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 1.159 by root, Thu Oct 29 05:13:56 2020 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) 2007,2008,2009,2010,2011,2012,2013,2016,2017,2018,2019 Marc Alexander Lehmann <libeio@schmorp.de>
5 * All rights reserved. 5 * All rights reserved.
6 * 6 *
7 * Redistribution and use in source and binary forms, with or without modifica- 7 * Redistribution and use in source and binary forms, with or without modifica-
8 * tion, are permitted provided that the following conditions are met: 8 * tion, are permitted provided that the following conditions are met:
9 * 9 *
10 * 1. Redistributions of source code must retain the above copyright notice, 10 * 1. Redistributions of source code must retain the above copyright notice,
11 * this list of conditions and the following disclaimer. 11 * this list of conditions and the following disclaimer.
12 * 12 *
13 * 2. Redistributions in binary form must reproduce the above copyright 13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the 14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution. 15 * documentation and/or other materials provided with the distribution.
16 * 16 *
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR IMPLIED 17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR IMPLIED
18 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MER- 18 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MER-
19 * CHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO 19 * CHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO
20 * EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPE- 20 * EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPE-
21 * CIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, 21 * CIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
35 * and other provisions required by the GPL. If you do not delete the 35 * and other provisions required by the GPL. If you do not delete the
36 * provisions above, a recipient may use your version of this file under 36 * provisions above, a recipient may use your version of this file under
37 * either the BSD or the GPL. 37 * either the BSD or the GPL.
38 */ 38 */
39 39
40#ifndef _WIN32
41# include "config.h"
42#endif
43
40#include "eio.h" 44#include "eio.h"
41
42#ifdef EIO_STACKSIZE
43# define XTHREAD_STACKSIZE EIO_STACKSIZE
44#endif
45#include "xthread.h" 45#include "ecb.h"
46 46
47#include <errno.h> 47#include <errno.h>
48#include <stddef.h> 48#include <stddef.h>
49#include <stdlib.h> 49#include <stdlib.h>
50#include <string.h> 50#include <string.h>
53#include <sys/stat.h> 53#include <sys/stat.h>
54#include <limits.h> 54#include <limits.h>
55#include <fcntl.h> 55#include <fcntl.h>
56#include <assert.h> 56#include <assert.h>
57 57
58/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
59/* intptr_t only comes from stdint.h, says idiot openbsd coder */
60#if HAVE_STDINT_H
61# include <stdint.h>
62#endif
63
64#ifndef ECANCELED
65# define ECANCELED EDOM
66#endif
67#ifndef ELOOP
68# define ELOOP EDOM
69#endif
70
71#if !defined(ENOTSOCK) && defined(WSAENOTSOCK)
72# define ENOTSOCK WSAENOTSOCK
73#endif
74
75static void eio_destroy (eio_req *req);
76
58#ifndef EIO_FINISH 77#ifndef EIO_FINISH
59# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0 78# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
60#endif 79#endif
61 80
62#ifndef EIO_DESTROY 81#ifndef EIO_DESTROY
65 84
66#ifndef EIO_FEED 85#ifndef EIO_FEED
67# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0) 86# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)
68#endif 87#endif
69 88
89#ifndef EIO_FD_TO_WIN32_HANDLE
90# define EIO_FD_TO_WIN32_HANDLE(fd) _get_osfhandle (fd)
91#endif
92#ifndef EIO_WIN32_HANDLE_TO_FD
93# define EIO_WIN32_HANDLE_TO_FD(handle) _open_osfhandle (handle, 0)
94#endif
95
96#define EIO_ERRNO(errval,retval) ((errno = errval), retval)
97
98#define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1)
99
70#ifdef _WIN32 100#ifdef _WIN32
71 101
72 /*doh*/ 102 #undef PAGESIZE
103 #define PAGESIZE 4096 /* GetSystemInfo? */
104
105 /* TODO: look at how perl does stat (non-sloppy), unlink (ro-files), utime, link */
106
107 #ifdef EIO_STRUCT_STATI64
108 /* look at perl's non-sloppy stat */
109 #define stat(path,buf) _stati64 (path,buf)
110 #define fstat(fd,buf) _fstati64 (fd,buf)
111 #endif
112 #define lstat(path,buf) stat (path,buf)
113 #define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
114 #define mkdir(path,mode) _mkdir (path)
115 #define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
116
117 #define chmod(path,mode) _chmod (path, mode)
118 #define dup(fd) _dup (fd)
119 #define dup2(fd1,fd2) _dup2 (fd1, fd2)
120 #define pipe(fds) _pipe (fds, 4096, O_BINARY)
121
122 #define fcntl(fd,cmd,arg) EIO_ENOSYS ()
123 #define ioctl(fd,cmd,arg) EIO_ENOSYS ()
124 #define fchmod(fd,mode) EIO_ENOSYS ()
125 #define chown(path,uid,gid) EIO_ENOSYS ()
126 #define fchown(fd,uid,gid) EIO_ENOSYS ()
127 #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
128 #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
129 #define mknod(path,mode,dev) EIO_ENOSYS ()
130 #define sync() EIO_ENOSYS ()
131 #define readlink(path,buf,s) EIO_ENOSYS ()
132 #define statvfs(path,buf) EIO_ENOSYS ()
133 #define fstatvfs(fd,buf) EIO_ENOSYS ()
134
135 #define pread(fd,buf,count,offset) eio__pread (fd, buf, count, offset)
136 #define pwrite(fd,buf,count,offset) eio__pwrite (fd, buf, count, offset)
137
138 #if __GNUC__
139 typedef long long eio_off_t; /* signed for compatibility to msvc */
140 #else
141 typedef __int64 eio_off_t; /* unsigned not supported by msvc */
142 #endif
143
144 static eio_ssize_t
145 eio__pread (int fd, void *buf, eio_ssize_t count, eio_off_t offset)
146 {
147 OVERLAPPED o = { 0 };
148 DWORD got;
149
150 o.Offset = offset;
151 o.OffsetHigh = offset >> 32;
152
153 return ReadFile ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd), buf, count, &got, &o)
154 ? got : -1;
155 }
156
157 static eio_ssize_t
158 eio__pwrite (int fd, void *buf, eio_ssize_t count, eio_off_t offset)
159 {
160 OVERLAPPED o = { 0 };
161 DWORD got;
162
163 o.Offset = offset;
164 o.OffsetHigh = offset >> 32;
165
166 return WriteFile ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd), buf, count, &got, &o)
167 ? got : -1;
168 }
169
170 /* rename() uses MoveFile, which fails to overwrite */
171 #define rename(old,neu) eio__rename (old, neu)
172
173 static int
174 eio__rename (const char *old, const char *neu)
175 {
176 if (MoveFileEx (old, neu, MOVEFILE_REPLACE_EXISTING))
177 return 0;
178
179 /* should steal _dosmaperr */
180 switch (GetLastError ())
181 {
182 case ERROR_FILE_NOT_FOUND:
183 case ERROR_PATH_NOT_FOUND:
184 case ERROR_INVALID_DRIVE:
185 case ERROR_NO_MORE_FILES:
186 case ERROR_BAD_NETPATH:
187 case ERROR_BAD_NET_NAME:
188 case ERROR_BAD_PATHNAME:
189 case ERROR_FILENAME_EXCED_RANGE:
190 errno = ENOENT;
191 break;
192
193 default:
194 errno = EACCES;
195 break;
196 }
197
198 return -1;
199 }
200
201 /* we could even stat and see if it exists */
202 static int
203 symlink (const char *old, const char *neu)
204 {
205 #if WINVER >= 0x0600
206 int flags;
207
208 /* This tries out all combinations of SYMBOLIC_LINK_FLAG_DIRECTORY
209 * and SYMBOLIC_LINK_FLAG_ALLOW_UNPRIVILEGED_CREATE,
210 * with directory first.
211 */
212 for (flags = 3; flags >= 0; --flags)
213 if (CreateSymbolicLink (neu, old, flags))
214 return 0;
215 #endif
216
217 return EIO_ERRNO (ENOENT, -1);
218 }
219
220 /* POSIX API only, causing trouble for win32 apps */
221 #define CreateHardLink(neu,old,flags) 0 /* not really creating hardlink, still using relative paths? */
222 #define CreateSymbolicLink(neu,old,flags) 0 /* vista+ only */
223
224 struct statvfs
225 {
226 int dummy;
227 };
228
229 #define DT_DIR EIO_DT_DIR
230 #define DT_REG EIO_DT_REG
231 #define D_NAME(entp) entp.cFileName
232 #define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
233
73#else 234#else
74 235
75# include "config.h" 236 #include <sys/ioctl.h>
76# include <sys/time.h> 237 #include <sys/time.h>
77# include <sys/select.h> 238 #include <sys/select.h>
78# include <sys/mman.h>
79# include <unistd.h> 239 #include <unistd.h>
240 #include <signal.h>
241 #include <dirent.h>
242
243 #ifdef ANDROID
244 #include <sys/vfs.h>
245 #define statvfs statfs
246 #define fstatvfs fstatfs
247 #include <asm/page.h> /* supposedly limits.h does #define PAGESIZE PAGESIZE */
248 #else
249 #include <sys/statvfs.h>
250 #endif
251
252 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
253 #include <sys/mman.h>
254 #endif
255
256 #define D_NAME(entp) entp->d_name
257
258 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
259 #if __FreeBSD__ || __NetBSD__ || __OpenBSD__
260 #define _DIRENT_HAVE_D_TYPE /* sigh */
261 #define D_INO(de) (de)->d_fileno
262 #define D_NAMLEN(de) (de)->d_namlen
263 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
264 #define D_INO(de) (de)->d_ino
265 #endif
266
267 #ifdef _D_EXACT_NAMLEN
268 #undef D_NAMLEN
269 #define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
270 #endif
271
272 #ifdef _DIRENT_HAVE_D_TYPE
273 #define D_TYPE(de) (de)->d_type
274 #endif
275
276 #ifndef EIO_STRUCT_DIRENT
277 #define EIO_STRUCT_DIRENT struct dirent
278 #endif
279
280#endif
281
282#if HAVE_UTIMES
80# include <utime.h> 283# include <utime.h>
81# include <signal.h>
82# include <dirent.h>
83
84/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
85# if __freebsd || defined __NetBSD__ || defined __OpenBSD__
86# define _DIRENT_HAVE_D_TYPE /* sigh */
87# define D_INO(de) (de)->d_fileno
88# define D_NAMLEN(de) (de)->d_namlen
89# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
90# define D_INO(de) (de)->d_ino
91# endif 284#endif
92 285
93#ifdef _D_EXACT_NAMLEN 286#if HAVE_SYS_SYSCALL_H
94# undef D_NAMLEN 287# include <sys/syscall.h>
95# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
96#endif
97
98# ifdef _DIRENT_HAVE_D_TYPE
99# define D_TYPE(de) (de)->d_type
100# endif
101
102# ifndef EIO_STRUCT_DIRENT
103# define EIO_STRUCT_DIRENT struct dirent
104# endif
105
106#endif 288#endif
107 289
108#if HAVE_SENDFILE 290#if HAVE_SENDFILE
109# if __linux 291# if __linux
110# include <sys/sendfile.h> 292# include <sys/sendfile.h>
111# elif __freebsd || defined __APPLE__ 293# elif __FreeBSD__ || defined __APPLE__
112# include <sys/socket.h> 294# include <sys/socket.h>
113# include <sys/uio.h> 295# include <sys/uio.h>
114# elif __hpux 296# elif __hpux
115# include <sys/socket.h> 297# include <sys/socket.h>
116# elif __solaris 298# elif __solaris
118# else 300# else
119# error sendfile support requested but not available 301# error sendfile support requested but not available
120# endif 302# endif
121#endif 303#endif
122 304
305#if HAVE_RENAMEAT2
306# include <sys/syscall.h>
307# include <linux/fs.h>
308#endif
309
123#ifndef D_TYPE 310#ifndef D_TYPE
124# define D_TYPE(de) 0 311# define D_TYPE(de) 0
125#endif 312#endif
126#ifndef D_INO 313#ifndef D_INO
127# define D_INO(de) 0 314# define D_INO(de) 0
128#endif 315#endif
129#ifndef D_NAMLEN 316#ifndef D_NAMLEN
130# define D_NAMLEN(de) strlen ((de)->d_name) 317# define D_NAMLEN(entp) strlen (D_NAME (entp))
131#endif 318#endif
132
133/* number of seconds after which an idle threads exit */
134#define IDLE_TIMEOUT 10
135 319
136/* used for struct dirent, AIX doesn't provide it */ 320/* used for struct dirent, AIX doesn't provide it */
137#ifndef NAME_MAX 321#ifndef NAME_MAX
138# define NAME_MAX 4096 322# define NAME_MAX 4096
139#endif 323#endif
140 324
325/* used for readlink etc. */
326#ifndef PATH_MAX
327# define PATH_MAX 0
328#endif
329
330#ifndef O_CLOEXEC
331 #define O_CLOEXEC 0
332#endif
333
334#ifndef O_NONBLOCK
335 #define O_NONBLOCK 0
336#endif
337
338#ifndef O_SEARCH
339 #define O_SEARCH O_RDONLY
340#endif
341
342#ifndef O_DIRECTORY
343 #define O_DIRECTORY 0
344#endif
345
346#ifndef EIO_PATH_MIN
347# define EIO_PATH_MIN 8160
348#endif
349
350#define EIO_PATH_MAX (PATH_MAX <= EIO_PATH_MIN ? EIO_PATH_MIN : PATH_MAX)
351
141/* buffer size for various temporary buffers */ 352/* buffer size for various temporary buffers */
142#define EIO_BUFSIZE 65536 353#define EIO_BUFSIZE 65536
143 354
144#define dBUF \ 355#define dBUF \
145 char *eio_buf; \
146 ETP_WORKER_LOCK (self); \
147 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 356 char *eio_buf = malloc (EIO_BUFSIZE); \
148 ETP_WORKER_UNLOCK (self); \
149 errno = ENOMEM; \ 357 errno = ENOMEM; \
150 if (!eio_buf) \ 358 if (!eio_buf) \
151 return -1; 359 return -1
152 360
153#define EIO_TICKS ((1000000 + 1023) >> 10) 361#define FUBd \
362 free (eio_buf)
154 363
155/*****************************************************************************/ 364/*****************************************************************************/
156 365
157#if __GNUC__ >= 3 366struct etp_tmpbuf;
158# define expect(expr,value) __builtin_expect ((expr),(value)) 367
368#if _POSIX_VERSION >= 200809L
369 #define HAVE_AT 1
370 #define WD2FD(wd) ((wd) ? (wd)->fd : AT_FDCWD)
159#else 371#else
160# define expect(expr,value) (expr) 372 #define HAVE_AT 0
373 static const char *wd_expand (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path);
161#endif 374#endif
162 375
163#define expect_false(expr) expect ((expr) != 0, 0) 376struct eio_pwd
164#define expect_true(expr) expect ((expr) != 0, 1) 377{
378#if HAVE_AT
379 int fd;
380#endif
381 int len;
382 char str[1]; /* actually, a 0-terminated canonical path */
383};
165 384
166/*****************************************************************************/ 385/*****************************************************************************/
167 386
168#define ETP_PRI_MIN EIO_PRI_MIN 387#define ETP_PRI_MIN EIO_PRI_MIN
169#define ETP_PRI_MAX EIO_PRI_MAX 388#define ETP_PRI_MAX EIO_PRI_MAX
170 389
390#define ETP_TYPE_QUIT -1
391#define ETP_TYPE_GROUP EIO_GROUP
392
393static void eio_nop_callback (void) { }
394static void (*eio_want_poll_cb)(void) = eio_nop_callback;
395static void (*eio_done_poll_cb)(void) = eio_nop_callback;
396
397#define ETP_WANT_POLL(pool) eio_want_poll_cb ()
398#define ETP_DONE_POLL(pool) eio_done_poll_cb ()
399
171struct etp_worker; 400struct etp_worker;
172
173#define ETP_REQ eio_req 401#define ETP_REQ eio_req
174#define ETP_DESTROY(req) eio_destroy (req) 402#define ETP_DESTROY(req) eio_destroy (req)
175static int eio_finish (eio_req *req); 403static int eio_finish (eio_req *req);
176#define ETP_FINISH(req) eio_finish (req) 404#define ETP_FINISH(req) eio_finish (req)
177static void eio_execute (struct etp_worker *self, eio_req *req); 405static void eio_execute (struct etp_worker *self, eio_req *req);
178#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) 406#define ETP_EXECUTE(wrk,req) eio_execute (wrk, req)
179 407
180#define ETP_WORKER_CLEAR(req) \ 408#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 }
192 409
193#define ETP_WORKER_COMMON \ 410static struct etp_pool eio_pool;
194 void *dbuf; \ 411#define EIO_POOL (&eio_pool)
195 DIR *dirp;
196 412
197/*****************************************************************************/ 413/*****************************************************************************/
198 414
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 415static 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) 416grp_try_feed (eio_req *grp)
639{ 417{
640 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 418 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
641 { 419 {
642 grp->flags &= ~EIO_FLAG_GROUPADD; 420 grp->flags &= ~ETP_FLAG_GROUPADD;
643 421
644 EIO_FEED (grp); 422 EIO_FEED (grp);
645 423
646 /* stop if no progress has been made */ 424 /* stop if no progress has been made */
647 if (!(grp->flags & EIO_FLAG_GROUPADD)) 425 if (!(grp->flags & ETP_FLAG_GROUPADD))
648 { 426 {
649 grp->feed = 0; 427 grp->feed = 0;
650 break; 428 break;
651 } 429 }
652 } 430 }
653} 431}
654 432
433static int
655static int grp_dec (eio_req *grp) 434grp_dec (eio_req *grp)
656{ 435{
657 --grp->size; 436 --grp->size;
658 437
659 /* call feeder, if applicable */ 438 /* call feeder, if applicable */
660 grp_try_feed (grp); 439 grp_try_feed (grp);
661 440
662 /* finish, if done */ 441 /* finish, if done */
663 if (!grp->size && grp->int1) 442 if (!grp->size && grp->flags & ETP_FLAG_DELAYED)
664 return eio_finish (grp); 443 return eio_finish (grp);
665 else 444 else
666 return 0; 445 return 0;
667} 446}
668 447
448static void
669void eio_destroy (eio_req *req) 449eio_destroy (eio_req *req)
670{ 450{
671 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); 451 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
672 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); 452 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
673 453
674 EIO_DESTROY (req); 454 EIO_DESTROY (req);
675} 455}
676 456
457static int
677static int eio_finish (eio_req *req) 458eio_finish (eio_req *req)
678{ 459{
679 int res = EIO_FINISH (req); 460 int res = EIO_FINISH (req);
680 461
681 if (req->grp) 462 if (req->grp)
682 { 463 {
690 if (grp->grp_first == req) 471 if (grp->grp_first == req)
691 grp->grp_first = req->grp_next; 472 grp->grp_first = req->grp_next;
692 473
693 res2 = grp_dec (grp); 474 res2 = grp_dec (grp);
694 475
695 if (!res && res2) 476 if (!res)
696 res = res2; 477 res = res2;
697 } 478 }
698 479
699 eio_destroy (req); 480 eio_destroy (req);
700 481
701 return res; 482 return res;
702} 483}
703 484
485void
704void eio_grp_cancel (eio_req *grp) 486eio_grp_cancel (eio_req *grp)
705{ 487{
706 for (grp = grp->grp_first; grp; grp = grp->grp_next) 488 etp_grp_cancel (EIO_POOL, grp);
707 eio_cancel (grp);
708} 489}
709 490
491void
710void eio_cancel (eio_req *req) 492eio_cancel (eio_req *req)
711{ 493{
712 etp_cancel (req); 494 etp_cancel (EIO_POOL, req);
713} 495}
714 496
497void
715void eio_submit (eio_req *req) 498eio_submit (eio_req *req)
716{ 499{
717 etp_submit (req); 500 etp_submit (EIO_POOL, req);
718} 501}
719 502
720unsigned int eio_nreqs (void) 503unsigned int
504eio_nreqs (void)
721{ 505{
722 return etp_nreqs (); 506 return etp_nreqs (EIO_POOL);
723} 507}
724 508
725unsigned int eio_nready (void) 509unsigned int
510eio_nready (void)
726{ 511{
727 return etp_nready (); 512 return etp_nready (EIO_POOL);
728} 513}
729 514
730unsigned int eio_npending (void) 515unsigned int
516eio_npending (void)
731{ 517{
732 return etp_npending (); 518 return etp_npending (EIO_POOL);
733} 519}
734 520
735unsigned int eio_nthreads (void) 521unsigned int ecb_cold
522eio_nthreads (void)
736{ 523{
737 return etp_nthreads (); 524 return etp_nthreads (EIO_POOL);
738} 525}
739 526
527void ecb_cold
740void eio_set_max_poll_time (double nseconds) 528eio_set_max_poll_time (double nseconds)
741{ 529{
742 etp_set_max_poll_time (nseconds); 530 etp_set_max_poll_time (EIO_POOL, nseconds);
743} 531}
744 532
533void ecb_cold
745void eio_set_max_poll_reqs (unsigned int maxreqs) 534eio_set_max_poll_reqs (unsigned int maxreqs)
746{ 535{
747 etp_set_max_poll_reqs (maxreqs); 536 etp_set_max_poll_reqs (EIO_POOL, maxreqs);
748} 537}
749 538
539void ecb_cold
750void eio_set_max_idle (unsigned int nthreads) 540eio_set_max_idle (unsigned int nthreads)
751{ 541{
752 etp_set_max_idle (nthreads); 542 etp_set_max_idle (EIO_POOL, nthreads);
753} 543}
754 544
545void ecb_cold
546eio_set_idle_timeout (unsigned int seconds)
547{
548 etp_set_idle_timeout (EIO_POOL, seconds);
549}
550
551void ecb_cold
755void eio_set_min_parallel (unsigned int nthreads) 552eio_set_min_parallel (unsigned int nthreads)
756{ 553{
757 etp_set_min_parallel (nthreads); 554 etp_set_min_parallel (EIO_POOL, nthreads);
758} 555}
759 556
557void ecb_cold
760void eio_set_max_parallel (unsigned int nthreads) 558eio_set_max_parallel (unsigned int nthreads)
761{ 559{
762 etp_set_max_parallel (nthreads); 560 etp_set_max_parallel (EIO_POOL, nthreads);
763} 561}
764 562
765int eio_poll (void) 563int eio_poll (void)
766{ 564{
767 return etp_poll (); 565 return etp_poll (EIO_POOL);
768} 566}
769 567
770/*****************************************************************************/ 568/*****************************************************************************/
771/* work around various missing functions */ 569/* work around various missing functions */
772 570
773#if !HAVE_PREADWRITE 571#if HAVE_POSIX_CLOSE && !__linux
774# undef pread 572# define eio__close(fd) posix_close (fd, 0)
775# undef pwrite 573#else
776# define pread eio__pread 574# 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 575#endif
811 576
577/* close() without disturbing errno */
578static void
579silent_close (int fd)
580{
581 int saved_errno = errno;
582 eio__close (fd);
583 errno = saved_errno;
584}
585
812#ifndef HAVE_FUTIMES 586#ifndef HAVE_UTIMES
813 587
814# undef utimes 588# undef utimes
815# undef futimes
816# define utimes(path,times) eio__utimes (path, times) 589# define utimes(path,times) eio__utimes (path, times)
817# define futimes(fd,times) eio__futimes (fd, times)
818 590
819static int 591static int
820eio__utimes (const char *filename, const struct timeval times[2]) 592eio__utimes (const char *filename, const struct timeval times[2])
821{ 593{
822 if (times) 594 if (times)
830 } 602 }
831 else 603 else
832 return utime (filename, 0); 604 return utime (filename, 0);
833} 605}
834 606
607#endif
608
609#ifndef HAVE_FUTIMES
610
611# undef futimes
612# define futimes(fd,times) eio__futimes (fd, times)
613
614static int
835static int eio__futimes (int fd, const struct timeval tv[2]) 615eio__futimes (int fd, const struct timeval tv[2])
836{ 616{
837 errno = ENOSYS; 617 errno = ENOSYS;
838 return -1; 618 return -1;
839} 619}
840 620
843#if !HAVE_FDATASYNC 623#if !HAVE_FDATASYNC
844# undef fdatasync 624# undef fdatasync
845# define fdatasync(fd) fsync (fd) 625# define fdatasync(fd) fsync (fd)
846#endif 626#endif
847 627
628static int
629eio__syncfs (int fd)
630{
631 int res;
632
633#if HAVE_SYS_SYNCFS
634 res = (int)syscall (__NR_syncfs, (int)(fd));
635#else
636 res = EIO_ENOSYS ();
637#endif
638
639 if (res < 0 && errno == ENOSYS && fd >= 0)
640 sync ();
641
642 return res;
643}
644
848/* sync_file_range always needs emulation */ 645/* sync_file_range always needs emulation */
849int 646static int
850eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags) 647eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
851{ 648{
852#if HAVE_SYNC_FILE_RANGE 649#if HAVE_SYNC_FILE_RANGE
853 int res; 650 int res;
854 651
871 /* even though we could play tricks with the flags, it's better to always 668 /* even though we could play tricks with the flags, it's better to always
872 * call fdatasync, as that matches the expectation of its users best */ 669 * call fdatasync, as that matches the expectation of its users best */
873 return fdatasync (fd); 670 return fdatasync (fd);
874} 671}
875 672
673static int
674eio__fallocate (int fd, int mode, off_t offset, size_t len)
675{
676#if HAVE_LINUX_FALLOCATE
677 return fallocate (fd, mode, offset, len);
678#else
679 return EIO_ENOSYS ();
680#endif
681}
682
876#if !HAVE_READAHEAD 683#if !HAVE_READAHEAD
877# undef readahead 684# undef readahead
878# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 685# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
879 686
880static ssize_t 687static eio_ssize_t
881eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) 688eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
882{ 689{
883 size_t todo = count; 690 size_t todo = count;
884 dBUF; 691 dBUF;
885 692
890 pread (fd, eio_buf, len, offset); 697 pread (fd, eio_buf, len, offset);
891 offset += len; 698 offset += len;
892 todo -= len; 699 todo -= len;
893 } 700 }
894 701
895 errno = 0; 702 FUBd;
703
704 /* linux's readahead basically only fails for EBADF or EINVAL (not mmappable) */
705 /* but not for e.g. EIO or eof, so we also never fail */
896 return count; 706 return 0;
897} 707}
898 708
899#endif 709#endif
900 710
901/* sendfile always needs emulation */ 711/* sendfile always needs emulation */
902static ssize_t 712static eio_ssize_t
903eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) 713eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
904{ 714{
715 eio_ssize_t written = 0;
905 ssize_t res; 716 eio_ssize_t res;
906 717
907 if (!count) 718 if (!count)
908 return 0; 719 return 0;
909 720
721 for (;;)
722 {
723#ifdef __APPLE__
724# undef HAVE_SENDFILE /* broken, as everything on os x */
725#endif
910#if HAVE_SENDFILE 726#if HAVE_SENDFILE
911# if __linux 727# if __linux
728 off_t soffset = offset;
912 res = sendfile (ofd, ifd, &offset, count); 729 res = sendfile (ofd, ifd, &soffset, count);
913 730
914# elif __freebsd 731# elif __FreeBSD__
915 /* 732 /*
916 * Of course, the freebsd sendfile is a dire hack with no thoughts 733 * Of course, the freebsd sendfile is a dire hack with no thoughts
917 * wasted on making it similar to other I/O functions. 734 * wasted on making it similar to other I/O functions.
918 */ 735 */
919 {
920 off_t sbytes; 736 off_t sbytes;
921 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 737 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
922 738
923 if (res < 0 && sbytes) 739 #if 0 /* according to the manpage, this is correct, but broken behaviour */
924 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 740 /* freebsd' sendfile will return 0 on success */
741 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
742 /* not on e.g. EIO or EPIPE - sounds broken */
743 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
925 res = sbytes; 744 res = sbytes;
926 } 745 #endif
927 746
747 /* according to source inspection, this is correct, and useful behaviour */
748 if (sbytes)
749 res = sbytes;
750
928# elif defined (__APPLE__) 751# elif defined __APPLE__
929
930 {
931 off_t sbytes = count; 752 off_t sbytes = count;
932 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0); 753 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
933 754
934 if (res < 0 && errno == EAGAIN && sbytes) 755 /* according to the manpage, sbytes is always valid */
756 if (sbytes)
935 res = sbytes; 757 res = sbytes;
936 }
937 758
938# elif __hpux 759# elif __hpux
939 res = sendfile (ofd, ifd, offset, count, 0, 0); 760 res = sendfile (ofd, ifd, offset, count, 0, 0);
940 761
941# elif __solaris 762# elif __solaris
942 {
943 struct sendfilevec vec; 763 struct sendfilevec vec;
944 size_t sbytes; 764 size_t sbytes;
945 765
946 vec.sfv_fd = ifd; 766 vec.sfv_fd = ifd;
947 vec.sfv_flag = 0; 767 vec.sfv_flag = 0;
948 vec.sfv_off = offset; 768 vec.sfv_off = offset;
949 vec.sfv_len = count; 769 vec.sfv_len = count;
950 770
951 res = sendfilev (ofd, &vec, 1, &sbytes); 771 res = sendfilev (ofd, &vec, 1, &sbytes);
952 772
953 if (res < 0 && sbytes) 773 if (res < 0 && sbytes)
954 res = sbytes; 774 res = sbytes;
955 }
956 775
957# endif 776# endif
958 777
959#elif defined (_WIN32) 778#elif defined (_WIN32) && 0
960
961 /* does not work, just for documentation of what would need to be done */ 779 /* does not work, just for documentation of what would need to be done */
962 { 780 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
781 /* libeio guarantees that the file offset does not change, and windows */
782 /* has no way to get an independent handle to the same file description */
963 HANDLE h = TO_SOCKET (ifd); 783 HANDLE h = TO_SOCKET (ifd);
964 SetFilePointer (h, offset, 0, FILE_BEGIN); 784 SetFilePointer (h, offset, 0, FILE_BEGIN);
965 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0); 785 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
966 }
967 786
968#else 787#else
969 res = -1; 788 res = EIO_ENOSYS ();
970 errno = ENOSYS;
971#endif 789#endif
972 790
791 /* we assume sendfile can copy at least 128mb in one go */
792 if (res <= 128 * 1024 * 1024)
793 {
794 if (res > 0)
795 written += res;
796
797 if (written)
798 return written;
799
800 break;
801 }
802 else
803 {
804 /* if we requested more, then probably the kernel was lazy */
805 written += res;
806 offset += res;
807 count -= res;
808
809 if (!count)
810 return written;
811 }
812 }
813
973 if (res < 0 814 if (res < 0
974 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 815 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
975 /* BSDs */ 816 /* BSDs */
976#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */ 817#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
977 || errno == ENOTSUP 818 || errno == ENOTSUP
978#endif 819#endif
820#ifdef EOPNOTSUPP /* windows */
979 || errno == EOPNOTSUPP /* BSDs */ 821 || errno == EOPNOTSUPP /* BSDs */
822#endif
980#if __solaris 823#if __solaris
981 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 824 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
982#endif 825#endif
983 ) 826 )
984 ) 827 )
988 831
989 res = 0; 832 res = 0;
990 833
991 while (count) 834 while (count)
992 { 835 {
993 ssize_t cnt; 836 eio_ssize_t cnt;
994 837
995 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 838 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
996 839
997 if (cnt <= 0) 840 if (cnt <= 0)
998 { 841 {
1010 853
1011 offset += cnt; 854 offset += cnt;
1012 res += cnt; 855 res += cnt;
1013 count -= cnt; 856 count -= cnt;
1014 } 857 }
858
859 FUBd;
1015 } 860 }
1016 861
1017 return res; 862 return res;
863}
864
865#ifdef PAGESIZE
866# define eio_pagesize() PAGESIZE
867#else
868static intptr_t
869eio_pagesize (void)
870{
871 static intptr_t page;
872
873 if (!page)
874 page = sysconf (_SC_PAGESIZE);
875
876 return page;
877}
878#endif
879
880static void
881eio_page_align (void **addr, size_t *length)
882{
883 intptr_t mask = eio_pagesize () - 1;
884
885 /* round down addr */
886 intptr_t adj = mask & (intptr_t)*addr;
887
888 *addr = (void *)((intptr_t)*addr - adj);
889 *length += adj;
890
891 /* round up length */
892 *length = (*length + mask) & ~mask;
893}
894
895#if !_POSIX_MEMLOCK
896# define eio__mlockall(a) EIO_ENOSYS ()
897#else
898
899static int
900eio__mlockall (int flags)
901{
902 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
903 extern int mallopt (int, int);
904 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
905 #endif
906
907 #ifndef MCL_ONFAULT
908 if (flags & EIO_MCL_ONFAULT)
909 return EIO_ERRNO (EINVAL, -1);
910 #define MCL_ONFAULT 4
911 #endif
912
913 if (EIO_MCL_CURRENT != MCL_CURRENT
914 || EIO_MCL_FUTURE != MCL_FUTURE
915 || EIO_MCL_ONFAULT != MCL_ONFAULT)
916 {
917 flags = 0
918 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
919 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0)
920 | (flags & EIO_MCL_ONFAULT ? MCL_ONFAULT : 0)
921 ;
922 }
923
924 return mlockall (flags);
925}
926#endif
927
928#if !_POSIX_MEMLOCK_RANGE
929# define eio__mlock(a,b) EIO_ENOSYS ()
930#else
931
932static int
933eio__mlock (void *addr, size_t length)
934{
935 eio_page_align (&addr, &length);
936
937 return mlock (addr, length);
938}
939
940#endif
941
942#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
943# define eio__msync(a,b,c) EIO_ENOSYS ()
944#else
945
946static int
947eio__msync (void *mem, size_t len, int flags)
948{
949 eio_page_align (&mem, &len);
950
951 if (EIO_MS_ASYNC != MS_SYNC
952 || EIO_MS_INVALIDATE != MS_INVALIDATE
953 || EIO_MS_SYNC != MS_SYNC)
954 {
955 flags = 0
956 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
957 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
958 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
959 }
960
961 return msync (mem, len, flags);
962}
963
964#endif
965
966static int
967eio__mtouch (eio_req *req)
968{
969 void *mem = req->ptr2;
970 size_t len = req->size;
971 int flags = req->int1;
972
973 eio_page_align (&mem, &len);
974
975 {
976 intptr_t addr = (intptr_t)mem;
977 intptr_t end = addr + len;
978 intptr_t page = eio_pagesize ();
979
980 if (addr < end)
981 if (flags & EIO_MT_MODIFY) /* modify */
982 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < end && !EIO_CANCELLED (req));
983 else
984 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < end && !EIO_CANCELLED (req));
985 }
986
987 return 0;
988}
989
990/*****************************************************************************/
991/* requests implemented outside eio_execute, because they are so large */
992
993static void
994eio__lseek (eio_req *req)
995{
996 /* this usually gets optimised away completely, or your compiler sucks, */
997 /* or the whence constants really are not 0, 1, 2 */
998 int whence = req->int2 == EIO_SEEK_SET ? SEEK_SET
999 : req->int2 == EIO_SEEK_CUR ? SEEK_CUR
1000 : req->int2 == EIO_SEEK_END ? SEEK_END
1001 : req->int2;
1002
1003 req->offs = lseek (req->int1, req->offs, whence);
1004 req->result = req->offs == (off_t)-1 ? -1 : 0;
1005}
1006
1007/* result will always end up in tmpbuf, there is always space for adding a 0-byte */
1008static int
1009eio__realpath (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
1010{
1011 char *res;
1012 const char *rel = path;
1013 char *tmp1, *tmp2;
1014#if SYMLOOP_MAX > 32
1015 int symlinks = SYMLOOP_MAX;
1016#else
1017 int symlinks = 32;
1018#endif
1019
1020 errno = EINVAL;
1021 if (!rel)
1022 return -1;
1023
1024 errno = ENOENT;
1025 if (!*rel)
1026 return -1;
1027
1028 res = etp_tmpbuf_get (tmpbuf, EIO_PATH_MAX * 3);
1029#ifdef _WIN32
1030 if (_access (rel, 4) != 0)
1031 return -1;
1032
1033 symlinks = GetFullPathName (rel, EIO_PATH_MAX * 3, res, 0);
1034
1035 errno = ENAMETOOLONG;
1036 if (symlinks >= EIO_PATH_MAX * 3)
1037 return -1;
1038
1039 errno = EIO;
1040 if (symlinks <= 0)
1041 return -1;
1042
1043 return symlinks;
1044
1045#else
1046 tmp1 = res + EIO_PATH_MAX;
1047 tmp2 = tmp1 + EIO_PATH_MAX;
1048
1049#if 0 /* disabled, the musl way to do things is just too racy */
1050#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1051 /* on linux we may be able to ask the kernel */
1052 {
1053 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1054
1055 if (fd >= 0)
1056 {
1057 sprintf (tmp1, "/proc/self/fd/%d", fd);
1058 req->result = readlink (tmp1, res, EIO_PATH_MAX);
1059 /* here we should probably stat the open file and the disk file, to make sure they still match */
1060 eio__close (fd);
1061
1062 if (req->result > 0)
1063 goto done;
1064 }
1065 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1066 return -1;
1067 }
1068#endif
1069#endif
1070
1071 if (*rel != '/')
1072 {
1073 int len;
1074
1075 errno = ENOENT;
1076 if (wd == EIO_INVALID_WD)
1077 return -1;
1078
1079 if (wd == EIO_CWD)
1080 {
1081 if (!getcwd (res, EIO_PATH_MAX))
1082 return -1;
1083
1084 len = strlen (res);
1085 }
1086 else
1087 memcpy (res, wd->str, len = wd->len);
1088
1089 if (res [1]) /* only use if not / */
1090 res += len;
1091 }
1092
1093 while (*rel)
1094 {
1095 eio_ssize_t len, linklen;
1096 const char *beg = rel;
1097
1098 while (*rel && *rel != '/')
1099 ++rel;
1100
1101 len = rel - beg;
1102
1103 if (!len) /* skip slashes */
1104 {
1105 ++rel;
1106 continue;
1107 }
1108
1109 if (beg [0] == '.')
1110 {
1111 if (len == 1)
1112 continue; /* . - nop */
1113
1114 if (beg [1] == '.' && len == 2)
1115 {
1116 /* .. - back up one component, if possible */
1117
1118 while (res != tmpbuf->ptr)
1119 if (*--res == '/')
1120 break;
1121
1122 continue;
1123 }
1124 }
1125
1126 errno = ENAMETOOLONG;
1127 if (res + 1 + len + 1 >= tmp1)
1128 return -1;
1129
1130 /* copy one component */
1131 *res = '/';
1132 memcpy (res + 1, beg, len);
1133
1134 /* zero-terminate, for readlink */
1135 res [len + 1] = 0;
1136
1137 /* now check if it's a symlink */
1138 linklen = readlink (tmpbuf->ptr, tmp1, EIO_PATH_MAX);
1139
1140 if (linklen < 0)
1141 {
1142 if (errno != EINVAL)
1143 return -1;
1144
1145 /* it's a normal directory. hopefully */
1146 res += len + 1;
1147 }
1148 else
1149 {
1150 /* yay, it was a symlink - build new path in tmp2 */
1151 int rellen = strlen (rel);
1152
1153 errno = ENAMETOOLONG;
1154 if (linklen + 1 + rellen >= EIO_PATH_MAX) /* also catch linklen >= EIO_PATH_MAX */
1155 return -1;
1156
1157 errno = ELOOP;
1158 if (!--symlinks)
1159 return -1;
1160
1161 if (*tmp1 == '/')
1162 res = tmpbuf->ptr; /* symlink resolves to an absolute path */
1163
1164 /* we need to be careful, as rel might point into tmp2 already */
1165 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1166 tmp2 [linklen] = '/';
1167 memcpy (tmp2, tmp1, linklen);
1168
1169 rel = tmp2;
1170 }
1171 }
1172
1173 /* special case for the lone root path */
1174 if (res == tmpbuf->ptr)
1175 *res++ = '/';
1176
1177 return res - (char *)tmpbuf->ptr;
1178#endif
1018} 1179}
1019 1180
1020static signed char 1181static signed char
1021eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) 1182eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1022{ 1183{
1023 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */ 1184 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; 1185 : a->inode < b->inode ? -1
1186 : a->inode > b->inode ? 1
1187 : 0;
1025} 1188}
1026 1189
1027#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0 1190#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1028 1191
1029#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */ 1192#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 */ 1193#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1031 1194
1032static void 1195static void
1033eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1196eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1034{ 1197{
1035 unsigned char bits [9 + sizeof (ino_t) * 8]; 1198 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1036 unsigned char *bit = bits; 1199 unsigned char *bit = bits;
1037 1200
1038 assert (CHAR_BIT == 8); 1201 assert (CHAR_BIT == 8);
1039 assert (sizeof (eio_dirent) * 8 < 256); 1202 assert (sizeof (eio_dirent) * 8 < 256);
1040 assert (offsetof (eio_dirent, inode)); /* we use 0 as sentinel */ 1203 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1041 assert (offsetof (eio_dirent, score)); /* we use 0 as sentinel */ 1204 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1042 1205
1043 if (size <= EIO_SORT_FAST) 1206 if (size <= EIO_SORT_FAST)
1044 return; 1207 return;
1045 1208
1046 /* first prepare an array of bits to test in our radix sort */ 1209 /* 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 */ 1210 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1048 /* inode_bits must contain all inodes ORed together */ 1211 /* inode_bits must contain all inodes ORed together */
1049 /* which is used to skip bits that are 0 everywhere, which is very common */ 1212 /* which is used to skip bits that are 0 everywhere, which is very common */
1050 { 1213 {
1051 ino_t endianness; 1214 eio_ino_t endianness;
1052 int i, j; 1215 int i, j;
1053 1216
1054 /* we store the byte offset of byte n into byte n of "endianness" */ 1217 /* we store the byte offset of byte n into byte n of "endianness" */
1055 for (i = 0; i < sizeof (ino_t); ++i) 1218 for (i = 0; i < sizeof (eio_ino_t); ++i)
1056 ((unsigned char *)&endianness)[i] = i; 1219 ((unsigned char *)&endianness)[i] = i;
1057 1220
1058 *bit++ = 0; 1221 *bit++ = 0;
1059 1222
1060 for (i = 0; i < sizeof (ino_t); ++i) 1223 for (i = 0; i < sizeof (eio_ino_t); ++i)
1061 { 1224 {
1062 /* shifting off the byte offsets out of "endianness" */ 1225 /* shifting off the byte offsets out of "endianness" */
1063 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8; 1226 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1064 endianness >>= 8; 1227 endianness >>= 8;
1065 1228
1066 for (j = 0; j < 8; ++j) 1229 for (j = 0; j < 8; ++j)
1067 if (inode_bits & (((ino_t)1) << (i * 8 + j))) 1230 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1068 *bit++ = offs + j; 1231 *bit++ = offs + j;
1069 } 1232 }
1070 1233
1071 for (j = 0; j < 8; ++j) 1234 for (j = 0; j < 8; ++j)
1072 if (score_bits & (1 << j)) 1235 if (score_bits & (1 << j))
1073 *bit++ = offsetof (eio_dirent, score) * 8 + j; 1236 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1074 } 1237 }
1075 1238
1076 /* now actually do the sorting (a variant of MSD radix sort) */ 1239 /* now actually do the sorting (a variant of MSD radix sort) */
1077 { 1240 {
1078 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base; 1241 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1079 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end; 1242 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1080 unsigned char *bit_stk [9 + sizeof (ino_t) * 8]; 1243 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1081 int stk_idx = 0; 1244 int stk_idx = 0;
1082 1245
1083 base_stk [stk_idx] = dents; 1246 base_stk [stk_idx] = dents;
1084 end_stk [stk_idx] = dents + size; 1247 end_stk [stk_idx] = dents + size;
1085 bit_stk [stk_idx] = bit - 1; 1248 bit_stk [stk_idx] = bit - 1;
1164 } 1327 }
1165 } 1328 }
1166} 1329}
1167 1330
1168static void 1331static void
1169eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1332eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1170{ 1333{
1171 if (size <= 1) 1334 if (size <= 1)
1172 return; /* our insertion sort relies on size > 0 */ 1335 return; /* our insertion sort relies on size > 0 */
1173 1336
1174 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */ 1337 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1182 1345
1183/* read a full directory */ 1346/* read a full directory */
1184static void 1347static void
1185eio__scandir (eio_req *req, etp_worker *self) 1348eio__scandir (eio_req *req, etp_worker *self)
1186{ 1349{
1187 DIR *dirp;
1188 EIO_STRUCT_DIRENT *entp;
1189 char *name, *names; 1350 char *name, *names;
1190 int namesalloc = 4096; 1351 int namesalloc = 4096 - sizeof (void *) * 4;
1191 int namesoffs = 0; 1352 int namesoffs = 0;
1192 int flags = req->int1; 1353 int flags = req->int1;
1193 eio_dirent *dents = 0; 1354 eio_dirent *dents = 0;
1194 int dentalloc = 128; 1355 int dentalloc = 128;
1195 int dentoffs = 0; 1356 int dentoffs = 0;
1196 ino_t inode_bits = 0; 1357 eio_ino_t inode_bits = 0;
1358#ifdef _WIN32
1359 HANDLE dirp;
1360 WIN32_FIND_DATA entp;
1361#else
1362 DIR *dirp;
1363 EIO_STRUCT_DIRENT *entp;
1364#endif
1197 1365
1198 req->result = -1; 1366 req->result = -1;
1199 1367
1200 if (!(flags & EIO_READDIR_DENTS)) 1368 if (!(flags & EIO_READDIR_DENTS))
1201 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); 1369 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1202 1370
1203 X_LOCK (wrklock); 1371#ifdef _WIN32
1204 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 1372 {
1205 self->dirp = dirp = opendir (req->ptr1); 1373 int len = strlen ((const char *)req->ptr1);
1374 char *path = malloc (MAX_PATH);
1375 const char *fmt;
1376 const char *reqpath = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1377
1378 if (!len)
1379 fmt = "./*";
1380 else if (reqpath[len - 1] == '/' || reqpath[len - 1] == '\\')
1381 fmt = "%s*";
1382 else
1383 fmt = "%s/*";
1384
1385 _snprintf (path, MAX_PATH, fmt, reqpath);
1386 dirp = FindFirstFile (path, &entp);
1387 free (path);
1388
1389 if (dirp == INVALID_HANDLE_VALUE)
1390 {
1391 /* should steal _dosmaperr */
1392 switch (GetLastError ())
1393 {
1394 case ERROR_FILE_NOT_FOUND:
1395 req->result = 0;
1396 break;
1397
1398 case ERROR_INVALID_NAME:
1399 case ERROR_PATH_NOT_FOUND:
1400 case ERROR_NO_MORE_FILES:
1401 errno = ENOENT;
1402 break;
1403
1404 case ERROR_NOT_ENOUGH_MEMORY:
1405 errno = ENOMEM;
1406 break;
1407
1408 default:
1409 errno = EINVAL;
1410 break;
1411 }
1412
1413 return;
1414 }
1415 }
1416#else
1417 #if HAVE_AT
1418 {
1419 int fd = openat (WD2FD (req->wd), req->ptr1, O_CLOEXEC | O_SEARCH | O_DIRECTORY | O_NONBLOCK);
1420
1421 if (fd < 0)
1422 return;
1423
1424 dirp = fdopendir (fd);
1425
1426 if (!dirp)
1427 {
1428 silent_close (fd);
1429 return;
1430 }
1431 }
1432 #else
1433 dirp = opendir (wd_expand (&self->tmpbuf, req->wd, req->ptr1));
1434
1435 if (!dirp)
1436 return;
1437 #endif
1438#endif
1439
1440 if (req->flags & EIO_FLAG_PTR1_FREE)
1441 free (req->ptr1);
1442
1206 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1443 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1207 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; 1444 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1208 req->ptr2 = names = malloc (namesalloc); 1445 req->ptr2 = names = malloc (namesalloc);
1209 X_UNLOCK (wrklock);
1210 1446
1211 if (dirp && names && (!flags || dents)) 1447 if (!names || (flags && !dents))
1448 return;
1449
1212 for (;;) 1450 for (;;)
1213 { 1451 {
1452 int done;
1453
1454#ifdef _WIN32
1455 done = !dirp;
1456#else
1214 errno = 0; 1457 errno = 0;
1215 entp = readdir (dirp); 1458 entp = readdir (dirp);
1459 done = !entp;
1460#endif
1216 1461
1217 if (!entp) 1462 if (done)
1218 { 1463 {
1464#ifndef _WIN32
1465 int old_errno = errno;
1466 closedir (dirp);
1467 errno = old_errno;
1468
1219 if (errno) 1469 if (errno)
1220 break; 1470 break;
1471#endif
1221 1472
1222 /* sort etc. */ 1473 /* sort etc. */
1223 req->int1 = flags; 1474 req->int1 = flags;
1224 req->result = dentoffs; 1475 req->result = dentoffs;
1225 1476
1226 if (flags & EIO_READDIR_STAT_ORDER) 1477 if (flags & EIO_READDIR_STAT_ORDER)
1227 eio_dent_sort (dents, dentoffs, 0, inode_bits); /* sort by inode exclusively */ 1478 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1228 else if (flags & EIO_READDIR_DIRS_FIRST) 1479 else if (flags & EIO_READDIR_DIRS_FIRST)
1229 if (flags & EIO_READDIR_FOUND_UNKNOWN) 1480 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1230 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */ 1481 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1231 else 1482 else
1483 {
1484 /* in this case, all is known, and we just put dirs first and sort them */
1485 eio_dirent *oth = dents + dentoffs;
1486 eio_dirent *dir = dents;
1487
1488 /* now partition dirs to the front, and non-dirs to the back */
1489 /* by walking from both sides and swapping if necessary */
1490 while (oth > dir)
1491 {
1492 if (dir->type == EIO_DT_DIR)
1493 ++dir;
1494 else if ((--oth)->type == EIO_DT_DIR)
1495 {
1496 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1497
1498 ++dir;
1499 }
1500 }
1501
1502 /* now sort the dirs only (dirs all have the same score) */
1503 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1504 }
1505
1506 break;
1507 }
1508
1509 /* now add the entry to our list(s) */
1510 name = D_NAME (entp);
1511
1512 /* skip . and .. entries */
1513 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1514 {
1515 int len = D_NAMLEN (entp) + 1;
1516
1517 while (ecb_expect_false (namesoffs + len > namesalloc))
1518 {
1519 namesalloc *= 2;
1520 req->ptr2 = names = realloc (names, namesalloc);
1521
1522 if (!names)
1523 break;
1524 }
1525
1526 memcpy (names + namesoffs, name, len);
1527
1528 if (dents)
1529 {
1530 struct eio_dirent *ent;
1531
1532 if (ecb_expect_false (dentoffs == dentalloc))
1232 { 1533 {
1233 /* in this case, all is known, and we just put dirs first and sort them */ 1534 dentalloc *= 2;
1535 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1536
1537 if (!dents)
1538 break;
1539 }
1540
1234 eio_dirent *oth = dents + dentoffs; 1541 ent = dents + dentoffs;
1235 eio_dirent *dir = dents;
1236 1542
1237 /* now partition dirs to the front, and non-dirs to the back */ 1543 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1238 /* by walking from both sides and swapping if necessary */ 1544 ent->namelen = len - 1;
1239 /* also clear score, so it doesn't influence sorting */ 1545 ent->inode = D_INO (entp);
1240 while (oth > dir) 1546
1547 inode_bits |= ent->inode;
1548
1549 switch (D_TYPE (entp))
1550 {
1551 default:
1552 ent->type = EIO_DT_UNKNOWN;
1553 flags |= EIO_READDIR_FOUND_UNKNOWN;
1554 break;
1555
1556 #ifdef DT_FIFO
1557 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1558 #endif
1559 #ifdef DT_CHR
1560 case DT_CHR: ent->type = EIO_DT_CHR; break;
1561 #endif
1562 #ifdef DT_MPC
1563 case DT_MPC: ent->type = EIO_DT_MPC; break;
1564 #endif
1565 #ifdef DT_DIR
1566 case DT_DIR: ent->type = EIO_DT_DIR; break;
1567 #endif
1568 #ifdef DT_NAM
1569 case DT_NAM: ent->type = EIO_DT_NAM; break;
1570 #endif
1571 #ifdef DT_BLK
1572 case DT_BLK: ent->type = EIO_DT_BLK; break;
1573 #endif
1574 #ifdef DT_MPB
1575 case DT_MPB: ent->type = EIO_DT_MPB; break;
1576 #endif
1577 #ifdef DT_REG
1578 case DT_REG: ent->type = EIO_DT_REG; break;
1579 #endif
1580 #ifdef DT_NWK
1581 case DT_NWK: ent->type = EIO_DT_NWK; break;
1582 #endif
1583 #ifdef DT_CMP
1584 case DT_CMP: ent->type = EIO_DT_CMP; break;
1585 #endif
1586 #ifdef DT_LNK
1587 case DT_LNK: ent->type = EIO_DT_LNK; break;
1588 #endif
1589 #ifdef DT_SOCK
1590 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1591 #endif
1592 #ifdef DT_DOOR
1593 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1594 #endif
1595 #ifdef DT_WHT
1596 case DT_WHT: ent->type = EIO_DT_WHT; break;
1597 #endif
1598 }
1599
1600 ent->score = 7;
1601
1602 if (flags & EIO_READDIR_DIRS_FIRST)
1603 {
1604 if (ent->type == EIO_DT_UNKNOWN)
1241 { 1605 {
1242 if (dir->type == EIO_DT_DIR) 1606 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1243 ++dir; 1607 ent->score = 1;
1244 else if ((--oth)->type == EIO_DT_DIR) 1608 else if (!strchr (name, '.')) /* absence of dots indicate likely dirs */
1245 { 1609 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 } 1610 }
1251 1611 else if (ent->type == EIO_DT_DIR)
1252 /* now sort the dirs only */ 1612 ent->score = 0;
1253 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1254 } 1613 }
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 } 1614 }
1277 1615
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; 1616 namesoffs += len;
1371 ++dentoffs; 1617 ++dentoffs;
1372 } 1618 }
1373 1619
1374 if (EIO_CANCELLED (req)) 1620 if (EIO_CANCELLED (req))
1375 { 1621 {
1376 errno = ECANCELED; 1622 errno = ECANCELED;
1377 break; 1623 break;
1378 } 1624 }
1625
1626#ifdef _WIN32
1627 if (!FindNextFile (dirp, &entp))
1628 {
1629 FindClose (dirp);
1630 dirp = 0;
1379 } 1631 }
1632#endif
1633 }
1380} 1634}
1381 1635
1382#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) 1636/*****************************************************************************/
1383# define eio__msync(a,b,c) ((errno = ENOSYS), -1) 1637/* working directory stuff */
1638/* various deficiencies in the posix 2008 api force us to */
1639/* keep the absolute path in string form at all times */
1640/* fuck yeah. */
1641
1642#if !HAVE_AT
1643
1644/* a bit like realpath, but usually faster because it doesn't have to return */
1645/* an absolute or canonical path */
1646static const char *
1647wd_expand (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
1648{
1649 if (!wd || *path == '/')
1650 return path;
1651
1652 if (path [0] == '.' && !path [1])
1653 return wd->str;
1654
1655 {
1656 int l1 = wd->len;
1657 int l2 = strlen (path);
1658
1659 char *res = etp_tmpbuf_get (tmpbuf, l1 + l2 + 2);
1660
1661 memcpy (res, wd->str, l1);
1662 res [l1] = '/';
1663 memcpy (res + l1 + 1, path, l2 + 1);
1664
1665 return res;
1666 }
1667}
1668
1669#endif
1670
1671static eio_wd
1672eio__wd_open_sync (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
1673{
1674 int fd;
1675 eio_wd res;
1676 int len = eio__realpath (tmpbuf, wd, path);
1677
1678 if (len < 0)
1679 return EIO_INVALID_WD;
1680
1681#if HAVE_AT
1682 fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY | O_NONBLOCK);
1683
1684 /* 0 is a valid fd, but we use it for EIO_CWD, so in the very unlikely */
1685 /* case of fd 0 being available (almost certainly an a pplication bug) */
1686 /* make sure we use another fd value */
1687 #if EIO_CWD
1688 error EIO_CWD must be 0
1689 #endif
1690 if (ecb_expect_false (fd == 0))
1691 {
1692 int fd2 = fcntl (fd, F_DUPFD_CLOEXEC ? F_DUPFD_CLOEXEC : F_DUPFD);
1693 fcntl (fd2, F_SETFD, FD_CLOEXEC);
1694 eio__close (fd);
1695 fd = fd2;
1696 }
1697
1698 if (fd < 0)
1699 return EIO_INVALID_WD;
1700#endif
1701
1702 res = malloc (sizeof (*res) + len); /* one extra 0-byte */
1703
1704#if HAVE_AT
1705 res->fd = fd;
1706#endif
1707
1708 res->len = len;
1709 memcpy (res->str, tmpbuf->ptr, len);
1710 res->str [len] = 0;
1711
1712 return res;
1713}
1714
1715eio_wd
1716eio_wd_open_sync (eio_wd wd, const char *path)
1717{
1718 struct etp_tmpbuf tmpbuf = { 0 };
1719 wd = eio__wd_open_sync (&tmpbuf, wd, path);
1720 free (tmpbuf.ptr);
1721
1722 return wd;
1723}
1724
1725void
1726eio_wd_close_sync (eio_wd wd)
1727{
1728 if (wd != EIO_INVALID_WD && wd != EIO_CWD)
1729 {
1730 #if HAVE_AT
1731 eio__close (wd->fd);
1732 #endif
1733 free (wd);
1734 }
1735}
1736
1737#if HAVE_AT
1738
1739static int
1740eio__renameat2 (int olddirfd, const char *oldpath, int newdirfd, const char *newpath, unsigned int flags)
1741{
1742#if HAVE_RENAMEAT2
1743 return syscall (SYS_renameat2, olddirfd, oldpath, newdirfd, newpath, flags);
1384#else 1744#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) 1745 if (flags)
1415 page = sysconf (_SC_PAGESIZE); 1746 return EIO_ENOSYS ();
1416#endif
1417 1747
1418 /* round down to start of page, although this is probably useless */ 1748 return renameat (olddirfd, oldpath, newdirfd, newpath);
1419 addr &= ~(page - 1); /* assume page size is always a power of two */ 1749#endif
1750}
1420 1751
1421 if (addr < end) 1752/* 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 1753
1754static int
1755eio__truncateat (int dirfd, const char *path, off_t length)
1756{
1757 int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC | O_NONBLOCK);
1758 int res;
1759
1760 if (fd < 0)
1761 return fd;
1762
1763 res = ftruncate (fd, length);
1764 silent_close (fd);
1427 return 0; 1765 return res;
1428} 1766}
1767
1768static int
1769eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
1770{
1771 int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC | O_NONBLOCK);
1772 int res;
1773
1774 if (fd < 0)
1775 return fd;
1776
1777 res = fstatvfs (fd, buf);
1778 silent_close (fd);
1779 return res;
1780}
1781
1782#endif
1429 1783
1430/*****************************************************************************/ 1784/*****************************************************************************/
1431 1785
1432#define ALLOC(len) \ 1786#define ALLOC(len) \
1433 if (!req->ptr2) \ 1787 if (!req->ptr2) \
1434 { \ 1788 { \
1435 X_LOCK (wrklock); \ 1789 X_LOCK (EIO_POOL->wrklock); \
1436 req->flags |= EIO_FLAG_PTR2_FREE; \ 1790 req->flags |= EIO_FLAG_PTR2_FREE; \
1437 X_UNLOCK (wrklock); \ 1791 X_UNLOCK (EIO_POOL->wrklock); \
1438 req->ptr2 = malloc (len); \ 1792 req->ptr2 = malloc (len); \
1439 if (!req->ptr2) \ 1793 if (!req->ptr2) \
1440 { \ 1794 { \
1441 errno = ENOMEM; \ 1795 errno = ENOMEM; \
1442 req->result = -1; \ 1796 req->result = -1; \
1443 break; \ 1797 goto alloc_fail; \
1444 } \ 1798 } \
1445 } 1799 }
1446 1800
1447X_THREAD_PROC (etp_proc) 1801/*****************************************************************************/
1448{
1449 ETP_REQ *req;
1450 struct timespec ts;
1451 etp_worker *self = (etp_worker *)thr_arg;
1452 1802
1453 /* try to distribute timeouts somewhat randomly */ 1803static void
1454 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 1804eio__slurp (int fd, eio_req *req)
1805{
1806 req->result = fd;
1455 1807
1456 for (;;) 1808 if (fd < 0)
1809 return;
1810
1811 if (req->offs < 0 || !req->size) /* do we need the size? */
1457 { 1812 {
1458 X_LOCK (reqlock); 1813 off_t size = lseek (fd, 0, SEEK_END);
1459 1814
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) 1815 if (req->offs < 0)
1494 goto quit; 1816 req->offs += size;
1495 1817
1496 if (!EIO_CANCELLED (req)) 1818 if (!req->size)
1497 ETP_EXECUTE (self, req); 1819 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 } 1820 }
1511 1821
1512quit: 1822 ALLOC (req->size);
1513 X_LOCK (wrklock); 1823 req->result = pread (fd, req->ptr2, req->size, req->offs);
1514 etp_worker_free (self);
1515 X_UNLOCK (wrklock);
1516 1824
1517 return 0; 1825 silent_close (fd);
1518}
1519 1826
1520/*****************************************************************************/ 1827alloc_fail:
1828 ;
1829}
1521 1830
1831int ecb_cold
1522int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1832eio_init (void (*want_poll)(void), void (*done_poll)(void))
1523{ 1833{
1524 return etp_init (want_poll, done_poll); 1834 eio_want_poll_cb = want_poll;
1525} 1835 eio_done_poll_cb = done_poll;
1526 1836
1837 return etp_init (EIO_POOL, 0, 0, 0);
1838}
1839
1840ecb_inline void
1527static void eio_api_destroy (eio_req *req) 1841eio_api_destroy (eio_req *req)
1528{ 1842{
1529 free (req); 1843 free (req);
1530} 1844}
1531 1845
1532#define REQ(rtype) \ 1846#define REQ(rtype) \
1533 eio_req *req; \ 1847 eio_req *req; \
1534 \ 1848 \
1535 req = (eio_req *)calloc (1, sizeof *req); \ 1849 req = (eio_req *)calloc (1, sizeof *req); \
1536 if (!req) \ 1850 if (!req) \
1537 return 0; \ 1851 return 0; \
1551 { \ 1865 { \
1552 eio_api_destroy (req); \ 1866 eio_api_destroy (req); \
1553 return 0; \ 1867 return 0; \
1554 } 1868 }
1555 1869
1870#define SINGLEDOT(ptr) (0[(char *)(ptr)] == '.' && !1[(char *)(ptr)])
1871
1872static void
1556static void eio_execute (etp_worker *self, eio_req *req) 1873eio_execute (etp_worker *self, eio_req *req)
1557{ 1874{
1558 errno = 0; 1875#if HAVE_AT
1876 int dirfd;
1877#else
1878 const char *path;
1879#endif
1880
1881 if (ecb_expect_false (EIO_CANCELLED (req)))
1882 {
1883 req->result = -1;
1884 req->errorno = ECANCELED;
1885 return;
1886 }
1887
1888 if (ecb_expect_false (req->wd == EIO_INVALID_WD))
1889 {
1890 req->result = -1;
1891 req->errorno = ENOENT;
1892 return;
1893 }
1894
1895 if (req->type >= EIO_OPEN)
1896 {
1897 #if HAVE_AT
1898 dirfd = WD2FD (req->wd);
1899 #else
1900 path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1901 #endif
1902 }
1559 1903
1560 switch (req->type) 1904 switch (req->type)
1561 { 1905 {
1906 case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
1907 req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
1908 break;
1909 case EIO_WD_CLOSE: req->result = 0;
1910 eio_wd_close_sync (req->wd); break;
1911
1912 case EIO_SEEK: eio__lseek (req); break;
1562 case EIO_READ: ALLOC (req->size); 1913 case EIO_READ: ALLOC (req->size);
1563 req->result = req->offs >= 0 1914 req->result = req->offs >= 0
1564 ? pread (req->int1, req->ptr2, req->size, req->offs) 1915 ? pread (req->int1, req->ptr2, req->size, req->offs)
1565 : read (req->int1, req->ptr2, req->size); break; 1916 : read (req->int1, req->ptr2, req->size); break;
1566 case EIO_WRITE: req->result = req->offs >= 0 1917 case EIO_WRITE: req->result = req->offs >= 0
1567 ? pwrite (req->int1, req->ptr2, req->size, req->offs) 1918 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1568 : write (req->int1, req->ptr2, req->size); break; 1919 : write (req->int1, req->ptr2, req->size); break;
1569 1920
1921 case EIO_FCNTL: req->result = fcntl (req->int1, (int) req->int2, req->ptr2); break;
1922 case EIO_IOCTL: req->result = ioctl (req->int1, (unsigned long)req->int2, req->ptr2); break;
1923
1570 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break; 1924 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; 1925 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
1926
1927#if HAVE_AT
1572 1928
1573 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1929 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1574 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 1930 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break;
1575 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1931 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1576 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 1932 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; 1933 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; 1934 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; 1935 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; 1936 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; 1937 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; 1938
1590 case EIO_UNLINK: req->result = unlink (req->ptr1); break; 1939 case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
1591 case EIO_RMDIR: req->result = rmdir (req->ptr1); break; 1940 case EIO_RMDIR: /* complications arise because "." cannot be removed, so we might have to expand */
1941 req->result = req->wd && SINGLEDOT (req->ptr1)
1942 ? rmdir (req->wd->str)
1943 : unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break;
1592 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; 1944 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; 1945 case EIO_RENAME: req->result = eio__renameat2 (
1946 dirfd,
1947 /* complications arise because "." cannot be renamed, so we might have to expand */
1948 req->wd && SINGLEDOT (req->ptr1) ? req->wd->str : req->ptr1,
1949 WD2FD ((eio_wd)req->int3),
1950 req->ptr2,
1951 req->int2
1952 );
1953 break;
1594 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; 1954 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; 1955 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; 1956 case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1597 1957 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1958 req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1598 case EIO_READLINK: ALLOC (NAME_MAX); 1959 case EIO_READLINK: ALLOC (EIO_PATH_MAX);
1599 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; 1960 req->result = readlinkat (dirfd, req->ptr1, req->ptr2, EIO_PATH_MAX);
1600 1961 if (req->result == EIO_PATH_MAX)
1601 case EIO_SYNC: req->result = 0; sync (); break; 1962 {
1602 case EIO_FSYNC: req->result = fsync (req->int1); break; 1963 req->result = -1;
1603 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 1964 errno = ENAMETOOLONG;
1604 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break; 1965 }
1605 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break; 1966 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: 1967 case EIO_UTIME:
1611#ifdef _WIN32 1968 case EIO_FUTIME:
1612 Sleep (req->nv1 * 1000.); 1969 {
1970 struct timespec ts[2];
1971 struct timespec *times;
1972
1973 if (req->nv1 != -1. || req->nv2 != -1.)
1974 {
1975 ts[0].tv_sec = req->nv1;
1976 ts[0].tv_nsec = (req->nv1 - ts[0].tv_sec) * 1e9;
1977 ts[1].tv_sec = req->nv2;
1978 ts[1].tv_nsec = (req->nv2 - ts[1].tv_sec) * 1e9;
1979
1980 times = ts;
1981 }
1982 else
1983 times = 0;
1984
1985 req->result = req->type == EIO_FUTIME
1986 ? futimens (req->int1, times)
1987 : utimensat (dirfd, req->ptr1, times, 0);
1988 }
1989 break;
1990
1613#else 1991#else
1614 {
1615 struct timeval tv;
1616 1992
1617 tv.tv_sec = req->nv1; 1993 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1618 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; 1994 req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
1995 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1996 req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
1997 case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
1998 case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
1999 case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
2000 case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
2001 case EIO_SLURP: eio__slurp ( open (path , O_RDONLY | O_CLOEXEC), req); break;
1619 2002
1620 req->result = select (0, 0, 0, 0, &tv); 2003 case EIO_UNLINK: req->result = unlink (path ); break;
1621 } 2004 case EIO_RMDIR: req->result = rmdir (path ); break;
1622#endif 2005 case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
1623 break; 2006 case EIO_RENAME: req->result = req->int2 ? EIO_ENOSYS () : rename (path, req->ptr2); break;
2007 case EIO_LINK: req->result = link (path , req->ptr2); break;
2008 case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
2009 case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
2010 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2011 req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
2012 case EIO_READLINK: ALLOC (EIO_PATH_MAX);
2013 req->result = readlink (path, req->ptr2, EIO_PATH_MAX);
2014 if (req->result == EIO_PATH_MAX)
2015 {
2016 req->result = -1;
2017 errno = ENAMETOOLONG;
2018 }
2019 break;
1624 2020
1625 case EIO_UTIME: 2021 case EIO_UTIME:
1626 case EIO_FUTIME: 2022 case EIO_FUTIME:
1627 { 2023 {
1628 struct timeval tv[2]; 2024 struct timeval tv[2];
1629 struct timeval *times; 2025 struct timeval *times;
1630 2026
1631 if (req->nv1 != -1. || req->nv2 != -1.) 2027 if (req->nv1 != -1. || req->nv2 != -1.)
1632 { 2028 {
1633 tv[0].tv_sec = req->nv1; 2029 tv[0].tv_sec = req->nv1;
1634 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.; 2030 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
1635 tv[1].tv_sec = req->nv2; 2031 tv[1].tv_sec = req->nv2;
1636 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.; 2032 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
1637 2033
1638 times = tv; 2034 times = tv;
1639 } 2035 }
1640 else 2036 else
1641 times = 0; 2037 times = 0;
1642
1643 2038
1644 req->result = req->type == EIO_FUTIME 2039 req->result = req->type == EIO_FUTIME
1645 ? futimes (req->int1, times) 2040 ? futimes (req->int1, times)
1646 : utimes (req->ptr1, times); 2041 : utimes (req->ptr1, times);
1647 } 2042 }
1648 break; 2043 break;
1649 2044
2045#endif
2046
2047 case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
2048 {
2049 ALLOC (req->result);
2050 memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
2051 }
2052 break;
2053
2054 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2055 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
2056
2057 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2058 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2059
2060 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
2061 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
2062 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
2063
2064 case EIO_CLOSE: req->result = eio__close (req->int1); break;
2065 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
2066 case EIO_SYNC: req->result = 0; sync (); break;
2067 case EIO_FSYNC: req->result = fsync (req->int1); break;
2068 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
2069 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
2070 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
2071 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
2072 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
2073 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
2074 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
2075 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
2076
2077 case EIO_READDIR: eio__scandir (req, self); break;
2078
2079 case EIO_BUSY:
2080#ifdef _WIN32
2081 Sleep (req->nv1 * 1e3);
2082#else
2083 {
2084 struct timeval tv;
2085
2086 tv.tv_sec = req->nv1;
2087 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
2088
2089 req->result = select (0, 0, 0, 0, &tv);
2090 }
2091#endif
2092 break;
2093
2094#if 0
1650 case EIO_GROUP: 2095 case EIO_GROUP:
1651 abort (); /* handled in eio_request */ 2096 abort (); /* handled in eio_request */
2097#endif
1652 2098
1653 case EIO_NOP: 2099 case EIO_NOP:
1654 req->result = 0; 2100 req->result = 0;
1655 break; 2101 break;
1656 2102
1657 case EIO_CUSTOM: 2103 case EIO_CUSTOM:
1658 ((void (*)(eio_req *))req->feed) (req); 2104 req->feed (req);
1659 break; 2105 break;
1660 2106
1661 default: 2107 default:
1662 req->result = -1; 2108 req->result = EIO_ENOSYS ();
1663 break; 2109 break;
1664 } 2110 }
1665 2111
2112alloc_fail:
1666 req->errorno = errno; 2113 req->errorno = errno;
1667} 2114}
1668 2115
1669#ifndef EIO_NO_WRAPPERS 2116#ifndef EIO_NO_WRAPPERS
1670 2117
2118eio_req *eio_wd_open (const char *path, int pri, eio_cb cb, void *data)
2119{
2120 REQ (EIO_WD_OPEN); PATH; SEND;
2121}
2122
2123eio_req *eio_wd_close (eio_wd wd, int pri, eio_cb cb, void *data)
2124{
2125 REQ (EIO_WD_CLOSE); req->wd = wd; SEND;
2126}
2127
1671eio_req *eio_nop (int pri, eio_cb cb, void *data) 2128eio_req *eio_nop (int pri, eio_cb cb, void *data)
1672{ 2129{
1673 REQ (EIO_NOP); SEND; 2130 REQ (EIO_NOP); SEND;
1674} 2131}
1675 2132
1691eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2148eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1692{ 2149{
1693 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2150 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1694} 2151}
1695 2152
2153eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2154{
2155 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2156}
2157
2158eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
2159{
2160 REQ (EIO_SYNCFS); req->int1 = fd; SEND;
2161}
2162
2163eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2164{
2165 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2166}
2167
1696eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2168eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1697{ 2169{
1698 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2170 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1699} 2171}
1700 2172
1701eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) 2173eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
1702{ 2174{
1703 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; 2175 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
1704} 2176}
1705 2177
1706eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) 2178eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
1707{ 2179{
1708 REQ (EIO_FDATASYNC); req->int1 = fd; SEND; 2180 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2181}
2182
2183eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2184{
2185 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
1709} 2186}
1710 2187
1711eio_req *eio_close (int fd, int pri, eio_cb cb, void *data) 2188eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
1712{ 2189{
1713 REQ (EIO_CLOSE); req->int1 = fd; SEND; 2190 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) 2193eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
1717{ 2194{
1718 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND; 2195 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
1719} 2196}
1720 2197
2198eio_req *eio_seek (int fd, off_t offset, int whence, int pri, eio_cb cb, void *data)
2199{
2200 REQ (EIO_SEEK); req->int1 = fd; req->offs = offset; req->int2 = whence; SEND;
2201}
2202
1721eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data) 2203eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1722{ 2204{
1723 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND; 2205 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1724} 2206}
1725 2207
1726eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data) 2208eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1727{ 2209{
1728 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND; 2210 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1729} 2211}
1730 2212
2213eio_req *eio_fcntl (int fd, int cmd, void *arg, int pri, eio_cb cb, void *data)
2214{
2215 REQ (EIO_IOCTL); req->int1 = fd; req->int2 = cmd; req->ptr2 = arg; SEND;
2216}
2217
2218eio_req *eio_ioctl (int fd, unsigned long request, void *buf, int pri, eio_cb cb, void *data)
2219{
2220 REQ (EIO_IOCTL); req->int1 = fd; req->int2 = request; req->ptr2 = buf; SEND;
2221}
2222
1731eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) 2223eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1732{ 2224{
1733 REQ (EIO_FSTAT); req->int1 = fd; SEND; 2225 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1734} 2226}
1735 2227
2228eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
2229{
2230 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
2231}
2232
1736eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) 2233eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1737{ 2234{
1738 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; 2235 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1739} 2236}
1740 2237
1746eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) 2243eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1747{ 2244{
1748 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2245 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1749} 2246}
1750 2247
1751eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2248eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1752{ 2249{
1753 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2250 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1754} 2251}
1755 2252
1756eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) 2253eio_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) 2273eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1777{ 2274{
1778 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2275 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1779} 2276}
1780 2277
1781eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2278eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1782{ 2279{
1783 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2280 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1784} 2281}
1785 2282
1786eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) 2283eio_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) 2299eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1803{ 2300{
1804 return eio__1path (EIO_READLINK, path, pri, cb, data); 2301 return eio__1path (EIO_READLINK, path, pri, cb, data);
1805} 2302}
1806 2303
2304eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2305{
2306 return eio__1path (EIO_REALPATH, path, pri, cb, data);
2307}
2308
1807eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) 2309eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1808{ 2310{
1809 return eio__1path (EIO_STAT, path, pri, cb, data); 2311 return eio__1path (EIO_STAT, path, pri, cb, data);
1810} 2312}
1811 2313
1812eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) 2314eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1813{ 2315{
1814 return eio__1path (EIO_LSTAT, path, pri, cb, data); 2316 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1815} 2317}
1816 2318
2319eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
2320{
2321 return eio__1path (EIO_STATVFS, path, pri, cb, data);
2322}
2323
1817eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) 2324eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1818{ 2325{
1819 return eio__1path (EIO_UNLINK, path, pri, cb, data); 2326 return eio__1path (EIO_UNLINK, path, pri, cb, data);
1820} 2327}
1821 2328
1829 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND; 2336 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1830} 2337}
1831 2338
1832eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) 2339eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1833{ 2340{
1834 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; 2341 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1835} 2342}
1836 2343
1837static eio_req * 2344static eio_req *
1838eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2345eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1839{ 2346{
1863eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2370eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1864{ 2371{
1865 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2372 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1866} 2373}
1867 2374
2375eio_req *eio_slurp (const char *path, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
2376{
2377 REQ (EIO_SLURP); PATH; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
2378}
2379
1868eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) 2380eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
1869{ 2381{
1870 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND; 2382 REQ (EIO_CUSTOM); req->feed = execute; SEND;
1871} 2383}
1872 2384
1873#endif 2385#endif
1874 2386
1875eio_req *eio_grp (eio_cb cb, void *data) 2387eio_req *eio_grp (eio_cb cb, void *data)
1884#undef SEND 2396#undef SEND
1885 2397
1886/*****************************************************************************/ 2398/*****************************************************************************/
1887/* grp functions */ 2399/* grp functions */
1888 2400
2401void
1889void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2402eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1890{ 2403{
1891 grp->int2 = limit; 2404 grp->int2 = limit;
1892 grp->feed = feed; 2405 grp->feed = feed;
1893 2406
1894 grp_try_feed (grp); 2407 grp_try_feed (grp);
1895} 2408}
1896 2409
2410void
1897void eio_grp_limit (eio_req *grp, int limit) 2411eio_grp_limit (eio_req *grp, int limit)
1898{ 2412{
1899 grp->int2 = limit; 2413 grp->int2 = limit;
1900 2414
1901 grp_try_feed (grp); 2415 grp_try_feed (grp);
1902} 2416}
1903 2417
2418void
1904void eio_grp_add (eio_req *grp, eio_req *req) 2419eio_grp_add (eio_req *grp, eio_req *req)
1905{ 2420{
1906 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2421 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1907 2422
1908 grp->flags |= EIO_FLAG_GROUPADD; 2423 grp->flags |= ETP_FLAG_GROUPADD;
1909 2424
1910 ++grp->size; 2425 ++grp->size;
1911 req->grp = grp; 2426 req->grp = grp;
1912 2427
1913 req->grp_prev = 0; 2428 req->grp_prev = 0;
1920} 2435}
1921 2436
1922/*****************************************************************************/ 2437/*****************************************************************************/
1923/* misc garbage */ 2438/* misc garbage */
1924 2439
2440eio_ssize_t
1925ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2441eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1926{ 2442{
1927 etp_worker wrk;
1928 ssize_t ret;
1929
1930 wrk.dbuf = 0;
1931
1932 ret = eio__sendfile (ofd, ifd, offset, count, &wrk); 2443 return eio__sendfile (ofd, ifd, offset, count);
1933
1934 if (wrk.dbuf)
1935 free (wrk.dbuf);
1936
1937 return ret;
1938} 2444}
1939 2445
2446int eio_mlockall_sync (int flags)
2447{
2448 return eio__mlockall (flags);
2449}
2450

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines