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 1.148 by root, Tue Feb 20 06:05:19 2018 UTC

1/* 1/*
2 * libeio implementation 2 * libeio implementation
3 * 3 *
4 * Copyright (c) 2007,2008,2009,2010 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 #define O_CLOEXEC 0
106
107 /* TODO: look at how perl does stat (non-sloppy), unlink (ro-files), utime, link */
108
109 #ifdef EIO_STRUCT_STATI64
110 /* look at perl's non-sloppy stat */
111 #define stat(path,buf) _stati64 (path,buf)
112 #define fstat(fd,buf) _fstati64 (fd,buf)
113 #endif
114 #define lstat(path,buf) stat (path,buf)
115 #define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
116 #define mkdir(path,mode) _mkdir (path)
117 #define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
118
119 #define chmod(path,mode) _chmod (path, mode)
120 #define dup(fd) _dup (fd)
121 #define dup2(fd1,fd2) _dup2 (fd1, fd2)
122 #define pipe(fds) _pipe (fds, 4096, O_BINARY)
123
124 #define fcntl(fd,cmd,arg) EIO_ENOSYS ()
125 #define ioctl(fd,cmd,arg) EIO_ENOSYS ()
126 #define fchmod(fd,mode) EIO_ENOSYS ()
127 #define chown(path,uid,gid) EIO_ENOSYS ()
128 #define fchown(fd,uid,gid) EIO_ENOSYS ()
129 #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
130 #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
131 #define mknod(path,mode,dev) EIO_ENOSYS ()
132 #define sync() EIO_ENOSYS ()
133 #define readlink(path,buf,s) EIO_ENOSYS ()
134 #define statvfs(path,buf) EIO_ENOSYS ()
135 #define fstatvfs(fd,buf) EIO_ENOSYS ()
136
137 #define pread(fd,buf,count,offset) eio__pread (fd, buf, count, offset)
138 #define pwrite(fd,buf,count,offset) eio__pwrite (fd, buf, count, offset)
139
140 #if __GNUC__
141 typedef long long eio_off_t; /* signed for compatibility to msvc */
142 #else
143 typedef __int64 eio_off_t; /* unsigned not supported by msvc */
144 #endif
145
146 static eio_ssize_t
147 eio__pread (int fd, void *buf, eio_ssize_t count, eio_off_t offset)
148 {
149 OVERLAPPED o = { 0 };
150 DWORD got;
151
152 o.Offset = offset;
153 o.OffsetHigh = offset >> 32;
154
155 return ReadFile ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd), buf, count, &got, &o)
156 ? got : -1;
157 }
158
159 static eio_ssize_t
160 eio__pwrite (int fd, void *buf, eio_ssize_t count, eio_off_t offset)
161 {
162 OVERLAPPED o = { 0 };
163 DWORD got;
164
165 o.Offset = offset;
166 o.OffsetHigh = offset >> 32;
167
168 return WriteFile ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd), buf, count, &got, &o)
169 ? got : -1;
170 }
171
172 /* rename() uses MoveFile, which fails to overwrite */
173 #define rename(old,neu) eio__rename (old, neu)
174
175 static int
176 eio__rename (const char *old, const char *neu)
177 {
178 if (MoveFileEx (old, neu, MOVEFILE_REPLACE_EXISTING))
179 return 0;
180
181 /* should steal _dosmaperr */
182 switch (GetLastError ())
183 {
184 case ERROR_FILE_NOT_FOUND:
185 case ERROR_PATH_NOT_FOUND:
186 case ERROR_INVALID_DRIVE:
187 case ERROR_NO_MORE_FILES:
188 case ERROR_BAD_NETPATH:
189 case ERROR_BAD_NET_NAME:
190 case ERROR_BAD_PATHNAME:
191 case ERROR_FILENAME_EXCED_RANGE:
192 errno = ENOENT;
193 break;
194
195 default:
196 errno = EACCES;
197 break;
198 }
199
200 return -1;
201 }
202
203 /* we could even stat and see if it exists */
204 static int
205 symlink (const char *old, const char *neu)
206 {
207 #if WINVER >= 0x0600
208 int flags;
209
210 /* This tries out all combinations of SYMBOLIC_LINK_FLAG_DIRECTORY
211 * and SYMBOLIC_LINK_FLAG_ALLOW_UNPRIVILEGED_CREATE,
212 * with directory first.
213 */
214 for (flags = 3; flags >= 0; --flags)
215 if (CreateSymbolicLink (neu, old, flags))
216 return 0;
217 #endif
218
219 return EIO_ERRNO (ENOENT, -1);
220 }
221
222 /* POSIX API only, causing trouble for win32 apps */
223 #define CreateHardLink(neu,old,flags) 0 /* not really creating hardlink, still using relative paths? */
224 #define CreateSymbolicLink(neu,old,flags) 0 /* vista+ only */
225
226 struct statvfs
227 {
228 int dummy;
229 };
230
231 #define DT_DIR EIO_DT_DIR
232 #define DT_REG EIO_DT_REG
233 #define D_NAME(entp) entp.cFileName
234 #define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
235
73#else 236#else
74 237
75# include "config.h" 238 #include <sys/ioctl.h>
76# include <sys/time.h> 239 #include <sys/time.h>
77# include <sys/select.h> 240 #include <sys/select.h>
78# include <sys/mman.h>
79# include <unistd.h> 241 #include <unistd.h>
242 #include <signal.h>
243 #include <dirent.h>
244
245 #ifdef ANDROID
246 #include <sys/vfs.h>
247 #define statvfs statfs
248 #define fstatvfs fstatfs
249 #include <asm/page.h> /* supposedly limits.h does #define PAGESIZE PAGESIZE */
250 #else
251 #include <sys/statvfs.h>
252 #endif
253
254 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
255 #include <sys/mman.h>
256 #endif
257
258 #define D_NAME(entp) entp->d_name
259
260 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
261 #if __FreeBSD__ || __NetBSD__ || __OpenBSD__
262 #define _DIRENT_HAVE_D_TYPE /* sigh */
263 #define D_INO(de) (de)->d_fileno
264 #define D_NAMLEN(de) (de)->d_namlen
265 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
266 #define D_INO(de) (de)->d_ino
267 #endif
268
269 #ifdef _D_EXACT_NAMLEN
270 #undef D_NAMLEN
271 #define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
272 #endif
273
274 #ifdef _DIRENT_HAVE_D_TYPE
275 #define D_TYPE(de) (de)->d_type
276 #endif
277
278 #ifndef EIO_STRUCT_DIRENT
279 #define EIO_STRUCT_DIRENT struct dirent
280 #endif
281
282#endif
283
284#if HAVE_UTIMES
80# include <utime.h> 285# 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 286#endif
92 287
93#ifdef _D_EXACT_NAMLEN 288#if HAVE_SYS_SYSCALL_H
94# undef D_NAMLEN 289# 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 290#endif
107 291
108#if HAVE_SENDFILE 292#if HAVE_SENDFILE
109# if __linux 293# if __linux
110# include <sys/sendfile.h> 294# include <sys/sendfile.h>
111# elif __freebsd || defined __APPLE__ 295# elif __FreeBSD__ || defined __APPLE__
112# include <sys/socket.h> 296# include <sys/socket.h>
113# include <sys/uio.h> 297# include <sys/uio.h>
114# elif __hpux 298# elif __hpux
115# include <sys/socket.h> 299# include <sys/socket.h>
116# elif __solaris 300# elif __solaris
118# else 302# else
119# error sendfile support requested but not available 303# error sendfile support requested but not available
120# endif 304# endif
121#endif 305#endif
122 306
307#if HAVE_RENAMEAT2
308# include <sys/syscall.h>
309# include <linux/fs.h>
310#endif
311
123#ifndef D_TYPE 312#ifndef D_TYPE
124# define D_TYPE(de) 0 313# define D_TYPE(de) 0
125#endif 314#endif
126#ifndef D_INO 315#ifndef D_INO
127# define D_INO(de) 0 316# define D_INO(de) 0
128#endif 317#endif
129#ifndef D_NAMLEN 318#ifndef D_NAMLEN
130# define D_NAMLEN(de) strlen ((de)->d_name) 319# define D_NAMLEN(entp) strlen (D_NAME (entp))
131#endif 320#endif
132
133/* number of seconds after which an idle threads exit */
134#define IDLE_TIMEOUT 10
135 321
136/* used for struct dirent, AIX doesn't provide it */ 322/* used for struct dirent, AIX doesn't provide it */
137#ifndef NAME_MAX 323#ifndef NAME_MAX
138# define NAME_MAX 4096 324# define NAME_MAX 4096
139#endif 325#endif
140 326
327/* used for readlink etc. */
328#ifndef PATH_MAX
329# define PATH_MAX 0
330#endif
331
332#ifndef EIO_PATH_MIN
333# define EIO_PATH_MIN 8160
334#endif
335
336#define EIO_PATH_MAX (PATH_MAX <= EIO_PATH_MIN ? EIO_PATH_MIN : PATH_MAX)
337
141/* buffer size for various temporary buffers */ 338/* buffer size for various temporary buffers */
142#define EIO_BUFSIZE 65536 339#define EIO_BUFSIZE 65536
143 340
144#define dBUF \ 341#define dBUF \
145 char *eio_buf; \
146 ETP_WORKER_LOCK (self); \
147 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 342 char *eio_buf = malloc (EIO_BUFSIZE); \
148 ETP_WORKER_UNLOCK (self); \
149 errno = ENOMEM; \ 343 errno = ENOMEM; \
150 if (!eio_buf) \ 344 if (!eio_buf) \
151 return -1; 345 return -1
152 346
153#define EIO_TICKS ((1000000 + 1023) >> 10) 347#define FUBd \
348 free (eio_buf)
154 349
155/*****************************************************************************/ 350/*****************************************************************************/
156 351
157#if __GNUC__ >= 3 352struct etp_tmpbuf;
158# define expect(expr,value) __builtin_expect ((expr),(value)) 353
354#if _POSIX_VERSION >= 200809L
355 #define HAVE_AT 1
356 #define WD2FD(wd) ((wd) ? (wd)->fd : AT_FDCWD)
357 #ifndef O_SEARCH
358 #define O_SEARCH O_RDONLY
359 #endif
159#else 360#else
160# define expect(expr,value) (expr) 361 #define HAVE_AT 0
362 static const char *wd_expand (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path);
161#endif 363#endif
162 364
163#define expect_false(expr) expect ((expr) != 0, 0) 365struct eio_pwd
164#define expect_true(expr) expect ((expr) != 0, 1) 366{
367#if HAVE_AT
368 int fd;
369#endif
370 int len;
371 char str[1]; /* actually, a 0-terminated canonical path */
372};
165 373
166/*****************************************************************************/ 374/*****************************************************************************/
167 375
168#define ETP_PRI_MIN EIO_PRI_MIN 376#define ETP_PRI_MIN EIO_PRI_MIN
169#define ETP_PRI_MAX EIO_PRI_MAX 377#define ETP_PRI_MAX EIO_PRI_MAX
170 378
379#define ETP_TYPE_QUIT -1
380#define ETP_TYPE_GROUP EIO_GROUP
381
382static void eio_nop_callback (void) { }
383static void (*eio_want_poll_cb)(void) = eio_nop_callback;
384static void (*eio_done_poll_cb)(void) = eio_nop_callback;
385
386#define ETP_WANT_POLL(pool) eio_want_poll_cb ()
387#define ETP_DONE_POLL(pool) eio_done_poll_cb ()
388
171struct etp_worker; 389struct etp_worker;
172
173#define ETP_REQ eio_req 390#define ETP_REQ eio_req
174#define ETP_DESTROY(req) eio_destroy (req) 391#define ETP_DESTROY(req) eio_destroy (req)
175static int eio_finish (eio_req *req); 392static int eio_finish (eio_req *req);
176#define ETP_FINISH(req) eio_finish (req) 393#define ETP_FINISH(req) eio_finish (req)
177static void eio_execute (struct etp_worker *self, eio_req *req); 394static void eio_execute (struct etp_worker *self, eio_req *req);
178#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) 395#define ETP_EXECUTE(wrk,req) eio_execute (wrk, req)
179 396
180#define ETP_WORKER_CLEAR(req) \ 397#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 }
192 398
193#define ETP_WORKER_COMMON \ 399static struct etp_pool eio_pool;
194 void *dbuf; \ 400#define EIO_POOL (&eio_pool)
195 DIR *dirp;
196 401
197/*****************************************************************************/ 402/*****************************************************************************/
198 403
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 404static 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) 405grp_try_feed (eio_req *grp)
639{ 406{
640 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 407 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
641 { 408 {
642 grp->flags &= ~EIO_FLAG_GROUPADD; 409 grp->flags &= ~ETP_FLAG_GROUPADD;
643 410
644 EIO_FEED (grp); 411 EIO_FEED (grp);
645 412
646 /* stop if no progress has been made */ 413 /* stop if no progress has been made */
647 if (!(grp->flags & EIO_FLAG_GROUPADD)) 414 if (!(grp->flags & ETP_FLAG_GROUPADD))
648 { 415 {
649 grp->feed = 0; 416 grp->feed = 0;
650 break; 417 break;
651 } 418 }
652 } 419 }
653} 420}
654 421
422static int
655static int grp_dec (eio_req *grp) 423grp_dec (eio_req *grp)
656{ 424{
657 --grp->size; 425 --grp->size;
658 426
659 /* call feeder, if applicable */ 427 /* call feeder, if applicable */
660 grp_try_feed (grp); 428 grp_try_feed (grp);
661 429
662 /* finish, if done */ 430 /* finish, if done */
663 if (!grp->size && grp->int1) 431 if (!grp->size && grp->flags & ETP_FLAG_DELAYED)
664 return eio_finish (grp); 432 return eio_finish (grp);
665 else 433 else
666 return 0; 434 return 0;
667} 435}
668 436
437static void
669void eio_destroy (eio_req *req) 438eio_destroy (eio_req *req)
670{ 439{
671 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); 440 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
672 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); 441 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
673 442
674 EIO_DESTROY (req); 443 EIO_DESTROY (req);
675} 444}
676 445
446static int
677static int eio_finish (eio_req *req) 447eio_finish (eio_req *req)
678{ 448{
679 int res = EIO_FINISH (req); 449 int res = EIO_FINISH (req);
680 450
681 if (req->grp) 451 if (req->grp)
682 { 452 {
690 if (grp->grp_first == req) 460 if (grp->grp_first == req)
691 grp->grp_first = req->grp_next; 461 grp->grp_first = req->grp_next;
692 462
693 res2 = grp_dec (grp); 463 res2 = grp_dec (grp);
694 464
695 if (!res && res2) 465 if (!res)
696 res = res2; 466 res = res2;
697 } 467 }
698 468
699 eio_destroy (req); 469 eio_destroy (req);
700 470
701 return res; 471 return res;
702} 472}
703 473
474void
704void eio_grp_cancel (eio_req *grp) 475eio_grp_cancel (eio_req *grp)
705{ 476{
706 for (grp = grp->grp_first; grp; grp = grp->grp_next) 477 etp_grp_cancel (EIO_POOL, grp);
707 eio_cancel (grp);
708} 478}
709 479
480void
710void eio_cancel (eio_req *req) 481eio_cancel (eio_req *req)
711{ 482{
712 etp_cancel (req); 483 etp_cancel (EIO_POOL, req);
713} 484}
714 485
486void
715void eio_submit (eio_req *req) 487eio_submit (eio_req *req)
716{ 488{
717 etp_submit (req); 489 etp_submit (EIO_POOL, req);
718} 490}
719 491
720unsigned int eio_nreqs (void) 492unsigned int
493eio_nreqs (void)
721{ 494{
722 return etp_nreqs (); 495 return etp_nreqs (EIO_POOL);
723} 496}
724 497
725unsigned int eio_nready (void) 498unsigned int
499eio_nready (void)
726{ 500{
727 return etp_nready (); 501 return etp_nready (EIO_POOL);
728} 502}
729 503
730unsigned int eio_npending (void) 504unsigned int
505eio_npending (void)
731{ 506{
732 return etp_npending (); 507 return etp_npending (EIO_POOL);
733} 508}
734 509
735unsigned int eio_nthreads (void) 510unsigned int ecb_cold
511eio_nthreads (void)
736{ 512{
737 return etp_nthreads (); 513 return etp_nthreads (EIO_POOL);
738} 514}
739 515
516void ecb_cold
740void eio_set_max_poll_time (double nseconds) 517eio_set_max_poll_time (double nseconds)
741{ 518{
742 etp_set_max_poll_time (nseconds); 519 etp_set_max_poll_time (EIO_POOL, nseconds);
743} 520}
744 521
522void ecb_cold
745void eio_set_max_poll_reqs (unsigned int maxreqs) 523eio_set_max_poll_reqs (unsigned int maxreqs)
746{ 524{
747 etp_set_max_poll_reqs (maxreqs); 525 etp_set_max_poll_reqs (EIO_POOL, maxreqs);
748} 526}
749 527
528void ecb_cold
750void eio_set_max_idle (unsigned int nthreads) 529eio_set_max_idle (unsigned int nthreads)
751{ 530{
752 etp_set_max_idle (nthreads); 531 etp_set_max_idle (EIO_POOL, nthreads);
753} 532}
754 533
534void ecb_cold
535eio_set_idle_timeout (unsigned int seconds)
536{
537 etp_set_idle_timeout (EIO_POOL, seconds);
538}
539
540void ecb_cold
755void eio_set_min_parallel (unsigned int nthreads) 541eio_set_min_parallel (unsigned int nthreads)
756{ 542{
757 etp_set_min_parallel (nthreads); 543 etp_set_min_parallel (EIO_POOL, nthreads);
758} 544}
759 545
546void ecb_cold
760void eio_set_max_parallel (unsigned int nthreads) 547eio_set_max_parallel (unsigned int nthreads)
761{ 548{
762 etp_set_max_parallel (nthreads); 549 etp_set_max_parallel (EIO_POOL, nthreads);
763} 550}
764 551
765int eio_poll (void) 552int eio_poll (void)
766{ 553{
767 return etp_poll (); 554 return etp_poll (EIO_POOL);
768} 555}
769 556
770/*****************************************************************************/ 557/*****************************************************************************/
771/* work around various missing functions */ 558/* work around various missing functions */
772 559
773#if !HAVE_PREADWRITE 560#if HAVE_POSIX_CLOSE && !__linux
774# undef pread 561# define eio__close(fd) posix_close (fd, 0)
775# undef pwrite 562#else
776# define pread eio__pread 563# 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 564#endif
811 565
566/* close() without disturbing errno */
567static void
568silent_close (int fd)
569{
570 int saved_errno = errno;
571 eio__close (fd);
572 errno = saved_errno;
573}
574
812#ifndef HAVE_FUTIMES 575#ifndef HAVE_UTIMES
813 576
814# undef utimes 577# undef utimes
815# undef futimes
816# define utimes(path,times) eio__utimes (path, times) 578# define utimes(path,times) eio__utimes (path, times)
817# define futimes(fd,times) eio__futimes (fd, times)
818 579
819static int 580static int
820eio__utimes (const char *filename, const struct timeval times[2]) 581eio__utimes (const char *filename, const struct timeval times[2])
821{ 582{
822 if (times) 583 if (times)
830 } 591 }
831 else 592 else
832 return utime (filename, 0); 593 return utime (filename, 0);
833} 594}
834 595
596#endif
597
598#ifndef HAVE_FUTIMES
599
600# undef futimes
601# define futimes(fd,times) eio__futimes (fd, times)
602
603static int
835static int eio__futimes (int fd, const struct timeval tv[2]) 604eio__futimes (int fd, const struct timeval tv[2])
836{ 605{
837 errno = ENOSYS; 606 errno = ENOSYS;
838 return -1; 607 return -1;
839} 608}
840 609
843#if !HAVE_FDATASYNC 612#if !HAVE_FDATASYNC
844# undef fdatasync 613# undef fdatasync
845# define fdatasync(fd) fsync (fd) 614# define fdatasync(fd) fsync (fd)
846#endif 615#endif
847 616
617static int
618eio__syncfs (int fd)
619{
620 int res;
621
622#if HAVE_SYS_SYNCFS
623 res = (int)syscall (__NR_syncfs, (int)(fd));
624#else
625 res = EIO_ENOSYS ();
626#endif
627
628 if (res < 0 && errno == ENOSYS && fd >= 0)
629 sync ();
630
631 return res;
632}
633
848/* sync_file_range always needs emulation */ 634/* sync_file_range always needs emulation */
849int 635static int
850eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags) 636eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
851{ 637{
852#if HAVE_SYNC_FILE_RANGE 638#if HAVE_SYNC_FILE_RANGE
853 int res; 639 int res;
854 640
871 /* even though we could play tricks with the flags, it's better to always 657 /* even though we could play tricks with the flags, it's better to always
872 * call fdatasync, as that matches the expectation of its users best */ 658 * call fdatasync, as that matches the expectation of its users best */
873 return fdatasync (fd); 659 return fdatasync (fd);
874} 660}
875 661
662static int
663eio__fallocate (int fd, int mode, off_t offset, size_t len)
664{
665#if HAVE_LINUX_FALLOCATE
666 return fallocate (fd, mode, offset, len);
667#else
668 return EIO_ENOSYS ();
669#endif
670}
671
876#if !HAVE_READAHEAD 672#if !HAVE_READAHEAD
877# undef readahead 673# undef readahead
878# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 674# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
879 675
880static ssize_t 676static eio_ssize_t
881eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) 677eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
882{ 678{
883 size_t todo = count; 679 size_t todo = count;
884 dBUF; 680 dBUF;
885 681
890 pread (fd, eio_buf, len, offset); 686 pread (fd, eio_buf, len, offset);
891 offset += len; 687 offset += len;
892 todo -= len; 688 todo -= len;
893 } 689 }
894 690
895 errno = 0; 691 FUBd;
692
693 /* linux's readahead basically only fails for EBADF or EINVAL (not mmappable) */
694 /* but not for e.g. EIO or eof, so we also never fail */
896 return count; 695 return 0;
897} 696}
898 697
899#endif 698#endif
900 699
901/* sendfile always needs emulation */ 700/* sendfile always needs emulation */
902static ssize_t 701static eio_ssize_t
903eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) 702eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
904{ 703{
704 eio_ssize_t written = 0;
905 ssize_t res; 705 eio_ssize_t res;
906 706
907 if (!count) 707 if (!count)
908 return 0; 708 return 0;
909 709
710 for (;;)
711 {
712#ifdef __APPLE__
713# undef HAVE_SENDFILE /* broken, as everything on os x */
714#endif
910#if HAVE_SENDFILE 715#if HAVE_SENDFILE
911# if __linux 716# if __linux
717 off_t soffset = offset;
912 res = sendfile (ofd, ifd, &offset, count); 718 res = sendfile (ofd, ifd, &soffset, count);
913 719
914# elif __freebsd 720# elif __FreeBSD__
915 /* 721 /*
916 * Of course, the freebsd sendfile is a dire hack with no thoughts 722 * Of course, the freebsd sendfile is a dire hack with no thoughts
917 * wasted on making it similar to other I/O functions. 723 * wasted on making it similar to other I/O functions.
918 */ 724 */
919 {
920 off_t sbytes; 725 off_t sbytes;
921 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 726 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
922 727
923 if (res < 0 && sbytes) 728 #if 0 /* according to the manpage, this is correct, but broken behaviour */
924 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 729 /* freebsd' sendfile will return 0 on success */
730 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
731 /* not on e.g. EIO or EPIPE - sounds broken */
732 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
925 res = sbytes; 733 res = sbytes;
926 } 734 #endif
927 735
736 /* according to source inspection, this is correct, and useful behaviour */
737 if (sbytes)
738 res = sbytes;
739
928# elif defined (__APPLE__) 740# elif defined __APPLE__
929
930 {
931 off_t sbytes = count; 741 off_t sbytes = count;
932 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0); 742 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
933 743
934 if (res < 0 && errno == EAGAIN && sbytes) 744 /* according to the manpage, sbytes is always valid */
745 if (sbytes)
935 res = sbytes; 746 res = sbytes;
936 }
937 747
938# elif __hpux 748# elif __hpux
939 res = sendfile (ofd, ifd, offset, count, 0, 0); 749 res = sendfile (ofd, ifd, offset, count, 0, 0);
940 750
941# elif __solaris 751# elif __solaris
942 {
943 struct sendfilevec vec; 752 struct sendfilevec vec;
944 size_t sbytes; 753 size_t sbytes;
945 754
946 vec.sfv_fd = ifd; 755 vec.sfv_fd = ifd;
947 vec.sfv_flag = 0; 756 vec.sfv_flag = 0;
948 vec.sfv_off = offset; 757 vec.sfv_off = offset;
949 vec.sfv_len = count; 758 vec.sfv_len = count;
950 759
951 res = sendfilev (ofd, &vec, 1, &sbytes); 760 res = sendfilev (ofd, &vec, 1, &sbytes);
952 761
953 if (res < 0 && sbytes) 762 if (res < 0 && sbytes)
954 res = sbytes; 763 res = sbytes;
955 }
956 764
957# endif 765# endif
958 766
959#elif defined (_WIN32) 767#elif defined (_WIN32) && 0
960
961 /* does not work, just for documentation of what would need to be done */ 768 /* does not work, just for documentation of what would need to be done */
962 { 769 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
770 /* libeio guarantees that the file offset does not change, and windows */
771 /* has no way to get an independent handle to the same file description */
963 HANDLE h = TO_SOCKET (ifd); 772 HANDLE h = TO_SOCKET (ifd);
964 SetFilePointer (h, offset, 0, FILE_BEGIN); 773 SetFilePointer (h, offset, 0, FILE_BEGIN);
965 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0); 774 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
966 }
967 775
968#else 776#else
969 res = -1; 777 res = EIO_ENOSYS ();
970 errno = ENOSYS;
971#endif 778#endif
972 779
780 /* we assume sendfile can copy at least 128mb in one go */
781 if (res <= 128 * 1024 * 1024)
782 {
783 if (res > 0)
784 written += res;
785
786 if (written)
787 return written;
788
789 break;
790 }
791 else
792 {
793 /* if we requested more, then probably the kernel was lazy */
794 written += res;
795 offset += res;
796 count -= res;
797
798 if (!count)
799 return written;
800 }
801 }
802
973 if (res < 0 803 if (res < 0
974 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 804 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
975 /* BSDs */ 805 /* BSDs */
976#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */ 806#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
977 || errno == ENOTSUP 807 || errno == ENOTSUP
978#endif 808#endif
809#ifdef EOPNOTSUPP /* windows */
979 || errno == EOPNOTSUPP /* BSDs */ 810 || errno == EOPNOTSUPP /* BSDs */
811#endif
980#if __solaris 812#if __solaris
981 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 813 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
982#endif 814#endif
983 ) 815 )
984 ) 816 )
988 820
989 res = 0; 821 res = 0;
990 822
991 while (count) 823 while (count)
992 { 824 {
993 ssize_t cnt; 825 eio_ssize_t cnt;
994 826
995 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 827 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
996 828
997 if (cnt <= 0) 829 if (cnt <= 0)
998 { 830 {
1010 842
1011 offset += cnt; 843 offset += cnt;
1012 res += cnt; 844 res += cnt;
1013 count -= cnt; 845 count -= cnt;
1014 } 846 }
847
848 FUBd;
1015 } 849 }
1016 850
1017 return res; 851 return res;
852}
853
854#ifdef PAGESIZE
855# define eio_pagesize() PAGESIZE
856#else
857static intptr_t
858eio_pagesize (void)
859{
860 static intptr_t page;
861
862 if (!page)
863 page = sysconf (_SC_PAGESIZE);
864
865 return page;
866}
867#endif
868
869static void
870eio_page_align (void **addr, size_t *length)
871{
872 intptr_t mask = eio_pagesize () - 1;
873
874 /* round down addr */
875 intptr_t adj = mask & (intptr_t)*addr;
876
877 *addr = (void *)((intptr_t)*addr - adj);
878 *length += adj;
879
880 /* round up length */
881 *length = (*length + mask) & ~mask;
882}
883
884#if !_POSIX_MEMLOCK
885# define eio__mlockall(a) EIO_ENOSYS ()
886#else
887
888static int
889eio__mlockall (int flags)
890{
891 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
892 extern int mallopt (int, int);
893 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
894 #endif
895
896 if (EIO_MCL_CURRENT != MCL_CURRENT
897 || EIO_MCL_FUTURE != MCL_FUTURE)
898 {
899 flags = 0
900 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
901 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
902 }
903
904 return mlockall (flags);
905}
906#endif
907
908#if !_POSIX_MEMLOCK_RANGE
909# define eio__mlock(a,b) EIO_ENOSYS ()
910#else
911
912static int
913eio__mlock (void *addr, size_t length)
914{
915 eio_page_align (&addr, &length);
916
917 return mlock (addr, length);
918}
919
920#endif
921
922#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
923# define eio__msync(a,b,c) EIO_ENOSYS ()
924#else
925
926static int
927eio__msync (void *mem, size_t len, int flags)
928{
929 eio_page_align (&mem, &len);
930
931 if (EIO_MS_ASYNC != MS_SYNC
932 || EIO_MS_INVALIDATE != MS_INVALIDATE
933 || EIO_MS_SYNC != MS_SYNC)
934 {
935 flags = 0
936 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
937 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
938 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
939 }
940
941 return msync (mem, len, flags);
942}
943
944#endif
945
946static int
947eio__mtouch (eio_req *req)
948{
949 void *mem = req->ptr2;
950 size_t len = req->size;
951 int flags = req->int1;
952
953 eio_page_align (&mem, &len);
954
955 {
956 intptr_t addr = (intptr_t)mem;
957 intptr_t end = addr + len;
958 intptr_t page = eio_pagesize ();
959
960 if (addr < end)
961 if (flags & EIO_MT_MODIFY) /* modify */
962 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
963 else
964 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
965 }
966
967 return 0;
968}
969
970/*****************************************************************************/
971/* requests implemented outside eio_execute, because they are so large */
972
973static void
974eio__lseek (eio_req *req)
975{
976 /* this usually gets optimised away completely, or your compiler sucks, */
977 /* or the whence constants really are not 0, 1, 2 */
978 int whence = req->int2 == EIO_SEEK_SET ? SEEK_SET
979 : req->int2 == EIO_SEEK_CUR ? SEEK_CUR
980 : req->int2 == EIO_SEEK_END ? SEEK_END
981 : req->int2;
982
983 req->offs = lseek (req->int1, req->offs, whence);
984 req->result = req->offs == (off_t)-1 ? -1 : 0;
985}
986
987/* result will always end up in tmpbuf, there is always space for adding a 0-byte */
988static int
989eio__realpath (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
990{
991 char *res;
992 const char *rel = path;
993 char *tmp1, *tmp2;
994#if SYMLOOP_MAX > 32
995 int symlinks = SYMLOOP_MAX;
996#else
997 int symlinks = 32;
998#endif
999
1000 errno = EINVAL;
1001 if (!rel)
1002 return -1;
1003
1004 errno = ENOENT;
1005 if (!*rel)
1006 return -1;
1007
1008 res = etp_tmpbuf_get (tmpbuf, EIO_PATH_MAX * 3);
1009#ifdef _WIN32
1010 if (_access (rel, 4) != 0)
1011 return -1;
1012
1013 symlinks = GetFullPathName (rel, EIO_PATH_MAX * 3, res, 0);
1014
1015 errno = ENAMETOOLONG;
1016 if (symlinks >= EIO_PATH_MAX * 3)
1017 return -1;
1018
1019 errno = EIO;
1020 if (symlinks <= 0)
1021 return -1;
1022
1023 return symlinks;
1024
1025#else
1026 tmp1 = res + EIO_PATH_MAX;
1027 tmp2 = tmp1 + EIO_PATH_MAX;
1028
1029#if 0 /* disabled, the musl way to do things is just too racy */
1030#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1031 /* on linux we may be able to ask the kernel */
1032 {
1033 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1034
1035 if (fd >= 0)
1036 {
1037 sprintf (tmp1, "/proc/self/fd/%d", fd);
1038 req->result = readlink (tmp1, res, EIO_PATH_MAX);
1039 /* here we should probably stat the open file and the disk file, to make sure they still match */
1040 eio__close (fd);
1041
1042 if (req->result > 0)
1043 goto done;
1044 }
1045 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1046 return -1;
1047 }
1048#endif
1049#endif
1050
1051 if (*rel != '/')
1052 {
1053 int len;
1054
1055 errno = ENOENT;
1056 if (wd == EIO_INVALID_WD)
1057 return -1;
1058
1059 if (wd == EIO_CWD)
1060 {
1061 if (!getcwd (res, EIO_PATH_MAX))
1062 return -1;
1063
1064 len = strlen (res);
1065 }
1066 else
1067 memcpy (res, wd->str, len = wd->len);
1068
1069 if (res [1]) /* only use if not / */
1070 res += len;
1071 }
1072
1073 while (*rel)
1074 {
1075 eio_ssize_t len, linklen;
1076 const char *beg = rel;
1077
1078 while (*rel && *rel != '/')
1079 ++rel;
1080
1081 len = rel - beg;
1082
1083 if (!len) /* skip slashes */
1084 {
1085 ++rel;
1086 continue;
1087 }
1088
1089 if (beg [0] == '.')
1090 {
1091 if (len == 1)
1092 continue; /* . - nop */
1093
1094 if (beg [1] == '.' && len == 2)
1095 {
1096 /* .. - back up one component, if possible */
1097
1098 while (res != tmpbuf->ptr)
1099 if (*--res == '/')
1100 break;
1101
1102 continue;
1103 }
1104 }
1105
1106 errno = ENAMETOOLONG;
1107 if (res + 1 + len + 1 >= tmp1)
1108 return -1;
1109
1110 /* copy one component */
1111 *res = '/';
1112 memcpy (res + 1, beg, len);
1113
1114 /* zero-terminate, for readlink */
1115 res [len + 1] = 0;
1116
1117 /* now check if it's a symlink */
1118 linklen = readlink (tmpbuf->ptr, tmp1, EIO_PATH_MAX);
1119
1120 if (linklen < 0)
1121 {
1122 if (errno != EINVAL)
1123 return -1;
1124
1125 /* it's a normal directory. hopefully */
1126 res += len + 1;
1127 }
1128 else
1129 {
1130 /* yay, it was a symlink - build new path in tmp2 */
1131 int rellen = strlen (rel);
1132
1133 errno = ENAMETOOLONG;
1134 if (linklen + 1 + rellen >= EIO_PATH_MAX) /* also catch linklen >= EIO_PATH_MAX */
1135 return -1;
1136
1137 errno = ELOOP;
1138 if (!--symlinks)
1139 return -1;
1140
1141 if (*tmp1 == '/')
1142 res = tmpbuf->ptr; /* symlink resolves to an absolute path */
1143
1144 /* we need to be careful, as rel might point into tmp2 already */
1145 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1146 tmp2 [linklen] = '/';
1147 memcpy (tmp2, tmp1, linklen);
1148
1149 rel = tmp2;
1150 }
1151 }
1152
1153 /* special case for the lone root path */
1154 if (res == tmpbuf->ptr)
1155 *res++ = '/';
1156
1157 return res - (char *)tmpbuf->ptr;
1158#endif
1018} 1159}
1019 1160
1020static signed char 1161static signed char
1021eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) 1162eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1022{ 1163{
1023 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */ 1164 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; 1165 : a->inode < b->inode ? -1
1166 : a->inode > b->inode ? 1
1167 : 0;
1025} 1168}
1026 1169
1027#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0 1170#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1028 1171
1029#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */ 1172#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 */ 1173#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1031 1174
1032static void 1175static void
1033eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1176eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1034{ 1177{
1035 unsigned char bits [9 + sizeof (ino_t) * 8]; 1178 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1036 unsigned char *bit = bits; 1179 unsigned char *bit = bits;
1037 1180
1038 assert (CHAR_BIT == 8); 1181 assert (CHAR_BIT == 8);
1039 assert (sizeof (eio_dirent) * 8 < 256); 1182 assert (sizeof (eio_dirent) * 8 < 256);
1040 assert (offsetof (eio_dirent, inode)); /* we use 0 as sentinel */ 1183 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1041 assert (offsetof (eio_dirent, score)); /* we use 0 as sentinel */ 1184 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1042 1185
1043 if (size <= EIO_SORT_FAST) 1186 if (size <= EIO_SORT_FAST)
1044 return; 1187 return;
1045 1188
1046 /* first prepare an array of bits to test in our radix sort */ 1189 /* 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 */ 1190 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1048 /* inode_bits must contain all inodes ORed together */ 1191 /* inode_bits must contain all inodes ORed together */
1049 /* which is used to skip bits that are 0 everywhere, which is very common */ 1192 /* which is used to skip bits that are 0 everywhere, which is very common */
1050 { 1193 {
1051 ino_t endianness; 1194 eio_ino_t endianness;
1052 int i, j; 1195 int i, j;
1053 1196
1054 /* we store the byte offset of byte n into byte n of "endianness" */ 1197 /* we store the byte offset of byte n into byte n of "endianness" */
1055 for (i = 0; i < sizeof (ino_t); ++i) 1198 for (i = 0; i < sizeof (eio_ino_t); ++i)
1056 ((unsigned char *)&endianness)[i] = i; 1199 ((unsigned char *)&endianness)[i] = i;
1057 1200
1058 *bit++ = 0; 1201 *bit++ = 0;
1059 1202
1060 for (i = 0; i < sizeof (ino_t); ++i) 1203 for (i = 0; i < sizeof (eio_ino_t); ++i)
1061 { 1204 {
1062 /* shifting off the byte offsets out of "endianness" */ 1205 /* shifting off the byte offsets out of "endianness" */
1063 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8; 1206 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1064 endianness >>= 8; 1207 endianness >>= 8;
1065 1208
1066 for (j = 0; j < 8; ++j) 1209 for (j = 0; j < 8; ++j)
1067 if (inode_bits & (((ino_t)1) << (i * 8 + j))) 1210 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1068 *bit++ = offs + j; 1211 *bit++ = offs + j;
1069 } 1212 }
1070 1213
1071 for (j = 0; j < 8; ++j) 1214 for (j = 0; j < 8; ++j)
1072 if (score_bits & (1 << j)) 1215 if (score_bits & (1 << j))
1073 *bit++ = offsetof (eio_dirent, score) * 8 + j; 1216 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1074 } 1217 }
1075 1218
1076 /* now actually do the sorting (a variant of MSD radix sort) */ 1219 /* now actually do the sorting (a variant of MSD radix sort) */
1077 { 1220 {
1078 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base; 1221 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1079 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end; 1222 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1080 unsigned char *bit_stk [9 + sizeof (ino_t) * 8]; 1223 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1081 int stk_idx = 0; 1224 int stk_idx = 0;
1082 1225
1083 base_stk [stk_idx] = dents; 1226 base_stk [stk_idx] = dents;
1084 end_stk [stk_idx] = dents + size; 1227 end_stk [stk_idx] = dents + size;
1085 bit_stk [stk_idx] = bit - 1; 1228 bit_stk [stk_idx] = bit - 1;
1164 } 1307 }
1165 } 1308 }
1166} 1309}
1167 1310
1168static void 1311static void
1169eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1312eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1170{ 1313{
1171 if (size <= 1) 1314 if (size <= 1)
1172 return; /* our insertion sort relies on size > 0 */ 1315 return; /* our insertion sort relies on size > 0 */
1173 1316
1174 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */ 1317 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1182 1325
1183/* read a full directory */ 1326/* read a full directory */
1184static void 1327static void
1185eio__scandir (eio_req *req, etp_worker *self) 1328eio__scandir (eio_req *req, etp_worker *self)
1186{ 1329{
1187 DIR *dirp;
1188 EIO_STRUCT_DIRENT *entp;
1189 char *name, *names; 1330 char *name, *names;
1190 int namesalloc = 4096; 1331 int namesalloc = 4096 - sizeof (void *) * 4;
1191 int namesoffs = 0; 1332 int namesoffs = 0;
1192 int flags = req->int1; 1333 int flags = req->int1;
1193 eio_dirent *dents = 0; 1334 eio_dirent *dents = 0;
1194 int dentalloc = 128; 1335 int dentalloc = 128;
1195 int dentoffs = 0; 1336 int dentoffs = 0;
1196 ino_t inode_bits = 0; 1337 eio_ino_t inode_bits = 0;
1338#ifdef _WIN32
1339 HANDLE dirp;
1340 WIN32_FIND_DATA entp;
1341#else
1342 DIR *dirp;
1343 EIO_STRUCT_DIRENT *entp;
1344#endif
1197 1345
1198 req->result = -1; 1346 req->result = -1;
1199 1347
1200 if (!(flags & EIO_READDIR_DENTS)) 1348 if (!(flags & EIO_READDIR_DENTS))
1201 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); 1349 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1202 1350
1203 X_LOCK (wrklock); 1351#ifdef _WIN32
1204 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 1352 {
1353 int len = strlen ((const char *)req->ptr1);
1354 char *path = malloc (MAX_PATH);
1355 const char *fmt;
1356 const char *reqpath = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1357
1358 if (!len)
1359 fmt = "./*";
1360 else if (reqpath[len - 1] == '/' || reqpath[len - 1] == '\\')
1361 fmt = "%s*";
1362 else
1363 fmt = "%s/*";
1364
1365 _snprintf (path, MAX_PATH, fmt, reqpath);
1366 dirp = FindFirstFile (path, &entp);
1367 free (path);
1368
1369 if (dirp == INVALID_HANDLE_VALUE)
1370 {
1371 /* should steal _dosmaperr */
1372 switch (GetLastError ())
1373 {
1374 case ERROR_FILE_NOT_FOUND:
1375 req->result = 0;
1376 break;
1377
1378 case ERROR_INVALID_NAME:
1379 case ERROR_PATH_NOT_FOUND:
1380 case ERROR_NO_MORE_FILES:
1381 errno = ENOENT;
1382 break;
1383
1384 case ERROR_NOT_ENOUGH_MEMORY:
1385 errno = ENOMEM;
1386 break;
1387
1388 default:
1389 errno = EINVAL;
1390 break;
1391 }
1392
1393 return;
1394 }
1395 }
1396#else
1397 #if HAVE_AT
1398 if (req->wd)
1399 {
1400 int fd = openat (WD2FD (req->wd), req->ptr1, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1401
1402 if (fd < 0)
1403 return;
1404
1405 dirp = fdopendir (fd);
1406
1407 if (!dirp)
1408 silent_close (fd);
1409 }
1410 else
1205 self->dirp = dirp = opendir (req->ptr1); 1411 dirp = opendir (req->ptr1);
1412 #else
1413 dirp = opendir (wd_expand (&self->tmpbuf, req->wd, req->ptr1));
1414 #endif
1415
1416 if (!dirp)
1417 return;
1418#endif
1419
1420 if (req->flags & EIO_FLAG_PTR1_FREE)
1421 free (req->ptr1);
1422
1206 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1423 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1207 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; 1424 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1208 req->ptr2 = names = malloc (namesalloc); 1425 req->ptr2 = names = malloc (namesalloc);
1209 X_UNLOCK (wrklock);
1210 1426
1211 if (dirp && names && (!flags || dents)) 1427 if (!names || (flags && !dents))
1428 return;
1429
1212 for (;;) 1430 for (;;)
1213 { 1431 {
1432 int done;
1433
1434#ifdef _WIN32
1435 done = !dirp;
1436#else
1214 errno = 0; 1437 errno = 0;
1215 entp = readdir (dirp); 1438 entp = readdir (dirp);
1439 done = !entp;
1440#endif
1216 1441
1217 if (!entp) 1442 if (done)
1218 { 1443 {
1444#ifndef _WIN32
1445 int old_errno = errno;
1446 closedir (dirp);
1447 errno = old_errno;
1448
1219 if (errno) 1449 if (errno)
1220 break; 1450 break;
1451#endif
1221 1452
1222 /* sort etc. */ 1453 /* sort etc. */
1223 req->int1 = flags; 1454 req->int1 = flags;
1224 req->result = dentoffs; 1455 req->result = dentoffs;
1225 1456
1226 if (flags & EIO_READDIR_STAT_ORDER) 1457 if (flags & EIO_READDIR_STAT_ORDER)
1227 eio_dent_sort (dents, dentoffs, 0, inode_bits); /* sort by inode exclusively */ 1458 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1228 else if (flags & EIO_READDIR_DIRS_FIRST) 1459 else if (flags & EIO_READDIR_DIRS_FIRST)
1229 if (flags & EIO_READDIR_FOUND_UNKNOWN) 1460 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1230 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */ 1461 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1231 else 1462 else
1463 {
1464 /* in this case, all is known, and we just put dirs first and sort them */
1465 eio_dirent *oth = dents + dentoffs;
1466 eio_dirent *dir = dents;
1467
1468 /* now partition dirs to the front, and non-dirs to the back */
1469 /* by walking from both sides and swapping if necessary */
1470 while (oth > dir)
1471 {
1472 if (dir->type == EIO_DT_DIR)
1473 ++dir;
1474 else if ((--oth)->type == EIO_DT_DIR)
1475 {
1476 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1477
1478 ++dir;
1479 }
1480 }
1481
1482 /* now sort the dirs only (dirs all have the same score) */
1483 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1484 }
1485
1486 break;
1487 }
1488
1489 /* now add the entry to our list(s) */
1490 name = D_NAME (entp);
1491
1492 /* skip . and .. entries */
1493 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1494 {
1495 int len = D_NAMLEN (entp) + 1;
1496
1497 while (ecb_expect_false (namesoffs + len > namesalloc))
1498 {
1499 namesalloc *= 2;
1500 req->ptr2 = names = realloc (names, namesalloc);
1501
1502 if (!names)
1503 break;
1504 }
1505
1506 memcpy (names + namesoffs, name, len);
1507
1508 if (dents)
1509 {
1510 struct eio_dirent *ent;
1511
1512 if (ecb_expect_false (dentoffs == dentalloc))
1232 { 1513 {
1233 /* in this case, all is known, and we just put dirs first and sort them */ 1514 dentalloc *= 2;
1515 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1516
1517 if (!dents)
1518 break;
1519 }
1520
1234 eio_dirent *oth = dents + dentoffs; 1521 ent = dents + dentoffs;
1235 eio_dirent *dir = dents;
1236 1522
1237 /* now partition dirs to the front, and non-dirs to the back */ 1523 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1238 /* by walking from both sides and swapping if necessary */ 1524 ent->namelen = len - 1;
1239 /* also clear score, so it doesn't influence sorting */ 1525 ent->inode = D_INO (entp);
1240 while (oth > dir) 1526
1527 inode_bits |= ent->inode;
1528
1529 switch (D_TYPE (entp))
1530 {
1531 default:
1532 ent->type = EIO_DT_UNKNOWN;
1533 flags |= EIO_READDIR_FOUND_UNKNOWN;
1534 break;
1535
1536 #ifdef DT_FIFO
1537 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1538 #endif
1539 #ifdef DT_CHR
1540 case DT_CHR: ent->type = EIO_DT_CHR; break;
1541 #endif
1542 #ifdef DT_MPC
1543 case DT_MPC: ent->type = EIO_DT_MPC; break;
1544 #endif
1545 #ifdef DT_DIR
1546 case DT_DIR: ent->type = EIO_DT_DIR; break;
1547 #endif
1548 #ifdef DT_NAM
1549 case DT_NAM: ent->type = EIO_DT_NAM; break;
1550 #endif
1551 #ifdef DT_BLK
1552 case DT_BLK: ent->type = EIO_DT_BLK; break;
1553 #endif
1554 #ifdef DT_MPB
1555 case DT_MPB: ent->type = EIO_DT_MPB; break;
1556 #endif
1557 #ifdef DT_REG
1558 case DT_REG: ent->type = EIO_DT_REG; break;
1559 #endif
1560 #ifdef DT_NWK
1561 case DT_NWK: ent->type = EIO_DT_NWK; break;
1562 #endif
1563 #ifdef DT_CMP
1564 case DT_CMP: ent->type = EIO_DT_CMP; break;
1565 #endif
1566 #ifdef DT_LNK
1567 case DT_LNK: ent->type = EIO_DT_LNK; break;
1568 #endif
1569 #ifdef DT_SOCK
1570 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1571 #endif
1572 #ifdef DT_DOOR
1573 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1574 #endif
1575 #ifdef DT_WHT
1576 case DT_WHT: ent->type = EIO_DT_WHT; break;
1577 #endif
1578 }
1579
1580 ent->score = 7;
1581
1582 if (flags & EIO_READDIR_DIRS_FIRST)
1583 {
1584 if (ent->type == EIO_DT_UNKNOWN)
1241 { 1585 {
1242 if (dir->type == EIO_DT_DIR) 1586 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1243 ++dir; 1587 ent->score = 1;
1244 else if ((--oth)->type == EIO_DT_DIR) 1588 else if (!strchr (name, '.')) /* absence of dots indicate likely dirs */
1245 { 1589 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 } 1590 }
1251 1591 else if (ent->type == EIO_DT_DIR)
1252 /* now sort the dirs only */ 1592 ent->score = 0;
1253 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1254 } 1593 }
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 } 1594 }
1277 1595
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; 1596 namesoffs += len;
1371 ++dentoffs; 1597 ++dentoffs;
1372 } 1598 }
1373 1599
1374 if (EIO_CANCELLED (req)) 1600 if (EIO_CANCELLED (req))
1375 { 1601 {
1376 errno = ECANCELED; 1602 errno = ECANCELED;
1377 break; 1603 break;
1378 } 1604 }
1605
1606#ifdef _WIN32
1607 if (!FindNextFile (dirp, &entp))
1608 {
1609 FindClose (dirp);
1610 dirp = 0;
1379 } 1611 }
1612#endif
1613 }
1380} 1614}
1381 1615
1382#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) 1616/*****************************************************************************/
1383# define eio__msync(a,b,c) ((errno = ENOSYS), -1) 1617/* working directory stuff */
1618/* various deficiencies in the posix 2008 api force us to */
1619/* keep the absolute path in string form at all times */
1620/* fuck yeah. */
1621
1622#if !HAVE_AT
1623
1624/* a bit like realpath, but usually faster because it doesn'T have to return */
1625/* an absolute or canonical path */
1626static const char *
1627wd_expand (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
1628{
1629 if (!wd || *path == '/')
1630 return path;
1631
1632 if (path [0] == '.' && !path [1])
1633 return wd->str;
1634
1635 {
1636 int l1 = wd->len;
1637 int l2 = strlen (path);
1638
1639 char *res = etp_tmpbuf_get (tmpbuf, l1 + l2 + 2);
1640
1641 memcpy (res, wd->str, l1);
1642 res [l1] = '/';
1643 memcpy (res + l1 + 1, path, l2 + 1);
1644
1645 return res;
1646 }
1647}
1648
1649#endif
1650
1651static eio_wd
1652eio__wd_open_sync (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
1653{
1654 int fd;
1655 eio_wd res;
1656 int len = eio__realpath (tmpbuf, wd, path);
1657
1658 if (len < 0)
1659 return EIO_INVALID_WD;
1660
1661#if HAVE_AT
1662 fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1663
1664 if (fd < 0)
1665 return EIO_INVALID_WD;
1666#endif
1667
1668 res = malloc (sizeof (*res) + len); /* one extra 0-byte */
1669
1670#if HAVE_AT
1671 res->fd = fd;
1672#endif
1673
1674 res->len = len;
1675 memcpy (res->str, tmpbuf->ptr, len);
1676 res->str [len] = 0;
1677
1678 return res;
1679}
1680
1681eio_wd
1682eio_wd_open_sync (eio_wd wd, const char *path)
1683{
1684 struct etp_tmpbuf tmpbuf = { };
1685 wd = eio__wd_open_sync (&tmpbuf, wd, path);
1686 free (tmpbuf.ptr);
1687
1688 return wd;
1689}
1690
1691void
1692eio_wd_close_sync (eio_wd wd)
1693{
1694 if (wd != EIO_INVALID_WD && wd != EIO_CWD)
1695 {
1696 #if HAVE_AT
1697 eio__close (wd->fd);
1698 #endif
1699 free (wd);
1700 }
1701}
1702
1703#if HAVE_AT
1704
1705static int
1706eio__renameat2 (int olddirfd, const char *oldpath, int newdirfd, const char *newpath, unsigned int flags)
1707{
1708#if HAVE_RENAMEAT2
1709 return syscall (SYS_renameat2, olddirfd, oldpath, newdirfd, newpath, flags);
1384#else 1710#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) 1711 if (flags)
1415 page = sysconf (_SC_PAGESIZE); 1712 return EIO_ENOSYS ();
1416#endif
1417 1713
1418 /* round down to start of page, although this is probably useless */ 1714 return renameat (olddirfd, oldpath, newdirfd, newpath);
1419 addr &= ~(page - 1); /* assume page size is always a power of two */ 1715#endif
1716}
1420 1717
1421 if (addr < end) 1718/* 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 1719
1720static int
1721eio__truncateat (int dirfd, const char *path, off_t length)
1722{
1723 int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
1724 int res;
1725
1726 if (fd < 0)
1727 return fd;
1728
1729 res = ftruncate (fd, length);
1730 silent_close (fd);
1427 return 0; 1731 return res;
1428} 1732}
1733
1734static int
1735eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
1736{
1737 int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
1738 int res;
1739
1740 if (fd < 0)
1741 return fd;
1742
1743 res = fstatvfs (fd, buf);
1744 silent_close (fd);
1745 return res;
1746}
1747
1748#endif
1429 1749
1430/*****************************************************************************/ 1750/*****************************************************************************/
1431 1751
1432#define ALLOC(len) \ 1752#define ALLOC(len) \
1433 if (!req->ptr2) \ 1753 if (!req->ptr2) \
1434 { \ 1754 { \
1435 X_LOCK (wrklock); \ 1755 X_LOCK (EIO_POOL->wrklock); \
1436 req->flags |= EIO_FLAG_PTR2_FREE; \ 1756 req->flags |= EIO_FLAG_PTR2_FREE; \
1437 X_UNLOCK (wrklock); \ 1757 X_UNLOCK (EIO_POOL->wrklock); \
1438 req->ptr2 = malloc (len); \ 1758 req->ptr2 = malloc (len); \
1439 if (!req->ptr2) \ 1759 if (!req->ptr2) \
1440 { \ 1760 { \
1441 errno = ENOMEM; \ 1761 errno = ENOMEM; \
1442 req->result = -1; \ 1762 req->result = -1; \
1443 break; \ 1763 goto alloc_fail; \
1444 } \ 1764 } \
1445 } 1765 }
1446 1766
1447X_THREAD_PROC (etp_proc) 1767/*****************************************************************************/
1448{
1449 ETP_REQ *req;
1450 struct timespec ts;
1451 etp_worker *self = (etp_worker *)thr_arg;
1452 1768
1453 /* try to distribute timeouts somewhat randomly */ 1769static void
1454 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 1770eio__slurp (int fd, eio_req *req)
1771{
1772 req->result = fd;
1455 1773
1456 for (;;) 1774 if (fd < 0)
1775 return;
1776
1777 if (req->offs < 0 || !req->size) /* do we need the size? */
1457 { 1778 {
1458 X_LOCK (reqlock); 1779 off_t size = lseek (fd, 0, SEEK_END);
1459 1780
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) 1781 if (req->offs < 0)
1494 goto quit; 1782 req->offs += size;
1495 1783
1496 if (!EIO_CANCELLED (req)) 1784 if (!req->size)
1497 ETP_EXECUTE (self, req); 1785 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 } 1786 }
1511 1787
1512quit: 1788 ALLOC (req->size);
1513 X_LOCK (wrklock); 1789 req->result = pread (fd, req->ptr2, req->size, req->offs);
1514 etp_worker_free (self);
1515 X_UNLOCK (wrklock);
1516 1790
1517 return 0; 1791 silent_close (fd);
1518}
1519 1792
1520/*****************************************************************************/ 1793alloc_fail:
1794 ;
1795}
1521 1796
1797int ecb_cold
1522int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1798eio_init (void (*want_poll)(void), void (*done_poll)(void))
1523{ 1799{
1524 return etp_init (want_poll, done_poll); 1800 eio_want_poll_cb = want_poll;
1525} 1801 eio_done_poll_cb = done_poll;
1526 1802
1803 return etp_init (EIO_POOL, 0, 0, 0);
1804}
1805
1806ecb_inline void
1527static void eio_api_destroy (eio_req *req) 1807eio_api_destroy (eio_req *req)
1528{ 1808{
1529 free (req); 1809 free (req);
1530} 1810}
1531 1811
1532#define REQ(rtype) \ 1812#define REQ(rtype) \
1533 eio_req *req; \ 1813 eio_req *req; \
1534 \ 1814 \
1535 req = (eio_req *)calloc (1, sizeof *req); \ 1815 req = (eio_req *)calloc (1, sizeof *req); \
1536 if (!req) \ 1816 if (!req) \
1537 return 0; \ 1817 return 0; \
1551 { \ 1831 { \
1552 eio_api_destroy (req); \ 1832 eio_api_destroy (req); \
1553 return 0; \ 1833 return 0; \
1554 } 1834 }
1555 1835
1836#define SINGLEDOT(ptr) (0[(char *)(ptr)] == '.' && !1[(char *)(ptr)])
1837
1838static void
1556static void eio_execute (etp_worker *self, eio_req *req) 1839eio_execute (etp_worker *self, eio_req *req)
1557{ 1840{
1558 errno = 0; 1841#if HAVE_AT
1842 int dirfd;
1843#else
1844 const char *path;
1845#endif
1846
1847 if (ecb_expect_false (EIO_CANCELLED (req)))
1848 {
1849 req->result = -1;
1850 req->errorno = ECANCELED;
1851 return;
1852 }
1853
1854 if (ecb_expect_false (req->wd == EIO_INVALID_WD))
1855 {
1856 req->result = -1;
1857 req->errorno = ENOENT;
1858 return;
1859 }
1860
1861 if (req->type >= EIO_OPEN)
1862 {
1863 #if HAVE_AT
1864 dirfd = WD2FD (req->wd);
1865 #else
1866 path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1867 #endif
1868 }
1559 1869
1560 switch (req->type) 1870 switch (req->type)
1561 { 1871 {
1872 case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
1873 req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
1874 break;
1875 case EIO_WD_CLOSE: req->result = 0;
1876 eio_wd_close_sync (req->wd); break;
1877
1878 case EIO_SEEK: eio__lseek (req); break;
1562 case EIO_READ: ALLOC (req->size); 1879 case EIO_READ: ALLOC (req->size);
1563 req->result = req->offs >= 0 1880 req->result = req->offs >= 0
1564 ? pread (req->int1, req->ptr2, req->size, req->offs) 1881 ? pread (req->int1, req->ptr2, req->size, req->offs)
1565 : read (req->int1, req->ptr2, req->size); break; 1882 : read (req->int1, req->ptr2, req->size); break;
1566 case EIO_WRITE: req->result = req->offs >= 0 1883 case EIO_WRITE: req->result = req->offs >= 0
1567 ? pwrite (req->int1, req->ptr2, req->size, req->offs) 1884 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1568 : write (req->int1, req->ptr2, req->size); break; 1885 : write (req->int1, req->ptr2, req->size); break;
1569 1886
1887 case EIO_FCNTL: req->result = fcntl (req->int1, (int) req->int2, req->ptr2); break;
1888 case EIO_IOCTL: req->result = ioctl (req->int1, (unsigned long)req->int2, req->ptr2); break;
1889
1570 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break; 1890 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1571 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break; 1891 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
1892
1893#if HAVE_AT
1572 1894
1573 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1895 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1574 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 1896 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break;
1575 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1897 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1576 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 1898 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, AT_SYMLINK_NOFOLLOW); break;
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; 1899 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; 1900 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; 1901 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; 1902 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; 1903 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; 1904
1590 case EIO_UNLINK: req->result = unlink (req->ptr1); break; 1905 case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
1591 case EIO_RMDIR: req->result = rmdir (req->ptr1); break; 1906 case EIO_RMDIR: /* complications arise because "." cannot be removed, so we might have to expand */
1907 req->result = req->wd && SINGLEDOT (req->ptr1)
1908 ? rmdir (req->wd->str)
1909 : unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break;
1592 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; 1910 case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); break;
1593 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; 1911 case EIO_RENAME: req->result = eio__renameat2 (
1912 dirfd,
1913 /* complications arise because "." cannot be renamed, so we might have to expand */
1914 req->wd && SINGLEDOT (req->ptr1) ? req->wd->str : req->ptr1,
1915 WD2FD ((eio_wd)req->int3),
1916 req->ptr2,
1917 req->int2
1918 );
1919 break;
1594 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; 1920 case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2, 0); break;
1595 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; 1921 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; 1922 case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1597 1923 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1924 req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1598 case EIO_READLINK: ALLOC (NAME_MAX); 1925 case EIO_READLINK: ALLOC (EIO_PATH_MAX);
1599 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; 1926 req->result = readlinkat (dirfd, req->ptr1, req->ptr2, EIO_PATH_MAX);
1600 1927 if (req->result == EIO_PATH_MAX)
1601 case EIO_SYNC: req->result = 0; sync (); break; 1928 {
1602 case EIO_FSYNC: req->result = fsync (req->int1); break; 1929 req->result = -1;
1603 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 1930 errno = ENAMETOOLONG;
1604 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break; 1931 }
1605 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break; 1932 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: 1933 case EIO_UTIME:
1611#ifdef _WIN32 1934 case EIO_FUTIME:
1612 Sleep (req->nv1 * 1000.); 1935 {
1936 struct timespec ts[2];
1937 struct timespec *times;
1938
1939 if (req->nv1 != -1. || req->nv2 != -1.)
1940 {
1941 ts[0].tv_sec = req->nv1;
1942 ts[0].tv_nsec = (req->nv1 - ts[0].tv_sec) * 1e9;
1943 ts[1].tv_sec = req->nv2;
1944 ts[1].tv_nsec = (req->nv2 - ts[1].tv_sec) * 1e9;
1945
1946 times = ts;
1947 }
1948 else
1949 times = 0;
1950
1951 req->result = req->type == EIO_FUTIME
1952 ? futimens (req->int1, times)
1953 : utimensat (dirfd, req->ptr1, times, 0);
1954 }
1955 break;
1956
1613#else 1957#else
1614 {
1615 struct timeval tv;
1616 1958
1617 tv.tv_sec = req->nv1; 1959 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1618 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; 1960 req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
1961 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1962 req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
1963 case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
1964 case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
1965 case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
1966 case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
1967 case EIO_SLURP: eio__slurp ( open (path , O_RDONLY | O_CLOEXEC), req); break;
1619 1968
1620 req->result = select (0, 0, 0, 0, &tv); 1969 case EIO_UNLINK: req->result = unlink (path ); break;
1621 } 1970 case EIO_RMDIR: req->result = rmdir (path ); break;
1622#endif 1971 case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
1623 break; 1972 case EIO_RENAME: req->result = req->int2 ? EIO_ENOSYS () : rename (path, req->ptr2); break;
1973 case EIO_LINK: req->result = link (path , req->ptr2); break;
1974 case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
1975 case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
1976 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1977 req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
1978 case EIO_READLINK: ALLOC (EIO_PATH_MAX);
1979 req->result = readlink (path, req->ptr2, EIO_PATH_MAX);
1980 if (req->result == EIO_PATH_MAX)
1981 {
1982 req->result = -1;
1983 errno = ENAMETOOLONG;
1984 }
1985 break;
1624 1986
1625 case EIO_UTIME: 1987 case EIO_UTIME:
1626 case EIO_FUTIME: 1988 case EIO_FUTIME:
1627 { 1989 {
1628 struct timeval tv[2]; 1990 struct timeval tv[2];
1629 struct timeval *times; 1991 struct timeval *times;
1630 1992
1631 if (req->nv1 != -1. || req->nv2 != -1.) 1993 if (req->nv1 != -1. || req->nv2 != -1.)
1632 { 1994 {
1633 tv[0].tv_sec = req->nv1; 1995 tv[0].tv_sec = req->nv1;
1634 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.; 1996 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
1635 tv[1].tv_sec = req->nv2; 1997 tv[1].tv_sec = req->nv2;
1636 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.; 1998 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
1637 1999
1638 times = tv; 2000 times = tv;
1639 } 2001 }
1640 else 2002 else
1641 times = 0; 2003 times = 0;
1642
1643 2004
1644 req->result = req->type == EIO_FUTIME 2005 req->result = req->type == EIO_FUTIME
1645 ? futimes (req->int1, times) 2006 ? futimes (req->int1, times)
1646 : utimes (req->ptr1, times); 2007 : utimes (req->ptr1, times);
1647 } 2008 }
1648 break; 2009 break;
1649 2010
2011#endif
2012
2013 case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
2014 {
2015 ALLOC (req->result);
2016 memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
2017 }
2018 break;
2019
2020 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2021 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
2022
2023 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2024 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2025
2026 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
2027 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
2028 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
2029
2030 case EIO_CLOSE: req->result = eio__close (req->int1); break;
2031 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
2032 case EIO_SYNC: req->result = 0; sync (); break;
2033 case EIO_FSYNC: req->result = fsync (req->int1); break;
2034 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
2035 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
2036 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
2037 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
2038 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
2039 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
2040 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
2041 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
2042
2043 case EIO_READDIR: eio__scandir (req, self); break;
2044
2045 case EIO_BUSY:
2046#ifdef _WIN32
2047 Sleep (req->nv1 * 1e3);
2048#else
2049 {
2050 struct timeval tv;
2051
2052 tv.tv_sec = req->nv1;
2053 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
2054
2055 req->result = select (0, 0, 0, 0, &tv);
2056 }
2057#endif
2058 break;
2059
2060#if 0
1650 case EIO_GROUP: 2061 case EIO_GROUP:
1651 abort (); /* handled in eio_request */ 2062 abort (); /* handled in eio_request */
2063#endif
1652 2064
1653 case EIO_NOP: 2065 case EIO_NOP:
1654 req->result = 0; 2066 req->result = 0;
1655 break; 2067 break;
1656 2068
1657 case EIO_CUSTOM: 2069 case EIO_CUSTOM:
1658 ((void (*)(eio_req *))req->feed) (req); 2070 req->feed (req);
1659 break; 2071 break;
1660 2072
1661 default: 2073 default:
1662 req->result = -1; 2074 req->result = EIO_ENOSYS ();
1663 break; 2075 break;
1664 } 2076 }
1665 2077
2078alloc_fail:
1666 req->errorno = errno; 2079 req->errorno = errno;
1667} 2080}
1668 2081
1669#ifndef EIO_NO_WRAPPERS 2082#ifndef EIO_NO_WRAPPERS
1670 2083
2084eio_req *eio_wd_open (const char *path, int pri, eio_cb cb, void *data)
2085{
2086 REQ (EIO_WD_OPEN); PATH; SEND;
2087}
2088
2089eio_req *eio_wd_close (eio_wd wd, int pri, eio_cb cb, void *data)
2090{
2091 REQ (EIO_WD_CLOSE); req->wd = wd; SEND;
2092}
2093
1671eio_req *eio_nop (int pri, eio_cb cb, void *data) 2094eio_req *eio_nop (int pri, eio_cb cb, void *data)
1672{ 2095{
1673 REQ (EIO_NOP); SEND; 2096 REQ (EIO_NOP); SEND;
1674} 2097}
1675 2098
1691eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2114eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1692{ 2115{
1693 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2116 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1694} 2117}
1695 2118
2119eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2120{
2121 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2122}
2123
2124eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
2125{
2126 REQ (EIO_SYNCFS); req->int1 = fd; SEND;
2127}
2128
2129eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2130{
2131 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2132}
2133
1696eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2134eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1697{ 2135{
1698 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2136 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1699} 2137}
1700 2138
1701eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) 2139eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
1702{ 2140{
1703 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; 2141 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
1704} 2142}
1705 2143
1706eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) 2144eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
1707{ 2145{
1708 REQ (EIO_FDATASYNC); req->int1 = fd; SEND; 2146 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2147}
2148
2149eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2150{
2151 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
1709} 2152}
1710 2153
1711eio_req *eio_close (int fd, int pri, eio_cb cb, void *data) 2154eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
1712{ 2155{
1713 REQ (EIO_CLOSE); req->int1 = fd; SEND; 2156 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) 2159eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
1717{ 2160{
1718 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND; 2161 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
1719} 2162}
1720 2163
2164eio_req *eio_seek (int fd, off_t offset, int whence, int pri, eio_cb cb, void *data)
2165{
2166 REQ (EIO_SEEK); req->int1 = fd; req->offs = offset; req->int2 = whence; SEND;
2167}
2168
1721eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data) 2169eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1722{ 2170{
1723 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND; 2171 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1724} 2172}
1725 2173
1726eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data) 2174eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1727{ 2175{
1728 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND; 2176 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1729} 2177}
1730 2178
2179eio_req *eio_fcntl (int fd, int cmd, void *arg, int pri, eio_cb cb, void *data)
2180{
2181 REQ (EIO_IOCTL); req->int1 = fd; req->int2 = cmd; req->ptr2 = arg; SEND;
2182}
2183
2184eio_req *eio_ioctl (int fd, unsigned long request, void *buf, int pri, eio_cb cb, void *data)
2185{
2186 REQ (EIO_IOCTL); req->int1 = fd; req->int2 = request; req->ptr2 = buf; SEND;
2187}
2188
1731eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) 2189eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1732{ 2190{
1733 REQ (EIO_FSTAT); req->int1 = fd; SEND; 2191 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1734} 2192}
1735 2193
2194eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
2195{
2196 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
2197}
2198
1736eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) 2199eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1737{ 2200{
1738 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; 2201 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1739} 2202}
1740 2203
1746eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) 2209eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1747{ 2210{
1748 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2211 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1749} 2212}
1750 2213
1751eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2214eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1752{ 2215{
1753 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2216 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1754} 2217}
1755 2218
1756eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) 2219eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
1776eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data) 2239eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1777{ 2240{
1778 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2241 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1779} 2242}
1780 2243
1781eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2244eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1782{ 2245{
1783 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2246 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1784} 2247}
1785 2248
1786eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) 2249eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
1802eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data) 2265eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1803{ 2266{
1804 return eio__1path (EIO_READLINK, path, pri, cb, data); 2267 return eio__1path (EIO_READLINK, path, pri, cb, data);
1805} 2268}
1806 2269
2270eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2271{
2272 return eio__1path (EIO_REALPATH, path, pri, cb, data);
2273}
2274
1807eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) 2275eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1808{ 2276{
1809 return eio__1path (EIO_STAT, path, pri, cb, data); 2277 return eio__1path (EIO_STAT, path, pri, cb, data);
1810} 2278}
1811 2279
1812eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) 2280eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1813{ 2281{
1814 return eio__1path (EIO_LSTAT, path, pri, cb, data); 2282 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1815} 2283}
1816 2284
2285eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
2286{
2287 return eio__1path (EIO_STATVFS, path, pri, cb, data);
2288}
2289
1817eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) 2290eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1818{ 2291{
1819 return eio__1path (EIO_UNLINK, path, pri, cb, data); 2292 return eio__1path (EIO_UNLINK, path, pri, cb, data);
1820} 2293}
1821 2294
1829 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND; 2302 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1830} 2303}
1831 2304
1832eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) 2305eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1833{ 2306{
1834 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; 2307 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1835} 2308}
1836 2309
1837static eio_req * 2310static eio_req *
1838eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2311eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1839{ 2312{
1863eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2336eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1864{ 2337{
1865 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2338 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1866} 2339}
1867 2340
2341eio_req *eio_slurp (const char *path, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
2342{
2343 REQ (EIO_SLURP); PATH; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
2344}
2345
1868eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) 2346eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
1869{ 2347{
1870 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND; 2348 REQ (EIO_CUSTOM); req->feed = execute; SEND;
1871} 2349}
1872 2350
1873#endif 2351#endif
1874 2352
1875eio_req *eio_grp (eio_cb cb, void *data) 2353eio_req *eio_grp (eio_cb cb, void *data)
1884#undef SEND 2362#undef SEND
1885 2363
1886/*****************************************************************************/ 2364/*****************************************************************************/
1887/* grp functions */ 2365/* grp functions */
1888 2366
2367void
1889void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2368eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1890{ 2369{
1891 grp->int2 = limit; 2370 grp->int2 = limit;
1892 grp->feed = feed; 2371 grp->feed = feed;
1893 2372
1894 grp_try_feed (grp); 2373 grp_try_feed (grp);
1895} 2374}
1896 2375
2376void
1897void eio_grp_limit (eio_req *grp, int limit) 2377eio_grp_limit (eio_req *grp, int limit)
1898{ 2378{
1899 grp->int2 = limit; 2379 grp->int2 = limit;
1900 2380
1901 grp_try_feed (grp); 2381 grp_try_feed (grp);
1902} 2382}
1903 2383
2384void
1904void eio_grp_add (eio_req *grp, eio_req *req) 2385eio_grp_add (eio_req *grp, eio_req *req)
1905{ 2386{
1906 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2387 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1907 2388
1908 grp->flags |= EIO_FLAG_GROUPADD; 2389 grp->flags |= ETP_FLAG_GROUPADD;
1909 2390
1910 ++grp->size; 2391 ++grp->size;
1911 req->grp = grp; 2392 req->grp = grp;
1912 2393
1913 req->grp_prev = 0; 2394 req->grp_prev = 0;
1920} 2401}
1921 2402
1922/*****************************************************************************/ 2403/*****************************************************************************/
1923/* misc garbage */ 2404/* misc garbage */
1924 2405
2406eio_ssize_t
1925ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2407eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1926{ 2408{
1927 etp_worker wrk;
1928 ssize_t ret;
1929
1930 wrk.dbuf = 0;
1931
1932 ret = eio__sendfile (ofd, ifd, offset, count, &wrk); 2409 return eio__sendfile (ofd, ifd, offset, count);
1933
1934 if (wrk.dbuf)
1935 free (wrk.dbuf);
1936
1937 return ret;
1938} 2410}
1939 2411

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines