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

Comparing libeio/eio.c (file contents):
Revision 1.59 by root, Wed Dec 1 07:30:39 2010 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 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_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
137/* number of seconds after which an idle threads exit */
138#define IDLE_TIMEOUT 10
139 319
140/* used for struct dirent, AIX doesn't provide it */ 320/* used for struct dirent, AIX doesn't provide it */
141#ifndef NAME_MAX 321#ifndef NAME_MAX
142# define NAME_MAX 4096 322# define NAME_MAX 4096
143#endif 323#endif
144 324
145/* used for readlink etc. */ 325/* used for readlink etc. */
146#ifndef PATH_MAX 326#ifndef PATH_MAX
147# define PATH_MAX 4096 327# define PATH_MAX 0
148#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)
149 339
150/* buffer size for various temporary buffers */ 340/* buffer size for various temporary buffers */
151#define EIO_BUFSIZE 65536 341#define EIO_BUFSIZE 65536
152 342
153#define dBUF \ 343#define dBUF \
154 char *eio_buf; \
155 ETP_WORKER_LOCK (self); \
156 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 344 char *eio_buf = malloc (EIO_BUFSIZE); \
157 ETP_WORKER_UNLOCK (self); \
158 errno = ENOMEM; \ 345 errno = ENOMEM; \
159 if (!eio_buf) \ 346 if (!eio_buf) \
160 return -1; 347 return -1
161 348
162#define EIO_TICKS ((1000000 + 1023) >> 10) 349#define FUBd \
350 free (eio_buf)
163 351
164/*****************************************************************************/ 352/*****************************************************************************/
165 353
166#if __GNUC__ >= 3 354struct etp_tmpbuf;
167# 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
168#else 362#else
169# 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);
170#endif 365#endif
171 366
172#define expect_false(expr) expect ((expr) != 0, 0) 367struct eio_pwd
173#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};
174 375
175/*****************************************************************************/ 376/*****************************************************************************/
176 377
177#define ETP_PRI_MIN EIO_PRI_MIN 378#define ETP_PRI_MIN EIO_PRI_MIN
178#define ETP_PRI_MAX EIO_PRI_MAX 379#define ETP_PRI_MAX EIO_PRI_MAX
179 380
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
180struct etp_worker; 391struct etp_worker;
181
182#define ETP_REQ eio_req 392#define ETP_REQ eio_req
183#define ETP_DESTROY(req) eio_destroy (req) 393#define ETP_DESTROY(req) eio_destroy (req)
184static int eio_finish (eio_req *req); 394static int eio_finish (eio_req *req);
185#define ETP_FINISH(req) eio_finish (req) 395#define ETP_FINISH(req) eio_finish (req)
186static void eio_execute (struct etp_worker *self, eio_req *req); 396static void eio_execute (struct etp_worker *self, eio_req *req);
187#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) 397#define ETP_EXECUTE(wrk,req) eio_execute (wrk, req)
188 398
189#define ETP_WORKER_CLEAR(req) \ 399#include "etp.c"
190 if (wrk->dbuf) \
191 { \
192 free (wrk->dbuf); \
193 wrk->dbuf = 0; \
194 } \
195 \
196 if (wrk->dirp) \
197 { \
198 closedir (wrk->dirp); \
199 wrk->dirp = 0; \
200 }
201 400
202#define ETP_WORKER_COMMON \ 401static struct etp_pool eio_pool;
203 void *dbuf; \ 402#define EIO_POOL (&eio_pool)
204 DIR *dirp;
205 403
206/*****************************************************************************/ 404/*****************************************************************************/
207 405
208#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
209
210/* calculate time difference in ~1/EIO_TICKS of a second */
211static int tvdiff (struct timeval *tv1, struct timeval *tv2)
212{
213 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
214 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
215}
216
217static unsigned int started, idle, wanted = 4;
218
219static void (*want_poll_cb) (void);
220static void (*done_poll_cb) (void);
221
222static unsigned int max_poll_time; /* reslock */
223static unsigned int max_poll_reqs; /* reslock */
224
225static volatile unsigned int nreqs; /* reqlock */
226static volatile unsigned int nready; /* reqlock */
227static volatile unsigned int npending; /* reqlock */
228static volatile unsigned int max_idle = 4;
229
230static xmutex_t wrklock = X_MUTEX_INIT;
231static xmutex_t reslock = X_MUTEX_INIT;
232static xmutex_t reqlock = X_MUTEX_INIT;
233static xcond_t reqwait = X_COND_INIT;
234
235#if !HAVE_PREADWRITE
236/*
237 * make our pread/pwrite emulation safe against themselves, but not against
238 * normal read/write by using a mutex. slows down execution a lot,
239 * but that's your problem, not mine.
240 */
241static xmutex_t preadwritelock = X_MUTEX_INIT;
242#endif
243
244typedef struct etp_worker
245{
246 /* locked by wrklock */
247 struct etp_worker *prev, *next;
248
249 xthread_t tid;
250
251 /* locked by reslock, reqlock or wrklock */
252 ETP_REQ *req; /* currently processed request */
253
254 ETP_WORKER_COMMON
255} etp_worker;
256
257static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */
258
259#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
260#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
261
262/* worker threads management */
263
264static void etp_worker_clear (etp_worker *wrk)
265{
266 ETP_WORKER_CLEAR (wrk);
267}
268
269static void etp_worker_free (etp_worker *wrk)
270{
271 wrk->next->prev = wrk->prev;
272 wrk->prev->next = wrk->next;
273
274 free (wrk);
275}
276
277static unsigned int etp_nreqs (void)
278{
279 int retval;
280 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
281 retval = nreqs;
282 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
283 return retval;
284}
285
286static unsigned int etp_nready (void)
287{
288 unsigned int retval;
289
290 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
291 retval = nready;
292 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
293
294 return retval;
295}
296
297static unsigned int etp_npending (void)
298{
299 unsigned int retval;
300
301 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
302 retval = npending;
303 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
304
305 return retval;
306}
307
308static unsigned int etp_nthreads (void)
309{
310 unsigned int retval;
311
312 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
313 retval = started;
314 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
315
316 return retval;
317}
318
319/*
320 * a somewhat faster data structure might be nice, but
321 * with 8 priorities this actually needs <20 insns
322 * per shift, the most expensive operation.
323 */
324typedef struct {
325 ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */
326 int size;
327} etp_reqq;
328
329static etp_reqq req_queue;
330static etp_reqq res_queue;
331
332static int reqq_push (etp_reqq *q, ETP_REQ *req)
333{
334 int pri = req->pri;
335 req->next = 0;
336
337 if (q->qe[pri])
338 {
339 q->qe[pri]->next = req;
340 q->qe[pri] = req;
341 }
342 else
343 q->qe[pri] = q->qs[pri] = req;
344
345 return q->size++;
346}
347
348static ETP_REQ *reqq_shift (etp_reqq *q)
349{
350 int pri;
351
352 if (!q->size)
353 return 0;
354
355 --q->size;
356
357 for (pri = ETP_NUM_PRI; pri--; )
358 {
359 eio_req *req = q->qs[pri];
360
361 if (req)
362 {
363 if (!(q->qs[pri] = (eio_req *)req->next))
364 q->qe[pri] = 0;
365
366 return req;
367 }
368 }
369
370 abort ();
371}
372
373static void etp_atfork_prepare (void)
374{
375 X_LOCK (wrklock);
376 X_LOCK (reqlock);
377 X_LOCK (reslock);
378#if !HAVE_PREADWRITE
379 X_LOCK (preadwritelock);
380#endif
381}
382
383static void etp_atfork_parent (void)
384{
385#if !HAVE_PREADWRITE
386 X_UNLOCK (preadwritelock);
387#endif
388 X_UNLOCK (reslock);
389 X_UNLOCK (reqlock);
390 X_UNLOCK (wrklock);
391}
392
393static void etp_atfork_child (void)
394{
395 ETP_REQ *prv;
396
397 while ((prv = reqq_shift (&req_queue)))
398 ETP_DESTROY (prv);
399
400 while ((prv = reqq_shift (&res_queue)))
401 ETP_DESTROY (prv);
402
403 while (wrk_first.next != &wrk_first)
404 {
405 etp_worker *wrk = wrk_first.next;
406
407 if (wrk->req)
408 ETP_DESTROY (wrk->req);
409
410 etp_worker_clear (wrk);
411 etp_worker_free (wrk);
412 }
413
414 started = 0;
415 idle = 0;
416 nreqs = 0;
417 nready = 0;
418 npending = 0;
419
420 etp_atfork_parent ();
421}
422
423static void 406static void
424etp_once_init (void)
425{
426 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
427}
428
429static int
430etp_init (void (*want_poll)(void), void (*done_poll)(void))
431{
432 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
433
434 pthread_once (&doinit, etp_once_init);
435
436 want_poll_cb = want_poll;
437 done_poll_cb = done_poll;
438
439 return 0;
440}
441
442X_THREAD_PROC (etp_proc);
443
444static void etp_start_thread (void)
445{
446 etp_worker *wrk = calloc (1, sizeof (etp_worker));
447
448 /*TODO*/
449 assert (("unable to allocate worker thread data", wrk));
450
451 X_LOCK (wrklock);
452
453 if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
454 {
455 wrk->prev = &wrk_first;
456 wrk->next = wrk_first.next;
457 wrk_first.next->prev = wrk;
458 wrk_first.next = wrk;
459 ++started;
460 }
461 else
462 free (wrk);
463
464 X_UNLOCK (wrklock);
465}
466
467static void etp_maybe_start_thread (void)
468{
469 if (expect_true (etp_nthreads () >= wanted))
470 return;
471
472 /* todo: maybe use idle here, but might be less exact */
473 if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
474 return;
475
476 etp_start_thread ();
477}
478
479static void etp_end_thread (void)
480{
481 eio_req *req = calloc (1, sizeof (eio_req));
482
483 req->type = -1;
484 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
485
486 X_LOCK (reqlock);
487 reqq_push (&req_queue, req);
488 X_COND_SIGNAL (reqwait);
489 X_UNLOCK (reqlock);
490
491 X_LOCK (wrklock);
492 --started;
493 X_UNLOCK (wrklock);
494}
495
496static int etp_poll (void)
497{
498 unsigned int maxreqs;
499 unsigned int maxtime;
500 struct timeval tv_start, tv_now;
501
502 X_LOCK (reslock);
503 maxreqs = max_poll_reqs;
504 maxtime = max_poll_time;
505 X_UNLOCK (reslock);
506
507 if (maxtime)
508 gettimeofday (&tv_start, 0);
509
510 for (;;)
511 {
512 ETP_REQ *req;
513
514 etp_maybe_start_thread ();
515
516 X_LOCK (reslock);
517 req = reqq_shift (&res_queue);
518
519 if (req)
520 {
521 --npending;
522
523 if (!res_queue.size && done_poll_cb)
524 done_poll_cb ();
525 }
526
527 X_UNLOCK (reslock);
528
529 if (!req)
530 return 0;
531
532 X_LOCK (reqlock);
533 --nreqs;
534 X_UNLOCK (reqlock);
535
536 if (expect_false (req->type == EIO_GROUP && req->size))
537 {
538 req->int1 = 1; /* mark request as delayed */
539 continue;
540 }
541 else
542 {
543 int res = ETP_FINISH (req);
544 if (expect_false (res))
545 return res;
546 }
547
548 if (expect_false (maxreqs && !--maxreqs))
549 break;
550
551 if (maxtime)
552 {
553 gettimeofday (&tv_now, 0);
554
555 if (tvdiff (&tv_start, &tv_now) >= maxtime)
556 break;
557 }
558 }
559
560 errno = EAGAIN;
561 return -1;
562}
563
564static void etp_cancel (ETP_REQ *req)
565{
566 X_LOCK (wrklock);
567 req->flags |= EIO_FLAG_CANCELLED;
568 X_UNLOCK (wrklock);
569
570 eio_grp_cancel (req);
571}
572
573static void etp_submit (ETP_REQ *req)
574{
575 req->pri -= ETP_PRI_MIN;
576
577 if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
578 if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
579
580 if (expect_false (req->type == EIO_GROUP))
581 {
582 /* I hope this is worth it :/ */
583 X_LOCK (reqlock);
584 ++nreqs;
585 X_UNLOCK (reqlock);
586
587 X_LOCK (reslock);
588
589 ++npending;
590
591 if (!reqq_push (&res_queue, req) && want_poll_cb)
592 want_poll_cb ();
593
594 X_UNLOCK (reslock);
595 }
596 else
597 {
598 X_LOCK (reqlock);
599 ++nreqs;
600 ++nready;
601 reqq_push (&req_queue, req);
602 X_COND_SIGNAL (reqwait);
603 X_UNLOCK (reqlock);
604
605 etp_maybe_start_thread ();
606 }
607}
608
609static void etp_set_max_poll_time (double nseconds)
610{
611 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
612 max_poll_time = nseconds * EIO_TICKS;
613 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
614}
615
616static void etp_set_max_poll_reqs (unsigned int maxreqs)
617{
618 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
619 max_poll_reqs = maxreqs;
620 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
621}
622
623static void etp_set_max_idle (unsigned int nthreads)
624{
625 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
626 max_idle = nthreads <= 0 ? 1 : nthreads;
627 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
628}
629
630static void etp_set_min_parallel (unsigned int nthreads)
631{
632 if (wanted < nthreads)
633 wanted = nthreads;
634}
635
636static void etp_set_max_parallel (unsigned int nthreads)
637{
638 if (wanted > nthreads)
639 wanted = nthreads;
640
641 while (started > wanted)
642 etp_end_thread ();
643}
644
645/*****************************************************************************/
646
647static void grp_try_feed (eio_req *grp) 407grp_try_feed (eio_req *grp)
648{ 408{
649 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 409 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
650 { 410 {
651 grp->flags &= ~EIO_FLAG_GROUPADD; 411 grp->flags &= ~ETP_FLAG_GROUPADD;
652 412
653 EIO_FEED (grp); 413 EIO_FEED (grp);
654 414
655 /* stop if no progress has been made */ 415 /* stop if no progress has been made */
656 if (!(grp->flags & EIO_FLAG_GROUPADD)) 416 if (!(grp->flags & ETP_FLAG_GROUPADD))
657 { 417 {
658 grp->feed = 0; 418 grp->feed = 0;
659 break; 419 break;
660 } 420 }
661 } 421 }
662} 422}
663 423
424static int
664static int grp_dec (eio_req *grp) 425grp_dec (eio_req *grp)
665{ 426{
666 --grp->size; 427 --grp->size;
667 428
668 /* call feeder, if applicable */ 429 /* call feeder, if applicable */
669 grp_try_feed (grp); 430 grp_try_feed (grp);
670 431
671 /* finish, if done */ 432 /* finish, if done */
672 if (!grp->size && grp->int1) 433 if (!grp->size && grp->flags & ETP_FLAG_DELAYED)
673 return eio_finish (grp); 434 return eio_finish (grp);
674 else 435 else
675 return 0; 436 return 0;
676} 437}
677 438
439static void
678void eio_destroy (eio_req *req) 440eio_destroy (eio_req *req)
679{ 441{
680 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); 442 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
681 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); 443 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
682 444
683 EIO_DESTROY (req); 445 EIO_DESTROY (req);
684} 446}
685 447
448static int
686static int eio_finish (eio_req *req) 449eio_finish (eio_req *req)
687{ 450{
688 int res = EIO_FINISH (req); 451 int res = EIO_FINISH (req);
689 452
690 if (req->grp) 453 if (req->grp)
691 { 454 {
… …
699 if (grp->grp_first == req) 462 if (grp->grp_first == req)
700 grp->grp_first = req->grp_next; 463 grp->grp_first = req->grp_next;
701 464
702 res2 = grp_dec (grp); 465 res2 = grp_dec (grp);
703 466
704 if (!res && res2) 467 if (!res)
705 res = res2; 468 res = res2;
706 } 469 }
707 470
708 eio_destroy (req); 471 eio_destroy (req);
709 472
710 return res; 473 return res;
711} 474}
712 475
476void
713void eio_grp_cancel (eio_req *grp) 477eio_grp_cancel (eio_req *grp)
714{ 478{
715 for (grp = grp->grp_first; grp; grp = grp->grp_next) 479 etp_grp_cancel (EIO_POOL, grp);
716 eio_cancel (grp);
717} 480}
718 481
482void
719void eio_cancel (eio_req *req) 483eio_cancel (eio_req *req)
720{ 484{
721 etp_cancel (req); 485 etp_cancel (EIO_POOL, req);
722} 486}
723 487
488void
724void eio_submit (eio_req *req) 489eio_submit (eio_req *req)
725{ 490{
726 etp_submit (req); 491 etp_submit (EIO_POOL, req);
727} 492}
728 493
729unsigned int eio_nreqs (void) 494unsigned int
495eio_nreqs (void)
730{ 496{
731 return etp_nreqs (); 497 return etp_nreqs (EIO_POOL);
732} 498}
733 499
734unsigned int eio_nready (void) 500unsigned int
501eio_nready (void)
735{ 502{
736 return etp_nready (); 503 return etp_nready (EIO_POOL);
737} 504}
738 505
739unsigned int eio_npending (void) 506unsigned int
507eio_npending (void)
740{ 508{
741 return etp_npending (); 509 return etp_npending (EIO_POOL);
742} 510}
743 511
744unsigned int eio_nthreads (void) 512unsigned int ecb_cold
513eio_nthreads (void)
745{ 514{
746 return etp_nthreads (); 515 return etp_nthreads (EIO_POOL);
747} 516}
748 517
518void ecb_cold
749void eio_set_max_poll_time (double nseconds) 519eio_set_max_poll_time (double nseconds)
750{ 520{
751 etp_set_max_poll_time (nseconds); 521 etp_set_max_poll_time (EIO_POOL, nseconds);
752} 522}
753 523
524void ecb_cold
754void eio_set_max_poll_reqs (unsigned int maxreqs) 525eio_set_max_poll_reqs (unsigned int maxreqs)
755{ 526{
756 etp_set_max_poll_reqs (maxreqs); 527 etp_set_max_poll_reqs (EIO_POOL, maxreqs);
757} 528}
758 529
530void ecb_cold
759void eio_set_max_idle (unsigned int nthreads) 531eio_set_max_idle (unsigned int nthreads)
760{ 532{
761 etp_set_max_idle (nthreads); 533 etp_set_max_idle (EIO_POOL, nthreads);
762} 534}
763 535
536void ecb_cold
537eio_set_idle_timeout (unsigned int seconds)
538{
539 etp_set_idle_timeout (EIO_POOL, seconds);
540}
541
542void ecb_cold
764void eio_set_min_parallel (unsigned int nthreads) 543eio_set_min_parallel (unsigned int nthreads)
765{ 544{
766 etp_set_min_parallel (nthreads); 545 etp_set_min_parallel (EIO_POOL, nthreads);
767} 546}
768 547
548void ecb_cold
769void eio_set_max_parallel (unsigned int nthreads) 549eio_set_max_parallel (unsigned int nthreads)
770{ 550{
771 etp_set_max_parallel (nthreads); 551 etp_set_max_parallel (EIO_POOL, nthreads);
772} 552}
773 553
774int eio_poll (void) 554int eio_poll (void)
775{ 555{
776 return etp_poll (); 556 return etp_poll (EIO_POOL);
777} 557}
778 558
779/*****************************************************************************/ 559/*****************************************************************************/
780/* work around various missing functions */ 560/* work around various missing functions */
781 561
782#if !HAVE_PREADWRITE 562#if HAVE_POSIX_CLOSE && !__linux
783# undef pread 563# define eio__close(fd) posix_close (fd, 0)
784# undef pwrite 564#else
785# define pread eio__pread 565# define eio__close(fd) close (fd)
786# define pwrite eio__pwrite
787
788static ssize_t
789eio__pread (int fd, void *buf, size_t count, off_t offset)
790{
791 ssize_t res;
792 off_t ooffset;
793
794 X_LOCK (preadwritelock);
795 ooffset = lseek (fd, 0, SEEK_CUR);
796 lseek (fd, offset, SEEK_SET);
797 res = read (fd, buf, count);
798 lseek (fd, ooffset, SEEK_SET);
799 X_UNLOCK (preadwritelock);
800
801 return res;
802}
803
804static ssize_t
805eio__pwrite (int fd, void *buf, size_t count, off_t offset)
806{
807 ssize_t res;
808 off_t ooffset;
809
810 X_LOCK (preadwritelock);
811 ooffset = lseek (fd, 0, SEEK_CUR);
812 lseek (fd, offset, SEEK_SET);
813 res = write (fd, buf, count);
814 lseek (fd, ooffset, SEEK_SET);
815 X_UNLOCK (preadwritelock);
816
817 return res;
818}
819#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}
820 576
821#ifndef HAVE_UTIMES 577#ifndef HAVE_UTIMES
822 578
823# undef utimes 579# undef utimes
824# define utimes(path,times) eio__utimes (path, times) 580# define utimes(path,times) eio__utimes (path, times)
… …
844#ifndef HAVE_FUTIMES 600#ifndef HAVE_FUTIMES
845 601
846# undef futimes 602# undef futimes
847# define futimes(fd,times) eio__futimes (fd, times) 603# define futimes(fd,times) eio__futimes (fd, times)
848 604
605static int
849static int eio__futimes (int fd, const struct timeval tv[2]) 606eio__futimes (int fd, const struct timeval tv[2])
850{ 607{
851 errno = ENOSYS; 608 errno = ENOSYS;
852 return -1; 609 return -1;
853} 610}
854 611
… …
857#if !HAVE_FDATASYNC 614#if !HAVE_FDATASYNC
858# undef fdatasync 615# undef fdatasync
859# define fdatasync(fd) fsync (fd) 616# define fdatasync(fd) fsync (fd)
860#endif 617#endif
861 618
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
862/* sync_file_range always needs emulation */ 636/* sync_file_range always needs emulation */
863int 637static int
864eio__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)
865{ 639{
866#if HAVE_SYNC_FILE_RANGE 640#if HAVE_SYNC_FILE_RANGE
867 int res; 641 int res;
868 642
… …
885 /* 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
886 * call fdatasync, as that matches the expectation of its users best */ 660 * call fdatasync, as that matches the expectation of its users best */
887 return fdatasync (fd); 661 return fdatasync (fd);
888} 662}
889 663
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
890#if !HAVE_READAHEAD 674#if !HAVE_READAHEAD
891# undef readahead 675# undef readahead
892# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 676# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
893 677
894static ssize_t 678static eio_ssize_t
895eio__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)
896{ 680{
897 size_t todo = count; 681 size_t todo = count;
898 dBUF; 682 dBUF;
899 683
… …
904 pread (fd, eio_buf, len, offset); 688 pread (fd, eio_buf, len, offset);
905 offset += len; 689 offset += len;
906 todo -= len; 690 todo -= len;
907 } 691 }
908 692
909 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 */
910 return count; 697 return 0;
911} 698}
912 699
913#endif 700#endif
914 701
915/* sendfile always needs emulation */ 702/* sendfile always needs emulation */
916static ssize_t 703static eio_ssize_t
917eio__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)
918{ 705{
706 eio_ssize_t written = 0;
919 ssize_t res; 707 eio_ssize_t res;
920 708
921 if (!count) 709 if (!count)
922 return 0; 710 return 0;
923 711
712 for (;;)
713 {
714#ifdef __APPLE__
715# undef HAVE_SENDFILE /* broken, as everything on os x */
716#endif
924#if HAVE_SENDFILE 717#if HAVE_SENDFILE
925# if __linux 718# if __linux
719 off_t soffset = offset;
926 res = sendfile (ofd, ifd, &offset, count); 720 res = sendfile (ofd, ifd, &soffset, count);
927 721
928# elif __FreeBSD__ 722# elif __FreeBSD__
929 /* 723 /*
930 * 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
931 * wasted on making it similar to other I/O functions. 725 * wasted on making it similar to other I/O functions.
932 */ 726 */
933 {
934 off_t sbytes; 727 off_t sbytes;
935 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 728 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
936 729
937 #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 */
938 /* freebsd' sendfile will return 0 on success */ 731 /* freebsd' sendfile will return 0 on success */
939 /* 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 */
940 /* not on e.g. EIO or EPIPE - sounds broken */ 733 /* not on e.g. EIO or EPIPE - sounds broken */
941 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0) 734 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
942 res = sbytes; 735 res = sbytes;
943 #endif 736 #endif
944 737
945 /* according to source inspection, this is correct, and useful behaviour */ 738 /* according to source inspection, this is correct, and useful behaviour */
946 if (sbytes) 739 if (sbytes)
947 res = sbytes; 740 res = sbytes;
948 }
949 741
950# elif defined (__APPLE__) 742# elif defined __APPLE__
951
952 {
953 off_t sbytes = count; 743 off_t sbytes = count;
954 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0); 744 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
955 745
956 /* according to the manpage, sbytes is always valid */ 746 /* according to the manpage, sbytes is always valid */
957 if (sbytes) 747 if (sbytes)
958 res = sbytes; 748 res = sbytes;
959 }
960 749
961# elif __hpux 750# elif __hpux
962 res = sendfile (ofd, ifd, offset, count, 0, 0); 751 res = sendfile (ofd, ifd, offset, count, 0, 0);
963 752
964# elif __solaris 753# elif __solaris
965 {
966 struct sendfilevec vec; 754 struct sendfilevec vec;
967 size_t sbytes; 755 size_t sbytes;
968 756
969 vec.sfv_fd = ifd; 757 vec.sfv_fd = ifd;
970 vec.sfv_flag = 0; 758 vec.sfv_flag = 0;
971 vec.sfv_off = offset; 759 vec.sfv_off = offset;
972 vec.sfv_len = count; 760 vec.sfv_len = count;
973 761
974 res = sendfilev (ofd, &vec, 1, &sbytes); 762 res = sendfilev (ofd, &vec, 1, &sbytes);
975 763
976 if (res < 0 && sbytes) 764 if (res < 0 && sbytes)
977 res = sbytes; 765 res = sbytes;
978 }
979 766
980# endif 767# endif
981 768
982#elif defined (_WIN32) 769#elif defined (_WIN32) && 0
983
984 /* 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 */
985 { 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 */
986 HANDLE h = TO_SOCKET (ifd); 774 HANDLE h = TO_SOCKET (ifd);
987 SetFilePointer (h, offset, 0, FILE_BEGIN); 775 SetFilePointer (h, offset, 0, FILE_BEGIN);
988 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0); 776 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
989 }
990 777
991#else 778#else
992 res = -1; 779 res = EIO_ENOSYS ();
993 errno = ENOSYS;
994#endif 780#endif
995 781
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
996 if (res < 0 805 if (res < 0
997 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 806 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
998 /* BSDs */ 807 /* BSDs */
999#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... */
1000 || errno == ENOTSUP 809 || errno == ENOTSUP
1001#endif 810#endif
811#ifdef EOPNOTSUPP /* windows */
1002 || errno == EOPNOTSUPP /* BSDs */ 812 || errno == EOPNOTSUPP /* BSDs */
813#endif
1003#if __solaris 814#if __solaris
1004 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 815 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
1005#endif 816#endif
1006 ) 817 )
1007 ) 818 )
… …
1011 822
1012 res = 0; 823 res = 0;
1013 824
1014 while (count) 825 while (count)
1015 { 826 {
1016 ssize_t cnt; 827 eio_ssize_t cnt;
1017 828
1018 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);
1019 830
1020 if (cnt <= 0) 831 if (cnt <= 0)
1021 { 832 {
… …
1033 844
1034 offset += cnt; 845 offset += cnt;
1035 res += cnt; 846 res += cnt;
1036 count -= cnt; 847 count -= cnt;
1037 } 848 }
849
850 FUBd;
1038 } 851 }
1039 852
1040 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
1041} 1161}
1042 1162
1043static signed char 1163static signed char
1044eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) 1164eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1045{ 1165{
1046 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */ 1166 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1047 : a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0; 1167 : a->inode < b->inode ? -1
1168 : a->inode > b->inode ? 1
1169 : 0;
1048} 1170}
1049 1171
1050#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0 1172#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1051 1173
1052#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 */
1053#define EIO_SORT_FAST 60 /* when to only use insertion sort */ 1175#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1054 1176
1055static void 1177static void
1056eio_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)
1057{ 1179{
1058 unsigned char bits [9 + sizeof (ino_t) * 8]; 1180 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1059 unsigned char *bit = bits; 1181 unsigned char *bit = bits;
1060 1182
1061 assert (CHAR_BIT == 8); 1183 assert (CHAR_BIT == 8);
1062 assert (sizeof (eio_dirent) * 8 < 256); 1184 assert (sizeof (eio_dirent) * 8 < 256);
1063 assert (offsetof (eio_dirent, inode)); /* we use 0 as sentinel */ 1185 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1064 assert (offsetof (eio_dirent, score)); /* we use 0 as sentinel */ 1186 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1065 1187
1066 if (size <= EIO_SORT_FAST) 1188 if (size <= EIO_SORT_FAST)
1067 return; 1189 return;
1068 1190
1069 /* 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 */
1070 /* 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 */
1071 /* inode_bits must contain all inodes ORed together */ 1193 /* inode_bits must contain all inodes ORed together */
1072 /* 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 */
1073 { 1195 {
1074 ino_t endianness; 1196 eio_ino_t endianness;
1075 int i, j; 1197 int i, j;
1076 1198
1077 /* 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" */
1078 for (i = 0; i < sizeof (ino_t); ++i) 1200 for (i = 0; i < sizeof (eio_ino_t); ++i)
1079 ((unsigned char *)&endianness)[i] = i; 1201 ((unsigned char *)&endianness)[i] = i;
1080 1202
1081 *bit++ = 0; 1203 *bit++ = 0;
1082 1204
1083 for (i = 0; i < sizeof (ino_t); ++i) 1205 for (i = 0; i < sizeof (eio_ino_t); ++i)
1084 { 1206 {
1085 /* shifting off the byte offsets out of "endianness" */ 1207 /* shifting off the byte offsets out of "endianness" */
1086 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8; 1208 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1087 endianness >>= 8; 1209 endianness >>= 8;
1088 1210
1089 for (j = 0; j < 8; ++j) 1211 for (j = 0; j < 8; ++j)
1090 if (inode_bits & (((ino_t)1) << (i * 8 + j))) 1212 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1091 *bit++ = offs + j; 1213 *bit++ = offs + j;
1092 } 1214 }
1093 1215
1094 for (j = 0; j < 8; ++j) 1216 for (j = 0; j < 8; ++j)
1095 if (score_bits & (1 << j)) 1217 if (score_bits & (1 << j))
1096 *bit++ = offsetof (eio_dirent, score) * 8 + j; 1218 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1097 } 1219 }
1098 1220
1099 /* now actually do the sorting (a variant of MSD radix sort) */ 1221 /* now actually do the sorting (a variant of MSD radix sort) */
1100 { 1222 {
1101 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base; 1223 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1102 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end; 1224 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1103 unsigned char *bit_stk [9 + sizeof (ino_t) * 8]; 1225 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1104 int stk_idx = 0; 1226 int stk_idx = 0;
1105 1227
1106 base_stk [stk_idx] = dents; 1228 base_stk [stk_idx] = dents;
1107 end_stk [stk_idx] = dents + size; 1229 end_stk [stk_idx] = dents + size;
1108 bit_stk [stk_idx] = bit - 1; 1230 bit_stk [stk_idx] = bit - 1;
… …
1187 } 1309 }
1188 } 1310 }
1189} 1311}
1190 1312
1191static void 1313static void
1192eio_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)
1193{ 1315{
1194 if (size <= 1) 1316 if (size <= 1)
1195 return; /* our insertion sort relies on size > 0 */ 1317 return; /* our insertion sort relies on size > 0 */
1196 1318
1197 /* 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 */
… …
1205 1327
1206/* read a full directory */ 1328/* read a full directory */
1207static void 1329static void
1208eio__scandir (eio_req *req, etp_worker *self) 1330eio__scandir (eio_req *req, etp_worker *self)
1209{ 1331{
1210 DIR *dirp;
1211 EIO_STRUCT_DIRENT *entp;
1212 char *name, *names; 1332 char *name, *names;
1213 int namesalloc = 4096; 1333 int namesalloc = 4096 - sizeof (void *) * 4;
1214 int namesoffs = 0; 1334 int namesoffs = 0;
1215 int flags = req->int1; 1335 int flags = req->int1;
1216 eio_dirent *dents = 0; 1336 eio_dirent *dents = 0;
1217 int dentalloc = 128; 1337 int dentalloc = 128;
1218 int dentoffs = 0; 1338 int dentoffs = 0;
1219 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
1220 1347
1221 req->result = -1; 1348 req->result = -1;
1222 1349
1223 if (!(flags & EIO_READDIR_DENTS)) 1350 if (!(flags & EIO_READDIR_DENTS))
1224 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); 1351 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1225 1352
1226 X_LOCK (wrklock); 1353#ifdef _WIN32
1227 /* 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
1228 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);
1424
1229 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1425 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1230 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; 1426 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1231 req->ptr2 = names = malloc (namesalloc); 1427 req->ptr2 = names = malloc (namesalloc);
1232 X_UNLOCK (wrklock);
1233 1428
1234 if (dirp && names && (!flags || dents)) 1429 if (!names || (flags && !dents))
1430 return;
1431
1235 for (;;) 1432 for (;;)
1236 { 1433 {
1434 int done;
1435
1436#ifdef _WIN32
1437 done = !dirp;
1438#else
1237 errno = 0; 1439 errno = 0;
1238 entp = readdir (dirp); 1440 entp = readdir (dirp);
1441 done = !entp;
1442#endif
1239 1443
1240 if (!entp) 1444 if (done)
1241 { 1445 {
1446#ifndef _WIN32
1447 int old_errno = errno;
1448 closedir (dirp);
1449 errno = old_errno;
1450
1242 if (errno) 1451 if (errno)
1243 break; 1452 break;
1453#endif
1244 1454
1245 /* sort etc. */ 1455 /* sort etc. */
1246 req->int1 = flags; 1456 req->int1 = flags;
1247 req->result = dentoffs; 1457 req->result = dentoffs;
1248 1458
1249 if (flags & EIO_READDIR_STAT_ORDER) 1459 if (flags & EIO_READDIR_STAT_ORDER)
1250 eio_dent_sort (dents, dentoffs, 0, inode_bits); /* sort by inode exclusively */ 1460 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1251 else if (flags & EIO_READDIR_DIRS_FIRST) 1461 else if (flags & EIO_READDIR_DIRS_FIRST)
1252 if (flags & EIO_READDIR_FOUND_UNKNOWN) 1462 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1253 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 */
1254 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))
1255 { 1515 {
1256 /* 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
1257 eio_dirent *oth = dents + dentoffs; 1523 ent = dents + dentoffs;
1258 eio_dirent *dir = dents;
1259 1524
1260 /* 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 */
1261 /* by walking from both sides and swapping if necessary */ 1526 ent->namelen = len - 1;
1262 /* also clear score, so it doesn't influence sorting */ 1527 ent->inode = D_INO (entp);
1263 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)
1264 { 1587 {
1265 if (dir->type == EIO_DT_DIR) 1588 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1266 ++dir; 1589 ent->score = 1;
1267 else if ((--oth)->type == EIO_DT_DIR) 1590 else if (!strchr (name, '.')) /* absence of dots indicate likely dirs */
1268 { 1591 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1269 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1270
1271 ++dir;
1272 }
1273 } 1592 }
1274 1593 else if (ent->type == EIO_DT_DIR)
1275 /* now sort the dirs only */ 1594 ent->score = 0;
1276 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1277 } 1595 }
1278
1279 break;
1280 }
1281
1282 /* now add the entry to our list(s) */
1283 name = entp->d_name;
1284
1285 /* skip . and .. entries */
1286 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1287 {
1288 int len = D_NAMLEN (entp) + 1;
1289
1290 while (expect_false (namesoffs + len > namesalloc))
1291 {
1292 namesalloc *= 2;
1293 X_LOCK (wrklock);
1294 req->ptr2 = names = realloc (names, namesalloc);
1295 X_UNLOCK (wrklock);
1296
1297 if (!names)
1298 break;
1299 } 1596 }
1300 1597
1301 memcpy (names + namesoffs, name, len);
1302
1303 if (dents)
1304 {
1305 struct eio_dirent *ent;
1306
1307 if (expect_false (dentoffs == dentalloc))
1308 {
1309 dentalloc *= 2;
1310 X_LOCK (wrklock);
1311 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1312 X_UNLOCK (wrklock);
1313
1314 if (!dents)
1315 break;
1316 }
1317
1318 ent = dents + dentoffs;
1319
1320 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1321 ent->namelen = len - 1;
1322 ent->inode = D_INO (entp);
1323
1324 inode_bits |= ent->inode;
1325
1326 switch (D_TYPE (entp))
1327 {
1328 default:
1329 ent->type = EIO_DT_UNKNOWN;
1330 flags |= EIO_READDIR_FOUND_UNKNOWN;
1331 break;
1332
1333 #ifdef DT_FIFO
1334 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1335 #endif
1336 #ifdef DT_CHR
1337 case DT_CHR: ent->type = EIO_DT_CHR; break;
1338 #endif
1339 #ifdef DT_MPC
1340 case DT_MPC: ent->type = EIO_DT_MPC; break;
1341 #endif
1342 #ifdef DT_DIR
1343 case DT_DIR: ent->type = EIO_DT_DIR; break;
1344 #endif
1345 #ifdef DT_NAM
1346 case DT_NAM: ent->type = EIO_DT_NAM; break;
1347 #endif
1348 #ifdef DT_BLK
1349 case DT_BLK: ent->type = EIO_DT_BLK; break;
1350 #endif
1351 #ifdef DT_MPB
1352 case DT_MPB: ent->type = EIO_DT_MPB; break;
1353 #endif
1354 #ifdef DT_REG
1355 case DT_REG: ent->type = EIO_DT_REG; break;
1356 #endif
1357 #ifdef DT_NWK
1358 case DT_NWK: ent->type = EIO_DT_NWK; break;
1359 #endif
1360 #ifdef DT_CMP
1361 case DT_CMP: ent->type = EIO_DT_CMP; break;
1362 #endif
1363 #ifdef DT_LNK
1364 case DT_LNK: ent->type = EIO_DT_LNK; break;
1365 #endif
1366 #ifdef DT_SOCK
1367 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1368 #endif
1369 #ifdef DT_DOOR
1370 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1371 #endif
1372 #ifdef DT_WHT
1373 case DT_WHT: ent->type = EIO_DT_WHT; break;
1374 #endif
1375 }
1376
1377 ent->score = 7;
1378
1379 if (flags & EIO_READDIR_DIRS_FIRST)
1380 {
1381 if (ent->type == EIO_DT_UNKNOWN)
1382 {
1383 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1384 ent->score = 1;
1385 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1386 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1387 }
1388 else if (ent->type == EIO_DT_DIR)
1389 ent->score = 0;
1390 }
1391 }
1392
1393 namesoffs += len; 1598 namesoffs += len;
1394 ++dentoffs; 1599 ++dentoffs;
1395 } 1600 }
1396 1601
1397 if (EIO_CANCELLED (req)) 1602 if (EIO_CANCELLED (req))
1398 { 1603 {
1399 errno = ECANCELED; 1604 errno = ECANCELED;
1400 break; 1605 break;
1401 } 1606 }
1607
1608#ifdef _WIN32
1609 if (!FindNextFile (dirp, &entp))
1610 {
1611 FindClose (dirp);
1612 dirp = 0;
1402 } 1613 }
1614#endif
1615 }
1403} 1616}
1404 1617
1405#ifdef PAGESIZE 1618/*****************************************************************************/
1406# 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);
1407#else 1712#else
1408static intptr_t
1409eio_pagesize (void)
1410{
1411 static intptr_t page;
1412
1413 if (!page) 1713 if (flags)
1414 page = sysconf (_SC_PAGESIZE); 1714 return EIO_ENOSYS ();
1415 1715
1416 return page; 1716 return renameat (olddirfd, oldpath, newdirfd, newpath);
1417}
1418#endif 1717#endif
1419
1420static void
1421eio_page_align (void **addr, size_t *length)
1422{
1423 intptr_t mask = eio_pagesize () - 1;
1424
1425 /* round down addr */
1426 intptr_t adj = mask & (intptr_t)*addr;
1427
1428 *addr = (void *)((intptr_t)*addr - adj);
1429 *length += adj;
1430
1431 /* round up length */
1432 *length = (*length + mask) & ~mask;
1433} 1718}
1434 1719
1435#if !_POSIX_MEMLOCK 1720/* they forgot these */
1436# define eio__mlock(a,b) ((errno = ENOSYS), -1)
1437# define eio__mlockall(a) ((errno = ENOSYS), -1)
1438#else
1439 1721
1440static int 1722static int
1441eio__mlock (void *addr, size_t length) 1723eio__truncateat (int dirfd, const char *path, off_t length)
1442{ 1724{
1443 eio_page_align (&addr, &length); 1725 int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
1726 int res;
1444 1727
1445 return mlock (addr, length); 1728 if (fd < 0)
1729 return fd;
1730
1731 res = ftruncate (fd, length);
1732 silent_close (fd);
1733 return res;
1446} 1734}
1447 1735
1448static int 1736static int
1449eio__mlockall (int flags) 1737eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
1450{ 1738{
1451 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7 1739 int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
1452 extern int mallopt (int, int); 1740 int res;
1453 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */ 1741
1742 if (fd < 0)
1743 return fd;
1744
1745 res = fstatvfs (fd, buf);
1746 silent_close (fd);
1747 return res;
1748}
1749
1454 #endif 1750#endif
1455
1456 if (EIO_MCL_CURRENT != MCL_CURRENT
1457 || EIO_MCL_FUTURE != MCL_FUTURE)
1458 {
1459 flags = 0
1460 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1461 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1462 }
1463
1464 return mlockall (flags);
1465}
1466#endif
1467
1468#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1469# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
1470#else
1471
1472int
1473eio__msync (void *mem, size_t len, int flags)
1474{
1475 eio_page_align (&mem, &len);
1476
1477 if (EIO_MS_ASYNC != MS_SYNC
1478 || EIO_MS_INVALIDATE != MS_INVALIDATE
1479 || EIO_MS_SYNC != MS_SYNC)
1480 {
1481 flags = 0
1482 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1483 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1484 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1485 }
1486
1487 return msync (mem, len, flags);
1488}
1489
1490#endif
1491
1492int
1493eio__mtouch (void *mem, size_t len, int flags)
1494{
1495 eio_page_align (&mem, &len);
1496
1497 {
1498 intptr_t addr = (intptr_t)mem;
1499 intptr_t end = addr + len;
1500 intptr_t page = eio_pagesize ();
1501
1502 if (addr < end)
1503 if (flags & EIO_MT_MODIFY) /* modify */
1504 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
1505 else
1506 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
1507 }
1508
1509 return 0;
1510}
1511 1751
1512/*****************************************************************************/ 1752/*****************************************************************************/
1513 1753
1514#define ALLOC(len) \ 1754#define ALLOC(len) \
1515 if (!req->ptr2) \ 1755 if (!req->ptr2) \
1516 { \ 1756 { \
1517 X_LOCK (wrklock); \ 1757 X_LOCK (EIO_POOL->wrklock); \
1518 req->flags |= EIO_FLAG_PTR2_FREE; \ 1758 req->flags |= EIO_FLAG_PTR2_FREE; \
1519 X_UNLOCK (wrklock); \ 1759 X_UNLOCK (EIO_POOL->wrklock); \
1520 req->ptr2 = malloc (len); \ 1760 req->ptr2 = malloc (len); \
1521 if (!req->ptr2) \ 1761 if (!req->ptr2) \
1522 { \ 1762 { \
1523 errno = ENOMEM; \ 1763 errno = ENOMEM; \
1524 req->result = -1; \ 1764 req->result = -1; \
1525 break; \ 1765 goto alloc_fail; \
1526 } \ 1766 } \
1527 } 1767 }
1528 1768
1529X_THREAD_PROC (etp_proc) 1769/*****************************************************************************/
1530{
1531 ETP_REQ *req;
1532 struct timespec ts;
1533 etp_worker *self = (etp_worker *)thr_arg;
1534 1770
1535 /* try to distribute timeouts somewhat randomly */ 1771static void
1536 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 1772eio__slurp (int fd, eio_req *req)
1773{
1774 req->result = fd;
1537 1775
1538 for (;;) 1776 if (fd < 0)
1777 return;
1778
1779 if (req->offs < 0 || !req->size) /* do we need the size? */
1539 { 1780 {
1540 X_LOCK (reqlock); 1781 off_t size = lseek (fd, 0, SEEK_END);
1541 1782
1542 for (;;)
1543 {
1544 self->req = req = reqq_shift (&req_queue);
1545
1546 if (req)
1547 break;
1548
1549 ++idle;
1550
1551 ts.tv_sec = time (0) + IDLE_TIMEOUT;
1552 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
1553 {
1554 if (idle > max_idle)
1555 {
1556 --idle;
1557 X_UNLOCK (reqlock);
1558 X_LOCK (wrklock);
1559 --started;
1560 X_UNLOCK (wrklock);
1561 goto quit;
1562 }
1563
1564 /* we are allowed to idle, so do so without any timeout */
1565 X_COND_WAIT (reqwait, reqlock);
1566 }
1567
1568 --idle;
1569 }
1570
1571 --nready;
1572
1573 X_UNLOCK (reqlock);
1574
1575 if (req->type < 0) 1783 if (req->offs < 0)
1576 goto quit; 1784 req->offs += size;
1577 1785
1578 if (!EIO_CANCELLED (req)) 1786 if (!req->size)
1579 ETP_EXECUTE (self, req); 1787 req->size = size - req->offs;
1580
1581 X_LOCK (reslock);
1582
1583 ++npending;
1584
1585 if (!reqq_push (&res_queue, req) && want_poll_cb)
1586 want_poll_cb ();
1587
1588 self->req = 0;
1589 etp_worker_clear (self);
1590
1591 X_UNLOCK (reslock);
1592 } 1788 }
1593 1789
1594quit: 1790 ALLOC (req->size);
1595 X_LOCK (wrklock); 1791 req->result = pread (fd, req->ptr2, req->size, req->offs);
1596 etp_worker_free (self);
1597 X_UNLOCK (wrklock);
1598 1792
1599 return 0; 1793 silent_close (fd);
1600}
1601 1794
1602/*****************************************************************************/ 1795alloc_fail:
1796 ;
1797}
1603 1798
1799int ecb_cold
1604int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1800eio_init (void (*want_poll)(void), void (*done_poll)(void))
1605{ 1801{
1606 return etp_init (want_poll, done_poll); 1802 eio_want_poll_cb = want_poll;
1607} 1803 eio_done_poll_cb = done_poll;
1608 1804
1805 return etp_init (EIO_POOL, 0, 0, 0);
1806}
1807
1808ecb_inline void
1609static void eio_api_destroy (eio_req *req) 1809eio_api_destroy (eio_req *req)
1610{ 1810{
1611 free (req); 1811 free (req);
1612} 1812}
1613 1813
1614#define REQ(rtype) \ 1814#define REQ(rtype) \
1615 eio_req *req; \ 1815 eio_req *req; \
1616 \ 1816 \
1617 req = (eio_req *)calloc (1, sizeof *req); \ 1817 req = (eio_req *)calloc (1, sizeof *req); \
1618 if (!req) \ 1818 if (!req) \
1619 return 0; \ 1819 return 0; \
… …
1633 { \ 1833 { \
1634 eio_api_destroy (req); \ 1834 eio_api_destroy (req); \
1635 return 0; \ 1835 return 0; \
1636 } 1836 }
1637 1837
1838#define SINGLEDOT(ptr) (0[(char *)(ptr)] == '.' && !1[(char *)(ptr)])
1839
1840static void
1638static void eio_execute (etp_worker *self, eio_req *req) 1841eio_execute (etp_worker *self, eio_req *req)
1639{ 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
1640 switch (req->type) 1872 switch (req->type)
1641 { 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;
1642 case EIO_READ: ALLOC (req->size); 1881 case EIO_READ: ALLOC (req->size);
1643 req->result = req->offs >= 0 1882 req->result = req->offs >= 0
1644 ? pread (req->int1, req->ptr2, req->size, req->offs) 1883 ? pread (req->int1, req->ptr2, req->size, req->offs)
1645 : read (req->int1, req->ptr2, req->size); break; 1884 : read (req->int1, req->ptr2, req->size); break;
1646 case EIO_WRITE: req->result = req->offs >= 0 1885 case EIO_WRITE: req->result = req->offs >= 0
1647 ? pwrite (req->int1, req->ptr2, req->size, req->offs) 1886 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1648 : write (req->int1, req->ptr2, req->size); break; 1887 : write (req->int1, req->ptr2, req->size); break;
1649 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
1650 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;
1651 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
1652 1896
1653 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1897 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1654 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;
1655 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));
1656 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
1657 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2022 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1658 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; 2023 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1659 2024
1660 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1661 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1662 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS)); 2025 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1663 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break; 2026 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1664 2027
1665 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1666 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;
1667 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1668 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;
1669 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1670 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break; 2030 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1671 2031
1672 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
1673 case EIO_CLOSE: req->result = close (req->int1); break; 2032 case EIO_CLOSE: req->result = eio__close (req->int1); break;
1674 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break; 2033 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1675 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
1676 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1677 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1678 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1679 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1680 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1681 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break;
1682
1683 case EIO_READLINK: ALLOC (PATH_MAX);
1684 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1685
1686 case EIO_SYNC: req->result = 0; sync (); break; 2034 case EIO_SYNC: req->result = 0; sync (); break;
1687 case EIO_FSYNC: req->result = fsync (req->int1); break; 2035 case EIO_FSYNC: req->result = fsync (req->int1); break;
1688 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;
1689 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;
1690 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break; 2040 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1691 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;
1692 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break; 2042 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1693 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;
1694 2044
1695 case EIO_READDIR: eio__scandir (req, self); break; 2045 case EIO_READDIR: eio__scandir (req, self); break;
1696 2046
1697 case EIO_BUSY: 2047 case EIO_BUSY:
1698#ifdef _WIN32 2048#ifdef _WIN32
… …
1707 req->result = select (0, 0, 0, 0, &tv); 2057 req->result = select (0, 0, 0, 0, &tv);
1708 } 2058 }
1709#endif 2059#endif
1710 break; 2060 break;
1711 2061
1712 case EIO_UTIME: 2062#if 0
1713 case EIO_FUTIME:
1714 {
1715 struct timeval tv[2];
1716 struct timeval *times;
1717
1718 if (req->nv1 != -1. || req->nv2 != -1.)
1719 {
1720 tv[0].tv_sec = req->nv1;
1721 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
1722 tv[1].tv_sec = req->nv2;
1723 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
1724
1725 times = tv;
1726 }
1727 else
1728 times = 0;
1729
1730 req->result = req->type == EIO_FUTIME
1731 ? futimes (req->int1, times)
1732 : utimes (req->ptr1, times);
1733 }
1734 break;
1735
1736 case EIO_GROUP: 2063 case EIO_GROUP:
1737 abort (); /* handled in eio_request */ 2064 abort (); /* handled in eio_request */
2065#endif
1738 2066
1739 case EIO_NOP: 2067 case EIO_NOP:
1740 req->result = 0; 2068 req->result = 0;
1741 break; 2069 break;
1742 2070
1743 case EIO_CUSTOM: 2071 case EIO_CUSTOM:
1744 ((void (*)(eio_req *))req->feed) (req); 2072 req->feed (req);
1745 break; 2073 break;
1746 2074
1747 default: 2075 default:
1748 errno = ENOSYS;
1749 req->result = -1; 2076 req->result = EIO_ENOSYS ();
1750 break; 2077 break;
1751 } 2078 }
1752 2079
2080alloc_fail:
1753 req->errorno = errno; 2081 req->errorno = errno;
1754} 2082}
1755 2083
1756#ifndef EIO_NO_WRAPPERS 2084#ifndef EIO_NO_WRAPPERS
1757 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
1758eio_req *eio_nop (int pri, eio_cb cb, void *data) 2096eio_req *eio_nop (int pri, eio_cb cb, void *data)
1759{ 2097{
1760 REQ (EIO_NOP); SEND; 2098 REQ (EIO_NOP); SEND;
1761} 2099}
1762 2100
… …
1778eio_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)
1779{ 2117{
1780 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;
1781} 2119}
1782 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
1783eio_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)
1784{ 2137{
1785 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;
1786} 2139}
1787 2140
… …
1793eio_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)
1794{ 2147{
1795 REQ (EIO_MLOCKALL); req->int1 = flags; SEND; 2148 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
1796} 2149}
1797 2150
1798eio_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)
1799{ 2152{
1800 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;
1801}
1802
1803eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1804{
1805 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1806} 2154}
1807 2155
1808eio_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)
1809{ 2157{
1810 REQ (EIO_CLOSE); req->int1 = fd; SEND; 2158 REQ (EIO_CLOSE); req->int1 = fd; SEND;
… …
1813eio_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)
1814{ 2162{
1815 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;
1816} 2164}
1817 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
1818eio_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)
1819{ 2172{
1820 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;
1821} 2174}
1822 2175
1823eio_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)
1824{ 2177{
1825 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;
1826} 2179}
1827 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
1828eio_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)
1829{ 2192{
1830 REQ (EIO_FSTAT); req->int1 = fd; SEND; 2193 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1831} 2194}
1832 2195
… …
1848eio_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)
1849{ 2212{
1850 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2213 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1851} 2214}
1852 2215
1853eio_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)
1854{ 2217{
1855 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;
1856} 2219}
1857 2220
1858eio_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)
… …
1878eio_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)
1879{ 2242{
1880 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2243 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1881} 2244}
1882 2245
1883eio_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)
1884{ 2247{
1885 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;
1886} 2249}
1887 2250
1888eio_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)
… …
1904eio_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)
1905{ 2268{
1906 return eio__1path (EIO_READLINK, path, pri, cb, data); 2269 return eio__1path (EIO_READLINK, path, pri, cb, data);
1907} 2270}
1908 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);
2275}
2276
1909eio_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)
1910{ 2278{
1911 return eio__1path (EIO_STAT, path, pri, cb, data); 2279 return eio__1path (EIO_STAT, path, pri, cb, data);
1912} 2280}
1913 2281
… …
1936 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND; 2304 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1937} 2305}
1938 2306
1939eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) 2307eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1940{ 2308{
1941 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; 2309 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1942} 2310}
1943 2311
1944static eio_req * 2312static eio_req *
1945eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2313eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1946{ 2314{
… …
1970eio_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)
1971{ 2339{
1972 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2340 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1973} 2341}
1974 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
1975eio_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)
1976{ 2349{
1977 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND; 2350 REQ (EIO_CUSTOM); req->feed = execute; SEND;
1978} 2351}
1979 2352
1980#endif 2353#endif
1981 2354
1982eio_req *eio_grp (eio_cb cb, void *data) 2355eio_req *eio_grp (eio_cb cb, void *data)
… …
1991#undef SEND 2364#undef SEND
1992 2365
1993/*****************************************************************************/ 2366/*****************************************************************************/
1994/* grp functions */ 2367/* grp functions */
1995 2368
2369void
1996void 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)
1997{ 2371{
1998 grp->int2 = limit; 2372 grp->int2 = limit;
1999 grp->feed = feed; 2373 grp->feed = feed;
2000 2374
2001 grp_try_feed (grp); 2375 grp_try_feed (grp);
2002} 2376}
2003 2377
2378void
2004void eio_grp_limit (eio_req *grp, int limit) 2379eio_grp_limit (eio_req *grp, int limit)
2005{ 2380{
2006 grp->int2 = limit; 2381 grp->int2 = limit;
2007 2382
2008 grp_try_feed (grp); 2383 grp_try_feed (grp);
2009} 2384}
2010 2385
2386void
2011void eio_grp_add (eio_req *grp, eio_req *req) 2387eio_grp_add (eio_req *grp, eio_req *req)
2012{ 2388{
2013 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));
2014 2390
2015 grp->flags |= EIO_FLAG_GROUPADD; 2391 grp->flags |= ETP_FLAG_GROUPADD;
2016 2392
2017 ++grp->size; 2393 ++grp->size;
2018 req->grp = grp; 2394 req->grp = grp;
2019 2395
2020 req->grp_prev = 0; 2396 req->grp_prev = 0;
… …
2027} 2403}
2028 2404
2029/*****************************************************************************/ 2405/*****************************************************************************/
2030/* misc garbage */ 2406/* misc garbage */
2031 2407
2408eio_ssize_t
2032ssize_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)
2033{ 2410{
2034 etp_worker wrk;
2035 ssize_t ret;
2036
2037 wrk.dbuf = 0;
2038
2039 ret = eio__sendfile (ofd, ifd, offset, count, &wrk); 2411 return eio__sendfile (ofd, ifd, offset, count);
2040
2041 if (wrk.dbuf)
2042 free (wrk.dbuf);
2043
2044 return ret;
2045} 2412}
2046 2413

Diff Legend

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