifd); 770 HANDLE h = TO_SOCKET (ifd);
964 SetFilePointer (h, offset, 0, FILE_BEGIN); 771 SetFilePointer (h, offset, 0, FILE_BEGIN);
965 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0); 772 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
966 }
967 773
968#else 774#else
969 res = -1; 775 res = EIO_ENOSYS ();
970 errno = ENOSYS;
971#endif 776#endif
972 777
778 /* we assume sendfile can copy at least 128mb in one go */
779 if (res <= 128 * 1024 * 1024)
780 {
781 if (res > 0)
782 written += res;
783
784 if (written)
785 return written;
786
787 break;
788 }
789 else
790 {
791 /* if we requested more, then probably the kernel was lazy */
792 written += res;
793 offset += res;
794 count -= res;
795
796 if (!count)
797 return written;
798 }
799 }
800
973 if (res < 0 801 if (res < 0
974 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 802 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
975 /* BSDs */ 803 /* BSDs */
976#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */ 804#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
977 || errno == ENOTSUP 805 || errno == ENOTSUP
978#endif 806#endif
807#ifdef EOPNOTSUPP /* windows */
979 || errno == EOPNOTSUPP /* BSDs */ 808 || errno == EOPNOTSUPP /* BSDs */
809#endif
980#if __solaris 810#if __solaris
981 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 811 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
982#endif 812#endif
983 ) 813 )
984 ) 814 )
… …
988 818
989 res = 0; 819 res = 0;
990 820
991 while (count) 821 while (count)
992 { 822 {
993 ssize_t cnt; 823 eio_ssize_t cnt;
994 824
995 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 825 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
996 826
997 if (cnt <= 0) 827 if (cnt <= 0)
998 { 828 {
… …
1010 840
1011 offset += cnt; 841 offset += cnt;
1012 res += cnt; 842 res += cnt;
1013 count -= cnt; 843 count -= cnt;
1014 } 844 }
845
846 FUBd;
1015 } 847 }
1016 848
1017 return res; 849 return res;
850}
851
852#ifdef PAGESIZE
853# define eio_pagesize() PAGESIZE
854#else
855static intptr_t
856eio_pagesize (void)
857{
858 static intptr_t page;
859
860 if (!page)
861 page = sysconf (_SC_PAGESIZE);
862
863 return page;
864}
865#endif
866
867static void
868eio_page_align (void **addr, size_t *length)
869{
870 intptr_t mask = eio_pagesize () - 1;
871
872 /* round down addr */
873 intptr_t adj = mask & (intptr_t)*addr;
874
875 *addr = (void *)((intptr_t)*addr - adj);
876 *length += adj;
877
878 /* round up length */
879 *length = (*length + mask) & ~mask;
880}
881
882#if !_POSIX_MEMLOCK
883# define eio__mlockall(a) EIO_ENOSYS ()
884#else
885
886static int
887eio__mlockall (int flags)
888{
889 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
890 extern int mallopt (int, int);
891 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
892 #endif
893
894 if (EIO_MCL_CURRENT != MCL_CURRENT
895 || EIO_MCL_FUTURE != MCL_FUTURE)
896 {
897 flags = 0
898 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
899 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
900 }
901
902 return mlockall (flags);
903}
904#endif
905
906#if !_POSIX_MEMLOCK_RANGE
907# define eio__mlock(a,b) EIO_ENOSYS ()
908#else
909
910static int
911eio__mlock (void *addr, size_t length)
912{
913 eio_page_align (&addr, &length);
914
915 return mlock (addr, length);
916}
917
918#endif
919
920#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
921# define eio__msync(a,b,c) EIO_ENOSYS ()
922#else
923
924static int
925eio__msync (void *mem, size_t len, int flags)
926{
927 eio_page_align (&mem, &len);
928
929 if (EIO_MS_ASYNC != MS_SYNC
930 || EIO_MS_INVALIDATE != MS_INVALIDATE
931 || EIO_MS_SYNC != MS_SYNC)
932 {
933 flags = 0
934 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
935 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
936 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
937 }
938
939 return msync (mem, len, flags);
940}
941
942#endif
943
944static int
945eio__mtouch (eio_req *req)
946{
947 void *mem = req->ptr2;
948 size_t len = req->size;
949 int flags = req->int1;
950
951 eio_page_align (&mem, &len);
952
953 {
954 intptr_t addr = (intptr_t)mem;
955 intptr_t end = addr + len;
956 intptr_t page = eio_pagesize ();
957
958 if (addr < end)
959 if (flags & EIO_MT_MODIFY) /* modify */
960 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
961 else
962 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
963 }
964
965 return 0;
966}
967
968/*****************************************************************************/
969/* requests implemented outside eio_execute, because they are so large */
970
971static void
972eio__lseek (eio_req *req)
973{
974 /* this usually gets optimised away completely, or your compiler sucks, */
975 /* or the whence constants really are not 0, 1, 2 */
976 int whence = req->int2 == EIO_SEEK_SET ? SEEK_SET
977 : req->int2 == EIO_SEEK_CUR ? SEEK_CUR
978 : req->int2 == EIO_SEEK_END ? SEEK_END
979 : req->int2;
980
981 req->offs = lseek (req->int1, req->offs, whence);
982 req->result = req->offs == (off_t)-1 ? -1 : 0;
983}
984
985/* result will always end up in tmpbuf, there is always space for adding a 0-byte */
986static int
987eio__realpath (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
988{
989 char *res;
990 const char *rel = path;
991 char *tmp1, *tmp2;
992#if SYMLOOP_MAX > 32
993 int symlinks = SYMLOOP_MAX;
994#else
995 int symlinks = 32;
996#endif
997
998 errno = EINVAL;
999 if (!rel)
1000 return -1;
1001
1002 errno = ENOENT;
1003 if (!*rel)
1004 return -1;
1005
1006 res = etp_tmpbuf_get (tmpbuf, EIO_PATH_MAX * 3);
1007#ifdef _WIN32
1008 if (_access (rel, 4) != 0)
1009 return -1;
1010
1011 symlinks = GetFullPathName (rel, EIO_PATH_MAX * 3, res, 0);
1012
1013 errno = ENAMETOOLONG;
1014 if (symlinks >= EIO_PATH_MAX * 3)
1015 return -1;
1016
1017 errno = EIO;
1018 if (symlinks <= 0)
1019 return -1;
1020
1021 return symlinks;
1022
1023#else
1024 tmp1 = res + EIO_PATH_MAX;
1025 tmp2 = tmp1 + EIO_PATH_MAX;
1026
1027#if 0 /* disabled, the musl way to do things is just too racy */
1028#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1029 /* on linux we may be able to ask the kernel */
1030 {
1031 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1032
1033 if (fd >= 0)
1034 {
1035 sprintf (tmp1, "/proc/self/fd/%d", fd);
1036 req->result = readlink (tmp1, res, EIO_PATH_MAX);
1037 /* here we should probably stat the open file and the disk file, to make sure they still match */
1038 eio__close (fd);
1039
1040 if (req->result > 0)
1041 goto done;
1042 }
1043 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1044 return -1;
1045 }
1046#endif
1047#endif
1048
1049 if (*rel != '/')
1050 {
1051 int len;
1052
1053 errno = ENOENT;
1054 if (wd == EIO_INVALID_WD)
1055 return -1;
1056
1057 if (wd == EIO_CWD)
1058 {
1059 if (!getcwd (res, EIO_PATH_MAX))
1060 return -1;
1061
1062 len = strlen (res);
1063 }
1064 else
1065 memcpy (res, wd->str, len = wd->len);
1066
1067 if (res [1]) /* only use if not / */
1068 res += len;
1069 }
1070
1071 while (*rel)
1072 {
1073 eio_ssize_t len, linklen;
1074 const char *beg = rel;
1075
1076 while (*rel && *rel != '/')
1077 ++rel;
1078
1079 len = rel - beg;
1080
1081 if (!len) /* skip slashes */
1082 {
1083 ++rel;
1084 continue;
1085 }
1086
1087 if (beg [0] == '.')
1088 {
1089 if (len == 1)
1090 continue; /* . - nop */
1091
1092 if (beg [1] == '.' && len == 2)
1093 {
1094 /* .. - back up one component, if possible */
1095
1096 while (res != tmpbuf->ptr)
1097 if (*--res == '/')
1098 break;
1099
1100 continue;
1101 }
1102 }
1103
1104 errno = ENAMETOOLONG;
1105 if (res + 1 + len + 1 >= tmp1)
1106 return -1;
1107
1108 /* copy one component */
1109 *res = '/';
1110 memcpy (res + 1, beg, len);
1111
1112 /* zero-terminate, for readlink */
1113 res [len + 1] = 0;
1114
1115 /* now check if it's a symlink */
1116 linklen = readlink (tmpbuf->ptr, tmp1, EIO_PATH_MAX);
1117
1118 if (linklen < 0)
1119 {
1120 if (errno != EINVAL)
1121 return -1;
1122
1123 /* it's a normal directory. hopefully */
1124 res += len + 1;
1125 }
1126 else
1127 {
1128 /* yay, it was a symlink - build new path in tmp2 */
1129 int rellen = strlen (rel);
1130
1131 errno = ENAMETOOLONG;
1132 if (linklen + 1 + rellen >= EIO_PATH_MAX) /* also catch linklen >= EIO_PATH_MAX */
1133 return -1;
1134
1135 errno = ELOOP;
1136 if (!--symlinks)
1137 return -1;
1138
1139 if (*tmp1 == '/')
1140 res = tmpbuf->ptr; /* symlink resolves to an absolute path */
1141
1142 /* we need to be careful, as rel might point into tmp2 already */
1143 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1144 tmp2 [linklen] = '/';
1145 memcpy (tmp2, tmp1, linklen);
1146
1147 rel = tmp2;
1148 }
1149 }
1150
1151 /* special case for the lone root path */
1152 if (res == tmpbuf->ptr)
1153 *res++ = '/';
1154
1155 return res - (char *)tmpbuf->ptr;
1156#endif
1018} 1157}
1019 1158
1020static signed char 1159static signed char
1021eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) 1160eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1022{ 1161{
1023 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */ 1162 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1024 : a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0; 1163 : a->inode < b->inode ? -1
1164 : a->inode > b->inode ? 1
1165 : 0;
1025} 1166}
1026 1167
1027#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0 1168#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1028 1169
1029#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */ 1170#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1030#define EIO_SORT_FAST 60 /* when to only use insertion sort */ 1171#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1031 1172
1032static void 1173static void
1033eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1174eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1034{ 1175{
1035 unsigned char bits [9 + sizeof (ino_t) * 8]; 1176 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1036 unsigned char *bit = bits; 1177 unsigned char *bit = bits;
1037 1178
1038 assert (CHAR_BIT == 8); 1179 assert (CHAR_BIT == 8);
1039 assert (sizeof (eio_dirent) * 8 < 256); 1180 assert (sizeof (eio_dirent) * 8 < 256);
1040 assert (offsetof (eio_dirent, inode)); /* we use 0 as sentinel */ 1181 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1041 assert (offsetof (eio_dirent, score)); /* we use 0 as sentinel */ 1182 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1042 1183
1043 if (size <= EIO_SORT_FAST) 1184 if (size <= EIO_SORT_FAST)
1044 return; 1185 return;
1045 1186
1046 /* first prepare an array of bits to test in our radix sort */ 1187 /* first prepare an array of bits to test in our radix sort */
1047 /* try to take endianness into account, as well as differences in ino_t sizes */ 1188 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1048 /* inode_bits must contain all inodes ORed together */ 1189 /* inode_bits must contain all inodes ORed together */
1049 /* which is used to skip bits that are 0 everywhere, which is very common */ 1190 /* which is used to skip bits that are 0 everywhere, which is very common */
1050 { 1191 {
1051 ino_t endianness; 1192 eio_ino_t endianness;
1052 int i, j; 1193 int i, j;
1053 1194
1054 /* we store the byte offset of byte n into byte n of "endianness" */ 1195 /* we store the byte offset of byte n into byte n of "endianness" */
1055 for (i = 0; i < sizeof (ino_t); ++i) 1196 for (i = 0; i < sizeof (eio_ino_t); ++i)
1056 ((unsigned char *)&endianness)[i] = i; 1197 ((unsigned char *)&endianness)[i] = i;
1057 1198
1058 *bit++ = 0; 1199 *bit++ = 0;
1059 1200
1060 for (i = 0; i < sizeof (ino_t); ++i) 1201 for (i = 0; i < sizeof (eio_ino_t); ++i)
1061 { 1202 {
1062 /* shifting off the byte offsets out of "endianness" */ 1203 /* shifting off the byte offsets out of "endianness" */
1063 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8; 1204 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1064 endianness >>= 8; 1205 endianness >>= 8;
1065 1206
1066 for (j = 0; j < 8; ++j) 1207 for (j = 0; j < 8; ++j)
1067 if (inode_bits & (((ino_t)1) << (i * 8 + j))) 1208 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1068 *bit++ = offs + j; 1209 *bit++ = offs + j;
1069 } 1210 }
1070 1211
1071 for (j = 0; j < 8; ++j) 1212 for (j = 0; j < 8; ++j)
1072 if (score_bits & (1 << j)) 1213 if (score_bits & (1 << j))
1073 *bit++ = offsetof (eio_dirent, score) * 8 + j; 1214 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1074 } 1215 }
1075 1216
1076 /* now actually do the sorting (a variant of MSD radix sort) */ 1217 /* now actually do the sorting (a variant of MSD radix sort) */
1077 { 1218 {
1078 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base; 1219 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1079 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end; 1220 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1080 unsigned char *bit_stk [9 + sizeof (ino_t) * 8]; 1221 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1081 int stk_idx = 0; 1222 int stk_idx = 0;
1082 1223
1083 base_stk [stk_idx] = dents; 1224 base_stk [stk_idx] = dents;
1084 end_stk [stk_idx] = dents + size; 1225 end_stk [stk_idx] = dents + size;
1085 bit_stk [stk_idx] = bit - 1; 1226 bit_stk [stk_idx] = bit - 1;
… …
1164 } 1305 }
1165 } 1306 }
1166} 1307}
1167 1308
1168static void 1309static void
1169eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1310eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1170{ 1311{
1171 if (size <= 1) 1312 if (size <= 1)
1172 return; /* our insertion sort relies on size > 0 */ 1313 return; /* our insertion sort relies on size > 0 */
1173 1314
1174 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */ 1315 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
… …
1182 1323
1183/* read a full directory */ 1324/* read a full directory */
1184static void 1325static void
1185eio__scandir (eio_req *req, etp_worker *self) 1326eio__scandir (eio_req *req, etp_worker *self)
1186{ 1327{
1187 DIR *dirp;
1188 EIO_STRUCT_DIRENT *entp;
1189 char *name, *names; 1328 char *name, *names;
1190 int namesalloc = 4096; 1329 int namesalloc = 4096 - sizeof (void *) * 4;
1191 int namesoffs = 0; 1330 int namesoffs = 0;
1192 int flags = req->int1; 1331 int flags = req->int1;
1193 eio_dirent *dents = 0; 1332 eio_dirent *dents = 0;
1194 int dentalloc = 128; 1333 int dentalloc = 128;
1195 int dentoffs = 0; 1334 int dentoffs = 0;
1196 ino_t inode_bits = 0; 1335 eio_ino_t inode_bits = 0;
1336#ifdef _WIN32
1337 HANDLE dirp;
1338 WIN32_FIND_DATA entp;
1339#else
1340 DIR *dirp;
1341 EIO_STRUCT_DIRENT *entp;
1342#endif
1197 1343
1198 req->result = -1; 1344 req->result = -1;
1199 1345
1200 if (!(flags & EIO_READDIR_DENTS)) 1346 if (!(flags & EIO_READDIR_DENTS))
1201 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); 1347 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1202 1348
1203 X_LOCK (wrklock); 1349#ifdef _WIN32
1204 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 1350 {
1351 int len = strlen ((const char *)req->ptr1);
1352 char *path = malloc (MAX_PATH);
1353 const char *fmt;
1354 const char *reqpath = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1355
1356 if (!len)
1357 fmt = "./*";
1358 else if (reqpath[len - 1] == '/' || reqpath[len - 1] == '\\')
1359 fmt = "%s*";
1360 else
1361 fmt = "%s/*";
1362
1363 _snprintf (path, MAX_PATH, fmt, reqpath);
1364 dirp = FindFirstFile (path, &entp);
1365 free (path);
1366
1367 if (dirp == INVALID_HANDLE_VALUE)
1368 {
1369 /* should steal _dosmaperr */
1370 switch (GetLastError ())
1371 {
1372 case ERROR_FILE_NOT_FOUND:
1373 req->result = 0;
1374 break;
1375
1376 case ERROR_INVALID_NAME:
1377 case ERROR_PATH_NOT_FOUND:
1378 case ERROR_NO_MORE_FILES:
1379 errno = ENOENT;
1380 break;
1381
1382 case ERROR_NOT_ENOUGH_MEMORY:
1383 errno = ENOMEM;
1384 break;
1385
1386 default:
1387 errno = EINVAL;
1388 break;
1389 }
1390
1391 return;
1392 }
1393 }
1394#else
1395 #if HAVE_AT
1396 if (req->wd)
1397 {
1398 int fd = openat (WD2FD (req->wd), req->ptr1, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1399
1400 if (fd < 0)
1401 return;
1402
1403 dirp = fdopendir (fd);
1404
1405 if (!dirp)
1406 silent_close (fd);
1407 }
1408 else
1205 self->dirp = dirp = opendir (req->ptr1); 1409 dirp = opendir (req->ptr1);
1410 #else
1411 dirp = opendir (wd_expand (&self->tmpbuf, req->wd, req->ptr1));
1412 #endif
1413
1414 if (!dirp)
1415 return;
1416#endif
1417
1418 if (req->flags & EIO_FLAG_PTR1_FREE)
1419 free (req->ptr1);
1420
1206 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1421 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1207 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; 1422 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1208 req->ptr2 = names = malloc (namesalloc); 1423 req->ptr2 = names = malloc (namesalloc);
1209 X_UNLOCK (wrklock);
1210 1424
1211 if (dirp && names && (!flags || dents)) 1425 if (!names || (flags && !dents))
1426 return;
1427
1212 for (;;) 1428 for (;;)
1213 { 1429 {
1430 int done;
1431
1432#ifdef _WIN32
1433 done = !dirp;
1434#else
1214 errno = 0; 1435 errno = 0;
1215 entp = readdir (dirp); 1436 entp = readdir (dirp);
1437 done = !entp;
1438#endif
1216 1439
1217 if (!entp) 1440 if (done)
1218 { 1441 {
1442#ifndef _WIN32
1443 int old_errno = errno;
1444 closedir (dirp);
1445 errno = old_errno;
1446
1219 if (errno) 1447 if (errno)
1220 break; 1448 break;
1449#endif
1221 1450
1222 /* sort etc. */ 1451 /* sort etc. */
1223 req->int1 = flags; 1452 req->int1 = flags;
1224 req->result = dentoffs; 1453 req->result = dentoffs;
1225 1454
1226 if (flags & EIO_READDIR_STAT_ORDER) 1455 if (flags & EIO_READDIR_STAT_ORDER)
1227 eio_dent_sort (dents, dentoffs, 0, inode_bits); /* sort by inode exclusively */ 1456 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1228 else if (flags & EIO_READDIR_DIRS_FIRST) 1457 else if (flags & EIO_READDIR_DIRS_FIRST)
1229 if (flags & EIO_READDIR_FOUND_UNKNOWN) 1458 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1230 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */ 1459 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1231 else 1460 else
1461 {
1462 /* in this case, all is known, and we just put dirs first and sort them */
1463 eio_dirent *oth = dents + dentoffs;
1464 eio_dirent *dir = dents;
1465
1466 /* now partition dirs to the front, and non-dirs to the back */
1467 /* by walking from both sides and swapping if necessary */
1468 while (oth > dir)
1469 {
1470 if (dir->type == EIO_DT_DIR)
1471 ++dir;
1472 else if ((--oth)->type == EIO_DT_DIR)
1473 {
1474 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1475
1476 ++dir;
1477 }
1478 }
1479
1480 /* now sort the dirs only (dirs all have the same score) */
1481 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1482 }
1483
1484 break;
1485 }
1486
1487 /* now add the entry to our list(s) */
1488 name = D_NAME (entp);
1489
1490 /* skip . and .. entries */
1491 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1492 {
1493 int len = D_NAMLEN (entp) + 1;
1494
1495 while (ecb_expect_false (namesoffs + len > namesalloc))
1496 {
1497 namesalloc *= 2;
1498 req->ptr2 = names = realloc (names, namesalloc);
1499
1500 if (!names)
1501 break;
1502 }
1503
1504 memcpy (names + namesoffs, name, len);
1505
1506 if (dents)
1507 {
1508 struct eio_dirent *ent;
1509
1510 if (ecb_expect_false (dentoffs == dentalloc))
1232 { 1511 {
1233 /* in this case, all is known, and we just put dirs first and sort them */ 1512 dentalloc *= 2;
1513 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1514
1515 if (!dents)
1516 break;
1517 }
1518
1234 eio_dirent *oth = dents + dentoffs; 1519 ent = dents + dentoffs;
1235 eio_dirent *dir = dents;
1236 1520
1237 /* now partition dirs to the front, and non-dirs to the back */ 1521 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1238 /* by walking from both sides and swapping if necessary */ 1522 ent->namelen = len - 1;
1239 /* also clear score, so it doesn't influence sorting */ 1523 ent->inode = D_INO (entp);
1240 while (oth > dir) 1524
1525 inode_bits |= ent->inode;
1526
1527 switch (D_TYPE (entp))
1528 {
1529 default:
1530 ent->type = EIO_DT_UNKNOWN;
1531 flags |= EIO_READDIR_FOUND_UNKNOWN;
1532 break;
1533
1534 #ifdef DT_FIFO
1535 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1536 #endif
1537 #ifdef DT_CHR
1538 case DT_CHR: ent->type = EIO_DT_CHR; break;
1539 #endif
1540 #ifdef DT_MPC
1541 case DT_MPC: ent->type = EIO_DT_MPC; break;
1542 #endif
1543 #ifdef DT_DIR
1544 case DT_DIR: ent->type = EIO_DT_DIR; break;
1545 #endif
1546 #ifdef DT_NAM
1547 case DT_NAM: ent->type = EIO_DT_NAM; break;
1548 #endif
1549 #ifdef DT_BLK
1550 case DT_BLK: ent->type = EIO_DT_BLK; break;
1551 #endif
1552 #ifdef DT_MPB
1553 case DT_MPB: ent->type = EIO_DT_MPB; break;
1554 #endif
1555 #ifdef DT_REG
1556 case DT_REG: ent->type = EIO_DT_REG; break;
1557 #endif
1558 #ifdef DT_NWK
1559 case DT_NWK: ent->type = EIO_DT_NWK; break;
1560 #endif
1561 #ifdef DT_CMP
1562 case DT_CMP: ent->type = EIO_DT_CMP; break;
1563 #endif
1564 #ifdef DT_LNK
1565 case DT_LNK: ent->type = EIO_DT_LNK; break;
1566 #endif
1567 #ifdef DT_SOCK
1568 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1569 #endif
1570 #ifdef DT_DOOR
1571 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1572 #endif
1573 #ifdef DT_WHT
1574 case DT_WHT: ent->type = EIO_DT_WHT; break;
1575 #endif
1576 }
1577
1578 ent->score = 7;
1579
1580 if (flags & EIO_READDIR_DIRS_FIRST)
1581 {
1582 if (ent->type == EIO_DT_UNKNOWN)
1241 { 1583 {
1242 if (dir->type == EIO_DT_DIR) 1584 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1243 ++dir; 1585 ent->score = 1;
1244 else if ((--oth)->type == EIO_DT_DIR) 1586 else if (!strchr (name, '.')) /* absence of dots indicate likely dirs */
1245 { 1587 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1246 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1247
1248 ++dir;
1249 }
1250 } 1588 }
1251 1589 else if (ent->type == EIO_DT_DIR)
1252 /* now sort the dirs only */ 1590 ent->score = 0;
1253 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1254 } 1591 }
1255
1256 break;
1257 }
1258
1259 /* now add the entry to our list(s) */
1260 name = entp->d_name;
1261
1262 /* skip . and .. entries */
1263 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1264 {
1265 int len = D_NAMLEN (entp) + 1;
1266
1267 while (expect_false (namesoffs + len > namesalloc))
1268 {
1269 namesalloc *= 2;
1270 X_LOCK (wrklock);
1271 req->ptr2 = names = realloc (names, namesalloc);
1272 X_UNLOCK (wrklock);
1273
1274 if (!names)
1275 break;
1276 } 1592 }
1277 1593
1278 memcpy (names + namesoffs, name, len);
1279
1280 if (dents)
1281 {
1282 struct eio_dirent *ent;
1283
1284 if (expect_false (dentoffs == dentalloc))
1285 {
1286 dentalloc *= 2;
1287 X_LOCK (wrklock);
1288 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1289 X_UNLOCK (wrklock);
1290
1291 if (!dents)
1292 break;
1293 }
1294
1295 ent = dents + dentoffs;
1296
1297 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1298 ent->namelen = len - 1;
1299 ent->inode = D_INO (entp);
1300
1301 inode_bits |= ent->inode;
1302
1303 switch (D_TYPE (entp))
1304 {
1305 default:
1306 ent->type = EIO_DT_UNKNOWN;
1307 flags |= EIO_READDIR_FOUND_UNKNOWN;
1308 break;
1309
1310 #ifdef DT_FIFO
1311 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1312 #endif
1313 #ifdef DT_CHR
1314 case DT_CHR: ent->type = EIO_DT_CHR; break;
1315 #endif
1316 #ifdef DT_MPC
1317 case DT_MPC: ent->type = EIO_DT_MPC; break;
1318 #endif
1319 #ifdef DT_DIR
1320 case DT_DIR: ent->type = EIO_DT_DIR; break;
1321 #endif
1322 #ifdef DT_NAM
1323 case DT_NAM: ent->type = EIO_DT_NAM; break;
1324 #endif
1325 #ifdef DT_BLK
1326 case DT_BLK: ent->type = EIO_DT_BLK; break;
1327 #endif
1328 #ifdef DT_MPB
1329 case DT_MPB: ent->type = EIO_DT_MPB; break;
1330 #endif
1331 #ifdef DT_REG
1332 case DT_REG: ent->type = EIO_DT_REG; break;
1333 #endif
1334 #ifdef DT_NWK
1335 case DT_NWK: ent->type = EIO_DT_NWK; break;
1336 #endif
1337 #ifdef DT_CMP
1338 case DT_CMP: ent->type = EIO_DT_CMP; break;
1339 #endif
1340 #ifdef DT_LNK
1341 case DT_LNK: ent->type = EIO_DT_LNK; break;
1342 #endif
1343 #ifdef DT_SOCK
1344 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1345 #endif
1346 #ifdef DT_DOOR
1347 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1348 #endif
1349 #ifdef DT_WHT
1350 case DT_WHT: ent->type = EIO_DT_WHT; break;
1351 #endif
1352 }
1353
1354 ent->score = 7;
1355
1356 if (flags & EIO_READDIR_DIRS_FIRST)
1357 {
1358 if (ent->type == EIO_DT_UNKNOWN)
1359 {
1360 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1361 ent->score = 1;
1362 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1363 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1364 }
1365 else if (ent->type == EIO_DT_DIR)
1366 ent->score = 0;
1367 }
1368 }
1369
1370 namesoffs += len; 1594 namesoffs += len;
1371 ++dentoffs; 1595 ++dentoffs;
1372 } 1596 }
1373 1597
1374 if (EIO_CANCELLED (req)) 1598 if (EIO_CANCELLED (req))
1375 { 1599 {
1376 errno = ECANCELED; 1600 errno = ECANCELED;
1377 break; 1601 break;
1378 } 1602 }
1603
1604#ifdef _WIN32
1605 if (!FindNextFile (dirp, &entp))
1606 {
1607 FindClose (dirp);
1608 dirp = 0;
1379 } 1609 }
1610#endif
1611 }
1380} 1612}
1381 1613
1382#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) 1614/*****************************************************************************/
1383# define eio__msync(a,b,c) ((errno = ENOSYS), -1) 1615/* working directory stuff */
1616/* various deficiencies in the posix 2008 api force us to */
1617/* keep the absolute path in string form at all times */
1618/* fuck yeah. */
1619
1620#if !HAVE_AT
1621
1622/* a bit like realpath, but usually faster because it doesn'T have to return */
1623/* an absolute or canonical path */
1624static const char *
1625wd_expand (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
1626{
1627 if (!wd || *path == '/')
1628 return path;
1629
1630 if (path [0] == '.' && !path [1])
1631 return wd->str;
1632
1633 {
1634 int l1 = wd->len;
1635 int l2 = strlen (path);
1636
1637 char *res = etp_tmpbuf_get (tmpbuf, l1 + l2 + 2);
1638
1639 memcpy (res, wd->str, l1);
1640 res [l1] = '/';
1641 memcpy (res + l1 + 1, path, l2 + 1);
1642
1643 return res;
1644 }
1645}
1646
1647#endif
1648
1649static eio_wd
1650eio__wd_open_sync (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
1651{
1652 int fd;
1653 eio_wd res;
1654 int len = eio__realpath (tmpbuf, wd, path);
1655
1656 if (len < 0)
1657 return EIO_INVALID_WD;
1658
1659#if HAVE_AT
1660 fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1661
1662 if (fd < 0)
1663 return EIO_INVALID_WD;
1664#endif
1665
1666 res = malloc (sizeof (*res) + len); /* one extra 0-byte */
1667
1668#if HAVE_AT
1669 res->fd = fd;
1670#endif
1671
1672 res->len = len;
1673 memcpy (res->str, tmpbuf->ptr, len);
1674 res->str [len] = 0;
1675
1676 return res;
1677}
1678
1679eio_wd
1680eio_wd_open_sync (eio_wd wd, const char *path)
1681{
1682 struct etp_tmpbuf tmpbuf = { };
1683 wd = eio__wd_open_sync (&tmpbuf, wd, path);
1684 free (tmpbuf.ptr);
1685
1686 return wd;
1687}
1688
1689void
1690eio_wd_close_sync (eio_wd wd)
1691{
1692 if (wd != EIO_INVALID_WD && wd != EIO_CWD)
1693 {
1694 #if HAVE_AT
1695 eio__close (wd->fd);
1696 #endif
1697 free (wd);
1698 }
1699}
1700
1701#if HAVE_AT
1702
1703static int
1704eio__renameat2 (int olddirfd, const char *oldpath, int newdirfd, const char *newpath, unsigned int flags)
1705{
1706#if HAVE_RENAMEAT2
1707 return syscall (SYS_renameat2, olddirfd, oldpath, newdirfd, newpath, flags);
1384#else 1708#else
1385
1386int
1387eio__msync (void *mem, size_t len, int flags)
1388{
1389 if (EIO_MS_ASYNC != MS_SYNC
1390 || EIO_MS_INVALIDATE != MS_INVALIDATE
1391 || EIO_MS_SYNC != MS_SYNC)
1392 {
1393 flags = 0
1394 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1395 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1396 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1397 }
1398
1399 return msync (mem, len, flags);
1400}
1401
1402#endif
1403
1404int
1405eio__mtouch (void *mem, size_t len, int flags)
1406{
1407 intptr_t addr = (intptr_t)mem;
1408 intptr_t end = addr + len;
1409#ifdef PAGESIZE
1410 const intptr_t page = PAGESIZE;
1411#else
1412 static intptr_t page;
1413
1414 if (!page) 1709 if (flags)
1415 page = sysconf (_SC_PAGESIZE); 1710 return EIO_ENOSYS ();
1416#endif
1417 1711
1418 /* round down to start of page, although this is probably useless */ 1712 return renameat (olddirfd, oldpath, newdirfd, newpath);
1419 addr &= ~(page - 1); /* assume page size is always a power of two */ 1713#endif
1714}
1420 1715
1421 if (addr < end) 1716/* they forgot these */
1422 if (flags & EIO_MT_MODIFY) /* modify */
1423 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
1424 else
1425 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
1426 1717
1718static int
1719eio__truncateat (int dirfd, const char *path, off_t length)
1720{
1721 int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
1722 int res;
1723
1724 if (fd < 0)
1725 return fd;
1726
1727 res = ftruncate (fd, length);
1728 silent_close (fd);
1427 return 0; 1729 return res;
1428} 1730}
1731
1732static int
1733eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
1734{
1735 int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
1736 int res;
1737
1738 if (fd < 0)
1739 return fd;
1740
1741 res = fstatvfs (fd, buf);
1742 silent_close (fd);
1743 return res;
1744}
1745
1746#endif
1429 1747
1430/*****************************************************************************/ 1748/*****************************************************************************/
1431 1749
1432#define ALLOC(len) \ 1750#define ALLOC(len) \
1433 if (!req->ptr2) \ 1751 if (!req->ptr2) \
1434 { \ 1752 { \
1435 X_LOCK (wrklock); \ 1753 X_LOCK (EIO_POOL->wrklock); \
1436 req->flags |= EIO_FLAG_PTR2_FREE; \ 1754 req->flags |= EIO_FLAG_PTR2_FREE; \
1437 X_UNLOCK (wrklock); \ 1755 X_UNLOCK (EIO_POOL->wrklock); \
1438 req->ptr2 = malloc (len); \ 1756 req->ptr2 = malloc (len); \
1439 if (!req->ptr2) \ 1757 if (!req->ptr2) \
1440 { \ 1758 { \
1441 errno = ENOMEM; \ 1759 errno = ENOMEM; \
1442 req->result = -1; \ 1760 req->result = -1; \
1443 break; \ 1761 goto alloc_fail; \
1444 } \ 1762 } \
1445 } 1763 }
1446 1764
1447X_THREAD_PROC (etp_proc) 1765/*****************************************************************************/
1448{
1449 ETP_REQ *req;
1450 struct timespec ts;
1451 etp_worker *self = (etp_worker *)thr_arg;
1452 1766
1453 /* try to distribute timeouts somewhat randomly */ 1767static void
1454 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 1768eio__slurp (int fd, eio_req *req)
1769{
1770 req->result = fd;
1455 1771
1456 for (;;) 1772 if (fd < 0)
1773 return;
1774
1775 if (req->offs < 0 || !req->size) /* do we need the size? */
1457 { 1776 {
1458 X_LOCK (reqlock); 1777 off_t size = lseek (fd, 0, SEEK_END);
1459 1778
1460 for (;;)
1461 {
1462 self->req = req = reqq_shift (&req_queue);
1463
1464 if (req)
1465 break;
1466
1467 ++idle;
1468
1469 ts.tv_sec = time (0) + IDLE_TIMEOUT;
1470 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
1471 {
1472 if (idle > max_idle)
1473 {
1474 --idle;
1475 X_UNLOCK (reqlock);
1476 X_LOCK (wrklock);
1477 --started;
1478 X_UNLOCK (wrklock);
1479 goto quit;
1480 }
1481
1482 /* we are allowed to idle, so do so without any timeout */
1483 X_COND_WAIT (reqwait, reqlock);
1484 }
1485
1486 --idle;
1487 }
1488
1489 --nready;
1490
1491 X_UNLOCK (reqlock);
1492
1493 if (req->type < 0) 1779 if (req->offs < 0)
1494 goto quit; 1780 req->offs += size;
1495 1781
1496 if (!EIO_CANCELLED (req)) 1782 if (!req->size)
1497 ETP_EXECUTE (self, req); 1783 req->size = size - req->offs;
1498
1499 X_LOCK (reslock);
1500
1501 ++npending;
1502
1503 if (!reqq_push (&res_queue, req) && want_poll_cb)
1504 want_poll_cb ();
1505
1506 self->req = 0;
1507 etp_worker_clear (self);
1508
1509 X_UNLOCK (reslock);
1510 } 1784 }
1511 1785
1512quit: 1786 ALLOC (req->size);
1513 X_LOCK (wrklock); 1787 req->result = pread (fd, req->ptr2, req->size, req->offs);
1514 etp_worker_free (self);
1515 X_UNLOCK (wrklock);
1516 1788
1517 return 0; 1789 silent_close (fd);
1518}
1519 1790
1520/*****************************************************************************/ 1791alloc_fail:
1792 ;
1793}
1521 1794
1795int ecb_cold
1522int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1796eio_init (void (*want_poll)(void), void (*done_poll)(void))
1523{ 1797{
1524 return etp_init (want_poll, done_poll); 1798 eio_want_poll_cb = want_poll;
1525} 1799 eio_done_poll_cb = done_poll;
1526 1800
1801 return etp_init (EIO_POOL, 0, 0, 0);
1802}
1803
1804ecb_inline void
1527static void eio_api_destroy (eio_req *req) 1805eio_api_destroy (eio_req *req)
1528{ 1806{
1529 free (req); 1807 free (req);
1530} 1808}
1531 1809
1532#define REQ(rtype) \ 1810#define REQ(rtype) \
1533 eio_req *req; \ 1811 eio_req *req; \
1534 \ 1812 \
1535 req = (eio_req *)calloc (1, sizeof *req); \ 1813 req = (eio_req *)calloc (1, sizeof *req); \
1536 if (!req) \ 1814 if (!req) \
1537 return 0; \ 1815 return 0; \
… …
1551 { \ 1829 { \
1552 eio_api_destroy (req); \ 1830 eio_api_destroy (req); \
1553 return 0; \ 1831 return 0; \
1554 } 1832 }
1555 1833
1834#define SINGLEDOT(ptr) (0[(char *)(ptr)] == '.' && !1[(char *)(ptr)])
1835
1836static void
1556static void eio_execute (etp_worker *self, eio_req *req) 1837eio_execute (etp_worker *self, eio_req *req)
1557{ 1838{
1558 errno = 0; 1839#if HAVE_AT
1840 int dirfd;
1841#else
1842 const char *path;
1843#endif
1844
1845 if (ecb_expect_false (EIO_CANCELLED (req)))
1846 {
1847 req->result = -1;
1848 req->errorno = ECANCELED;
1849 return;
1850 }
1851
1852 if (ecb_expect_false (req->wd == EIO_INVALID_WD))
1853 {
1854 req->result = -1;
1855 req->errorno = ENOENT;
1856 return;
1857 }
1858
1859 if (req->type >= EIO_OPEN)
1860 {
1861 #if HAVE_AT
1862 dirfd = WD2FD (req->wd);
1863 #else
1864 path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1865 #endif
1866 }
1559 1867
1560 switch (req->type) 1868 switch (req->type)
1561 { 1869 {
1870 case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
1871 req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
1872 break;
1873 case EIO_WD_CLOSE: req->result = 0;
1874 eio_wd_close_sync (req->wd); break;
1875
1876 case EIO_SEEK: eio__lseek (req); break;
1562 case EIO_READ: ALLOC (req->size); 1877 case EIO_READ: ALLOC (req->size);
1563 req->result = req->offs >= 0 1878 req->result = req->offs >= 0
1564 ? pread (req->int1, req->ptr2, req->size, req->offs) 1879 ? pread (req->int1, req->ptr2, req->size, req->offs)
1565 : read (req->int1, req->ptr2, req->size); break; 1880 : read (req->int1, req->ptr2, req->size); break;
1566 case EIO_WRITE: req->result = req->offs >= 0 1881 case EIO_WRITE: req->result = req->offs >= 0
1567 ? pwrite (req->int1, req->ptr2, req->size, req->offs) 1882 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1568 : write (req->int1, req->ptr2, req->size); break; 1883 : write (req->int1, req->ptr2, req->size); break;
1569 1884
1885 case EIO_FCNTL: req->result = fcntl (req->int1, (int) req->int2, req->ptr2); break;
1886 case EIO_IOCTL: req->result = ioctl (req->int1, (unsigned long)req->int2, req->ptr2); break;
1887
1570 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break; 1888 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1571 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break; 1889 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
1890
1891#if HAVE_AT
1572 1892
1573 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1893 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1574 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 1894 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break;
1575 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1895 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1576 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 1896 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, AT_SYMLINK_NOFOLLOW); break;
1577 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1578 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1579
1580 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; 1897 case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break;
1581 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1582 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break; 1898 case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break;
1583 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1584 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break; 1899 case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break;
1585 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1586
1587 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break; 1900 case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break;
1588 case EIO_CLOSE: req->result = close (req->int1); break; 1901 case EIO_SLURP: eio__slurp ( openat (dirfd, req->ptr1, O_RDONLY | O_CLOEXEC), req); break;
1589 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break; 1902
1590 case EIO_UNLINK: req->result = unlink (req->ptr1); break; 1903 case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
1591 case EIO_RMDIR: req->result = rmdir (req->ptr1); break; 1904 case EIO_RMDIR: /* complications arise because "." cannot be removed, so we might have to expand */
1905 req->result = req->wd && SINGLEDOT (req->ptr1)
1906 ? rmdir (req->wd->str)
1907 : unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break;
1592 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; 1908 case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); break;
1593 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; 1909 case EIO_RENAME: req->result = eio__renameat2 (
1910 dirfd,
1911 /* complications arise because "." cannot be renamed, so we might have to expand */
1912 req->wd && SINGLEDOT (req->ptr1) ? req->wd->str : req->ptr1,
1913 WD2FD ((eio_wd)req->int3),
1914 req->ptr2,
1915 req->int2
1916 );
1917 break;
1594 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; 1918 case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2, 0); break;
1595 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; 1919 case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
1596 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break; 1920 case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1597 1921 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1922 req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1598 case EIO_READLINK: ALLOC (NAME_MAX); 1923 case EIO_READLINK: ALLOC (EIO_PATH_MAX);
1599 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; 1924 req->result = readlinkat (dirfd, req->ptr1, req->ptr2, EIO_PATH_MAX);
1600 1925 if (req->result == EIO_PATH_MAX)
1601 case EIO_SYNC: req->result = 0; sync (); break; 1926 {
1602 case EIO_FSYNC: req->result = fsync (req->int1); break; 1927 req->result = -1;
1603 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 1928 errno = ENAMETOOLONG;
1604 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break; 1929 }
1605 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break; 1930 break;
1606 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1607
1608 case EIO_READDIR: eio__scandir (req, self); break;
1609
1610 case EIO_BUSY: 1931 case EIO_UTIME:
1611#ifdef _WIN32 1932 case EIO_FUTIME:
1612 Sleep (req->nv1 * 1000.); 1933 {
1934 struct timespec ts[2];
1935 struct timespec *times;
1936
1937 if (req->nv1 != -1. || req->nv2 != -1.)
1938 {
1939 ts[0].tv_sec = req->nv1;
1940 ts[0].tv_nsec = (req->nv1 - ts[0].tv_sec) * 1e9;
1941 ts[1].tv_sec = req->nv2;
1942 ts[1].tv_nsec = (req->nv2 - ts[1].tv_sec) * 1e9;
1943
1944 times = ts;
1945 }
1946 else
1947 times = 0;
1948
1949 req->result = req->type == EIO_FUTIME
1950 ? futimens (req->int1, times)
1951 : utimensat (dirfd, req->ptr1, times, 0);
1952 }
1953 break;
1954
1613#else 1955#else
1614 {
1615 struct timeval tv;
1616 1956
1617 tv.tv_sec = req->nv1; 1957 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1618 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; 1958 req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
1959 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1960 req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
1961 case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
1962 case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
1963 case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
1964 case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
1965 case EIO_SLURP: eio__slurp ( open (path , O_RDONLY | O_CLOEXEC), req); break;
1619 1966
1620 req->result = select (0, 0, 0, 0, &tv); 1967 case EIO_UNLINK: req->result = unlink (path ); break;
1621 } 1968 case EIO_RMDIR: req->result = rmdir (path ); break;
1622#endif 1969 case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
1623 break; 1970 case EIO_RENAME: req->result = req->int2 ? EIO_ENOSYS () : rename (path, req->ptr2); break;
1971 case EIO_LINK: req->result = link (path , req->ptr2); break;
1972 case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
1973 case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
1974 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1975 req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
1976 case EIO_READLINK: ALLOC (EIO_PATH_MAX);
1977 req->result = readlink (path, req->ptr2, EIO_PATH_MAX);
1978 if (req->result == EIO_PATH_MAX)
1979 {
1980 req->result = -1;
1981 errno = ENAMETOOLONG;
1982 }
1983 break;
1624 1984
1625 case EIO_UTIME: 1985 case EIO_UTIME:
1626 case EIO_FUTIME: 1986 case EIO_FUTIME:
1627 { 1987 {
1628 struct timeval tv[2]; 1988 struct timeval tv[2];
1629 struct timeval *times; 1989 struct timeval *times;
1630 1990
1631 if (req->nv1 != -1. || req->nv2 != -1.) 1991 if (req->nv1 != -1. || req->nv2 != -1.)
1632 { 1992 {
1633 tv[0].tv_sec = req->nv1; 1993 tv[0].tv_sec = req->nv1;
1634 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.; 1994 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
1635 tv[1].tv_sec = req->nv2; 1995 tv[1].tv_sec = req->nv2;
1636 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.; 1996 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
1637 1997
1638 times = tv; 1998 times = tv;
1639 } 1999 }
1640 else 2000 else
1641 times = 0; 2001 times = 0;
1642
1643 2002
1644 req->result = req->type == EIO_FUTIME 2003 req->result = req->type == EIO_FUTIME
1645 ? futimes (req->int1, times) 2004 ? futimes (req->int1, times)
1646 : utimes (req->ptr1, times); 2005 : utimes (req->ptr1, times);
1647 } 2006 }
1648 break; 2007 break;
1649 2008
2009#endif
2010
2011 case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
2012 {
2013 ALLOC (req->result);
2014 memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
2015 }
2016 break;
2017
2018 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2019 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
2020
2021 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2022 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2023
2024 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
2025 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
2026 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
2027
2028 case EIO_CLOSE: req->result = eio__close (req->int1); break;
2029 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
2030 case EIO_SYNC: req->result = 0; sync (); break;
2031 case EIO_FSYNC: req->result = fsync (req->int1); break;
2032 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
2033 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
2034 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
2035 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
2036 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
2037 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
2038 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
2039 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
2040
2041 case EIO_READDIR: eio__scandir (req, self); break;
2042
2043 case EIO_BUSY:
2044#ifdef _WIN32
2045 Sleep (req->nv1 * 1e3);
2046#else
2047 {
2048 struct timeval tv;
2049
2050 tv.tv_sec = req->nv1;
2051 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
2052
2053 req->result = select (0, 0, 0, 0, &tv);
2054 }
2055#endif
2056 break;
2057
2058#if 0
1650 case EIO_GROUP: 2059 case EIO_GROUP:
1651 abort (); /* handled in eio_request */ 2060 abort (); /* handled in eio_request */
2061#endif
1652 2062
1653 case EIO_NOP: 2063 case EIO_NOP:
1654 req->result = 0; 2064 req->result = 0;
1655 break; 2065 break;
1656 2066
1657 case EIO_CUSTOM: 2067 case EIO_CUSTOM:
1658 ((void (*)(eio_req *))req->feed) (req); 2068 req->feed (req);
1659 break; 2069 break;
1660 2070
1661 default: 2071 default:
1662 req->result = -1; 2072 req->result = EIO_ENOSYS ();
1663 break; 2073 break;
1664 } 2074 }
1665 2075
2076alloc_fail:
1666 req->errorno = errno; 2077 req->errorno = errno;
1667} 2078}
1668 2079
1669#ifndef EIO_NO_WRAPPERS 2080#ifndef EIO_NO_WRAPPERS
1670 2081
2082eio_req *eio_wd_open (const char *path, int pri, eio_cb cb, void *data)
2083{
2084 REQ (EIO_WD_OPEN); PATH; SEND;
2085}
2086
2087eio_req *eio_wd_close (eio_wd wd, int pri, eio_cb cb, void *data)
2088{
2089 REQ (EIO_WD_CLOSE); req->wd = wd; SEND;
2090}
2091
1671eio_req *eio_nop (int pri, eio_cb cb, void *data) 2092eio_req *eio_nop (int pri, eio_cb cb, void *data)
1672{ 2093{
1673 REQ (EIO_NOP); SEND; 2094 REQ (EIO_NOP); SEND;
1674} 2095}
1675 2096
… …
1691eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2112eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1692{ 2113{
1693 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2114 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1694} 2115}
1695 2116
2117eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2118{
2119 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2120}
2121
2122eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
2123{
2124 REQ (EIO_SYNCFS); req->int1 = fd; SEND;
2125}
2126
2127eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2128{
2129 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2130}
2131
1696eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2132eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1697{ 2133{
1698 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2134 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1699} 2135}
1700 2136
1701eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) 2137eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
1702{ 2138{
1703 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; 2139 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
1704} 2140}
1705 2141
1706eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) 2142eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
1707{ 2143{
1708 REQ (EIO_FDATASYNC); req->int1 = fd; SEND; 2144 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2145}
2146
2147eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2148{
2149 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
1709} 2150}
1710 2151
1711eio_req *eio_close (int fd, int pri, eio_cb cb, void *data) 2152eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
1712{ 2153{
1713 REQ (EIO_CLOSE); req->int1 = fd; SEND; 2154 REQ (EIO_CLOSE); req->int1 = fd; SEND;
… …
1716eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data) 2157eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
1717{ 2158{
1718 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND; 2159 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
1719} 2160}
1720 2161
2162eio_req *eio_seek (int fd, off_t offset, int whence, int pri, eio_cb cb, void *data)
2163{
2164 REQ (EIO_SEEK); req->int1 = fd; req->offs = offset; req->int2 = whence; SEND;
2165}
2166
1721eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data) 2167eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1722{ 2168{
1723 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND; 2169 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1724} 2170}
1725 2171
1726eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data) 2172eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1727{ 2173{
1728 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND; 2174 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1729} 2175}
1730 2176
2177eio_req *eio_fcntl (int fd, int cmd, void *arg, int pri, eio_cb cb, void *data)
2178{
2179 REQ (EIO_IOCTL); req->int1 = fd; req->int2 = cmd; req->ptr2 = arg; SEND;
2180}
2181
2182eio_req *eio_ioctl (int fd, unsigned long request, void *buf, int pri, eio_cb cb, void *data)
2183{
2184 REQ (EIO_IOCTL); req->int1 = fd; req->int2 = request; req->ptr2 = buf; SEND;
2185}
2186
1731eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) 2187eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1732{ 2188{
1733 REQ (EIO_FSTAT); req->int1 = fd; SEND; 2189 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1734} 2190}
1735 2191
2192eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
2193{
2194 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
2195}
2196
1736eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) 2197eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1737{ 2198{
1738 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; 2199 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1739} 2200}
1740 2201
… …
1746eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) 2207eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1747{ 2208{
1748 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2209 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1749} 2210}
1750 2211
1751eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2212eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1752{ 2213{
1753 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2214 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1754} 2215}
1755 2216
1756eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) 2217eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
… …
1776eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data) 2237eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1777{ 2238{
1778 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2239 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1779} 2240}
1780 2241
1781eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2242eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1782{ 2243{
1783 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2244 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1784} 2245}
1785 2246
1786eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) 2247eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
… …
1802eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data) 2263eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1803{ 2264{
1804 return eio__1path (EIO_READLINK, path, pri, cb, data); 2265 return eio__1path (EIO_READLINK, path, pri, cb, data);
1805} 2266}
1806 2267
2268eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2269{
2270 return eio__1path (EIO_REALPATH, path, pri, cb, data);
2271}
2272
1807eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) 2273eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1808{ 2274{
1809 return eio__1path (EIO_STAT, path, pri, cb, data); 2275 return eio__1path (EIO_STAT, path, pri, cb, data);
1810} 2276}
1811 2277
1812eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) 2278eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1813{ 2279{
1814 return eio__1path (EIO_LSTAT, path, pri, cb, data); 2280 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1815} 2281}
1816 2282
2283eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
2284{
2285 return eio__1path (EIO_STATVFS, path, pri, cb, data);
2286}
2287
1817eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) 2288eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1818{ 2289{
1819 return eio__1path (EIO_UNLINK, path, pri, cb, data); 2290 return eio__1path (EIO_UNLINK, path, pri, cb, data);
1820} 2291}
1821 2292
… …
1829 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND; 2300 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1830} 2301}
1831 2302
1832eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) 2303eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1833{ 2304{
1834 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; 2305 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1835} 2306}
1836 2307
1837static eio_req * 2308static eio_req *
1838eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2309eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1839{ 2310{
… …
1863eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2334eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1864{ 2335{
1865 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2336 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1866} 2337}
1867 2338
2339eio_req *eio_slurp (const char *path, void *buf, ssize_t length, off_t offset, int pri, eio_cb cb, void *data)
2340{
2341 REQ (EIO_SLURP); PATH; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
2342}
2343
1868eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) 2344eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
1869{ 2345{
1870 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND; 2346 REQ (EIO_CUSTOM); req->feed = execute; SEND;
1871} 2347}
1872 2348
1873#endif 2349#endif
1874 2350
1875eio_req *eio_grp (eio_cb cb, void *data) 2351eio_req *eio_grp (eio_cb cb, void *data)
… …
1884#undef SEND 2360#undef SEND
1885 2361
1886/*****************************************************************************/ 2362/*****************************************************************************/
1887/* grp functions */ 2363/* grp functions */
1888 2364
2365void
1889void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2366eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1890{ 2367{
1891 grp->int2 = limit; 2368 grp->int2 = limit;
1892 grp->feed = feed; 2369 grp->feed = feed;
1893 2370
1894 grp_try_feed (grp); 2371 grp_try_feed (grp);
1895} 2372}
1896 2373
2374void
1897void eio_grp_limit (eio_req *grp, int limit) 2375eio_grp_limit (eio_req *grp, int limit)
1898{ 2376{
1899 grp->int2 = limit; 2377 grp->int2 = limit;
1900 2378
1901 grp_try_feed (grp); 2379 grp_try_feed (grp);
1902} 2380}
1903 2381
2382void
1904void eio_grp_add (eio_req *grp, eio_req *req) 2383eio_grp_add (eio_req *grp, eio_req *req)
1905{ 2384{
1906 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2385 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1907 2386
1908 grp->flags |= EIO_FLAG_GROUPADD; 2387 grp->flags |= ETP_FLAG_GROUPADD;
1909 2388
1910 ++grp->size; 2389 ++grp->size;
1911 req->grp = grp; 2390 req->grp = grp;
1912 2391
1913 req->grp_prev = 0; 2392 req->grp_prev = 0;
… …
1920} 2399}
1921 2400
1922/*****************************************************************************/ 2401/*****************************************************************************/
1923/* misc garbage */ 2402/* misc garbage */
1924 2403
2404eio_ssize_t
1925ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2405eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1926{ 2406{
1927 etp_worker wrk;
1928 ssize_t ret;
1929
1930 wrk.dbuf = 0;
1931
1932 ret = eio__sendfile (ofd, ifd, offset, count, &wrk); 2407 return eio__sendfile (ofd, ifd, offset, count);
1933
1934 if (wrk.dbuf)
1935 free (wrk.dbuf);
1936
1937 return ret;
1938} 2408}
1939 2409

Diff Legend

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