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

Comparing libeio/eio.c (file contents):
Revision 1.63 by root, Thu May 26 03:51:55 2011 UTC vs.
Revision 1.150 by root, Fri Jun 15 02:57:36 2018 UTC

1/* 1/*
2 * libeio implementation 2 * libeio implementation
3 * 3 *
4 * Copyright (c) 2007,2008,2009,2010,2011 Marc Alexander Lehmann <libeio@schmorp.de> 4 * Copyright (c) 2007,2008,2009,2010,2011,2012,2013,2016,2017 Marc Alexander Lehmann <libeio@schmorp.de>
5 * All rights reserved. 5 * All rights reserved.
6 * 6 *
7 * Redistribution and use in source and binary forms, with or without modifica- 7 * Redistribution and use in source and binary forms, with or without modifica-
8 * tion, are permitted provided that the following conditions are met: 8 * tion, are permitted provided that the following conditions are met:
9 * 9 *
10 * 1. Redistributions of source code must retain the above copyright notice, 10 * 1. Redistributions of source code must retain the above copyright notice,
11 * this list of conditions and the following disclaimer. 11 * this list of conditions and the following disclaimer.
12 * 12 *
13 * 2. Redistributions in binary form must reproduce the above copyright 13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the 14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution. 15 * documentation and/or other materials provided with the distribution.
16 * 16 *
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR IMPLIED 17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR IMPLIED
18 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MER- 18 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MER-
19 * CHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO 19 * CHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO
20 * EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPE- 20 * EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPE-
21 * CIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, 21 * CIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
35 * and other provisions required by the GPL. If you do not delete the 35 * and other provisions required by the GPL. If you do not delete the
36 * provisions above, a recipient may use your version of this file under 36 * provisions above, a recipient may use your version of this file under
37 * either the BSD or the GPL. 37 * either the BSD or the GPL.
38 */ 38 */
39 39
40#ifndef _WIN32
41# include "config.h"
42#endif
43
40#include "eio.h" 44#include "eio.h"
41
42#ifdef EIO_STACKSIZE
43# define XTHREAD_STACKSIZE EIO_STACKSIZE
44#endif
45#include "xthread.h" 45#include "ecb.h"
46 46
47#include <errno.h> 47#include <errno.h>
48#include <stddef.h> 48#include <stddef.h>
49#include <stdlib.h> 49#include <stdlib.h>
50#include <string.h> 50#include <string.h>
51#include <errno.h> 51#include <errno.h>
52#include <sys/types.h> 52#include <sys/types.h>
53#include <sys/stat.h> 53#include <sys/stat.h>
54#include <sys/statvfs.h>
55#include <limits.h> 54#include <limits.h>
56#include <fcntl.h> 55#include <fcntl.h>
57#include <assert.h> 56#include <assert.h>
58 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
59#ifndef EIO_FINISH 77#ifndef EIO_FINISH
60# 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
61#endif 79#endif
62 80
63#ifndef EIO_DESTROY 81#ifndef EIO_DESTROY
66 84
67#ifndef EIO_FEED 85#ifndef EIO_FEED
68# 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)
69#endif 87#endif
70 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
71#ifdef _WIN32 100#ifdef _WIN32
72 101
73 /*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
74#else 234#else
75 235
76# include "config.h" 236 #include <sys/ioctl.h>
77# include <sys/time.h> 237 #include <sys/time.h>
78# include <sys/select.h> 238 #include <sys/select.h>
79# include <unistd.h> 239 #include <unistd.h>
240 #include <signal.h>
241 #include <dirent.h>
242
243 #ifdef ANDROID
244 #include <sys/vfs.h>
245 #define statvfs statfs
246 #define fstatvfs fstatfs
247 #include <asm/page.h> /* supposedly limits.h does #define PAGESIZE PAGESIZE */
248 #else
249 #include <sys/statvfs.h>
250 #endif
251
252 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
253 #include <sys/mman.h>
254 #endif
255
256 #define D_NAME(entp) entp->d_name
257
258 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
259 #if __FreeBSD__ || __NetBSD__ || __OpenBSD__
260 #define _DIRENT_HAVE_D_TYPE /* sigh */
261 #define D_INO(de) (de)->d_fileno
262 #define D_NAMLEN(de) (de)->d_namlen
263 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
264 #define D_INO(de) (de)->d_ino
265 #endif
266
267 #ifdef _D_EXACT_NAMLEN
268 #undef D_NAMLEN
269 #define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
270 #endif
271
272 #ifdef _DIRENT_HAVE_D_TYPE
273 #define D_TYPE(de) (de)->d_type
274 #endif
275
276 #ifndef EIO_STRUCT_DIRENT
277 #define EIO_STRUCT_DIRENT struct dirent
278 #endif
279
280#endif
281
282#if HAVE_UTIMES
80# include <utime.h> 283# include <utime.h>
81# include <signal.h>
82# include <dirent.h>
83
84#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
85# include <sys/mman.h>
86#endif
87
88/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
89# if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
90# define _DIRENT_HAVE_D_TYPE /* sigh */
91# define D_INO(de) (de)->d_fileno
92# define D_NAMLEN(de) (de)->d_namlen
93# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
94# define D_INO(de) (de)->d_ino
95# endif 284#endif
96 285
97#ifdef _D_EXACT_NAMLEN 286#if HAVE_SYS_SYSCALL_H
98# undef D_NAMLEN 287# include <sys/syscall.h>
99# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
100#endif
101
102# ifdef _DIRENT_HAVE_D_TYPE
103# define D_TYPE(de) (de)->d_type
104# endif
105
106# ifndef EIO_STRUCT_DIRENT
107# define EIO_STRUCT_DIRENT struct dirent
108# endif
109
110#endif 288#endif
111 289
112#if HAVE_SENDFILE 290#if HAVE_SENDFILE
113# if __linux 291# if __linux
114# include <sys/sendfile.h> 292# include <sys/sendfile.h>
122# else 300# else
123# error sendfile support requested but not available 301# error sendfile support requested but not available
124# endif 302# endif
125#endif 303#endif
126 304
305#if HAVE_RENAMEAT2
306# include <sys/syscall.h>
307# include <linux/fs.h>
308#endif
309
127#ifndef D_TYPE 310#ifndef D_TYPE
128# define D_TYPE(de) 0 311# define D_TYPE(de) 0
129#endif 312#endif
130#ifndef D_INO 313#ifndef D_INO
131# define D_INO(de) 0 314# define D_INO(de) 0
132#endif 315#endif
133#ifndef D_NAMLEN 316#ifndef D_NAMLEN
134# define D_NAMLEN(de) strlen ((de)->d_name) 317# define D_NAMLEN(entp) strlen (D_NAME (entp))
135#endif 318#endif
136 319
137/* used for struct dirent, AIX doesn't provide it */ 320/* used for struct dirent, AIX doesn't provide it */
138#ifndef NAME_MAX 321#ifndef NAME_MAX
139# define NAME_MAX 4096 322# define NAME_MAX 4096
140#endif 323#endif
141 324
142/* used for readlink etc. */ 325/* used for readlink etc. */
143#ifndef PATH_MAX 326#ifndef PATH_MAX
144# define PATH_MAX 4096 327# define PATH_MAX 0
145#endif 328#endif
329
330#ifndef O_CLOEXEC
331 #define O_CLOEXEC 0
332#endif
333
334#ifndef EIO_PATH_MIN
335# define EIO_PATH_MIN 8160
336#endif
337
338#define EIO_PATH_MAX (PATH_MAX <= EIO_PATH_MIN ? EIO_PATH_MIN : PATH_MAX)
146 339
147/* buffer size for various temporary buffers */ 340/* buffer size for various temporary buffers */
148#define EIO_BUFSIZE 65536 341#define EIO_BUFSIZE 65536
149 342
150#define dBUF \ 343#define dBUF \
151 char *eio_buf; \
152 ETP_WORKER_LOCK (self); \
153 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 344 char *eio_buf = malloc (EIO_BUFSIZE); \
154 ETP_WORKER_UNLOCK (self); \
155 errno = ENOMEM; \ 345 errno = ENOMEM; \
156 if (!eio_buf) \ 346 if (!eio_buf) \
157 return -1; 347 return -1
158 348
159#define EIO_TICKS ((1000000 + 1023) >> 10) 349#define FUBd \
350 free (eio_buf)
160 351
161/*****************************************************************************/ 352/*****************************************************************************/
162 353
163#if __GNUC__ >= 3 354struct etp_tmpbuf;
164# define expect(expr,value) __builtin_expect ((expr),(value)) 355
356#if _POSIX_VERSION >= 200809L
357 #define HAVE_AT 1
358 #define WD2FD(wd) ((wd) ? (wd)->fd : AT_FDCWD)
359 #ifndef O_SEARCH
360 #define O_SEARCH O_RDONLY
361 #endif
165#else 362#else
166# define expect(expr,value) (expr) 363 #define HAVE_AT 0
364 static const char *wd_expand (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path);
167#endif 365#endif
168 366
169#define expect_false(expr) expect ((expr) != 0, 0) 367struct eio_pwd
170#define expect_true(expr) expect ((expr) != 0, 1) 368{
369#if HAVE_AT
370 int fd;
371#endif
372 int len;
373 char str[1]; /* actually, a 0-terminated canonical path */
374};
171 375
172/*****************************************************************************/ 376/*****************************************************************************/
173 377
174#define ETP_PRI_MIN EIO_PRI_MIN 378#define ETP_PRI_MIN EIO_PRI_MIN
175#define ETP_PRI_MAX EIO_PRI_MAX 379#define ETP_PRI_MAX EIO_PRI_MAX
176 380
381#define ETP_TYPE_QUIT -1
382#define ETP_TYPE_GROUP EIO_GROUP
383
384static void eio_nop_callback (void) { }
385static void (*eio_want_poll_cb)(void) = eio_nop_callback;
386static void (*eio_done_poll_cb)(void) = eio_nop_callback;
387
388#define ETP_WANT_POLL(pool) eio_want_poll_cb ()
389#define ETP_DONE_POLL(pool) eio_done_poll_cb ()
390
177struct etp_worker; 391struct etp_worker;
178
179#define ETP_REQ eio_req 392#define ETP_REQ eio_req
180#define ETP_DESTROY(req) eio_destroy (req) 393#define ETP_DESTROY(req) eio_destroy (req)
181static int eio_finish (eio_req *req); 394static int eio_finish (eio_req *req);
182#define ETP_FINISH(req) eio_finish (req) 395#define ETP_FINISH(req) eio_finish (req)
183static void eio_execute (struct etp_worker *self, eio_req *req); 396static void eio_execute (struct etp_worker *self, eio_req *req);
184#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) 397#define ETP_EXECUTE(wrk,req) eio_execute (wrk, req)
185 398
186#define ETP_WORKER_CLEAR(req) \ 399#include "etp.c"
187 if (wrk->dbuf) \
188 { \
189 free (wrk->dbuf); \
190 wrk->dbuf = 0; \
191 } \
192 \
193 if (wrk->dirp) \
194 { \
195 closedir (wrk->dirp); \
196 wrk->dirp = 0; \
197 }
198 400
199#define ETP_WORKER_COMMON \ 401static struct etp_pool eio_pool;
200 void *dbuf; \ 402#define EIO_POOL (&eio_pool)
201 DIR *dirp;
202 403
203/*****************************************************************************/ 404/*****************************************************************************/
204 405
205#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
206
207/* calculate time difference in ~1/EIO_TICKS of a second */
208static int tvdiff (struct timeval *tv1, struct timeval *tv2)
209{
210 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
211 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
212}
213
214static unsigned int started, idle, wanted = 4;
215
216static void (*want_poll_cb) (void);
217static void (*done_poll_cb) (void);
218
219static unsigned int max_poll_time; /* reslock */
220static unsigned int max_poll_reqs; /* reslock */
221
222static volatile unsigned int nreqs; /* reqlock */
223static volatile unsigned int nready; /* reqlock */
224static volatile unsigned int npending; /* reqlock */
225static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
226static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
227
228static xmutex_t wrklock;
229static xmutex_t reslock;
230static xmutex_t reqlock;
231static xcond_t reqwait;
232
233#if !HAVE_PREADWRITE
234/*
235 * make our pread/pwrite emulation safe against themselves, but not against
236 * normal read/write by using a mutex. slows down execution a lot,
237 * but that's your problem, not mine.
238 */
239static xmutex_t preadwritelock = X_MUTEX_INIT;
240#endif
241
242typedef struct etp_worker
243{
244 /* locked by wrklock */
245 struct etp_worker *prev, *next;
246
247 xthread_t tid;
248
249 /* locked by reslock, reqlock or wrklock */
250 ETP_REQ *req; /* currently processed request */
251
252 ETP_WORKER_COMMON
253} etp_worker;
254
255static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */
256
257#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
258#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
259
260/* worker threads management */
261
262static void etp_worker_clear (etp_worker *wrk)
263{
264 ETP_WORKER_CLEAR (wrk);
265}
266
267static void etp_worker_free (etp_worker *wrk)
268{
269 wrk->next->prev = wrk->prev;
270 wrk->prev->next = wrk->next;
271
272 free (wrk);
273}
274
275static unsigned int etp_nreqs (void)
276{
277 int retval;
278 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
279 retval = nreqs;
280 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
281 return retval;
282}
283
284static unsigned int etp_nready (void)
285{
286 unsigned int retval;
287
288 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
289 retval = nready;
290 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
291
292 return retval;
293}
294
295static unsigned int etp_npending (void)
296{
297 unsigned int retval;
298
299 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
300 retval = npending;
301 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
302
303 return retval;
304}
305
306static unsigned int etp_nthreads (void)
307{
308 unsigned int retval;
309
310 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
311 retval = started;
312 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
313
314 return retval;
315}
316
317/*
318 * a somewhat faster data structure might be nice, but
319 * with 8 priorities this actually needs <20 insns
320 * per shift, the most expensive operation.
321 */
322typedef struct {
323 ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */
324 int size;
325} etp_reqq;
326
327static etp_reqq req_queue;
328static etp_reqq res_queue;
329
330static int reqq_push (etp_reqq *q, ETP_REQ *req)
331{
332 int pri = req->pri;
333 req->next = 0;
334
335 if (q->qe[pri])
336 {
337 q->qe[pri]->next = req;
338 q->qe[pri] = req;
339 }
340 else
341 q->qe[pri] = q->qs[pri] = req;
342
343 return q->size++;
344}
345
346static ETP_REQ *reqq_shift (etp_reqq *q)
347{
348 int pri;
349
350 if (!q->size)
351 return 0;
352
353 --q->size;
354
355 for (pri = ETP_NUM_PRI; pri--; )
356 {
357 eio_req *req = q->qs[pri];
358
359 if (req)
360 {
361 if (!(q->qs[pri] = (eio_req *)req->next))
362 q->qe[pri] = 0;
363
364 return req;
365 }
366 }
367
368 abort ();
369}
370
371static void etp_thread_init (void)
372{
373 X_MUTEX_CREATE (wrklock);
374 X_MUTEX_CREATE (reslock);
375 X_MUTEX_CREATE (reqlock);
376 X_COND_CREATE (reqwait);
377}
378
379static void etp_atfork_prepare (void)
380{
381 X_LOCK (wrklock);
382 X_LOCK (reqlock);
383 X_LOCK (reslock);
384#if !HAVE_PREADWRITE
385 X_LOCK (preadwritelock);
386#endif
387}
388
389static void etp_atfork_parent (void)
390{
391#if !HAVE_PREADWRITE
392 X_UNLOCK (preadwritelock);
393#endif
394 X_UNLOCK (reslock);
395 X_UNLOCK (reqlock);
396 X_UNLOCK (wrklock);
397}
398
399static void etp_atfork_child (void)
400{
401 ETP_REQ *prv;
402
403 while ((prv = reqq_shift (&req_queue)))
404 ETP_DESTROY (prv);
405
406 while ((prv = reqq_shift (&res_queue)))
407 ETP_DESTROY (prv);
408
409 while (wrk_first.next != &wrk_first)
410 {
411 etp_worker *wrk = wrk_first.next;
412
413 if (wrk->req)
414 ETP_DESTROY (wrk->req);
415
416 etp_worker_clear (wrk);
417 etp_worker_free (wrk);
418 }
419
420 started = 0;
421 idle = 0;
422 nreqs = 0;
423 nready = 0;
424 npending = 0;
425
426 etp_thread_init ();
427}
428
429static void 406static void
430etp_once_init (void)
431{
432 etp_thread_init ();
433 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
434}
435
436static int
437etp_init (void (*want_poll)(void), void (*done_poll)(void))
438{
439 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
440
441 pthread_once (&doinit, etp_once_init);
442
443 want_poll_cb = want_poll;
444 done_poll_cb = done_poll;
445
446 return 0;
447}
448
449X_THREAD_PROC (etp_proc);
450
451static void etp_start_thread (void)
452{
453 etp_worker *wrk = calloc (1, sizeof (etp_worker));
454
455 /*TODO*/
456 assert (("unable to allocate worker thread data", wrk));
457
458 X_LOCK (wrklock);
459
460 if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
461 {
462 wrk->prev = &wrk_first;
463 wrk->next = wrk_first.next;
464 wrk_first.next->prev = wrk;
465 wrk_first.next = wrk;
466 ++started;
467 }
468 else
469 free (wrk);
470
471 X_UNLOCK (wrklock);
472}
473
474static void etp_maybe_start_thread (void)
475{
476 if (expect_true (etp_nthreads () >= wanted))
477 return;
478
479 /* todo: maybe use idle here, but might be less exact */
480 if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
481 return;
482
483 etp_start_thread ();
484}
485
486static void etp_end_thread (void)
487{
488 eio_req *req = calloc (1, sizeof (eio_req));
489
490 req->type = -1;
491 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
492
493 X_LOCK (reqlock);
494 reqq_push (&req_queue, req);
495 X_COND_SIGNAL (reqwait);
496 X_UNLOCK (reqlock);
497
498 X_LOCK (wrklock);
499 --started;
500 X_UNLOCK (wrklock);
501}
502
503static int etp_poll (void)
504{
505 unsigned int maxreqs;
506 unsigned int maxtime;
507 struct timeval tv_start, tv_now;
508
509 X_LOCK (reslock);
510 maxreqs = max_poll_reqs;
511 maxtime = max_poll_time;
512 X_UNLOCK (reslock);
513
514 if (maxtime)
515 gettimeofday (&tv_start, 0);
516
517 for (;;)
518 {
519 ETP_REQ *req;
520
521 etp_maybe_start_thread ();
522
523 X_LOCK (reslock);
524 req = reqq_shift (&res_queue);
525
526 if (req)
527 {
528 --npending;
529
530 if (!res_queue.size && done_poll_cb)
531 done_poll_cb ();
532 }
533
534 X_UNLOCK (reslock);
535
536 if (!req)
537 return 0;
538
539 X_LOCK (reqlock);
540 --nreqs;
541 X_UNLOCK (reqlock);
542
543 if (expect_false (req->type == EIO_GROUP && req->size))
544 {
545 req->int1 = 1; /* mark request as delayed */
546 continue;
547 }
548 else
549 {
550 int res = ETP_FINISH (req);
551 if (expect_false (res))
552 return res;
553 }
554
555 if (expect_false (maxreqs && !--maxreqs))
556 break;
557
558 if (maxtime)
559 {
560 gettimeofday (&tv_now, 0);
561
562 if (tvdiff (&tv_start, &tv_now) >= maxtime)
563 break;
564 }
565 }
566
567 errno = EAGAIN;
568 return -1;
569}
570
571static void etp_cancel (ETP_REQ *req)
572{
573 X_LOCK (wrklock);
574 req->flags |= EIO_FLAG_CANCELLED;
575 X_UNLOCK (wrklock);
576
577 eio_grp_cancel (req);
578}
579
580static void etp_submit (ETP_REQ *req)
581{
582 req->pri -= ETP_PRI_MIN;
583
584 if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
585 if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
586
587 if (expect_false (req->type == EIO_GROUP))
588 {
589 /* I hope this is worth it :/ */
590 X_LOCK (reqlock);
591 ++nreqs;
592 X_UNLOCK (reqlock);
593
594 X_LOCK (reslock);
595
596 ++npending;
597
598 if (!reqq_push (&res_queue, req) && want_poll_cb)
599 want_poll_cb ();
600
601 X_UNLOCK (reslock);
602 }
603 else
604 {
605 X_LOCK (reqlock);
606 ++nreqs;
607 ++nready;
608 reqq_push (&req_queue, req);
609 X_COND_SIGNAL (reqwait);
610 X_UNLOCK (reqlock);
611
612 etp_maybe_start_thread ();
613 }
614}
615
616static void etp_set_max_poll_time (double nseconds)
617{
618 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
619 max_poll_time = nseconds * EIO_TICKS;
620 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
621}
622
623static void etp_set_max_poll_reqs (unsigned int maxreqs)
624{
625 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
626 max_poll_reqs = maxreqs;
627 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
628}
629
630static void etp_set_max_idle (unsigned int nthreads)
631{
632 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
633 max_idle = nthreads;
634 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
635}
636
637static void etp_set_idle_timeout (unsigned int seconds)
638{
639 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
640 idle_timeout = seconds;
641 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
642}
643
644static void etp_set_min_parallel (unsigned int nthreads)
645{
646 if (wanted < nthreads)
647 wanted = nthreads;
648}
649
650static void etp_set_max_parallel (unsigned int nthreads)
651{
652 if (wanted > nthreads)
653 wanted = nthreads;
654
655 while (started > wanted)
656 etp_end_thread ();
657}
658
659/*****************************************************************************/
660
661static void grp_try_feed (eio_req *grp) 407grp_try_feed (eio_req *grp)
662{ 408{
663 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 409 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
664 { 410 {
665 grp->flags &= ~EIO_FLAG_GROUPADD; 411 grp->flags &= ~ETP_FLAG_GROUPADD;
666 412
667 EIO_FEED (grp); 413 EIO_FEED (grp);
668 414
669 /* stop if no progress has been made */ 415 /* stop if no progress has been made */
670 if (!(grp->flags & EIO_FLAG_GROUPADD)) 416 if (!(grp->flags & ETP_FLAG_GROUPADD))
671 { 417 {
672 grp->feed = 0; 418 grp->feed = 0;
673 break; 419 break;
674 } 420 }
675 } 421 }
676} 422}
677 423
424static int
678static int grp_dec (eio_req *grp) 425grp_dec (eio_req *grp)
679{ 426{
680 --grp->size; 427 --grp->size;
681 428
682 /* call feeder, if applicable */ 429 /* call feeder, if applicable */
683 grp_try_feed (grp); 430 grp_try_feed (grp);
684 431
685 /* finish, if done */ 432 /* finish, if done */
686 if (!grp->size && grp->int1) 433 if (!grp->size && grp->flags & ETP_FLAG_DELAYED)
687 return eio_finish (grp); 434 return eio_finish (grp);
688 else 435 else
689 return 0; 436 return 0;
690} 437}
691 438
439static void
692void eio_destroy (eio_req *req) 440eio_destroy (eio_req *req)
693{ 441{
694 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); 442 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
695 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); 443 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
696 444
697 EIO_DESTROY (req); 445 EIO_DESTROY (req);
698} 446}
699 447
448static int
700static int eio_finish (eio_req *req) 449eio_finish (eio_req *req)
701{ 450{
702 int res = EIO_FINISH (req); 451 int res = EIO_FINISH (req);
703 452
704 if (req->grp) 453 if (req->grp)
705 { 454 {
713 if (grp->grp_first == req) 462 if (grp->grp_first == req)
714 grp->grp_first = req->grp_next; 463 grp->grp_first = req->grp_next;
715 464
716 res2 = grp_dec (grp); 465 res2 = grp_dec (grp);
717 466
718 if (!res && res2) 467 if (!res)
719 res = res2; 468 res = res2;
720 } 469 }
721 470
722 eio_destroy (req); 471 eio_destroy (req);
723 472
724 return res; 473 return res;
725} 474}
726 475
476void
727void eio_grp_cancel (eio_req *grp) 477eio_grp_cancel (eio_req *grp)
728{ 478{
729 for (grp = grp->grp_first; grp; grp = grp->grp_next) 479 etp_grp_cancel (EIO_POOL, grp);
730 eio_cancel (grp);
731} 480}
732 481
482void
733void eio_cancel (eio_req *req) 483eio_cancel (eio_req *req)
734{ 484{
735 etp_cancel (req); 485 etp_cancel (EIO_POOL, req);
736} 486}
737 487
488void
738void eio_submit (eio_req *req) 489eio_submit (eio_req *req)
739{ 490{
740 etp_submit (req); 491 etp_submit (EIO_POOL, req);
741} 492}
742 493
743unsigned int eio_nreqs (void) 494unsigned int
495eio_nreqs (void)
744{ 496{
745 return etp_nreqs (); 497 return etp_nreqs (EIO_POOL);
746} 498}
747 499
748unsigned int eio_nready (void) 500unsigned int
501eio_nready (void)
749{ 502{
750 return etp_nready (); 503 return etp_nready (EIO_POOL);
751} 504}
752 505
753unsigned int eio_npending (void) 506unsigned int
507eio_npending (void)
754{ 508{
755 return etp_npending (); 509 return etp_npending (EIO_POOL);
756} 510}
757 511
758unsigned int eio_nthreads (void) 512unsigned int ecb_cold
513eio_nthreads (void)
759{ 514{
760 return etp_nthreads (); 515 return etp_nthreads (EIO_POOL);
761} 516}
762 517
518void ecb_cold
763void eio_set_max_poll_time (double nseconds) 519eio_set_max_poll_time (double nseconds)
764{ 520{
765 etp_set_max_poll_time (nseconds); 521 etp_set_max_poll_time (EIO_POOL, nseconds);
766} 522}
767 523
524void ecb_cold
768void eio_set_max_poll_reqs (unsigned int maxreqs) 525eio_set_max_poll_reqs (unsigned int maxreqs)
769{ 526{
770 etp_set_max_poll_reqs (maxreqs); 527 etp_set_max_poll_reqs (EIO_POOL, maxreqs);
771} 528}
772 529
530void ecb_cold
773void eio_set_max_idle (unsigned int nthreads) 531eio_set_max_idle (unsigned int nthreads)
774{ 532{
775 etp_set_max_idle (nthreads); 533 etp_set_max_idle (EIO_POOL, nthreads);
776} 534}
777 535
536void ecb_cold
778void eio_set_idle_timeout (unsigned int seconds) 537eio_set_idle_timeout (unsigned int seconds)
779{ 538{
780 etp_set_idle_timeout (seconds); 539 etp_set_idle_timeout (EIO_POOL, seconds);
781} 540}
782 541
542void ecb_cold
783void eio_set_min_parallel (unsigned int nthreads) 543eio_set_min_parallel (unsigned int nthreads)
784{ 544{
785 etp_set_min_parallel (nthreads); 545 etp_set_min_parallel (EIO_POOL, nthreads);
786} 546}
787 547
548void ecb_cold
788void eio_set_max_parallel (unsigned int nthreads) 549eio_set_max_parallel (unsigned int nthreads)
789{ 550{
790 etp_set_max_parallel (nthreads); 551 etp_set_max_parallel (EIO_POOL, nthreads);
791} 552}
792 553
793int eio_poll (void) 554int eio_poll (void)
794{ 555{
795 return etp_poll (); 556 return etp_poll (EIO_POOL);
796} 557}
797 558
798/*****************************************************************************/ 559/*****************************************************************************/
799/* work around various missing functions */ 560/* work around various missing functions */
800 561
801#if !HAVE_PREADWRITE 562#if HAVE_POSIX_CLOSE && !__linux
802# undef pread 563# define eio__close(fd) posix_close (fd, 0)
803# undef pwrite 564#else
804# define pread eio__pread 565# define eio__close(fd) close (fd)
805# define pwrite eio__pwrite
806
807static ssize_t
808eio__pread (int fd, void *buf, size_t count, off_t offset)
809{
810 ssize_t res;
811 off_t ooffset;
812
813 X_LOCK (preadwritelock);
814 ooffset = lseek (fd, 0, SEEK_CUR);
815 lseek (fd, offset, SEEK_SET);
816 res = read (fd, buf, count);
817 lseek (fd, ooffset, SEEK_SET);
818 X_UNLOCK (preadwritelock);
819
820 return res;
821}
822
823static ssize_t
824eio__pwrite (int fd, void *buf, size_t count, off_t offset)
825{
826 ssize_t res;
827 off_t ooffset;
828
829 X_LOCK (preadwritelock);
830 ooffset = lseek (fd, 0, SEEK_CUR);
831 lseek (fd, offset, SEEK_SET);
832 res = write (fd, buf, count);
833 lseek (fd, ooffset, SEEK_SET);
834 X_UNLOCK (preadwritelock);
835
836 return res;
837}
838#endif 566#endif
567
568/* close() without disturbing errno */
569static void
570silent_close (int fd)
571{
572 int saved_errno = errno;
573 eio__close (fd);
574 errno = saved_errno;
575}
839 576
840#ifndef HAVE_UTIMES 577#ifndef HAVE_UTIMES
841 578
842# undef utimes 579# undef utimes
843# define utimes(path,times) eio__utimes (path, times) 580# define utimes(path,times) eio__utimes (path, times)
863#ifndef HAVE_FUTIMES 600#ifndef HAVE_FUTIMES
864 601
865# undef futimes 602# undef futimes
866# define futimes(fd,times) eio__futimes (fd, times) 603# define futimes(fd,times) eio__futimes (fd, times)
867 604
605static int
868static int eio__futimes (int fd, const struct timeval tv[2]) 606eio__futimes (int fd, const struct timeval tv[2])
869{ 607{
870 errno = ENOSYS; 608 errno = ENOSYS;
871 return -1; 609 return -1;
872} 610}
873 611
876#if !HAVE_FDATASYNC 614#if !HAVE_FDATASYNC
877# undef fdatasync 615# undef fdatasync
878# define fdatasync(fd) fsync (fd) 616# define fdatasync(fd) fsync (fd)
879#endif 617#endif
880 618
619static int
620eio__syncfs (int fd)
621{
622 int res;
623
624#if HAVE_SYS_SYNCFS
625 res = (int)syscall (__NR_syncfs, (int)(fd));
626#else
627 res = EIO_ENOSYS ();
628#endif
629
630 if (res < 0 && errno == ENOSYS && fd >= 0)
631 sync ();
632
633 return res;
634}
635
881/* sync_file_range always needs emulation */ 636/* sync_file_range always needs emulation */
882int 637static int
883eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags) 638eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
884{ 639{
885#if HAVE_SYNC_FILE_RANGE 640#if HAVE_SYNC_FILE_RANGE
886 int res; 641 int res;
887 642
904 /* even though we could play tricks with the flags, it's better to always 659 /* even though we could play tricks with the flags, it's better to always
905 * call fdatasync, as that matches the expectation of its users best */ 660 * call fdatasync, as that matches the expectation of its users best */
906 return fdatasync (fd); 661 return fdatasync (fd);
907} 662}
908 663
664static int
665eio__fallocate (int fd, int mode, off_t offset, size_t len)
666{
667#if HAVE_LINUX_FALLOCATE
668 return fallocate (fd, mode, offset, len);
669#else
670 return EIO_ENOSYS ();
671#endif
672}
673
909#if !HAVE_READAHEAD 674#if !HAVE_READAHEAD
910# undef readahead 675# undef readahead
911# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 676# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
912 677
913static ssize_t 678static eio_ssize_t
914eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) 679eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
915{ 680{
916 size_t todo = count; 681 size_t todo = count;
917 dBUF; 682 dBUF;
918 683
923 pread (fd, eio_buf, len, offset); 688 pread (fd, eio_buf, len, offset);
924 offset += len; 689 offset += len;
925 todo -= len; 690 todo -= len;
926 } 691 }
927 692
928 errno = 0; 693 FUBd;
694
695 /* linux's readahead basically only fails for EBADF or EINVAL (not mmappable) */
696 /* but not for e.g. EIO or eof, so we also never fail */
929 return count; 697 return 0;
930} 698}
931 699
932#endif 700#endif
933 701
934/* sendfile always needs emulation */ 702/* sendfile always needs emulation */
935static ssize_t 703static eio_ssize_t
936eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) 704eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
937{ 705{
706 eio_ssize_t written = 0;
938 ssize_t res; 707 eio_ssize_t res;
939 708
940 if (!count) 709 if (!count)
941 return 0; 710 return 0;
942 711
712 for (;;)
713 {
714#ifdef __APPLE__
715# undef HAVE_SENDFILE /* broken, as everything on os x */
716#endif
943#if HAVE_SENDFILE 717#if HAVE_SENDFILE
944# if __linux 718# if __linux
719 off_t soffset = offset;
945 res = sendfile (ofd, ifd, &offset, count); 720 res = sendfile (ofd, ifd, &soffset, count);
946 721
947# elif __FreeBSD__ 722# elif __FreeBSD__
948 /* 723 /*
949 * Of course, the freebsd sendfile is a dire hack with no thoughts 724 * Of course, the freebsd sendfile is a dire hack with no thoughts
950 * wasted on making it similar to other I/O functions. 725 * wasted on making it similar to other I/O functions.
951 */ 726 */
952 {
953 off_t sbytes; 727 off_t sbytes;
954 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 728 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
955 729
956 #if 0 /* according to the manpage, this is correct, but broken behaviour */ 730 #if 0 /* according to the manpage, this is correct, but broken behaviour */
957 /* freebsd' sendfile will return 0 on success */ 731 /* freebsd' sendfile will return 0 on success */
958 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */ 732 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
959 /* not on e.g. EIO or EPIPE - sounds broken */ 733 /* not on e.g. EIO or EPIPE - sounds broken */
960 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0) 734 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
961 res = sbytes; 735 res = sbytes;
962 #endif 736 #endif
963 737
964 /* according to source inspection, this is correct, and useful behaviour */ 738 /* according to source inspection, this is correct, and useful behaviour */
965 if (sbytes) 739 if (sbytes)
966 res = sbytes; 740 res = sbytes;
967 }
968 741
969# elif defined (__APPLE__) 742# elif defined __APPLE__
970
971 {
972 off_t sbytes = count; 743 off_t sbytes = count;
973 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0); 744 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
974 745
975 /* according to the manpage, sbytes is always valid */ 746 /* according to the manpage, sbytes is always valid */
976 if (sbytes) 747 if (sbytes)
977 res = sbytes; 748 res = sbytes;
978 }
979 749
980# elif __hpux 750# elif __hpux
981 res = sendfile (ofd, ifd, offset, count, 0, 0); 751 res = sendfile (ofd, ifd, offset, count, 0, 0);
982 752
983# elif __solaris 753# elif __solaris
984 {
985 struct sendfilevec vec; 754 struct sendfilevec vec;
986 size_t sbytes; 755 size_t sbytes;
987 756
988 vec.sfv_fd = ifd; 757 vec.sfv_fd = ifd;
989 vec.sfv_flag = 0; 758 vec.sfv_flag = 0;
990 vec.sfv_off = offset; 759 vec.sfv_off = offset;
991 vec.sfv_len = count; 760 vec.sfv_len = count;
992 761
993 res = sendfilev (ofd, &vec, 1, &sbytes); 762 res = sendfilev (ofd, &vec, 1, &sbytes);
994 763
995 if (res < 0 && sbytes) 764 if (res < 0 && sbytes)
996 res = sbytes; 765 res = sbytes;
997 }
998 766
999# endif 767# endif
1000 768
1001#elif defined (_WIN32) 769#elif defined (_WIN32) && 0
1002
1003 /* does not work, just for documentation of what would need to be done */ 770 /* does not work, just for documentation of what would need to be done */
1004 { 771 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
772 /* libeio guarantees that the file offset does not change, and windows */
773 /* has no way to get an independent handle to the same file description */
1005 HANDLE h = TO_SOCKET (ifd); 774 HANDLE h = TO_SOCKET (ifd);
1006 SetFilePointer (h, offset, 0, FILE_BEGIN); 775 SetFilePointer (h, offset, 0, FILE_BEGIN);
1007 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0); 776 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1008 }
1009 777
1010#else 778#else
1011 res = -1; 779 res = EIO_ENOSYS ();
1012 errno = ENOSYS;
1013#endif 780#endif
1014 781
782 /* we assume sendfile can copy at least 128mb in one go */
783 if (res <= 128 * 1024 * 1024)
784 {
785 if (res > 0)
786 written += res;
787
788 if (written)
789 return written;
790
791 break;
792 }
793 else
794 {
795 /* if we requested more, then probably the kernel was lazy */
796 written += res;
797 offset += res;
798 count -= res;
799
800 if (!count)
801 return written;
802 }
803 }
804
1015 if (res < 0 805 if (res < 0
1016 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 806 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1017 /* BSDs */ 807 /* BSDs */
1018#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */ 808#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1019 || errno == ENOTSUP 809 || errno == ENOTSUP
1020#endif 810#endif
811#ifdef EOPNOTSUPP /* windows */
1021 || errno == EOPNOTSUPP /* BSDs */ 812 || errno == EOPNOTSUPP /* BSDs */
813#endif
1022#if __solaris 814#if __solaris
1023 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 815 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
1024#endif 816#endif
1025 ) 817 )
1026 ) 818 )
1030 822
1031 res = 0; 823 res = 0;
1032 824
1033 while (count) 825 while (count)
1034 { 826 {
1035 ssize_t cnt; 827 eio_ssize_t cnt;
1036 828
1037 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 829 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
1038 830
1039 if (cnt <= 0) 831 if (cnt <= 0)
1040 { 832 {
1052 844
1053 offset += cnt; 845 offset += cnt;
1054 res += cnt; 846 res += cnt;
1055 count -= cnt; 847 count -= cnt;
1056 } 848 }
849
850 FUBd;
1057 } 851 }
1058 852
1059 return res; 853 return res;
854}
855
856#ifdef PAGESIZE
857# define eio_pagesize() PAGESIZE
858#else
859static intptr_t
860eio_pagesize (void)
861{
862 static intptr_t page;
863
864 if (!page)
865 page = sysconf (_SC_PAGESIZE);
866
867 return page;
868}
869#endif
870
871static void
872eio_page_align (void **addr, size_t *length)
873{
874 intptr_t mask = eio_pagesize () - 1;
875
876 /* round down addr */
877 intptr_t adj = mask & (intptr_t)*addr;
878
879 *addr = (void *)((intptr_t)*addr - adj);
880 *length += adj;
881
882 /* round up length */
883 *length = (*length + mask) & ~mask;
884}
885
886#if !_POSIX_MEMLOCK
887# define eio__mlockall(a) EIO_ENOSYS ()
888#else
889
890static int
891eio__mlockall (int flags)
892{
893 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
894 extern int mallopt (int, int);
895 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
896 #endif
897
898 if (EIO_MCL_CURRENT != MCL_CURRENT
899 || EIO_MCL_FUTURE != MCL_FUTURE)
900 {
901 flags = 0
902 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
903 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
904 }
905
906 return mlockall (flags);
907}
908#endif
909
910#if !_POSIX_MEMLOCK_RANGE
911# define eio__mlock(a,b) EIO_ENOSYS ()
912#else
913
914static int
915eio__mlock (void *addr, size_t length)
916{
917 eio_page_align (&addr, &length);
918
919 return mlock (addr, length);
920}
921
922#endif
923
924#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
925# define eio__msync(a,b,c) EIO_ENOSYS ()
926#else
927
928static int
929eio__msync (void *mem, size_t len, int flags)
930{
931 eio_page_align (&mem, &len);
932
933 if (EIO_MS_ASYNC != MS_SYNC
934 || EIO_MS_INVALIDATE != MS_INVALIDATE
935 || EIO_MS_SYNC != MS_SYNC)
936 {
937 flags = 0
938 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
939 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
940 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
941 }
942
943 return msync (mem, len, flags);
944}
945
946#endif
947
948static int
949eio__mtouch (eio_req *req)
950{
951 void *mem = req->ptr2;
952 size_t len = req->size;
953 int flags = req->int1;
954
955 eio_page_align (&mem, &len);
956
957 {
958 intptr_t addr = (intptr_t)mem;
959 intptr_t end = addr + len;
960 intptr_t page = eio_pagesize ();
961
962 if (addr < end)
963 if (flags & EIO_MT_MODIFY) /* modify */
964 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < end && !EIO_CANCELLED (req));
965 else
966 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < end && !EIO_CANCELLED (req));
967 }
968
969 return 0;
970}
971
972/*****************************************************************************/
973/* requests implemented outside eio_execute, because they are so large */
974
975static void
976eio__lseek (eio_req *req)
977{
978 /* this usually gets optimised away completely, or your compiler sucks, */
979 /* or the whence constants really are not 0, 1, 2 */
980 int whence = req->int2 == EIO_SEEK_SET ? SEEK_SET
981 : req->int2 == EIO_SEEK_CUR ? SEEK_CUR
982 : req->int2 == EIO_SEEK_END ? SEEK_END
983 : req->int2;
984
985 req->offs = lseek (req->int1, req->offs, whence);
986 req->result = req->offs == (off_t)-1 ? -1 : 0;
987}
988
989/* result will always end up in tmpbuf, there is always space for adding a 0-byte */
990static int
991eio__realpath (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
992{
993 char *res;
994 const char *rel = path;
995 char *tmp1, *tmp2;
996#if SYMLOOP_MAX > 32
997 int symlinks = SYMLOOP_MAX;
998#else
999 int symlinks = 32;
1000#endif
1001
1002 errno = EINVAL;
1003 if (!rel)
1004 return -1;
1005
1006 errno = ENOENT;
1007 if (!*rel)
1008 return -1;
1009
1010 res = etp_tmpbuf_get (tmpbuf, EIO_PATH_MAX * 3);
1011#ifdef _WIN32
1012 if (_access (rel, 4) != 0)
1013 return -1;
1014
1015 symlinks = GetFullPathName (rel, EIO_PATH_MAX * 3, res, 0);
1016
1017 errno = ENAMETOOLONG;
1018 if (symlinks >= EIO_PATH_MAX * 3)
1019 return -1;
1020
1021 errno = EIO;
1022 if (symlinks <= 0)
1023 return -1;
1024
1025 return symlinks;
1026
1027#else
1028 tmp1 = res + EIO_PATH_MAX;
1029 tmp2 = tmp1 + EIO_PATH_MAX;
1030
1031#if 0 /* disabled, the musl way to do things is just too racy */
1032#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1033 /* on linux we may be able to ask the kernel */
1034 {
1035 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1036
1037 if (fd >= 0)
1038 {
1039 sprintf (tmp1, "/proc/self/fd/%d", fd);
1040 req->result = readlink (tmp1, res, EIO_PATH_MAX);
1041 /* here we should probably stat the open file and the disk file, to make sure they still match */
1042 eio__close (fd);
1043
1044 if (req->result > 0)
1045 goto done;
1046 }
1047 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1048 return -1;
1049 }
1050#endif
1051#endif
1052
1053 if (*rel != '/')
1054 {
1055 int len;
1056
1057 errno = ENOENT;
1058 if (wd == EIO_INVALID_WD)
1059 return -1;
1060
1061 if (wd == EIO_CWD)
1062 {
1063 if (!getcwd (res, EIO_PATH_MAX))
1064 return -1;
1065
1066 len = strlen (res);
1067 }
1068 else
1069 memcpy (res, wd->str, len = wd->len);
1070
1071 if (res [1]) /* only use if not / */
1072 res += len;
1073 }
1074
1075 while (*rel)
1076 {
1077 eio_ssize_t len, linklen;
1078 const char *beg = rel;
1079
1080 while (*rel && *rel != '/')
1081 ++rel;
1082
1083 len = rel - beg;
1084
1085 if (!len) /* skip slashes */
1086 {
1087 ++rel;
1088 continue;
1089 }
1090
1091 if (beg [0] == '.')
1092 {
1093 if (len == 1)
1094 continue; /* . - nop */
1095
1096 if (beg [1] == '.' && len == 2)
1097 {
1098 /* .. - back up one component, if possible */
1099
1100 while (res != tmpbuf->ptr)
1101 if (*--res == '/')
1102 break;
1103
1104 continue;
1105 }
1106 }
1107
1108 errno = ENAMETOOLONG;
1109 if (res + 1 + len + 1 >= tmp1)
1110 return -1;
1111
1112 /* copy one component */
1113 *res = '/';
1114 memcpy (res + 1, beg, len);
1115
1116 /* zero-terminate, for readlink */
1117 res [len + 1] = 0;
1118
1119 /* now check if it's a symlink */
1120 linklen = readlink (tmpbuf->ptr, tmp1, EIO_PATH_MAX);
1121
1122 if (linklen < 0)
1123 {
1124 if (errno != EINVAL)
1125 return -1;
1126
1127 /* it's a normal directory. hopefully */
1128 res += len + 1;
1129 }
1130 else
1131 {
1132 /* yay, it was a symlink - build new path in tmp2 */
1133 int rellen = strlen (rel);
1134
1135 errno = ENAMETOOLONG;
1136 if (linklen + 1 + rellen >= EIO_PATH_MAX) /* also catch linklen >= EIO_PATH_MAX */
1137 return -1;
1138
1139 errno = ELOOP;
1140 if (!--symlinks)
1141 return -1;
1142
1143 if (*tmp1 == '/')
1144 res = tmpbuf->ptr; /* symlink resolves to an absolute path */
1145
1146 /* we need to be careful, as rel might point into tmp2 already */
1147 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1148 tmp2 [linklen] = '/';
1149 memcpy (tmp2, tmp1, linklen);
1150
1151 rel = tmp2;
1152 }
1153 }
1154
1155 /* special case for the lone root path */
1156 if (res == tmpbuf->ptr)
1157 *res++ = '/';
1158
1159 return res - (char *)tmpbuf->ptr;
1160#endif
1060} 1161}
1061 1162
1062static signed char 1163static signed char
1063eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) 1164eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1064{ 1165{
1072 1173
1073#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */ 1174#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1074#define EIO_SORT_FAST 60 /* when to only use insertion sort */ 1175#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1075 1176
1076static void 1177static void
1077eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1178eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1078{ 1179{
1079 unsigned char bits [9 + sizeof (ino_t) * 8]; 1180 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1080 unsigned char *bit = bits; 1181 unsigned char *bit = bits;
1081 1182
1082 assert (CHAR_BIT == 8); 1183 assert (CHAR_BIT == 8);
1083 assert (sizeof (eio_dirent) * 8 < 256); 1184 assert (sizeof (eio_dirent) * 8 < 256);
1084 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */ 1185 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1086 1187
1087 if (size <= EIO_SORT_FAST) 1188 if (size <= EIO_SORT_FAST)
1088 return; 1189 return;
1089 1190
1090 /* first prepare an array of bits to test in our radix sort */ 1191 /* first prepare an array of bits to test in our radix sort */
1091 /* try to take endianness into account, as well as differences in ino_t sizes */ 1192 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1092 /* inode_bits must contain all inodes ORed together */ 1193 /* inode_bits must contain all inodes ORed together */
1093 /* which is used to skip bits that are 0 everywhere, which is very common */ 1194 /* which is used to skip bits that are 0 everywhere, which is very common */
1094 { 1195 {
1095 ino_t endianness; 1196 eio_ino_t endianness;
1096 int i, j; 1197 int i, j;
1097 1198
1098 /* we store the byte offset of byte n into byte n of "endianness" */ 1199 /* we store the byte offset of byte n into byte n of "endianness" */
1099 for (i = 0; i < sizeof (ino_t); ++i) 1200 for (i = 0; i < sizeof (eio_ino_t); ++i)
1100 ((unsigned char *)&endianness)[i] = i; 1201 ((unsigned char *)&endianness)[i] = i;
1101 1202
1102 *bit++ = 0; 1203 *bit++ = 0;
1103 1204
1104 for (i = 0; i < sizeof (ino_t); ++i) 1205 for (i = 0; i < sizeof (eio_ino_t); ++i)
1105 { 1206 {
1106 /* shifting off the byte offsets out of "endianness" */ 1207 /* shifting off the byte offsets out of "endianness" */
1107 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8; 1208 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1108 endianness >>= 8; 1209 endianness >>= 8;
1109 1210
1110 for (j = 0; j < 8; ++j) 1211 for (j = 0; j < 8; ++j)
1111 if (inode_bits & (((ino_t)1) << (i * 8 + j))) 1212 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1112 *bit++ = offs + j; 1213 *bit++ = offs + j;
1113 } 1214 }
1114 1215
1115 for (j = 0; j < 8; ++j) 1216 for (j = 0; j < 8; ++j)
1116 if (score_bits & (1 << j)) 1217 if (score_bits & (1 << j))
1117 *bit++ = offsetof (eio_dirent, score) * 8 + j; 1218 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1118 } 1219 }
1119 1220
1120 /* now actually do the sorting (a variant of MSD radix sort) */ 1221 /* now actually do the sorting (a variant of MSD radix sort) */
1121 { 1222 {
1122 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base; 1223 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1123 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end; 1224 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1124 unsigned char *bit_stk [9 + sizeof (ino_t) * 8]; 1225 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1125 int stk_idx = 0; 1226 int stk_idx = 0;
1126 1227
1127 base_stk [stk_idx] = dents; 1228 base_stk [stk_idx] = dents;
1128 end_stk [stk_idx] = dents + size; 1229 end_stk [stk_idx] = dents + size;
1129 bit_stk [stk_idx] = bit - 1; 1230 bit_stk [stk_idx] = bit - 1;
1208 } 1309 }
1209 } 1310 }
1210} 1311}
1211 1312
1212static void 1313static void
1213eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1314eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1214{ 1315{
1215 if (size <= 1) 1316 if (size <= 1)
1216 return; /* our insertion sort relies on size > 0 */ 1317 return; /* our insertion sort relies on size > 0 */
1217 1318
1218 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */ 1319 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1226 1327
1227/* read a full directory */ 1328/* read a full directory */
1228static void 1329static void
1229eio__scandir (eio_req *req, etp_worker *self) 1330eio__scandir (eio_req *req, etp_worker *self)
1230{ 1331{
1231 DIR *dirp;
1232 EIO_STRUCT_DIRENT *entp;
1233 char *name, *names; 1332 char *name, *names;
1234 int namesalloc = 4096; 1333 int namesalloc = 4096 - sizeof (void *) * 4;
1235 int namesoffs = 0; 1334 int namesoffs = 0;
1236 int flags = req->int1; 1335 int flags = req->int1;
1237 eio_dirent *dents = 0; 1336 eio_dirent *dents = 0;
1238 int dentalloc = 128; 1337 int dentalloc = 128;
1239 int dentoffs = 0; 1338 int dentoffs = 0;
1240 ino_t inode_bits = 0; 1339 eio_ino_t inode_bits = 0;
1340#ifdef _WIN32
1341 HANDLE dirp;
1342 WIN32_FIND_DATA entp;
1343#else
1344 DIR *dirp;
1345 EIO_STRUCT_DIRENT *entp;
1346#endif
1241 1347
1242 req->result = -1; 1348 req->result = -1;
1243 1349
1244 if (!(flags & EIO_READDIR_DENTS)) 1350 if (!(flags & EIO_READDIR_DENTS))
1245 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); 1351 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1246 1352
1247 X_LOCK (wrklock); 1353#ifdef _WIN32
1248 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 1354 {
1355 int len = strlen ((const char *)req->ptr1);
1356 char *path = malloc (MAX_PATH);
1357 const char *fmt;
1358 const char *reqpath = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1359
1360 if (!len)
1361 fmt = "./*";
1362 else if (reqpath[len - 1] == '/' || reqpath[len - 1] == '\\')
1363 fmt = "%s*";
1364 else
1365 fmt = "%s/*";
1366
1367 _snprintf (path, MAX_PATH, fmt, reqpath);
1368 dirp = FindFirstFile (path, &entp);
1369 free (path);
1370
1371 if (dirp == INVALID_HANDLE_VALUE)
1372 {
1373 /* should steal _dosmaperr */
1374 switch (GetLastError ())
1375 {
1376 case ERROR_FILE_NOT_FOUND:
1377 req->result = 0;
1378 break;
1379
1380 case ERROR_INVALID_NAME:
1381 case ERROR_PATH_NOT_FOUND:
1382 case ERROR_NO_MORE_FILES:
1383 errno = ENOENT;
1384 break;
1385
1386 case ERROR_NOT_ENOUGH_MEMORY:
1387 errno = ENOMEM;
1388 break;
1389
1390 default:
1391 errno = EINVAL;
1392 break;
1393 }
1394
1395 return;
1396 }
1397 }
1398#else
1399 #if HAVE_AT
1400 if (req->wd)
1401 {
1402 int fd = openat (WD2FD (req->wd), req->ptr1, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1403
1404 if (fd < 0)
1405 return;
1406
1407 dirp = fdopendir (fd);
1408
1409 if (!dirp)
1410 silent_close (fd);
1411 }
1412 else
1249 self->dirp = dirp = opendir (req->ptr1); 1413 dirp = opendir (req->ptr1);
1414 #else
1415 dirp = opendir (wd_expand (&self->tmpbuf, req->wd, req->ptr1));
1416 #endif
1417
1418 if (!dirp)
1419 return;
1420#endif
1421
1422 if (req->flags & EIO_FLAG_PTR1_FREE)
1423 free (req->ptr1);
1250 1424
1251 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1425 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1252 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; 1426 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1253 req->ptr2 = names = malloc (namesalloc); 1427 req->ptr2 = names = malloc (namesalloc);
1254 X_UNLOCK (wrklock);
1255 1428
1256 if (dirp && names && (!flags || dents)) 1429 if (!names || (flags && !dents))
1430 return;
1431
1257 for (;;) 1432 for (;;)
1258 { 1433 {
1434 int done;
1435
1436#ifdef _WIN32
1437 done = !dirp;
1438#else
1259 errno = 0; 1439 errno = 0;
1260 entp = readdir (dirp); 1440 entp = readdir (dirp);
1441 done = !entp;
1442#endif
1261 1443
1262 if (!entp) 1444 if (done)
1263 { 1445 {
1446#ifndef _WIN32
1447 int old_errno = errno;
1448 closedir (dirp);
1449 errno = old_errno;
1450
1264 if (errno) 1451 if (errno)
1265 break; 1452 break;
1453#endif
1266 1454
1267 /* sort etc. */ 1455 /* sort etc. */
1268 req->int1 = flags; 1456 req->int1 = flags;
1269 req->result = dentoffs; 1457 req->result = dentoffs;
1270 1458
1271 if (flags & EIO_READDIR_STAT_ORDER) 1459 if (flags & EIO_READDIR_STAT_ORDER)
1272 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits); 1460 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1273 else if (flags & EIO_READDIR_DIRS_FIRST) 1461 else if (flags & EIO_READDIR_DIRS_FIRST)
1274 if (flags & EIO_READDIR_FOUND_UNKNOWN) 1462 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1275 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */ 1463 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1276 else 1464 else
1465 {
1466 /* in this case, all is known, and we just put dirs first and sort them */
1467 eio_dirent *oth = dents + dentoffs;
1468 eio_dirent *dir = dents;
1469
1470 /* now partition dirs to the front, and non-dirs to the back */
1471 /* by walking from both sides and swapping if necessary */
1472 while (oth > dir)
1473 {
1474 if (dir->type == EIO_DT_DIR)
1475 ++dir;
1476 else if ((--oth)->type == EIO_DT_DIR)
1477 {
1478 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1479
1480 ++dir;
1481 }
1482 }
1483
1484 /* now sort the dirs only (dirs all have the same score) */
1485 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1486 }
1487
1488 break;
1489 }
1490
1491 /* now add the entry to our list(s) */
1492 name = D_NAME (entp);
1493
1494 /* skip . and .. entries */
1495 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1496 {
1497 int len = D_NAMLEN (entp) + 1;
1498
1499 while (ecb_expect_false (namesoffs + len > namesalloc))
1500 {
1501 namesalloc *= 2;
1502 req->ptr2 = names = realloc (names, namesalloc);
1503
1504 if (!names)
1505 break;
1506 }
1507
1508 memcpy (names + namesoffs, name, len);
1509
1510 if (dents)
1511 {
1512 struct eio_dirent *ent;
1513
1514 if (ecb_expect_false (dentoffs == dentalloc))
1277 { 1515 {
1278 /* in this case, all is known, and we just put dirs first and sort them */ 1516 dentalloc *= 2;
1517 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1518
1519 if (!dents)
1520 break;
1521 }
1522
1279 eio_dirent *oth = dents + dentoffs; 1523 ent = dents + dentoffs;
1280 eio_dirent *dir = dents;
1281 1524
1282 /* now partition dirs to the front, and non-dirs to the back */ 1525 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1283 /* by walking from both sides and swapping if necessary */ 1526 ent->namelen = len - 1;
1284 /* also clear score, so it doesn't influence sorting */ 1527 ent->inode = D_INO (entp);
1285 while (oth > dir) 1528
1529 inode_bits |= ent->inode;
1530
1531 switch (D_TYPE (entp))
1532 {
1533 default:
1534 ent->type = EIO_DT_UNKNOWN;
1535 flags |= EIO_READDIR_FOUND_UNKNOWN;
1536 break;
1537
1538 #ifdef DT_FIFO
1539 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1540 #endif
1541 #ifdef DT_CHR
1542 case DT_CHR: ent->type = EIO_DT_CHR; break;
1543 #endif
1544 #ifdef DT_MPC
1545 case DT_MPC: ent->type = EIO_DT_MPC; break;
1546 #endif
1547 #ifdef DT_DIR
1548 case DT_DIR: ent->type = EIO_DT_DIR; break;
1549 #endif
1550 #ifdef DT_NAM
1551 case DT_NAM: ent->type = EIO_DT_NAM; break;
1552 #endif
1553 #ifdef DT_BLK
1554 case DT_BLK: ent->type = EIO_DT_BLK; break;
1555 #endif
1556 #ifdef DT_MPB
1557 case DT_MPB: ent->type = EIO_DT_MPB; break;
1558 #endif
1559 #ifdef DT_REG
1560 case DT_REG: ent->type = EIO_DT_REG; break;
1561 #endif
1562 #ifdef DT_NWK
1563 case DT_NWK: ent->type = EIO_DT_NWK; break;
1564 #endif
1565 #ifdef DT_CMP
1566 case DT_CMP: ent->type = EIO_DT_CMP; break;
1567 #endif
1568 #ifdef DT_LNK
1569 case DT_LNK: ent->type = EIO_DT_LNK; break;
1570 #endif
1571 #ifdef DT_SOCK
1572 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1573 #endif
1574 #ifdef DT_DOOR
1575 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1576 #endif
1577 #ifdef DT_WHT
1578 case DT_WHT: ent->type = EIO_DT_WHT; break;
1579 #endif
1580 }
1581
1582 ent->score = 7;
1583
1584 if (flags & EIO_READDIR_DIRS_FIRST)
1585 {
1586 if (ent->type == EIO_DT_UNKNOWN)
1286 { 1587 {
1287 if (dir->type == EIO_DT_DIR) 1588 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1288 ++dir; 1589 ent->score = 1;
1289 else if ((--oth)->type == EIO_DT_DIR) 1590 else if (!strchr (name, '.')) /* absence of dots indicate likely dirs */
1290 { 1591 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1291 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1292
1293 ++dir;
1294 }
1295 } 1592 }
1296 1593 else if (ent->type == EIO_DT_DIR)
1297 /* now sort the dirs only */ 1594 ent->score = 0;
1298 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1299 } 1595 }
1300
1301 break;
1302 }
1303
1304 /* now add the entry to our list(s) */
1305 name = entp->d_name;
1306
1307 /* skip . and .. entries */
1308 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1309 {
1310 int len = D_NAMLEN (entp) + 1;
1311
1312 while (expect_false (namesoffs + len > namesalloc))
1313 {
1314 namesalloc *= 2;
1315 X_LOCK (wrklock);
1316 req->ptr2 = names = realloc (names, namesalloc);
1317 X_UNLOCK (wrklock);
1318
1319 if (!names)
1320 break;
1321 } 1596 }
1322 1597
1323 memcpy (names + namesoffs, name, len);
1324
1325 if (dents)
1326 {
1327 struct eio_dirent *ent;
1328
1329 if (expect_false (dentoffs == dentalloc))
1330 {
1331 dentalloc *= 2;
1332 X_LOCK (wrklock);
1333 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1334 X_UNLOCK (wrklock);
1335
1336 if (!dents)
1337 break;
1338 }
1339
1340 ent = dents + dentoffs;
1341
1342 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1343 ent->namelen = len - 1;
1344 ent->inode = D_INO (entp);
1345
1346 inode_bits |= ent->inode;
1347
1348 switch (D_TYPE (entp))
1349 {
1350 default:
1351 ent->type = EIO_DT_UNKNOWN;
1352 flags |= EIO_READDIR_FOUND_UNKNOWN;
1353 break;
1354
1355 #ifdef DT_FIFO
1356 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1357 #endif
1358 #ifdef DT_CHR
1359 case DT_CHR: ent->type = EIO_DT_CHR; break;
1360 #endif
1361 #ifdef DT_MPC
1362 case DT_MPC: ent->type = EIO_DT_MPC; break;
1363 #endif
1364 #ifdef DT_DIR
1365 case DT_DIR: ent->type = EIO_DT_DIR; break;
1366 #endif
1367 #ifdef DT_NAM
1368 case DT_NAM: ent->type = EIO_DT_NAM; break;
1369 #endif
1370 #ifdef DT_BLK
1371 case DT_BLK: ent->type = EIO_DT_BLK; break;
1372 #endif
1373 #ifdef DT_MPB
1374 case DT_MPB: ent->type = EIO_DT_MPB; break;
1375 #endif
1376 #ifdef DT_REG
1377 case DT_REG: ent->type = EIO_DT_REG; break;
1378 #endif
1379 #ifdef DT_NWK
1380 case DT_NWK: ent->type = EIO_DT_NWK; break;
1381 #endif
1382 #ifdef DT_CMP
1383 case DT_CMP: ent->type = EIO_DT_CMP; break;
1384 #endif
1385 #ifdef DT_LNK
1386 case DT_LNK: ent->type = EIO_DT_LNK; break;
1387 #endif
1388 #ifdef DT_SOCK
1389 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1390 #endif
1391 #ifdef DT_DOOR
1392 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1393 #endif
1394 #ifdef DT_WHT
1395 case DT_WHT: ent->type = EIO_DT_WHT; break;
1396 #endif
1397 }
1398
1399 ent->score = 7;
1400
1401 if (flags & EIO_READDIR_DIRS_FIRST)
1402 {
1403 if (ent->type == EIO_DT_UNKNOWN)
1404 {
1405 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1406 ent->score = 1;
1407 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1408 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1409 }
1410 else if (ent->type == EIO_DT_DIR)
1411 ent->score = 0;
1412 }
1413 }
1414
1415 namesoffs += len; 1598 namesoffs += len;
1416 ++dentoffs; 1599 ++dentoffs;
1417 } 1600 }
1418 1601
1419 if (EIO_CANCELLED (req)) 1602 if (EIO_CANCELLED (req))
1420 { 1603 {
1421 errno = ECANCELED; 1604 errno = ECANCELED;
1422 break; 1605 break;
1423 } 1606 }
1607
1608#ifdef _WIN32
1609 if (!FindNextFile (dirp, &entp))
1610 {
1611 FindClose (dirp);
1612 dirp = 0;
1424 } 1613 }
1614#endif
1615 }
1425} 1616}
1426 1617
1427#ifdef PAGESIZE 1618/*****************************************************************************/
1428# define eio_pagesize() PAGESIZE 1619/* working directory stuff */
1620/* various deficiencies in the posix 2008 api force us to */
1621/* keep the absolute path in string form at all times */
1622/* fuck yeah. */
1623
1624#if !HAVE_AT
1625
1626/* a bit like realpath, but usually faster because it doesn'T have to return */
1627/* an absolute or canonical path */
1628static const char *
1629wd_expand (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
1630{
1631 if (!wd || *path == '/')
1632 return path;
1633
1634 if (path [0] == '.' && !path [1])
1635 return wd->str;
1636
1637 {
1638 int l1 = wd->len;
1639 int l2 = strlen (path);
1640
1641 char *res = etp_tmpbuf_get (tmpbuf, l1 + l2 + 2);
1642
1643 memcpy (res, wd->str, l1);
1644 res [l1] = '/';
1645 memcpy (res + l1 + 1, path, l2 + 1);
1646
1647 return res;
1648 }
1649}
1650
1651#endif
1652
1653static eio_wd
1654eio__wd_open_sync (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
1655{
1656 int fd;
1657 eio_wd res;
1658 int len = eio__realpath (tmpbuf, wd, path);
1659
1660 if (len < 0)
1661 return EIO_INVALID_WD;
1662
1663#if HAVE_AT
1664 fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1665
1666 if (fd < 0)
1667 return EIO_INVALID_WD;
1668#endif
1669
1670 res = malloc (sizeof (*res) + len); /* one extra 0-byte */
1671
1672#if HAVE_AT
1673 res->fd = fd;
1674#endif
1675
1676 res->len = len;
1677 memcpy (res->str, tmpbuf->ptr, len);
1678 res->str [len] = 0;
1679
1680 return res;
1681}
1682
1683eio_wd
1684eio_wd_open_sync (eio_wd wd, const char *path)
1685{
1686 struct etp_tmpbuf tmpbuf = { };
1687 wd = eio__wd_open_sync (&tmpbuf, wd, path);
1688 free (tmpbuf.ptr);
1689
1690 return wd;
1691}
1692
1693void
1694eio_wd_close_sync (eio_wd wd)
1695{
1696 if (wd != EIO_INVALID_WD && wd != EIO_CWD)
1697 {
1698 #if HAVE_AT
1699 eio__close (wd->fd);
1700 #endif
1701 free (wd);
1702 }
1703}
1704
1705#if HAVE_AT
1706
1707static int
1708eio__renameat2 (int olddirfd, const char *oldpath, int newdirfd, const char *newpath, unsigned int flags)
1709{
1710#if HAVE_RENAMEAT2
1711 return syscall (SYS_renameat2, olddirfd, oldpath, newdirfd, newpath, flags);
1429#else 1712#else
1430static intptr_t
1431eio_pagesize (void)
1432{
1433 static intptr_t page;
1434
1435 if (!page) 1713 if (flags)
1436 page = sysconf (_SC_PAGESIZE); 1714 return EIO_ENOSYS ();
1437 1715
1438 return page; 1716 return renameat (olddirfd, oldpath, newdirfd, newpath);
1439}
1440#endif 1717#endif
1441
1442static void
1443eio_page_align (void **addr, size_t *length)
1444{
1445 intptr_t mask = eio_pagesize () - 1;
1446
1447 /* round down addr */
1448 intptr_t adj = mask & (intptr_t)*addr;
1449
1450 *addr = (void *)((intptr_t)*addr - adj);
1451 *length += adj;
1452
1453 /* round up length */
1454 *length = (*length + mask) & ~mask;
1455} 1718}
1456 1719
1457#if !_POSIX_MEMLOCK 1720/* they forgot these */
1458# define eio__mlockall(a) ((errno = ENOSYS), -1)
1459#else
1460 1721
1461static int 1722static int
1462eio__mlockall (int flags) 1723eio__truncateat (int dirfd, const char *path, off_t length)
1463{ 1724{
1464 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7 1725 int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
1465 extern int mallopt (int, int); 1726 int res;
1466 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1467 #endif
1468 1727
1469 if (EIO_MCL_CURRENT != MCL_CURRENT 1728 if (fd < 0)
1470 || EIO_MCL_FUTURE != MCL_FUTURE) 1729 return fd;
1471 {
1472 flags = 0
1473 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1474 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1475 }
1476 1730
1477 return mlockall (flags); 1731 res = ftruncate (fd, length);
1732 silent_close (fd);
1733 return res;
1478} 1734}
1479#endif
1480
1481#if !_POSIX_MEMLOCK_RANGE
1482# define eio__mlock(a,b) ((errno = ENOSYS), -1)
1483#else
1484 1735
1485static int 1736static int
1486eio__mlock (void *addr, size_t length) 1737eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
1487{ 1738{
1488 eio_page_align (&addr, &length); 1739 int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
1740 int res;
1489 1741
1490 return mlock (addr, length); 1742 if (fd < 0)
1491} 1743 return fd;
1492 1744
1493#endif 1745 res = fstatvfs (fd, buf);
1494 1746 silent_close (fd);
1495#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1496# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
1497#else
1498
1499int
1500eio__msync (void *mem, size_t len, int flags)
1501{
1502 eio_page_align (&mem, &len);
1503
1504 if (EIO_MS_ASYNC != MS_SYNC
1505 || EIO_MS_INVALIDATE != MS_INVALIDATE
1506 || EIO_MS_SYNC != MS_SYNC)
1507 {
1508 flags = 0
1509 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1510 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1511 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1512 }
1513
1514 return msync (mem, len, flags);
1515}
1516
1517#endif
1518
1519int
1520eio__mtouch (void *mem, size_t len, int flags)
1521{
1522 eio_page_align (&mem, &len);
1523
1524 {
1525 intptr_t addr = (intptr_t)mem;
1526 intptr_t end = addr + len;
1527 intptr_t page = eio_pagesize ();
1528
1529 if (addr < end)
1530 if (flags & EIO_MT_MODIFY) /* modify */
1531 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
1532 else
1533 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
1534 }
1535
1536 return 0; 1747 return res;
1537} 1748}
1749
1750#endif
1538 1751
1539/*****************************************************************************/ 1752/*****************************************************************************/
1540 1753
1541#define ALLOC(len) \ 1754#define ALLOC(len) \
1542 if (!req->ptr2) \ 1755 if (!req->ptr2) \
1543 { \ 1756 { \
1544 X_LOCK (wrklock); \ 1757 X_LOCK (EIO_POOL->wrklock); \
1545 req->flags |= EIO_FLAG_PTR2_FREE; \ 1758 req->flags |= EIO_FLAG_PTR2_FREE; \
1546 X_UNLOCK (wrklock); \ 1759 X_UNLOCK (EIO_POOL->wrklock); \
1547 req->ptr2 = malloc (len); \ 1760 req->ptr2 = malloc (len); \
1548 if (!req->ptr2) \ 1761 if (!req->ptr2) \
1549 { \ 1762 { \
1550 errno = ENOMEM; \ 1763 errno = ENOMEM; \
1551 req->result = -1; \ 1764 req->result = -1; \
1552 break; \ 1765 goto alloc_fail; \
1553 } \ 1766 } \
1554 } 1767 }
1555 1768
1556X_THREAD_PROC (etp_proc) 1769/*****************************************************************************/
1557{
1558 ETP_REQ *req;
1559 struct timespec ts;
1560 etp_worker *self = (etp_worker *)thr_arg;
1561 1770
1562 /* try to distribute timeouts somewhat randomly */ 1771static void
1563 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 1772eio__slurp (int fd, eio_req *req)
1773{
1774 req->result = fd;
1564 1775
1565 for (;;) 1776 if (fd < 0)
1777 return;
1778
1779 if (req->offs < 0 || !req->size) /* do we need the size? */
1566 { 1780 {
1567 X_LOCK (reqlock); 1781 off_t size = lseek (fd, 0, SEEK_END);
1568 1782
1569 for (;;)
1570 {
1571 self->req = req = reqq_shift (&req_queue);
1572
1573 if (req)
1574 break;
1575
1576 ++idle;
1577
1578 ts.tv_sec = time (0) + idle_timeout;
1579 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
1580 {
1581 if (idle > max_idle)
1582 {
1583 --idle;
1584 X_UNLOCK (reqlock);
1585 X_LOCK (wrklock);
1586 --started;
1587 X_UNLOCK (wrklock);
1588 goto quit;
1589 }
1590
1591 /* we are allowed to idle, so do so without any timeout */
1592 X_COND_WAIT (reqwait, reqlock);
1593 }
1594
1595 --idle;
1596 }
1597
1598 --nready;
1599
1600 X_UNLOCK (reqlock);
1601
1602 if (req->type < 0) 1783 if (req->offs < 0)
1603 goto quit; 1784 req->offs += size;
1604 1785
1605 if (!EIO_CANCELLED (req)) 1786 if (!req->size)
1606 ETP_EXECUTE (self, req); 1787 req->size = size - req->offs;
1607
1608 X_LOCK (reslock);
1609
1610 ++npending;
1611
1612 if (!reqq_push (&res_queue, req) && want_poll_cb)
1613 want_poll_cb ();
1614
1615 self->req = 0;
1616 etp_worker_clear (self);
1617
1618 X_UNLOCK (reslock);
1619 } 1788 }
1620 1789
1621quit: 1790 ALLOC (req->size);
1622 X_LOCK (wrklock); 1791 req->result = pread (fd, req->ptr2, req->size, req->offs);
1623 etp_worker_free (self);
1624 X_UNLOCK (wrklock);
1625 1792
1626 return 0; 1793 silent_close (fd);
1627}
1628 1794
1629/*****************************************************************************/ 1795alloc_fail:
1796 ;
1797}
1630 1798
1799int ecb_cold
1631int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1800eio_init (void (*want_poll)(void), void (*done_poll)(void))
1632{ 1801{
1633 return etp_init (want_poll, done_poll); 1802 eio_want_poll_cb = want_poll;
1634} 1803 eio_done_poll_cb = done_poll;
1635 1804
1805 return etp_init (EIO_POOL, 0, 0, 0);
1806}
1807
1808ecb_inline void
1636static void eio_api_destroy (eio_req *req) 1809eio_api_destroy (eio_req *req)
1637{ 1810{
1638 free (req); 1811 free (req);
1639} 1812}
1640 1813
1641#define REQ(rtype) \ 1814#define REQ(rtype) \
1642 eio_req *req; \ 1815 eio_req *req; \
1643 \ 1816 \
1644 req = (eio_req *)calloc (1, sizeof *req); \ 1817 req = (eio_req *)calloc (1, sizeof *req); \
1645 if (!req) \ 1818 if (!req) \
1646 return 0; \ 1819 return 0; \
1660 { \ 1833 { \
1661 eio_api_destroy (req); \ 1834 eio_api_destroy (req); \
1662 return 0; \ 1835 return 0; \
1663 } 1836 }
1664 1837
1838#define SINGLEDOT(ptr) (0[(char *)(ptr)] == '.' && !1[(char *)(ptr)])
1839
1840static void
1665static void eio_execute (etp_worker *self, eio_req *req) 1841eio_execute (etp_worker *self, eio_req *req)
1666{ 1842{
1843#if HAVE_AT
1844 int dirfd;
1845#else
1846 const char *path;
1847#endif
1848
1849 if (ecb_expect_false (EIO_CANCELLED (req)))
1850 {
1851 req->result = -1;
1852 req->errorno = ECANCELED;
1853 return;
1854 }
1855
1856 if (ecb_expect_false (req->wd == EIO_INVALID_WD))
1857 {
1858 req->result = -1;
1859 req->errorno = ENOENT;
1860 return;
1861 }
1862
1863 if (req->type >= EIO_OPEN)
1864 {
1865 #if HAVE_AT
1866 dirfd = WD2FD (req->wd);
1867 #else
1868 path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1869 #endif
1870 }
1871
1667 switch (req->type) 1872 switch (req->type)
1668 { 1873 {
1874 case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
1875 req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
1876 break;
1877 case EIO_WD_CLOSE: req->result = 0;
1878 eio_wd_close_sync (req->wd); break;
1879
1880 case EIO_SEEK: eio__lseek (req); break;
1669 case EIO_READ: ALLOC (req->size); 1881 case EIO_READ: ALLOC (req->size);
1670 req->result = req->offs >= 0 1882 req->result = req->offs >= 0
1671 ? pread (req->int1, req->ptr2, req->size, req->offs) 1883 ? pread (req->int1, req->ptr2, req->size, req->offs)
1672 : read (req->int1, req->ptr2, req->size); break; 1884 : read (req->int1, req->ptr2, req->size); break;
1673 case EIO_WRITE: req->result = req->offs >= 0 1885 case EIO_WRITE: req->result = req->offs >= 0
1674 ? pwrite (req->int1, req->ptr2, req->size, req->offs) 1886 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1675 : write (req->int1, req->ptr2, req->size); break; 1887 : write (req->int1, req->ptr2, req->size); break;
1676 1888
1889 case EIO_FCNTL: req->result = fcntl (req->int1, (int) req->int2, req->ptr2); break;
1890 case EIO_IOCTL: req->result = ioctl (req->int1, (unsigned long)req->int2, req->ptr2); break;
1891
1677 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break; 1892 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1678 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break; 1893 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
1894
1895#if HAVE_AT
1679 1896
1680 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1897 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1681 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 1898 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break;
1682 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1899 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1900 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, AT_SYMLINK_NOFOLLOW); break;
1901 case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break;
1902 case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break;
1903 case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break;
1904 case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break;
1905 case EIO_SLURP: eio__slurp ( openat (dirfd, req->ptr1, O_RDONLY | O_CLOEXEC), req); break;
1906
1907 case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
1908 case EIO_RMDIR: /* complications arise because "." cannot be removed, so we might have to expand */
1909 req->result = req->wd && SINGLEDOT (req->ptr1)
1910 ? rmdir (req->wd->str)
1911 : unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break;
1912 case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); break;
1913 case EIO_RENAME: req->result = eio__renameat2 (
1914 dirfd,
1915 /* complications arise because "." cannot be renamed, so we might have to expand */
1916 req->wd && SINGLEDOT (req->ptr1) ? req->wd->str : req->ptr1,
1917 WD2FD ((eio_wd)req->int3),
1918 req->ptr2,
1919 req->int2
1920 );
1921 break;
1922 case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2, 0); break;
1923 case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
1924 case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1925 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1926 req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1927 case EIO_READLINK: ALLOC (EIO_PATH_MAX);
1928 req->result = readlinkat (dirfd, req->ptr1, req->ptr2, EIO_PATH_MAX);
1929 if (req->result == EIO_PATH_MAX)
1930 {
1931 req->result = -1;
1932 errno = ENAMETOOLONG;
1933 }
1934 break;
1935 case EIO_UTIME:
1936 case EIO_FUTIME:
1937 {
1938 struct timespec ts[2];
1939 struct timespec *times;
1940
1941 if (req->nv1 != -1. || req->nv2 != -1.)
1942 {
1943 ts[0].tv_sec = req->nv1;
1944 ts[0].tv_nsec = (req->nv1 - ts[0].tv_sec) * 1e9;
1945 ts[1].tv_sec = req->nv2;
1946 ts[1].tv_nsec = (req->nv2 - ts[1].tv_sec) * 1e9;
1947
1948 times = ts;
1949 }
1950 else
1951 times = 0;
1952
1953 req->result = req->type == EIO_FUTIME
1954 ? futimens (req->int1, times)
1955 : utimensat (dirfd, req->ptr1, times, 0);
1956 }
1957 break;
1958
1959#else
1960
1961 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1962 req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
1963 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1683 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 1964 req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
1965 case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
1966 case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
1967 case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
1968 case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
1969 case EIO_SLURP: eio__slurp ( open (path , O_RDONLY | O_CLOEXEC), req); break;
1970
1971 case EIO_UNLINK: req->result = unlink (path ); break;
1972 case EIO_RMDIR: req->result = rmdir (path ); break;
1973 case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
1974 case EIO_RENAME: req->result = req->int2 ? EIO_ENOSYS () : rename (path, req->ptr2); break;
1975 case EIO_LINK: req->result = link (path , req->ptr2); break;
1976 case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
1977 case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
1978 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1979 req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
1980 case EIO_READLINK: ALLOC (EIO_PATH_MAX);
1981 req->result = readlink (path, req->ptr2, EIO_PATH_MAX);
1982 if (req->result == EIO_PATH_MAX)
1983 {
1984 req->result = -1;
1985 errno = ENAMETOOLONG;
1986 }
1987 break;
1988
1989 case EIO_UTIME:
1990 case EIO_FUTIME:
1991 {
1992 struct timeval tv[2];
1993 struct timeval *times;
1994
1995 if (req->nv1 != -1. || req->nv2 != -1.)
1996 {
1997 tv[0].tv_sec = req->nv1;
1998 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
1999 tv[1].tv_sec = req->nv2;
2000 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
2001
2002 times = tv;
2003 }
2004 else
2005 times = 0;
2006
2007 req->result = req->type == EIO_FUTIME
2008 ? futimes (req->int1, times)
2009 : utimes (req->ptr1, times);
2010 }
2011 break;
2012
2013#endif
2014
2015 case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
2016 {
2017 ALLOC (req->result);
2018 memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
2019 }
2020 break;
2021
1684 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2022 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1685 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; 2023 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1686 2024
1687 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1688 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1689 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS)); 2025 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1690 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break; 2026 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1691 2027
1692 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1693 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; 2028 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1694 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1695 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break; 2029 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1696 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1697 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break; 2030 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1698 2031
1699 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
1700 case EIO_CLOSE: req->result = close (req->int1); break; 2032 case EIO_CLOSE: req->result = eio__close (req->int1); break;
1701 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break; 2033 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1702 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
1703 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1704 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1705 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1706 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1707 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1708 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1709
1710 case EIO_READLINK: ALLOC (PATH_MAX);
1711 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1712
1713 case EIO_SYNC: req->result = 0; sync (); break; 2034 case EIO_SYNC: req->result = 0; sync (); break;
1714 case EIO_FSYNC: req->result = fsync (req->int1); break; 2035 case EIO_FSYNC: req->result = fsync (req->int1); break;
1715 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 2036 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
2037 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
2038 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1716 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break; 2039 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1717 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break; 2040 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1718 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break; 2041 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1719 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break; 2042 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1720 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break; 2043 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
1721 2044
1722 case EIO_READDIR: eio__scandir (req, self); break; 2045 case EIO_READDIR: eio__scandir (req, self); break;
1723 2046
1724 case EIO_BUSY: 2047 case EIO_BUSY:
1725#ifdef _WIN32 2048#ifdef _WIN32
1734 req->result = select (0, 0, 0, 0, &tv); 2057 req->result = select (0, 0, 0, 0, &tv);
1735 } 2058 }
1736#endif 2059#endif
1737 break; 2060 break;
1738 2061
1739 case EIO_UTIME: 2062#if 0
1740 case EIO_FUTIME:
1741 {
1742 struct timeval tv[2];
1743 struct timeval *times;
1744
1745 if (req->nv1 != -1. || req->nv2 != -1.)
1746 {
1747 tv[0].tv_sec = req->nv1;
1748 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
1749 tv[1].tv_sec = req->nv2;
1750 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
1751
1752 times = tv;
1753 }
1754 else
1755 times = 0;
1756
1757 req->result = req->type == EIO_FUTIME
1758 ? futimes (req->int1, times)
1759 : utimes (req->ptr1, times);
1760 }
1761 break;
1762
1763 case EIO_GROUP: 2063 case EIO_GROUP:
1764 abort (); /* handled in eio_request */ 2064 abort (); /* handled in eio_request */
2065#endif
1765 2066
1766 case EIO_NOP: 2067 case EIO_NOP:
1767 req->result = 0; 2068 req->result = 0;
1768 break; 2069 break;
1769 2070
1770 case EIO_CUSTOM: 2071 case EIO_CUSTOM:
1771 ((void (*)(eio_req *))req->feed) (req); 2072 req->feed (req);
1772 break; 2073 break;
1773 2074
1774 default: 2075 default:
1775 errno = ENOSYS;
1776 req->result = -1; 2076 req->result = EIO_ENOSYS ();
1777 break; 2077 break;
1778 } 2078 }
1779 2079
2080alloc_fail:
1780 req->errorno = errno; 2081 req->errorno = errno;
1781} 2082}
1782 2083
1783#ifndef EIO_NO_WRAPPERS 2084#ifndef EIO_NO_WRAPPERS
1784 2085
2086eio_req *eio_wd_open (const char *path, int pri, eio_cb cb, void *data)
2087{
2088 REQ (EIO_WD_OPEN); PATH; SEND;
2089}
2090
2091eio_req *eio_wd_close (eio_wd wd, int pri, eio_cb cb, void *data)
2092{
2093 REQ (EIO_WD_CLOSE); req->wd = wd; SEND;
2094}
2095
1785eio_req *eio_nop (int pri, eio_cb cb, void *data) 2096eio_req *eio_nop (int pri, eio_cb cb, void *data)
1786{ 2097{
1787 REQ (EIO_NOP); SEND; 2098 REQ (EIO_NOP); SEND;
1788} 2099}
1789 2100
1805eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2116eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1806{ 2117{
1807 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2118 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1808} 2119}
1809 2120
2121eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2122{
2123 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2124}
2125
2126eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
2127{
2128 REQ (EIO_SYNCFS); req->int1 = fd; SEND;
2129}
2130
2131eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2132{
2133 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2134}
2135
1810eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2136eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1811{ 2137{
1812 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2138 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1813} 2139}
1814 2140
1820eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data) 2146eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
1821{ 2147{
1822 REQ (EIO_MLOCKALL); req->int1 = flags; SEND; 2148 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
1823} 2149}
1824 2150
1825eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) 2151eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
1826{ 2152{
1827 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; 2153 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
1828}
1829
1830eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1831{
1832 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1833} 2154}
1834 2155
1835eio_req *eio_close (int fd, int pri, eio_cb cb, void *data) 2156eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
1836{ 2157{
1837 REQ (EIO_CLOSE); req->int1 = fd; SEND; 2158 REQ (EIO_CLOSE); req->int1 = fd; SEND;
1840eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data) 2161eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
1841{ 2162{
1842 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND; 2163 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
1843} 2164}
1844 2165
2166eio_req *eio_seek (int fd, off_t offset, int whence, int pri, eio_cb cb, void *data)
2167{
2168 REQ (EIO_SEEK); req->int1 = fd; req->offs = offset; req->int2 = whence; SEND;
2169}
2170
1845eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data) 2171eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1846{ 2172{
1847 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND; 2173 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1848} 2174}
1849 2175
1850eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data) 2176eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1851{ 2177{
1852 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND; 2178 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1853} 2179}
1854 2180
2181eio_req *eio_fcntl (int fd, int cmd, void *arg, int pri, eio_cb cb, void *data)
2182{
2183 REQ (EIO_IOCTL); req->int1 = fd; req->int2 = cmd; req->ptr2 = arg; SEND;
2184}
2185
2186eio_req *eio_ioctl (int fd, unsigned long request, void *buf, int pri, eio_cb cb, void *data)
2187{
2188 REQ (EIO_IOCTL); req->int1 = fd; req->int2 = request; req->ptr2 = buf; SEND;
2189}
2190
1855eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) 2191eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1856{ 2192{
1857 REQ (EIO_FSTAT); req->int1 = fd; SEND; 2193 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1858} 2194}
1859 2195
1875eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) 2211eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1876{ 2212{
1877 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2213 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1878} 2214}
1879 2215
1880eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2216eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1881{ 2217{
1882 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2218 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1883} 2219}
1884 2220
1885eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) 2221eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
1905eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data) 2241eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1906{ 2242{
1907 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2243 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1908} 2244}
1909 2245
1910eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2246eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1911{ 2247{
1912 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2248 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1913} 2249}
1914 2250
1915eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) 2251eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
1929} 2265}
1930 2266
1931eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data) 2267eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1932{ 2268{
1933 return eio__1path (EIO_READLINK, path, pri, cb, data); 2269 return eio__1path (EIO_READLINK, path, pri, cb, data);
2270}
2271
2272eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2273{
2274 return eio__1path (EIO_REALPATH, path, pri, cb, data);
1934} 2275}
1935 2276
1936eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) 2277eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1937{ 2278{
1938 return eio__1path (EIO_STAT, path, pri, cb, data); 2279 return eio__1path (EIO_STAT, path, pri, cb, data);
1997eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2338eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1998{ 2339{
1999 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2340 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
2000} 2341}
2001 2342
2343eio_req *eio_slurp (const char *path, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
2344{
2345 REQ (EIO_SLURP); PATH; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
2346}
2347
2002eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) 2348eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
2003{ 2349{
2004 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND; 2350 REQ (EIO_CUSTOM); req->feed = execute; SEND;
2005} 2351}
2006 2352
2007#endif 2353#endif
2008 2354
2009eio_req *eio_grp (eio_cb cb, void *data) 2355eio_req *eio_grp (eio_cb cb, void *data)
2018#undef SEND 2364#undef SEND
2019 2365
2020/*****************************************************************************/ 2366/*****************************************************************************/
2021/* grp functions */ 2367/* grp functions */
2022 2368
2369void
2023void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2370eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
2024{ 2371{
2025 grp->int2 = limit; 2372 grp->int2 = limit;
2026 grp->feed = feed; 2373 grp->feed = feed;
2027 2374
2028 grp_try_feed (grp); 2375 grp_try_feed (grp);
2029} 2376}
2030 2377
2378void
2031void eio_grp_limit (eio_req *grp, int limit) 2379eio_grp_limit (eio_req *grp, int limit)
2032{ 2380{
2033 grp->int2 = limit; 2381 grp->int2 = limit;
2034 2382
2035 grp_try_feed (grp); 2383 grp_try_feed (grp);
2036} 2384}
2037 2385
2386void
2038void eio_grp_add (eio_req *grp, eio_req *req) 2387eio_grp_add (eio_req *grp, eio_req *req)
2039{ 2388{
2040 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2389 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
2041 2390
2042 grp->flags |= EIO_FLAG_GROUPADD; 2391 grp->flags |= ETP_FLAG_GROUPADD;
2043 2392
2044 ++grp->size; 2393 ++grp->size;
2045 req->grp = grp; 2394 req->grp = grp;
2046 2395
2047 req->grp_prev = 0; 2396 req->grp_prev = 0;
2054} 2403}
2055 2404
2056/*****************************************************************************/ 2405/*****************************************************************************/
2057/* misc garbage */ 2406/* misc garbage */
2058 2407
2408eio_ssize_t
2059ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2409eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2060{ 2410{
2061 etp_worker wrk;
2062 ssize_t ret;
2063
2064 wrk.dbuf = 0;
2065
2066 ret = eio__sendfile (ofd, ifd, offset, count, &wrk); 2411 return eio__sendfile (ofd, ifd, offset, count);
2067
2068 if (wrk.dbuf)
2069 free (wrk.dbuf);
2070
2071 return ret;
2072} 2412}
2073 2413

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines