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

Comparing libeio/eio.c (file contents):
Revision 1.53 by root, Sat Jan 9 10:03:09 2010 UTC vs.
Revision 1.148 by root, Tue Feb 20 06:05:19 2018 UTC

1/* 1/*
2 * libeio implementation 2 * libeio implementation
3 * 3 *
4 * Copyright (c) 2007,2008,2009,2010 Marc Alexander Lehmann <libeio@schmorp.de> 4 * Copyright (c) 2007,2008,2009,2010,2011,2012,2013,2016,2017 Marc Alexander Lehmann <libeio@schmorp.de>
5 * All rights reserved. 5 * All rights reserved.
6 * 6 *
7 * Redistribution and use in source and binary forms, with or without modifica- 7 * Redistribution and use in source and binary forms, with or without modifica-
8 * tion, are permitted provided that the following conditions are met: 8 * tion, are permitted provided that the following conditions are met:
9 * 9 *
10 * 1. Redistributions of source code must retain the above copyright notice, 10 * 1. Redistributions of source code must retain the above copyright notice,
11 * this list of conditions and the following disclaimer. 11 * this list of conditions and the following disclaimer.
12 * 12 *
13 * 2. Redistributions in binary form must reproduce the above copyright 13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the 14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution. 15 * documentation and/or other materials provided with the distribution.
16 * 16 *
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR IMPLIED 17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR IMPLIED
18 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MER- 18 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MER-
19 * CHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO 19 * CHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO
20 * EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPE- 20 * EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPE-
21 * CIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, 21 * CIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
35 * and other provisions required by the GPL. If you do not delete the 35 * and other provisions required by the GPL. If you do not delete the
36 * provisions above, a recipient may use your version of this file under 36 * provisions above, a recipient may use your version of this file under
37 * either the BSD or the GPL. 37 * either the BSD or the GPL.
38 */ 38 */
39 39
40#ifndef _WIN32
41# include "config.h"
42#endif
43
40#include "eio.h" 44#include "eio.h"
41
42#ifdef EIO_STACKSIZE
43# define XTHREAD_STACKSIZE EIO_STACKSIZE
44#endif
45#include "xthread.h" 45#include "ecb.h"
46 46
47#include <errno.h> 47#include <errno.h>
48#include <stddef.h> 48#include <stddef.h>
49#include <stdlib.h> 49#include <stdlib.h>
50#include <string.h> 50#include <string.h>
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 #define O_CLOEXEC 0
106
107 /* TODO: look at how perl does stat (non-sloppy), unlink (ro-files), utime, link */
108
109 #ifdef EIO_STRUCT_STATI64
110 /* look at perl's non-sloppy stat */
111 #define stat(path,buf) _stati64 (path,buf)
112 #define fstat(fd,buf) _fstati64 (fd,buf)
113 #endif
114 #define lstat(path,buf) stat (path,buf)
115 #define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
116 #define mkdir(path,mode) _mkdir (path)
117 #define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
118
119 #define chmod(path,mode) _chmod (path, mode)
120 #define dup(fd) _dup (fd)
121 #define dup2(fd1,fd2) _dup2 (fd1, fd2)
122 #define pipe(fds) _pipe (fds, 4096, O_BINARY)
123
124 #define fcntl(fd,cmd,arg) EIO_ENOSYS ()
125 #define ioctl(fd,cmd,arg) EIO_ENOSYS ()
126 #define fchmod(fd,mode) EIO_ENOSYS ()
127 #define chown(path,uid,gid) EIO_ENOSYS ()
128 #define fchown(fd,uid,gid) EIO_ENOSYS ()
129 #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
130 #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
131 #define mknod(path,mode,dev) EIO_ENOSYS ()
132 #define sync() EIO_ENOSYS ()
133 #define readlink(path,buf,s) EIO_ENOSYS ()
134 #define statvfs(path,buf) EIO_ENOSYS ()
135 #define fstatvfs(fd,buf) EIO_ENOSYS ()
136
137 #define pread(fd,buf,count,offset) eio__pread (fd, buf, count, offset)
138 #define pwrite(fd,buf,count,offset) eio__pwrite (fd, buf, count, offset)
139
140 #if __GNUC__
141 typedef long long eio_off_t; /* signed for compatibility to msvc */
142 #else
143 typedef __int64 eio_off_t; /* unsigned not supported by msvc */
144 #endif
145
146 static eio_ssize_t
147 eio__pread (int fd, void *buf, eio_ssize_t count, eio_off_t offset)
148 {
149 OVERLAPPED o = { 0 };
150 DWORD got;
151
152 o.Offset = offset;
153 o.OffsetHigh = offset >> 32;
154
155 return ReadFile ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd), buf, count, &got, &o)
156 ? got : -1;
157 }
158
159 static eio_ssize_t
160 eio__pwrite (int fd, void *buf, eio_ssize_t count, eio_off_t offset)
161 {
162 OVERLAPPED o = { 0 };
163 DWORD got;
164
165 o.Offset = offset;
166 o.OffsetHigh = offset >> 32;
167
168 return WriteFile ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd), buf, count, &got, &o)
169 ? got : -1;
170 }
171
172 /* rename() uses MoveFile, which fails to overwrite */
173 #define rename(old,neu) eio__rename (old, neu)
174
175 static int
176 eio__rename (const char *old, const char *neu)
177 {
178 if (MoveFileEx (old, neu, MOVEFILE_REPLACE_EXISTING))
179 return 0;
180
181 /* should steal _dosmaperr */
182 switch (GetLastError ())
183 {
184 case ERROR_FILE_NOT_FOUND:
185 case ERROR_PATH_NOT_FOUND:
186 case ERROR_INVALID_DRIVE:
187 case ERROR_NO_MORE_FILES:
188 case ERROR_BAD_NETPATH:
189 case ERROR_BAD_NET_NAME:
190 case ERROR_BAD_PATHNAME:
191 case ERROR_FILENAME_EXCED_RANGE:
192 errno = ENOENT;
193 break;
194
195 default:
196 errno = EACCES;
197 break;
198 }
199
200 return -1;
201 }
202
203 /* we could even stat and see if it exists */
204 static int
205 symlink (const char *old, const char *neu)
206 {
207 #if WINVER >= 0x0600
208 int flags;
209
210 /* This tries out all combinations of SYMBOLIC_LINK_FLAG_DIRECTORY
211 * and SYMBOLIC_LINK_FLAG_ALLOW_UNPRIVILEGED_CREATE,
212 * with directory first.
213 */
214 for (flags = 3; flags >= 0; --flags)
215 if (CreateSymbolicLink (neu, old, flags))
216 return 0;
217 #endif
218
219 return EIO_ERRNO (ENOENT, -1);
220 }
221
222 /* POSIX API only, causing trouble for win32 apps */
223 #define CreateHardLink(neu,old,flags) 0 /* not really creating hardlink, still using relative paths? */
224 #define CreateSymbolicLink(neu,old,flags) 0 /* vista+ only */
225
226 struct statvfs
227 {
228 int dummy;
229 };
230
231 #define DT_DIR EIO_DT_DIR
232 #define DT_REG EIO_DT_REG
233 #define D_NAME(entp) entp.cFileName
234 #define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
235
74#else 236#else
75 237
76# include "config.h" 238 #include <sys/ioctl.h>
77# include <sys/time.h> 239 #include <sys/time.h>
78# include <sys/select.h> 240 #include <sys/select.h>
79# include <sys/mman.h>
80# include <unistd.h> 241 #include <unistd.h>
242 #include <signal.h>
243 #include <dirent.h>
244
245 #ifdef ANDROID
246 #include <sys/vfs.h>
247 #define statvfs statfs
248 #define fstatvfs fstatfs
249 #include <asm/page.h> /* supposedly limits.h does #define PAGESIZE PAGESIZE */
250 #else
251 #include <sys/statvfs.h>
252 #endif
253
254 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
255 #include <sys/mman.h>
256 #endif
257
258 #define D_NAME(entp) entp->d_name
259
260 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
261 #if __FreeBSD__ || __NetBSD__ || __OpenBSD__
262 #define _DIRENT_HAVE_D_TYPE /* sigh */
263 #define D_INO(de) (de)->d_fileno
264 #define D_NAMLEN(de) (de)->d_namlen
265 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
266 #define D_INO(de) (de)->d_ino
267 #endif
268
269 #ifdef _D_EXACT_NAMLEN
270 #undef D_NAMLEN
271 #define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
272 #endif
273
274 #ifdef _DIRENT_HAVE_D_TYPE
275 #define D_TYPE(de) (de)->d_type
276 #endif
277
278 #ifndef EIO_STRUCT_DIRENT
279 #define EIO_STRUCT_DIRENT struct dirent
280 #endif
281
282#endif
283
284#if HAVE_UTIMES
81# include <utime.h> 285# include <utime.h>
82# include <signal.h>
83# include <dirent.h>
84
85/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
86# if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
87# define _DIRENT_HAVE_D_TYPE /* sigh */
88# define D_INO(de) (de)->d_fileno
89# define D_NAMLEN(de) (de)->d_namlen
90# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
91# define D_INO(de) (de)->d_ino
92# endif 286#endif
93 287
94#ifdef _D_EXACT_NAMLEN 288#if HAVE_SYS_SYSCALL_H
95# undef D_NAMLEN 289# include <sys/syscall.h>
96# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
97#endif
98
99# ifdef _DIRENT_HAVE_D_TYPE
100# define D_TYPE(de) (de)->d_type
101# endif
102
103# ifndef EIO_STRUCT_DIRENT
104# define EIO_STRUCT_DIRENT struct dirent
105# endif
106
107#endif 290#endif
108 291
109#if HAVE_SENDFILE 292#if HAVE_SENDFILE
110# if __linux 293# if __linux
111# include <sys/sendfile.h> 294# include <sys/sendfile.h>
119# else 302# else
120# error sendfile support requested but not available 303# error sendfile support requested but not available
121# endif 304# endif
122#endif 305#endif
123 306
307#if HAVE_RENAMEAT2
308# include <sys/syscall.h>
309# include <linux/fs.h>
310#endif
311
124#ifndef D_TYPE 312#ifndef D_TYPE
125# define D_TYPE(de) 0 313# define D_TYPE(de) 0
126#endif 314#endif
127#ifndef D_INO 315#ifndef D_INO
128# define D_INO(de) 0 316# define D_INO(de) 0
129#endif 317#endif
130#ifndef D_NAMLEN 318#ifndef D_NAMLEN
131# define D_NAMLEN(de) strlen ((de)->d_name) 319# define D_NAMLEN(entp) strlen (D_NAME (entp))
132#endif 320#endif
133
134/* number of seconds after which an idle threads exit */
135#define IDLE_TIMEOUT 10
136 321
137/* used for struct dirent, AIX doesn't provide it */ 322/* used for struct dirent, AIX doesn't provide it */
138#ifndef NAME_MAX 323#ifndef NAME_MAX
139# define NAME_MAX 4096 324# define NAME_MAX 4096
140#endif 325#endif
141 326
327/* used for readlink etc. */
328#ifndef PATH_MAX
329# define PATH_MAX 0
330#endif
331
332#ifndef EIO_PATH_MIN
333# define EIO_PATH_MIN 8160
334#endif
335
336#define EIO_PATH_MAX (PATH_MAX <= EIO_PATH_MIN ? EIO_PATH_MIN : PATH_MAX)
337
142/* buffer size for various temporary buffers */ 338/* buffer size for various temporary buffers */
143#define EIO_BUFSIZE 65536 339#define EIO_BUFSIZE 65536
144 340
145#define dBUF \ 341#define dBUF \
146 char *eio_buf; \
147 ETP_WORKER_LOCK (self); \
148 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 342 char *eio_buf = malloc (EIO_BUFSIZE); \
149 ETP_WORKER_UNLOCK (self); \
150 errno = ENOMEM; \ 343 errno = ENOMEM; \
151 if (!eio_buf) \ 344 if (!eio_buf) \
152 return -1; 345 return -1
153 346
154#define EIO_TICKS ((1000000 + 1023) >> 10) 347#define FUBd \
348 free (eio_buf)
155 349
156/*****************************************************************************/ 350/*****************************************************************************/
157 351
158#if __GNUC__ >= 3 352struct etp_tmpbuf;
159# define expect(expr,value) __builtin_expect ((expr),(value)) 353
354#if _POSIX_VERSION >= 200809L
355 #define HAVE_AT 1
356 #define WD2FD(wd) ((wd) ? (wd)->fd : AT_FDCWD)
357 #ifndef O_SEARCH
358 #define O_SEARCH O_RDONLY
359 #endif
160#else 360#else
161# define expect(expr,value) (expr) 361 #define HAVE_AT 0
362 static const char *wd_expand (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path);
162#endif 363#endif
163 364
164#define expect_false(expr) expect ((expr) != 0, 0) 365struct eio_pwd
165#define expect_true(expr) expect ((expr) != 0, 1) 366{
367#if HAVE_AT
368 int fd;
369#endif
370 int len;
371 char str[1]; /* actually, a 0-terminated canonical path */
372};
166 373
167/*****************************************************************************/ 374/*****************************************************************************/
168 375
169#define ETP_PRI_MIN EIO_PRI_MIN 376#define ETP_PRI_MIN EIO_PRI_MIN
170#define ETP_PRI_MAX EIO_PRI_MAX 377#define ETP_PRI_MAX EIO_PRI_MAX
171 378
379#define ETP_TYPE_QUIT -1
380#define ETP_TYPE_GROUP EIO_GROUP
381
382static void eio_nop_callback (void) { }
383static void (*eio_want_poll_cb)(void) = eio_nop_callback;
384static void (*eio_done_poll_cb)(void) = eio_nop_callback;
385
386#define ETP_WANT_POLL(pool) eio_want_poll_cb ()
387#define ETP_DONE_POLL(pool) eio_done_poll_cb ()
388
172struct etp_worker; 389struct etp_worker;
173
174#define ETP_REQ eio_req 390#define ETP_REQ eio_req
175#define ETP_DESTROY(req) eio_destroy (req) 391#define ETP_DESTROY(req) eio_destroy (req)
176static int eio_finish (eio_req *req); 392static int eio_finish (eio_req *req);
177#define ETP_FINISH(req) eio_finish (req) 393#define ETP_FINISH(req) eio_finish (req)
178static void eio_execute (struct etp_worker *self, eio_req *req); 394static void eio_execute (struct etp_worker *self, eio_req *req);
179#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) 395#define ETP_EXECUTE(wrk,req) eio_execute (wrk, req)
180 396
181#define ETP_WORKER_CLEAR(req) \ 397#include "etp.c"
182 if (wrk->dbuf) \
183 { \
184 free (wrk->dbuf); \
185 wrk->dbuf = 0; \
186 } \
187 \
188 if (wrk->dirp) \
189 { \
190 closedir (wrk->dirp); \
191 wrk->dirp = 0; \
192 }
193 398
194#define ETP_WORKER_COMMON \ 399static struct etp_pool eio_pool;
195 void *dbuf; \ 400#define EIO_POOL (&eio_pool)
196 DIR *dirp;
197 401
198/*****************************************************************************/ 402/*****************************************************************************/
199 403
200#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
201
202/* calculate time difference in ~1/EIO_TICKS of a second */
203static int tvdiff (struct timeval *tv1, struct timeval *tv2)
204{
205 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
206 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
207}
208
209static unsigned int started, idle, wanted = 4;
210
211static void (*want_poll_cb) (void);
212static void (*done_poll_cb) (void);
213
214static unsigned int max_poll_time; /* reslock */
215static unsigned int max_poll_reqs; /* reslock */
216
217static volatile unsigned int nreqs; /* reqlock */
218static volatile unsigned int nready; /* reqlock */
219static volatile unsigned int npending; /* reqlock */
220static volatile unsigned int max_idle = 4;
221
222static mutex_t wrklock = X_MUTEX_INIT;
223static mutex_t reslock = X_MUTEX_INIT;
224static mutex_t reqlock = X_MUTEX_INIT;
225static cond_t reqwait = X_COND_INIT;
226
227#if !HAVE_PREADWRITE
228/*
229 * make our pread/pwrite emulation safe against themselves, but not against
230 * normal read/write by using a mutex. slows down execution a lot,
231 * but that's your problem, not mine.
232 */
233static mutex_t preadwritelock = X_MUTEX_INIT;
234#endif
235
236typedef struct etp_worker
237{
238 /* locked by wrklock */
239 struct etp_worker *prev, *next;
240
241 thread_t tid;
242
243 /* locked by reslock, reqlock or wrklock */
244 ETP_REQ *req; /* currently processed request */
245
246 ETP_WORKER_COMMON
247} etp_worker;
248
249static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */
250
251#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
252#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
253
254/* worker threads management */
255
256static void etp_worker_clear (etp_worker *wrk)
257{
258 ETP_WORKER_CLEAR (wrk);
259}
260
261static void etp_worker_free (etp_worker *wrk)
262{
263 wrk->next->prev = wrk->prev;
264 wrk->prev->next = wrk->next;
265
266 free (wrk);
267}
268
269static unsigned int etp_nreqs (void)
270{
271 int retval;
272 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
273 retval = nreqs;
274 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
275 return retval;
276}
277
278static unsigned int etp_nready (void)
279{
280 unsigned int retval;
281
282 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
283 retval = nready;
284 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
285
286 return retval;
287}
288
289static unsigned int etp_npending (void)
290{
291 unsigned int retval;
292
293 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
294 retval = npending;
295 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
296
297 return retval;
298}
299
300static unsigned int etp_nthreads (void)
301{
302 unsigned int retval;
303
304 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
305 retval = started;
306 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
307
308 return retval;
309}
310
311/*
312 * a somewhat faster data structure might be nice, but
313 * with 8 priorities this actually needs <20 insns
314 * per shift, the most expensive operation.
315 */
316typedef struct {
317 ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */
318 int size;
319} etp_reqq;
320
321static etp_reqq req_queue;
322static etp_reqq res_queue;
323
324static int reqq_push (etp_reqq *q, ETP_REQ *req)
325{
326 int pri = req->pri;
327 req->next = 0;
328
329 if (q->qe[pri])
330 {
331 q->qe[pri]->next = req;
332 q->qe[pri] = req;
333 }
334 else
335 q->qe[pri] = q->qs[pri] = req;
336
337 return q->size++;
338}
339
340static ETP_REQ *reqq_shift (etp_reqq *q)
341{
342 int pri;
343
344 if (!q->size)
345 return 0;
346
347 --q->size;
348
349 for (pri = ETP_NUM_PRI; pri--; )
350 {
351 eio_req *req = q->qs[pri];
352
353 if (req)
354 {
355 if (!(q->qs[pri] = (eio_req *)req->next))
356 q->qe[pri] = 0;
357
358 return req;
359 }
360 }
361
362 abort ();
363}
364
365static void etp_atfork_prepare (void)
366{
367 X_LOCK (wrklock);
368 X_LOCK (reqlock);
369 X_LOCK (reslock);
370#if !HAVE_PREADWRITE
371 X_LOCK (preadwritelock);
372#endif
373}
374
375static void etp_atfork_parent (void)
376{
377#if !HAVE_PREADWRITE
378 X_UNLOCK (preadwritelock);
379#endif
380 X_UNLOCK (reslock);
381 X_UNLOCK (reqlock);
382 X_UNLOCK (wrklock);
383}
384
385static void etp_atfork_child (void)
386{
387 ETP_REQ *prv;
388
389 while ((prv = reqq_shift (&req_queue)))
390 ETP_DESTROY (prv);
391
392 while ((prv = reqq_shift (&res_queue)))
393 ETP_DESTROY (prv);
394
395 while (wrk_first.next != &wrk_first)
396 {
397 etp_worker *wrk = wrk_first.next;
398
399 if (wrk->req)
400 ETP_DESTROY (wrk->req);
401
402 etp_worker_clear (wrk);
403 etp_worker_free (wrk);
404 }
405
406 started = 0;
407 idle = 0;
408 nreqs = 0;
409 nready = 0;
410 npending = 0;
411
412 etp_atfork_parent ();
413}
414
415static void 404static void
416etp_once_init (void)
417{
418 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
419}
420
421static int
422etp_init (void (*want_poll)(void), void (*done_poll)(void))
423{
424 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
425
426 pthread_once (&doinit, etp_once_init);
427
428 want_poll_cb = want_poll;
429 done_poll_cb = done_poll;
430
431 return 0;
432}
433
434X_THREAD_PROC (etp_proc);
435
436static void etp_start_thread (void)
437{
438 etp_worker *wrk = calloc (1, sizeof (etp_worker));
439
440 /*TODO*/
441 assert (("unable to allocate worker thread data", wrk));
442
443 X_LOCK (wrklock);
444
445 if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
446 {
447 wrk->prev = &wrk_first;
448 wrk->next = wrk_first.next;
449 wrk_first.next->prev = wrk;
450 wrk_first.next = wrk;
451 ++started;
452 }
453 else
454 free (wrk);
455
456 X_UNLOCK (wrklock);
457}
458
459static void etp_maybe_start_thread (void)
460{
461 if (expect_true (etp_nthreads () >= wanted))
462 return;
463
464 /* todo: maybe use idle here, but might be less exact */
465 if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
466 return;
467
468 etp_start_thread ();
469}
470
471static void etp_end_thread (void)
472{
473 eio_req *req = calloc (1, sizeof (eio_req));
474
475 req->type = -1;
476 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
477
478 X_LOCK (reqlock);
479 reqq_push (&req_queue, req);
480 X_COND_SIGNAL (reqwait);
481 X_UNLOCK (reqlock);
482
483 X_LOCK (wrklock);
484 --started;
485 X_UNLOCK (wrklock);
486}
487
488static int etp_poll (void)
489{
490 unsigned int maxreqs;
491 unsigned int maxtime;
492 struct timeval tv_start, tv_now;
493
494 X_LOCK (reslock);
495 maxreqs = max_poll_reqs;
496 maxtime = max_poll_time;
497 X_UNLOCK (reslock);
498
499 if (maxtime)
500 gettimeofday (&tv_start, 0);
501
502 for (;;)
503 {
504 ETP_REQ *req;
505
506 etp_maybe_start_thread ();
507
508 X_LOCK (reslock);
509 req = reqq_shift (&res_queue);
510
511 if (req)
512 {
513 --npending;
514
515 if (!res_queue.size && done_poll_cb)
516 done_poll_cb ();
517 }
518
519 X_UNLOCK (reslock);
520
521 if (!req)
522 return 0;
523
524 X_LOCK (reqlock);
525 --nreqs;
526 X_UNLOCK (reqlock);
527
528 if (expect_false (req->type == EIO_GROUP && req->size))
529 {
530 req->int1 = 1; /* mark request as delayed */
531 continue;
532 }
533 else
534 {
535 int res = ETP_FINISH (req);
536 if (expect_false (res))
537 return res;
538 }
539
540 if (expect_false (maxreqs && !--maxreqs))
541 break;
542
543 if (maxtime)
544 {
545 gettimeofday (&tv_now, 0);
546
547 if (tvdiff (&tv_start, &tv_now) >= maxtime)
548 break;
549 }
550 }
551
552 errno = EAGAIN;
553 return -1;
554}
555
556static void etp_cancel (ETP_REQ *req)
557{
558 X_LOCK (wrklock);
559 req->flags |= EIO_FLAG_CANCELLED;
560 X_UNLOCK (wrklock);
561
562 eio_grp_cancel (req);
563}
564
565static void etp_submit (ETP_REQ *req)
566{
567 req->pri -= ETP_PRI_MIN;
568
569 if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
570 if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
571
572 if (expect_false (req->type == EIO_GROUP))
573 {
574 /* I hope this is worth it :/ */
575 X_LOCK (reqlock);
576 ++nreqs;
577 X_UNLOCK (reqlock);
578
579 X_LOCK (reslock);
580
581 ++npending;
582
583 if (!reqq_push (&res_queue, req) && want_poll_cb)
584 want_poll_cb ();
585
586 X_UNLOCK (reslock);
587 }
588 else
589 {
590 X_LOCK (reqlock);
591 ++nreqs;
592 ++nready;
593 reqq_push (&req_queue, req);
594 X_COND_SIGNAL (reqwait);
595 X_UNLOCK (reqlock);
596
597 etp_maybe_start_thread ();
598 }
599}
600
601static void etp_set_max_poll_time (double nseconds)
602{
603 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
604 max_poll_time = nseconds * EIO_TICKS;
605 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
606}
607
608static void etp_set_max_poll_reqs (unsigned int maxreqs)
609{
610 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
611 max_poll_reqs = maxreqs;
612 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
613}
614
615static void etp_set_max_idle (unsigned int nthreads)
616{
617 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
618 max_idle = nthreads <= 0 ? 1 : nthreads;
619 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
620}
621
622static void etp_set_min_parallel (unsigned int nthreads)
623{
624 if (wanted < nthreads)
625 wanted = nthreads;
626}
627
628static void etp_set_max_parallel (unsigned int nthreads)
629{
630 if (wanted > nthreads)
631 wanted = nthreads;
632
633 while (started > wanted)
634 etp_end_thread ();
635}
636
637/*****************************************************************************/
638
639static void grp_try_feed (eio_req *grp) 405grp_try_feed (eio_req *grp)
640{ 406{
641 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 407 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
642 { 408 {
643 grp->flags &= ~EIO_FLAG_GROUPADD; 409 grp->flags &= ~ETP_FLAG_GROUPADD;
644 410
645 EIO_FEED (grp); 411 EIO_FEED (grp);
646 412
647 /* stop if no progress has been made */ 413 /* stop if no progress has been made */
648 if (!(grp->flags & EIO_FLAG_GROUPADD)) 414 if (!(grp->flags & ETP_FLAG_GROUPADD))
649 { 415 {
650 grp->feed = 0; 416 grp->feed = 0;
651 break; 417 break;
652 } 418 }
653 } 419 }
654} 420}
655 421
422static int
656static int grp_dec (eio_req *grp) 423grp_dec (eio_req *grp)
657{ 424{
658 --grp->size; 425 --grp->size;
659 426
660 /* call feeder, if applicable */ 427 /* call feeder, if applicable */
661 grp_try_feed (grp); 428 grp_try_feed (grp);
662 429
663 /* finish, if done */ 430 /* finish, if done */
664 if (!grp->size && grp->int1) 431 if (!grp->size && grp->flags & ETP_FLAG_DELAYED)
665 return eio_finish (grp); 432 return eio_finish (grp);
666 else 433 else
667 return 0; 434 return 0;
668} 435}
669 436
437static void
670void eio_destroy (eio_req *req) 438eio_destroy (eio_req *req)
671{ 439{
672 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); 440 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
673 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); 441 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
674 442
675 EIO_DESTROY (req); 443 EIO_DESTROY (req);
676} 444}
677 445
446static int
678static int eio_finish (eio_req *req) 447eio_finish (eio_req *req)
679{ 448{
680 int res = EIO_FINISH (req); 449 int res = EIO_FINISH (req);
681 450
682 if (req->grp) 451 if (req->grp)
683 { 452 {
691 if (grp->grp_first == req) 460 if (grp->grp_first == req)
692 grp->grp_first = req->grp_next; 461 grp->grp_first = req->grp_next;
693 462
694 res2 = grp_dec (grp); 463 res2 = grp_dec (grp);
695 464
696 if (!res && res2) 465 if (!res)
697 res = res2; 466 res = res2;
698 } 467 }
699 468
700 eio_destroy (req); 469 eio_destroy (req);
701 470
702 return res; 471 return res;
703} 472}
704 473
474void
705void eio_grp_cancel (eio_req *grp) 475eio_grp_cancel (eio_req *grp)
706{ 476{
707 for (grp = grp->grp_first; grp; grp = grp->grp_next) 477 etp_grp_cancel (EIO_POOL, grp);
708 eio_cancel (grp);
709} 478}
710 479
480void
711void eio_cancel (eio_req *req) 481eio_cancel (eio_req *req)
712{ 482{
713 etp_cancel (req); 483 etp_cancel (EIO_POOL, req);
714} 484}
715 485
486void
716void eio_submit (eio_req *req) 487eio_submit (eio_req *req)
717{ 488{
718 etp_submit (req); 489 etp_submit (EIO_POOL, req);
719} 490}
720 491
721unsigned int eio_nreqs (void) 492unsigned int
493eio_nreqs (void)
722{ 494{
723 return etp_nreqs (); 495 return etp_nreqs (EIO_POOL);
724} 496}
725 497
726unsigned int eio_nready (void) 498unsigned int
499eio_nready (void)
727{ 500{
728 return etp_nready (); 501 return etp_nready (EIO_POOL);
729} 502}
730 503
731unsigned int eio_npending (void) 504unsigned int
505eio_npending (void)
732{ 506{
733 return etp_npending (); 507 return etp_npending (EIO_POOL);
734} 508}
735 509
736unsigned int eio_nthreads (void) 510unsigned int ecb_cold
511eio_nthreads (void)
737{ 512{
738 return etp_nthreads (); 513 return etp_nthreads (EIO_POOL);
739} 514}
740 515
516void ecb_cold
741void eio_set_max_poll_time (double nseconds) 517eio_set_max_poll_time (double nseconds)
742{ 518{
743 etp_set_max_poll_time (nseconds); 519 etp_set_max_poll_time (EIO_POOL, nseconds);
744} 520}
745 521
522void ecb_cold
746void eio_set_max_poll_reqs (unsigned int maxreqs) 523eio_set_max_poll_reqs (unsigned int maxreqs)
747{ 524{
748 etp_set_max_poll_reqs (maxreqs); 525 etp_set_max_poll_reqs (EIO_POOL, maxreqs);
749} 526}
750 527
528void ecb_cold
751void eio_set_max_idle (unsigned int nthreads) 529eio_set_max_idle (unsigned int nthreads)
752{ 530{
753 etp_set_max_idle (nthreads); 531 etp_set_max_idle (EIO_POOL, nthreads);
754} 532}
755 533
534void ecb_cold
535eio_set_idle_timeout (unsigned int seconds)
536{
537 etp_set_idle_timeout (EIO_POOL, seconds);
538}
539
540void ecb_cold
756void eio_set_min_parallel (unsigned int nthreads) 541eio_set_min_parallel (unsigned int nthreads)
757{ 542{
758 etp_set_min_parallel (nthreads); 543 etp_set_min_parallel (EIO_POOL, nthreads);
759} 544}
760 545
546void ecb_cold
761void eio_set_max_parallel (unsigned int nthreads) 547eio_set_max_parallel (unsigned int nthreads)
762{ 548{
763 etp_set_max_parallel (nthreads); 549 etp_set_max_parallel (EIO_POOL, nthreads);
764} 550}
765 551
766int eio_poll (void) 552int eio_poll (void)
767{ 553{
768 return etp_poll (); 554 return etp_poll (EIO_POOL);
769} 555}
770 556
771/*****************************************************************************/ 557/*****************************************************************************/
772/* work around various missing functions */ 558/* work around various missing functions */
773 559
774#if !HAVE_PREADWRITE 560#if HAVE_POSIX_CLOSE && !__linux
775# undef pread 561# define eio__close(fd) posix_close (fd, 0)
776# undef pwrite 562#else
777# define pread eio__pread 563# define eio__close(fd) close (fd)
778# define pwrite eio__pwrite
779
780static ssize_t
781eio__pread (int fd, void *buf, size_t count, off_t offset)
782{
783 ssize_t res;
784 off_t ooffset;
785
786 X_LOCK (preadwritelock);
787 ooffset = lseek (fd, 0, SEEK_CUR);
788 lseek (fd, offset, SEEK_SET);
789 res = read (fd, buf, count);
790 lseek (fd, ooffset, SEEK_SET);
791 X_UNLOCK (preadwritelock);
792
793 return res;
794}
795
796static ssize_t
797eio__pwrite (int fd, void *buf, size_t count, off_t offset)
798{
799 ssize_t res;
800 off_t ooffset;
801
802 X_LOCK (preadwritelock);
803 ooffset = lseek (fd, 0, SEEK_CUR);
804 lseek (fd, offset, SEEK_SET);
805 res = write (fd, buf, count);
806 lseek (fd, ooffset, SEEK_SET);
807 X_UNLOCK (preadwritelock);
808
809 return res;
810}
811#endif 564#endif
812 565
566/* close() without disturbing errno */
567static void
568silent_close (int fd)
569{
570 int saved_errno = errno;
571 eio__close (fd);
572 errno = saved_errno;
573}
574
813#ifndef HAVE_FUTIMES 575#ifndef HAVE_UTIMES
814 576
815# undef utimes 577# undef utimes
816# undef futimes
817# define utimes(path,times) eio__utimes (path, times) 578# define utimes(path,times) eio__utimes (path, times)
818# define futimes(fd,times) eio__futimes (fd, times)
819 579
820static int 580static int
821eio__utimes (const char *filename, const struct timeval times[2]) 581eio__utimes (const char *filename, const struct timeval times[2])
822{ 582{
823 if (times) 583 if (times)
831 } 591 }
832 else 592 else
833 return utime (filename, 0); 593 return utime (filename, 0);
834} 594}
835 595
596#endif
597
598#ifndef HAVE_FUTIMES
599
600# undef futimes
601# define futimes(fd,times) eio__futimes (fd, times)
602
603static int
836static int eio__futimes (int fd, const struct timeval tv[2]) 604eio__futimes (int fd, const struct timeval tv[2])
837{ 605{
838 errno = ENOSYS; 606 errno = ENOSYS;
839 return -1; 607 return -1;
840} 608}
841 609
844#if !HAVE_FDATASYNC 612#if !HAVE_FDATASYNC
845# undef fdatasync 613# undef fdatasync
846# define fdatasync(fd) fsync (fd) 614# define fdatasync(fd) fsync (fd)
847#endif 615#endif
848 616
617static int
618eio__syncfs (int fd)
619{
620 int res;
621
622#if HAVE_SYS_SYNCFS
623 res = (int)syscall (__NR_syncfs, (int)(fd));
624#else
625 res = EIO_ENOSYS ();
626#endif
627
628 if (res < 0 && errno == ENOSYS && fd >= 0)
629 sync ();
630
631 return res;
632}
633
849/* sync_file_range always needs emulation */ 634/* sync_file_range always needs emulation */
850int 635static int
851eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags) 636eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
852{ 637{
853#if HAVE_SYNC_FILE_RANGE 638#if HAVE_SYNC_FILE_RANGE
854 int res; 639 int res;
855 640
872 /* even though we could play tricks with the flags, it's better to always 657 /* even though we could play tricks with the flags, it's better to always
873 * call fdatasync, as that matches the expectation of its users best */ 658 * call fdatasync, as that matches the expectation of its users best */
874 return fdatasync (fd); 659 return fdatasync (fd);
875} 660}
876 661
662static int
663eio__fallocate (int fd, int mode, off_t offset, size_t len)
664{
665#if HAVE_LINUX_FALLOCATE
666 return fallocate (fd, mode, offset, len);
667#else
668 return EIO_ENOSYS ();
669#endif
670}
671
877#if !HAVE_READAHEAD 672#if !HAVE_READAHEAD
878# undef readahead 673# undef readahead
879# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 674# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
880 675
881static ssize_t 676static eio_ssize_t
882eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) 677eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
883{ 678{
884 size_t todo = count; 679 size_t todo = count;
885 dBUF; 680 dBUF;
886 681
891 pread (fd, eio_buf, len, offset); 686 pread (fd, eio_buf, len, offset);
892 offset += len; 687 offset += len;
893 todo -= len; 688 todo -= len;
894 } 689 }
895 690
896 errno = 0; 691 FUBd;
692
693 /* linux's readahead basically only fails for EBADF or EINVAL (not mmappable) */
694 /* but not for e.g. EIO or eof, so we also never fail */
897 return count; 695 return 0;
898} 696}
899 697
900#endif 698#endif
901 699
902/* sendfile always needs emulation */ 700/* sendfile always needs emulation */
903static ssize_t 701static eio_ssize_t
904eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) 702eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
905{ 703{
704 eio_ssize_t written = 0;
906 ssize_t res; 705 eio_ssize_t res;
907 706
908 if (!count) 707 if (!count)
909 return 0; 708 return 0;
910 709
710 for (;;)
711 {
712#ifdef __APPLE__
713# undef HAVE_SENDFILE /* broken, as everything on os x */
714#endif
911#if HAVE_SENDFILE 715#if HAVE_SENDFILE
912# if __linux 716# if __linux
717 off_t soffset = offset;
913 res = sendfile (ofd, ifd, &offset, count); 718 res = sendfile (ofd, ifd, &soffset, count);
914 719
915# elif __FreeBSD__ 720# elif __FreeBSD__
916 /* 721 /*
917 * Of course, the freebsd sendfile is a dire hack with no thoughts 722 * Of course, the freebsd sendfile is a dire hack with no thoughts
918 * wasted on making it similar to other I/O functions. 723 * wasted on making it similar to other I/O functions.
919 */ 724 */
920 {
921 off_t sbytes; 725 off_t sbytes;
922 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 726 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
923 727
924 #if 0 /* according to the manpage, this is correct, but broken behaviour */ 728 #if 0 /* according to the manpage, this is correct, but broken behaviour */
925 /* freebsd' sendfile will return 0 on success */ 729 /* freebsd' sendfile will return 0 on success */
926 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */ 730 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
927 /* not on e.g. EIO or EPIPE - sounds broken */ 731 /* not on e.g. EIO or EPIPE - sounds broken */
928 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0) 732 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
929 res = sbytes; 733 res = sbytes;
930 #endif 734 #endif
931 735
932 /* according to source inspection, this is correct, and useful behaviour */ 736 /* according to source inspection, this is correct, and useful behaviour */
933 if (sbytes) 737 if (sbytes)
934 res = sbytes; 738 res = sbytes;
935 }
936 739
937# elif defined (__APPLE__) 740# elif defined __APPLE__
938
939 {
940 off_t sbytes = count; 741 off_t sbytes = count;
941 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0); 742 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
942 743
943 /* according to the manpage, sbytes is always valid */ 744 /* according to the manpage, sbytes is always valid */
944 if (sbytes) 745 if (sbytes)
945 res = sbytes; 746 res = sbytes;
946 }
947 747
948# elif __hpux 748# elif __hpux
949 res = sendfile (ofd, ifd, offset, count, 0, 0); 749 res = sendfile (ofd, ifd, offset, count, 0, 0);
950 750
951# elif __solaris 751# elif __solaris
952 {
953 struct sendfilevec vec; 752 struct sendfilevec vec;
954 size_t sbytes; 753 size_t sbytes;
955 754
956 vec.sfv_fd = ifd; 755 vec.sfv_fd = ifd;
957 vec.sfv_flag = 0; 756 vec.sfv_flag = 0;
958 vec.sfv_off = offset; 757 vec.sfv_off = offset;
959 vec.sfv_len = count; 758 vec.sfv_len = count;
960 759
961 res = sendfilev (ofd, &vec, 1, &sbytes); 760 res = sendfilev (ofd, &vec, 1, &sbytes);
962 761
963 if (res < 0 && sbytes) 762 if (res < 0 && sbytes)
964 res = sbytes; 763 res = sbytes;
965 }
966 764
967# endif 765# endif
968 766
969#elif defined (_WIN32) 767#elif defined (_WIN32) && 0
970
971 /* does not work, just for documentation of what would need to be done */ 768 /* does not work, just for documentation of what would need to be done */
972 { 769 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
770 /* libeio guarantees that the file offset does not change, and windows */
771 /* has no way to get an independent handle to the same file description */
973 HANDLE h = TO_SOCKET (ifd); 772 HANDLE h = TO_SOCKET (ifd);
974 SetFilePointer (h, offset, 0, FILE_BEGIN); 773 SetFilePointer (h, offset, 0, FILE_BEGIN);
975 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0); 774 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
976 }
977 775
978#else 776#else
979 res = -1; 777 res = EIO_ENOSYS ();
980 errno = ENOSYS;
981#endif 778#endif
982 779
780 /* we assume sendfile can copy at least 128mb in one go */
781 if (res <= 128 * 1024 * 1024)
782 {
783 if (res > 0)
784 written += res;
785
786 if (written)
787 return written;
788
789 break;
790 }
791 else
792 {
793 /* if we requested more, then probably the kernel was lazy */
794 written += res;
795 offset += res;
796 count -= res;
797
798 if (!count)
799 return written;
800 }
801 }
802
983 if (res < 0 803 if (res < 0
984 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 804 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
985 /* BSDs */ 805 /* BSDs */
986#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */ 806#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
987 || errno == ENOTSUP 807 || errno == ENOTSUP
988#endif 808#endif
809#ifdef EOPNOTSUPP /* windows */
989 || errno == EOPNOTSUPP /* BSDs */ 810 || errno == EOPNOTSUPP /* BSDs */
811#endif
990#if __solaris 812#if __solaris
991 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 813 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
992#endif 814#endif
993 ) 815 )
994 ) 816 )
998 820
999 res = 0; 821 res = 0;
1000 822
1001 while (count) 823 while (count)
1002 { 824 {
1003 ssize_t cnt; 825 eio_ssize_t cnt;
1004 826
1005 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 827 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
1006 828
1007 if (cnt <= 0) 829 if (cnt <= 0)
1008 { 830 {
1020 842
1021 offset += cnt; 843 offset += cnt;
1022 res += cnt; 844 res += cnt;
1023 count -= cnt; 845 count -= cnt;
1024 } 846 }
847
848 FUBd;
1025 } 849 }
1026 850
1027 return res; 851 return res;
852}
853
854#ifdef PAGESIZE
855# define eio_pagesize() PAGESIZE
856#else
857static intptr_t
858eio_pagesize (void)
859{
860 static intptr_t page;
861
862 if (!page)
863 page = sysconf (_SC_PAGESIZE);
864
865 return page;
866}
867#endif
868
869static void
870eio_page_align (void **addr, size_t *length)
871{
872 intptr_t mask = eio_pagesize () - 1;
873
874 /* round down addr */
875 intptr_t adj = mask & (intptr_t)*addr;
876
877 *addr = (void *)((intptr_t)*addr - adj);
878 *length += adj;
879
880 /* round up length */
881 *length = (*length + mask) & ~mask;
882}
883
884#if !_POSIX_MEMLOCK
885# define eio__mlockall(a) EIO_ENOSYS ()
886#else
887
888static int
889eio__mlockall (int flags)
890{
891 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
892 extern int mallopt (int, int);
893 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
894 #endif
895
896 if (EIO_MCL_CURRENT != MCL_CURRENT
897 || EIO_MCL_FUTURE != MCL_FUTURE)
898 {
899 flags = 0
900 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
901 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
902 }
903
904 return mlockall (flags);
905}
906#endif
907
908#if !_POSIX_MEMLOCK_RANGE
909# define eio__mlock(a,b) EIO_ENOSYS ()
910#else
911
912static int
913eio__mlock (void *addr, size_t length)
914{
915 eio_page_align (&addr, &length);
916
917 return mlock (addr, length);
918}
919
920#endif
921
922#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
923# define eio__msync(a,b,c) EIO_ENOSYS ()
924#else
925
926static int
927eio__msync (void *mem, size_t len, int flags)
928{
929 eio_page_align (&mem, &len);
930
931 if (EIO_MS_ASYNC != MS_SYNC
932 || EIO_MS_INVALIDATE != MS_INVALIDATE
933 || EIO_MS_SYNC != MS_SYNC)
934 {
935 flags = 0
936 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
937 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
938 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
939 }
940
941 return msync (mem, len, flags);
942}
943
944#endif
945
946static int
947eio__mtouch (eio_req *req)
948{
949 void *mem = req->ptr2;
950 size_t len = req->size;
951 int flags = req->int1;
952
953 eio_page_align (&mem, &len);
954
955 {
956 intptr_t addr = (intptr_t)mem;
957 intptr_t end = addr + len;
958 intptr_t page = eio_pagesize ();
959
960 if (addr < end)
961 if (flags & EIO_MT_MODIFY) /* modify */
962 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
963 else
964 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
965 }
966
967 return 0;
968}
969
970/*****************************************************************************/
971/* requests implemented outside eio_execute, because they are so large */
972
973static void
974eio__lseek (eio_req *req)
975{
976 /* this usually gets optimised away completely, or your compiler sucks, */
977 /* or the whence constants really are not 0, 1, 2 */
978 int whence = req->int2 == EIO_SEEK_SET ? SEEK_SET
979 : req->int2 == EIO_SEEK_CUR ? SEEK_CUR
980 : req->int2 == EIO_SEEK_END ? SEEK_END
981 : req->int2;
982
983 req->offs = lseek (req->int1, req->offs, whence);
984 req->result = req->offs == (off_t)-1 ? -1 : 0;
985}
986
987/* result will always end up in tmpbuf, there is always space for adding a 0-byte */
988static int
989eio__realpath (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
990{
991 char *res;
992 const char *rel = path;
993 char *tmp1, *tmp2;
994#if SYMLOOP_MAX > 32
995 int symlinks = SYMLOOP_MAX;
996#else
997 int symlinks = 32;
998#endif
999
1000 errno = EINVAL;
1001 if (!rel)
1002 return -1;
1003
1004 errno = ENOENT;
1005 if (!*rel)
1006 return -1;
1007
1008 res = etp_tmpbuf_get (tmpbuf, EIO_PATH_MAX * 3);
1009#ifdef _WIN32
1010 if (_access (rel, 4) != 0)
1011 return -1;
1012
1013 symlinks = GetFullPathName (rel, EIO_PATH_MAX * 3, res, 0);
1014
1015 errno = ENAMETOOLONG;
1016 if (symlinks >= EIO_PATH_MAX * 3)
1017 return -1;
1018
1019 errno = EIO;
1020 if (symlinks <= 0)
1021 return -1;
1022
1023 return symlinks;
1024
1025#else
1026 tmp1 = res + EIO_PATH_MAX;
1027 tmp2 = tmp1 + EIO_PATH_MAX;
1028
1029#if 0 /* disabled, the musl way to do things is just too racy */
1030#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1031 /* on linux we may be able to ask the kernel */
1032 {
1033 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1034
1035 if (fd >= 0)
1036 {
1037 sprintf (tmp1, "/proc/self/fd/%d", fd);
1038 req->result = readlink (tmp1, res, EIO_PATH_MAX);
1039 /* here we should probably stat the open file and the disk file, to make sure they still match */
1040 eio__close (fd);
1041
1042 if (req->result > 0)
1043 goto done;
1044 }
1045 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1046 return -1;
1047 }
1048#endif
1049#endif
1050
1051 if (*rel != '/')
1052 {
1053 int len;
1054
1055 errno = ENOENT;
1056 if (wd == EIO_INVALID_WD)
1057 return -1;
1058
1059 if (wd == EIO_CWD)
1060 {
1061 if (!getcwd (res, EIO_PATH_MAX))
1062 return -1;
1063
1064 len = strlen (res);
1065 }
1066 else
1067 memcpy (res, wd->str, len = wd->len);
1068
1069 if (res [1]) /* only use if not / */
1070 res += len;
1071 }
1072
1073 while (*rel)
1074 {
1075 eio_ssize_t len, linklen;
1076 const char *beg = rel;
1077
1078 while (*rel && *rel != '/')
1079 ++rel;
1080
1081 len = rel - beg;
1082
1083 if (!len) /* skip slashes */
1084 {
1085 ++rel;
1086 continue;
1087 }
1088
1089 if (beg [0] == '.')
1090 {
1091 if (len == 1)
1092 continue; /* . - nop */
1093
1094 if (beg [1] == '.' && len == 2)
1095 {
1096 /* .. - back up one component, if possible */
1097
1098 while (res != tmpbuf->ptr)
1099 if (*--res == '/')
1100 break;
1101
1102 continue;
1103 }
1104 }
1105
1106 errno = ENAMETOOLONG;
1107 if (res + 1 + len + 1 >= tmp1)
1108 return -1;
1109
1110 /* copy one component */
1111 *res = '/';
1112 memcpy (res + 1, beg, len);
1113
1114 /* zero-terminate, for readlink */
1115 res [len + 1] = 0;
1116
1117 /* now check if it's a symlink */
1118 linklen = readlink (tmpbuf->ptr, tmp1, EIO_PATH_MAX);
1119
1120 if (linklen < 0)
1121 {
1122 if (errno != EINVAL)
1123 return -1;
1124
1125 /* it's a normal directory. hopefully */
1126 res += len + 1;
1127 }
1128 else
1129 {
1130 /* yay, it was a symlink - build new path in tmp2 */
1131 int rellen = strlen (rel);
1132
1133 errno = ENAMETOOLONG;
1134 if (linklen + 1 + rellen >= EIO_PATH_MAX) /* also catch linklen >= EIO_PATH_MAX */
1135 return -1;
1136
1137 errno = ELOOP;
1138 if (!--symlinks)
1139 return -1;
1140
1141 if (*tmp1 == '/')
1142 res = tmpbuf->ptr; /* symlink resolves to an absolute path */
1143
1144 /* we need to be careful, as rel might point into tmp2 already */
1145 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1146 tmp2 [linklen] = '/';
1147 memcpy (tmp2, tmp1, linklen);
1148
1149 rel = tmp2;
1150 }
1151 }
1152
1153 /* special case for the lone root path */
1154 if (res == tmpbuf->ptr)
1155 *res++ = '/';
1156
1157 return res - (char *)tmpbuf->ptr;
1158#endif
1028} 1159}
1029 1160
1030static signed char 1161static signed char
1031eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) 1162eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1032{ 1163{
1033 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */ 1164 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1034 : a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0; 1165 : a->inode < b->inode ? -1
1166 : a->inode > b->inode ? 1
1167 : 0;
1035} 1168}
1036 1169
1037#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0 1170#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1038 1171
1039#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */ 1172#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1040#define EIO_SORT_FAST 60 /* when to only use insertion sort */ 1173#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1041 1174
1042static void 1175static void
1043eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1176eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1044{ 1177{
1045 unsigned char bits [9 + sizeof (ino_t) * 8]; 1178 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1046 unsigned char *bit = bits; 1179 unsigned char *bit = bits;
1047 1180
1048 assert (CHAR_BIT == 8); 1181 assert (CHAR_BIT == 8);
1049 assert (sizeof (eio_dirent) * 8 < 256); 1182 assert (sizeof (eio_dirent) * 8 < 256);
1050 assert (offsetof (eio_dirent, inode)); /* we use 0 as sentinel */ 1183 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1051 assert (offsetof (eio_dirent, score)); /* we use 0 as sentinel */ 1184 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1052 1185
1053 if (size <= EIO_SORT_FAST) 1186 if (size <= EIO_SORT_FAST)
1054 return; 1187 return;
1055 1188
1056 /* first prepare an array of bits to test in our radix sort */ 1189 /* first prepare an array of bits to test in our radix sort */
1057 /* try to take endianness into account, as well as differences in ino_t sizes */ 1190 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1058 /* inode_bits must contain all inodes ORed together */ 1191 /* inode_bits must contain all inodes ORed together */
1059 /* which is used to skip bits that are 0 everywhere, which is very common */ 1192 /* which is used to skip bits that are 0 everywhere, which is very common */
1060 { 1193 {
1061 ino_t endianness; 1194 eio_ino_t endianness;
1062 int i, j; 1195 int i, j;
1063 1196
1064 /* we store the byte offset of byte n into byte n of "endianness" */ 1197 /* we store the byte offset of byte n into byte n of "endianness" */
1065 for (i = 0; i < sizeof (ino_t); ++i) 1198 for (i = 0; i < sizeof (eio_ino_t); ++i)
1066 ((unsigned char *)&endianness)[i] = i; 1199 ((unsigned char *)&endianness)[i] = i;
1067 1200
1068 *bit++ = 0; 1201 *bit++ = 0;
1069 1202
1070 for (i = 0; i < sizeof (ino_t); ++i) 1203 for (i = 0; i < sizeof (eio_ino_t); ++i)
1071 { 1204 {
1072 /* shifting off the byte offsets out of "endianness" */ 1205 /* shifting off the byte offsets out of "endianness" */
1073 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8; 1206 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1074 endianness >>= 8; 1207 endianness >>= 8;
1075 1208
1076 for (j = 0; j < 8; ++j) 1209 for (j = 0; j < 8; ++j)
1077 if (inode_bits & (((ino_t)1) << (i * 8 + j))) 1210 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1078 *bit++ = offs + j; 1211 *bit++ = offs + j;
1079 } 1212 }
1080 1213
1081 for (j = 0; j < 8; ++j) 1214 for (j = 0; j < 8; ++j)
1082 if (score_bits & (1 << j)) 1215 if (score_bits & (1 << j))
1083 *bit++ = offsetof (eio_dirent, score) * 8 + j; 1216 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1084 } 1217 }
1085 1218
1086 /* now actually do the sorting (a variant of MSD radix sort) */ 1219 /* now actually do the sorting (a variant of MSD radix sort) */
1087 { 1220 {
1088 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base; 1221 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1089 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end; 1222 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1090 unsigned char *bit_stk [9 + sizeof (ino_t) * 8]; 1223 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1091 int stk_idx = 0; 1224 int stk_idx = 0;
1092 1225
1093 base_stk [stk_idx] = dents; 1226 base_stk [stk_idx] = dents;
1094 end_stk [stk_idx] = dents + size; 1227 end_stk [stk_idx] = dents + size;
1095 bit_stk [stk_idx] = bit - 1; 1228 bit_stk [stk_idx] = bit - 1;
1174 } 1307 }
1175 } 1308 }
1176} 1309}
1177 1310
1178static void 1311static void
1179eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1312eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1180{ 1313{
1181 if (size <= 1) 1314 if (size <= 1)
1182 return; /* our insertion sort relies on size > 0 */ 1315 return; /* our insertion sort relies on size > 0 */
1183 1316
1184 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */ 1317 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1192 1325
1193/* read a full directory */ 1326/* read a full directory */
1194static void 1327static void
1195eio__scandir (eio_req *req, etp_worker *self) 1328eio__scandir (eio_req *req, etp_worker *self)
1196{ 1329{
1197 DIR *dirp;
1198 EIO_STRUCT_DIRENT *entp;
1199 char *name, *names; 1330 char *name, *names;
1200 int namesalloc = 4096; 1331 int namesalloc = 4096 - sizeof (void *) * 4;
1201 int namesoffs = 0; 1332 int namesoffs = 0;
1202 int flags = req->int1; 1333 int flags = req->int1;
1203 eio_dirent *dents = 0; 1334 eio_dirent *dents = 0;
1204 int dentalloc = 128; 1335 int dentalloc = 128;
1205 int dentoffs = 0; 1336 int dentoffs = 0;
1206 ino_t inode_bits = 0; 1337 eio_ino_t inode_bits = 0;
1338#ifdef _WIN32
1339 HANDLE dirp;
1340 WIN32_FIND_DATA entp;
1341#else
1342 DIR *dirp;
1343 EIO_STRUCT_DIRENT *entp;
1344#endif
1207 1345
1208 req->result = -1; 1346 req->result = -1;
1209 1347
1210 if (!(flags & EIO_READDIR_DENTS)) 1348 if (!(flags & EIO_READDIR_DENTS))
1211 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); 1349 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1212 1350
1213 X_LOCK (wrklock); 1351#ifdef _WIN32
1214 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 1352 {
1353 int len = strlen ((const char *)req->ptr1);
1354 char *path = malloc (MAX_PATH);
1355 const char *fmt;
1356 const char *reqpath = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1357
1358 if (!len)
1359 fmt = "./*";
1360 else if (reqpath[len - 1] == '/' || reqpath[len - 1] == '\\')
1361 fmt = "%s*";
1362 else
1363 fmt = "%s/*";
1364
1365 _snprintf (path, MAX_PATH, fmt, reqpath);
1366 dirp = FindFirstFile (path, &entp);
1367 free (path);
1368
1369 if (dirp == INVALID_HANDLE_VALUE)
1370 {
1371 /* should steal _dosmaperr */
1372 switch (GetLastError ())
1373 {
1374 case ERROR_FILE_NOT_FOUND:
1375 req->result = 0;
1376 break;
1377
1378 case ERROR_INVALID_NAME:
1379 case ERROR_PATH_NOT_FOUND:
1380 case ERROR_NO_MORE_FILES:
1381 errno = ENOENT;
1382 break;
1383
1384 case ERROR_NOT_ENOUGH_MEMORY:
1385 errno = ENOMEM;
1386 break;
1387
1388 default:
1389 errno = EINVAL;
1390 break;
1391 }
1392
1393 return;
1394 }
1395 }
1396#else
1397 #if HAVE_AT
1398 if (req->wd)
1399 {
1400 int fd = openat (WD2FD (req->wd), req->ptr1, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1401
1402 if (fd < 0)
1403 return;
1404
1405 dirp = fdopendir (fd);
1406
1407 if (!dirp)
1408 silent_close (fd);
1409 }
1410 else
1215 self->dirp = dirp = opendir (req->ptr1); 1411 dirp = opendir (req->ptr1);
1412 #else
1413 dirp = opendir (wd_expand (&self->tmpbuf, req->wd, req->ptr1));
1414 #endif
1415
1416 if (!dirp)
1417 return;
1418#endif
1419
1420 if (req->flags & EIO_FLAG_PTR1_FREE)
1421 free (req->ptr1);
1422
1216 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1423 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1217 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; 1424 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1218 req->ptr2 = names = malloc (namesalloc); 1425 req->ptr2 = names = malloc (namesalloc);
1219 X_UNLOCK (wrklock);
1220 1426
1221 if (dirp && names && (!flags || dents)) 1427 if (!names || (flags && !dents))
1428 return;
1429
1222 for (;;) 1430 for (;;)
1223 { 1431 {
1432 int done;
1433
1434#ifdef _WIN32
1435 done = !dirp;
1436#else
1224 errno = 0; 1437 errno = 0;
1225 entp = readdir (dirp); 1438 entp = readdir (dirp);
1439 done = !entp;
1440#endif
1226 1441
1227 if (!entp) 1442 if (done)
1228 { 1443 {
1444#ifndef _WIN32
1445 int old_errno = errno;
1446 closedir (dirp);
1447 errno = old_errno;
1448
1229 if (errno) 1449 if (errno)
1230 break; 1450 break;
1451#endif
1231 1452
1232 /* sort etc. */ 1453 /* sort etc. */
1233 req->int1 = flags; 1454 req->int1 = flags;
1234 req->result = dentoffs; 1455 req->result = dentoffs;
1235 1456
1236 if (flags & EIO_READDIR_STAT_ORDER) 1457 if (flags & EIO_READDIR_STAT_ORDER)
1237 eio_dent_sort (dents, dentoffs, 0, inode_bits); /* sort by inode exclusively */ 1458 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1238 else if (flags & EIO_READDIR_DIRS_FIRST) 1459 else if (flags & EIO_READDIR_DIRS_FIRST)
1239 if (flags & EIO_READDIR_FOUND_UNKNOWN) 1460 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1240 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */ 1461 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1241 else 1462 else
1463 {
1464 /* in this case, all is known, and we just put dirs first and sort them */
1465 eio_dirent *oth = dents + dentoffs;
1466 eio_dirent *dir = dents;
1467
1468 /* now partition dirs to the front, and non-dirs to the back */
1469 /* by walking from both sides and swapping if necessary */
1470 while (oth > dir)
1471 {
1472 if (dir->type == EIO_DT_DIR)
1473 ++dir;
1474 else if ((--oth)->type == EIO_DT_DIR)
1475 {
1476 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1477
1478 ++dir;
1479 }
1480 }
1481
1482 /* now sort the dirs only (dirs all have the same score) */
1483 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1484 }
1485
1486 break;
1487 }
1488
1489 /* now add the entry to our list(s) */
1490 name = D_NAME (entp);
1491
1492 /* skip . and .. entries */
1493 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1494 {
1495 int len = D_NAMLEN (entp) + 1;
1496
1497 while (ecb_expect_false (namesoffs + len > namesalloc))
1498 {
1499 namesalloc *= 2;
1500 req->ptr2 = names = realloc (names, namesalloc);
1501
1502 if (!names)
1503 break;
1504 }
1505
1506 memcpy (names + namesoffs, name, len);
1507
1508 if (dents)
1509 {
1510 struct eio_dirent *ent;
1511
1512 if (ecb_expect_false (dentoffs == dentalloc))
1242 { 1513 {
1243 /* in this case, all is known, and we just put dirs first and sort them */ 1514 dentalloc *= 2;
1515 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1516
1517 if (!dents)
1518 break;
1519 }
1520
1244 eio_dirent *oth = dents + dentoffs; 1521 ent = dents + dentoffs;
1245 eio_dirent *dir = dents;
1246 1522
1247 /* now partition dirs to the front, and non-dirs to the back */ 1523 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1248 /* by walking from both sides and swapping if necessary */ 1524 ent->namelen = len - 1;
1249 /* also clear score, so it doesn't influence sorting */ 1525 ent->inode = D_INO (entp);
1250 while (oth > dir) 1526
1527 inode_bits |= ent->inode;
1528
1529 switch (D_TYPE (entp))
1530 {
1531 default:
1532 ent->type = EIO_DT_UNKNOWN;
1533 flags |= EIO_READDIR_FOUND_UNKNOWN;
1534 break;
1535
1536 #ifdef DT_FIFO
1537 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1538 #endif
1539 #ifdef DT_CHR
1540 case DT_CHR: ent->type = EIO_DT_CHR; break;
1541 #endif
1542 #ifdef DT_MPC
1543 case DT_MPC: ent->type = EIO_DT_MPC; break;
1544 #endif
1545 #ifdef DT_DIR
1546 case DT_DIR: ent->type = EIO_DT_DIR; break;
1547 #endif
1548 #ifdef DT_NAM
1549 case DT_NAM: ent->type = EIO_DT_NAM; break;
1550 #endif
1551 #ifdef DT_BLK
1552 case DT_BLK: ent->type = EIO_DT_BLK; break;
1553 #endif
1554 #ifdef DT_MPB
1555 case DT_MPB: ent->type = EIO_DT_MPB; break;
1556 #endif
1557 #ifdef DT_REG
1558 case DT_REG: ent->type = EIO_DT_REG; break;
1559 #endif
1560 #ifdef DT_NWK
1561 case DT_NWK: ent->type = EIO_DT_NWK; break;
1562 #endif
1563 #ifdef DT_CMP
1564 case DT_CMP: ent->type = EIO_DT_CMP; break;
1565 #endif
1566 #ifdef DT_LNK
1567 case DT_LNK: ent->type = EIO_DT_LNK; break;
1568 #endif
1569 #ifdef DT_SOCK
1570 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1571 #endif
1572 #ifdef DT_DOOR
1573 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1574 #endif
1575 #ifdef DT_WHT
1576 case DT_WHT: ent->type = EIO_DT_WHT; break;
1577 #endif
1578 }
1579
1580 ent->score = 7;
1581
1582 if (flags & EIO_READDIR_DIRS_FIRST)
1583 {
1584 if (ent->type == EIO_DT_UNKNOWN)
1251 { 1585 {
1252 if (dir->type == EIO_DT_DIR) 1586 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1253 ++dir; 1587 ent->score = 1;
1254 else if ((--oth)->type == EIO_DT_DIR) 1588 else if (!strchr (name, '.')) /* absence of dots indicate likely dirs */
1255 { 1589 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1256 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1257
1258 ++dir;
1259 }
1260 } 1590 }
1261 1591 else if (ent->type == EIO_DT_DIR)
1262 /* now sort the dirs only */ 1592 ent->score = 0;
1263 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1264 } 1593 }
1265
1266 break;
1267 }
1268
1269 /* now add the entry to our list(s) */
1270 name = entp->d_name;
1271
1272 /* skip . and .. entries */
1273 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1274 {
1275 int len = D_NAMLEN (entp) + 1;
1276
1277 while (expect_false (namesoffs + len > namesalloc))
1278 {
1279 namesalloc *= 2;
1280 X_LOCK (wrklock);
1281 req->ptr2 = names = realloc (names, namesalloc);
1282 X_UNLOCK (wrklock);
1283
1284 if (!names)
1285 break;
1286 } 1594 }
1287 1595
1288 memcpy (names + namesoffs, name, len);
1289
1290 if (dents)
1291 {
1292 struct eio_dirent *ent;
1293
1294 if (expect_false (dentoffs == dentalloc))
1295 {
1296 dentalloc *= 2;
1297 X_LOCK (wrklock);
1298 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1299 X_UNLOCK (wrklock);
1300
1301 if (!dents)
1302 break;
1303 }
1304
1305 ent = dents + dentoffs;
1306
1307 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1308 ent->namelen = len - 1;
1309 ent->inode = D_INO (entp);
1310
1311 inode_bits |= ent->inode;
1312
1313 switch (D_TYPE (entp))
1314 {
1315 default:
1316 ent->type = EIO_DT_UNKNOWN;
1317 flags |= EIO_READDIR_FOUND_UNKNOWN;
1318 break;
1319
1320 #ifdef DT_FIFO
1321 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1322 #endif
1323 #ifdef DT_CHR
1324 case DT_CHR: ent->type = EIO_DT_CHR; break;
1325 #endif
1326 #ifdef DT_MPC
1327 case DT_MPC: ent->type = EIO_DT_MPC; break;
1328 #endif
1329 #ifdef DT_DIR
1330 case DT_DIR: ent->type = EIO_DT_DIR; break;
1331 #endif
1332 #ifdef DT_NAM
1333 case DT_NAM: ent->type = EIO_DT_NAM; break;
1334 #endif
1335 #ifdef DT_BLK
1336 case DT_BLK: ent->type = EIO_DT_BLK; break;
1337 #endif
1338 #ifdef DT_MPB
1339 case DT_MPB: ent->type = EIO_DT_MPB; break;
1340 #endif
1341 #ifdef DT_REG
1342 case DT_REG: ent->type = EIO_DT_REG; break;
1343 #endif
1344 #ifdef DT_NWK
1345 case DT_NWK: ent->type = EIO_DT_NWK; break;
1346 #endif
1347 #ifdef DT_CMP
1348 case DT_CMP: ent->type = EIO_DT_CMP; break;
1349 #endif
1350 #ifdef DT_LNK
1351 case DT_LNK: ent->type = EIO_DT_LNK; break;
1352 #endif
1353 #ifdef DT_SOCK
1354 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1355 #endif
1356 #ifdef DT_DOOR
1357 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1358 #endif
1359 #ifdef DT_WHT
1360 case DT_WHT: ent->type = EIO_DT_WHT; break;
1361 #endif
1362 }
1363
1364 ent->score = 7;
1365
1366 if (flags & EIO_READDIR_DIRS_FIRST)
1367 {
1368 if (ent->type == EIO_DT_UNKNOWN)
1369 {
1370 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1371 ent->score = 1;
1372 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1373 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1374 }
1375 else if (ent->type == EIO_DT_DIR)
1376 ent->score = 0;
1377 }
1378 }
1379
1380 namesoffs += len; 1596 namesoffs += len;
1381 ++dentoffs; 1597 ++dentoffs;
1382 } 1598 }
1383 1599
1384 if (EIO_CANCELLED (req)) 1600 if (EIO_CANCELLED (req))
1385 { 1601 {
1386 errno = ECANCELED; 1602 errno = ECANCELED;
1387 break; 1603 break;
1388 } 1604 }
1605
1606#ifdef _WIN32
1607 if (!FindNextFile (dirp, &entp))
1608 {
1609 FindClose (dirp);
1610 dirp = 0;
1389 } 1611 }
1612#endif
1613 }
1390} 1614}
1391 1615
1392#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) 1616/*****************************************************************************/
1393# define eio__msync(a,b,c) ((errno = ENOSYS), -1) 1617/* working directory stuff */
1618/* various deficiencies in the posix 2008 api force us to */
1619/* keep the absolute path in string form at all times */
1620/* fuck yeah. */
1621
1622#if !HAVE_AT
1623
1624/* a bit like realpath, but usually faster because it doesn'T have to return */
1625/* an absolute or canonical path */
1626static const char *
1627wd_expand (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
1628{
1629 if (!wd || *path == '/')
1630 return path;
1631
1632 if (path [0] == '.' && !path [1])
1633 return wd->str;
1634
1635 {
1636 int l1 = wd->len;
1637 int l2 = strlen (path);
1638
1639 char *res = etp_tmpbuf_get (tmpbuf, l1 + l2 + 2);
1640
1641 memcpy (res, wd->str, l1);
1642 res [l1] = '/';
1643 memcpy (res + l1 + 1, path, l2 + 1);
1644
1645 return res;
1646 }
1647}
1648
1649#endif
1650
1651static eio_wd
1652eio__wd_open_sync (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
1653{
1654 int fd;
1655 eio_wd res;
1656 int len = eio__realpath (tmpbuf, wd, path);
1657
1658 if (len < 0)
1659 return EIO_INVALID_WD;
1660
1661#if HAVE_AT
1662 fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1663
1664 if (fd < 0)
1665 return EIO_INVALID_WD;
1666#endif
1667
1668 res = malloc (sizeof (*res) + len); /* one extra 0-byte */
1669
1670#if HAVE_AT
1671 res->fd = fd;
1672#endif
1673
1674 res->len = len;
1675 memcpy (res->str, tmpbuf->ptr, len);
1676 res->str [len] = 0;
1677
1678 return res;
1679}
1680
1681eio_wd
1682eio_wd_open_sync (eio_wd wd, const char *path)
1683{
1684 struct etp_tmpbuf tmpbuf = { };
1685 wd = eio__wd_open_sync (&tmpbuf, wd, path);
1686 free (tmpbuf.ptr);
1687
1688 return wd;
1689}
1690
1691void
1692eio_wd_close_sync (eio_wd wd)
1693{
1694 if (wd != EIO_INVALID_WD && wd != EIO_CWD)
1695 {
1696 #if HAVE_AT
1697 eio__close (wd->fd);
1698 #endif
1699 free (wd);
1700 }
1701}
1702
1703#if HAVE_AT
1704
1705static int
1706eio__renameat2 (int olddirfd, const char *oldpath, int newdirfd, const char *newpath, unsigned int flags)
1707{
1708#if HAVE_RENAMEAT2
1709 return syscall (SYS_renameat2, olddirfd, oldpath, newdirfd, newpath, flags);
1394#else 1710#else
1395
1396int
1397eio__msync (void *mem, size_t len, int flags)
1398{
1399 if (EIO_MS_ASYNC != MS_SYNC
1400 || EIO_MS_INVALIDATE != MS_INVALIDATE
1401 || EIO_MS_SYNC != MS_SYNC)
1402 {
1403 flags = 0
1404 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1405 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1406 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1407 }
1408
1409 return msync (mem, len, flags);
1410}
1411
1412#endif
1413
1414int
1415eio__mtouch (void *mem, size_t len, int flags)
1416{
1417 intptr_t addr = (intptr_t)mem;
1418 intptr_t end = addr + len;
1419#ifdef PAGESIZE
1420 const intptr_t page = PAGESIZE;
1421#else
1422 static intptr_t page;
1423
1424 if (!page) 1711 if (flags)
1425 page = sysconf (_SC_PAGESIZE); 1712 return EIO_ENOSYS ();
1426#endif
1427 1713
1428 /* round down to start of page, although this is probably useless */ 1714 return renameat (olddirfd, oldpath, newdirfd, newpath);
1429 addr &= ~(page - 1); /* assume page size is always a power of two */ 1715#endif
1716}
1430 1717
1431 if (addr < end) 1718/* they forgot these */
1432 if (flags & EIO_MT_MODIFY) /* modify */
1433 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
1434 else
1435 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
1436 1719
1720static int
1721eio__truncateat (int dirfd, const char *path, off_t length)
1722{
1723 int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
1724 int res;
1725
1726 if (fd < 0)
1727 return fd;
1728
1729 res = ftruncate (fd, length);
1730 silent_close (fd);
1437 return 0; 1731 return res;
1438} 1732}
1733
1734static int
1735eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
1736{
1737 int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
1738 int res;
1739
1740 if (fd < 0)
1741 return fd;
1742
1743 res = fstatvfs (fd, buf);
1744 silent_close (fd);
1745 return res;
1746}
1747
1748#endif
1439 1749
1440/*****************************************************************************/ 1750/*****************************************************************************/
1441 1751
1442#define ALLOC(len) \ 1752#define ALLOC(len) \
1443 if (!req->ptr2) \ 1753 if (!req->ptr2) \
1444 { \ 1754 { \
1445 X_LOCK (wrklock); \ 1755 X_LOCK (EIO_POOL->wrklock); \
1446 req->flags |= EIO_FLAG_PTR2_FREE; \ 1756 req->flags |= EIO_FLAG_PTR2_FREE; \
1447 X_UNLOCK (wrklock); \ 1757 X_UNLOCK (EIO_POOL->wrklock); \
1448 req->ptr2 = malloc (len); \ 1758 req->ptr2 = malloc (len); \
1449 if (!req->ptr2) \ 1759 if (!req->ptr2) \
1450 { \ 1760 { \
1451 errno = ENOMEM; \ 1761 errno = ENOMEM; \
1452 req->result = -1; \ 1762 req->result = -1; \
1453 break; \ 1763 goto alloc_fail; \
1454 } \ 1764 } \
1455 } 1765 }
1456 1766
1457X_THREAD_PROC (etp_proc) 1767/*****************************************************************************/
1458{
1459 ETP_REQ *req;
1460 struct timespec ts;
1461 etp_worker *self = (etp_worker *)thr_arg;
1462 1768
1463 /* try to distribute timeouts somewhat randomly */ 1769static void
1464 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 1770eio__slurp (int fd, eio_req *req)
1771{
1772 req->result = fd;
1465 1773
1466 for (;;) 1774 if (fd < 0)
1775 return;
1776
1777 if (req->offs < 0 || !req->size) /* do we need the size? */
1467 { 1778 {
1468 X_LOCK (reqlock); 1779 off_t size = lseek (fd, 0, SEEK_END);
1469 1780
1470 for (;;)
1471 {
1472 self->req = req = reqq_shift (&req_queue);
1473
1474 if (req)
1475 break;
1476
1477 ++idle;
1478
1479 ts.tv_sec = time (0) + IDLE_TIMEOUT;
1480 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
1481 {
1482 if (idle > max_idle)
1483 {
1484 --idle;
1485 X_UNLOCK (reqlock);
1486 X_LOCK (wrklock);
1487 --started;
1488 X_UNLOCK (wrklock);
1489 goto quit;
1490 }
1491
1492 /* we are allowed to idle, so do so without any timeout */
1493 X_COND_WAIT (reqwait, reqlock);
1494 }
1495
1496 --idle;
1497 }
1498
1499 --nready;
1500
1501 X_UNLOCK (reqlock);
1502
1503 if (req->type < 0) 1781 if (req->offs < 0)
1504 goto quit; 1782 req->offs += size;
1505 1783
1506 if (!EIO_CANCELLED (req)) 1784 if (!req->size)
1507 ETP_EXECUTE (self, req); 1785 req->size = size - req->offs;
1508
1509 X_LOCK (reslock);
1510
1511 ++npending;
1512
1513 if (!reqq_push (&res_queue, req) && want_poll_cb)
1514 want_poll_cb ();
1515
1516 self->req = 0;
1517 etp_worker_clear (self);
1518
1519 X_UNLOCK (reslock);
1520 } 1786 }
1521 1787
1522quit: 1788 ALLOC (req->size);
1523 X_LOCK (wrklock); 1789 req->result = pread (fd, req->ptr2, req->size, req->offs);
1524 etp_worker_free (self);
1525 X_UNLOCK (wrklock);
1526 1790
1527 return 0; 1791 silent_close (fd);
1528}
1529 1792
1530/*****************************************************************************/ 1793alloc_fail:
1794 ;
1795}
1531 1796
1797int ecb_cold
1532int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1798eio_init (void (*want_poll)(void), void (*done_poll)(void))
1533{ 1799{
1534 return etp_init (want_poll, done_poll); 1800 eio_want_poll_cb = want_poll;
1535} 1801 eio_done_poll_cb = done_poll;
1536 1802
1803 return etp_init (EIO_POOL, 0, 0, 0);
1804}
1805
1806ecb_inline void
1537static void eio_api_destroy (eio_req *req) 1807eio_api_destroy (eio_req *req)
1538{ 1808{
1539 free (req); 1809 free (req);
1540} 1810}
1541 1811
1542#define REQ(rtype) \ 1812#define REQ(rtype) \
1543 eio_req *req; \ 1813 eio_req *req; \
1544 \ 1814 \
1545 req = (eio_req *)calloc (1, sizeof *req); \ 1815 req = (eio_req *)calloc (1, sizeof *req); \
1546 if (!req) \ 1816 if (!req) \
1547 return 0; \ 1817 return 0; \
1561 { \ 1831 { \
1562 eio_api_destroy (req); \ 1832 eio_api_destroy (req); \
1563 return 0; \ 1833 return 0; \
1564 } 1834 }
1565 1835
1836#define SINGLEDOT(ptr) (0[(char *)(ptr)] == '.' && !1[(char *)(ptr)])
1837
1838static void
1566static void eio_execute (etp_worker *self, eio_req *req) 1839eio_execute (etp_worker *self, eio_req *req)
1567{ 1840{
1568 errno = 0; 1841#if HAVE_AT
1842 int dirfd;
1843#else
1844 const char *path;
1845#endif
1846
1847 if (ecb_expect_false (EIO_CANCELLED (req)))
1848 {
1849 req->result = -1;
1850 req->errorno = ECANCELED;
1851 return;
1852 }
1853
1854 if (ecb_expect_false (req->wd == EIO_INVALID_WD))
1855 {
1856 req->result = -1;
1857 req->errorno = ENOENT;
1858 return;
1859 }
1860
1861 if (req->type >= EIO_OPEN)
1862 {
1863 #if HAVE_AT
1864 dirfd = WD2FD (req->wd);
1865 #else
1866 path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1867 #endif
1868 }
1569 1869
1570 switch (req->type) 1870 switch (req->type)
1571 { 1871 {
1872 case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
1873 req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
1874 break;
1875 case EIO_WD_CLOSE: req->result = 0;
1876 eio_wd_close_sync (req->wd); break;
1877
1878 case EIO_SEEK: eio__lseek (req); break;
1572 case EIO_READ: ALLOC (req->size); 1879 case EIO_READ: ALLOC (req->size);
1573 req->result = req->offs >= 0 1880 req->result = req->offs >= 0
1574 ? pread (req->int1, req->ptr2, req->size, req->offs) 1881 ? pread (req->int1, req->ptr2, req->size, req->offs)
1575 : read (req->int1, req->ptr2, req->size); break; 1882 : read (req->int1, req->ptr2, req->size); break;
1576 case EIO_WRITE: req->result = req->offs >= 0 1883 case EIO_WRITE: req->result = req->offs >= 0
1577 ? pwrite (req->int1, req->ptr2, req->size, req->offs) 1884 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1578 : write (req->int1, req->ptr2, req->size); break; 1885 : write (req->int1, req->ptr2, req->size); break;
1579 1886
1887 case EIO_FCNTL: req->result = fcntl (req->int1, (int) req->int2, req->ptr2); break;
1888 case EIO_IOCTL: req->result = ioctl (req->int1, (unsigned long)req->int2, req->ptr2); break;
1889
1580 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break; 1890 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1581 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break; 1891 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
1892
1893#if HAVE_AT
1582 1894
1583 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1895 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1584 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 1896 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break;
1585 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1897 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1586 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 1898 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, AT_SYMLINK_NOFOLLOW); break;
1587 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1899 case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break;
1588 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; 1900 case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break;
1901 case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break;
1902 case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break;
1903 case EIO_SLURP: eio__slurp ( openat (dirfd, req->ptr1, O_RDONLY | O_CLOEXEC), req); break;
1589 1904
1905 case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
1906 case EIO_RMDIR: /* complications arise because "." cannot be removed, so we might have to expand */
1907 req->result = req->wd && SINGLEDOT (req->ptr1)
1908 ? rmdir (req->wd->str)
1909 : unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break;
1910 case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); break;
1911 case EIO_RENAME: req->result = eio__renameat2 (
1912 dirfd,
1913 /* complications arise because "." cannot be renamed, so we might have to expand */
1914 req->wd && SINGLEDOT (req->ptr1) ? req->wd->str : req->ptr1,
1915 WD2FD ((eio_wd)req->int3),
1916 req->ptr2,
1917 req->int2
1918 );
1919 break;
1920 case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2, 0); break;
1921 case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
1922 case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1590 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS)); 1923 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1591 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break; 1924 req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1592 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1593 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1594
1595 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1596 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1597 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1598 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1599 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1600 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1601
1602 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
1603 case EIO_CLOSE: req->result = close (req->int1); break;
1604 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1605 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
1606 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1607 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1608 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1609 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1610 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1611 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break;
1612
1613 case EIO_READLINK: ALLOC (NAME_MAX); 1925 case EIO_READLINK: ALLOC (EIO_PATH_MAX);
1614 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; 1926 req->result = readlinkat (dirfd, req->ptr1, req->ptr2, EIO_PATH_MAX);
1615 1927 if (req->result == EIO_PATH_MAX)
1616 case EIO_SYNC: req->result = 0; sync (); break; 1928 {
1617 case EIO_FSYNC: req->result = fsync (req->int1); break; 1929 req->result = -1;
1618 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 1930 errno = ENAMETOOLONG;
1619 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break; 1931 }
1620 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break; 1932 break;
1621 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1622
1623 case EIO_READDIR: eio__scandir (req, self); break;
1624
1625 case EIO_BUSY: 1933 case EIO_UTIME:
1626#ifdef _WIN32 1934 case EIO_FUTIME:
1627 Sleep (req->nv1 * 1000.); 1935 {
1936 struct timespec ts[2];
1937 struct timespec *times;
1938
1939 if (req->nv1 != -1. || req->nv2 != -1.)
1940 {
1941 ts[0].tv_sec = req->nv1;
1942 ts[0].tv_nsec = (req->nv1 - ts[0].tv_sec) * 1e9;
1943 ts[1].tv_sec = req->nv2;
1944 ts[1].tv_nsec = (req->nv2 - ts[1].tv_sec) * 1e9;
1945
1946 times = ts;
1947 }
1948 else
1949 times = 0;
1950
1951 req->result = req->type == EIO_FUTIME
1952 ? futimens (req->int1, times)
1953 : utimensat (dirfd, req->ptr1, times, 0);
1954 }
1955 break;
1956
1628#else 1957#else
1629 {
1630 struct timeval tv;
1631 1958
1632 tv.tv_sec = req->nv1; 1959 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1633 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; 1960 req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
1961 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1962 req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
1963 case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
1964 case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
1965 case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
1966 case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
1967 case EIO_SLURP: eio__slurp ( open (path , O_RDONLY | O_CLOEXEC), req); break;
1634 1968
1635 req->result = select (0, 0, 0, 0, &tv); 1969 case EIO_UNLINK: req->result = unlink (path ); break;
1636 } 1970 case EIO_RMDIR: req->result = rmdir (path ); break;
1637#endif 1971 case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
1638 break; 1972 case EIO_RENAME: req->result = req->int2 ? EIO_ENOSYS () : rename (path, req->ptr2); break;
1973 case EIO_LINK: req->result = link (path , req->ptr2); break;
1974 case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
1975 case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
1976 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1977 req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
1978 case EIO_READLINK: ALLOC (EIO_PATH_MAX);
1979 req->result = readlink (path, req->ptr2, EIO_PATH_MAX);
1980 if (req->result == EIO_PATH_MAX)
1981 {
1982 req->result = -1;
1983 errno = ENAMETOOLONG;
1984 }
1985 break;
1639 1986
1640 case EIO_UTIME: 1987 case EIO_UTIME:
1641 case EIO_FUTIME: 1988 case EIO_FUTIME:
1642 { 1989 {
1643 struct timeval tv[2]; 1990 struct timeval tv[2];
1644 struct timeval *times; 1991 struct timeval *times;
1645 1992
1646 if (req->nv1 != -1. || req->nv2 != -1.) 1993 if (req->nv1 != -1. || req->nv2 != -1.)
1647 { 1994 {
1648 tv[0].tv_sec = req->nv1; 1995 tv[0].tv_sec = req->nv1;
1649 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.; 1996 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
1650 tv[1].tv_sec = req->nv2; 1997 tv[1].tv_sec = req->nv2;
1651 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.; 1998 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
1652 1999
1653 times = tv; 2000 times = tv;
1654 } 2001 }
1655 else 2002 else
1656 times = 0; 2003 times = 0;
1657
1658 2004
1659 req->result = req->type == EIO_FUTIME 2005 req->result = req->type == EIO_FUTIME
1660 ? futimes (req->int1, times) 2006 ? futimes (req->int1, times)
1661 : utimes (req->ptr1, times); 2007 : utimes (req->ptr1, times);
1662 } 2008 }
1663 break; 2009 break;
1664 2010
2011#endif
2012
2013 case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
2014 {
2015 ALLOC (req->result);
2016 memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
2017 }
2018 break;
2019
2020 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2021 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
2022
2023 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2024 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2025
2026 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
2027 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
2028 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
2029
2030 case EIO_CLOSE: req->result = eio__close (req->int1); break;
2031 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
2032 case EIO_SYNC: req->result = 0; sync (); break;
2033 case EIO_FSYNC: req->result = fsync (req->int1); break;
2034 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
2035 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
2036 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
2037 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
2038 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
2039 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
2040 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
2041 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
2042
2043 case EIO_READDIR: eio__scandir (req, self); break;
2044
2045 case EIO_BUSY:
2046#ifdef _WIN32
2047 Sleep (req->nv1 * 1e3);
2048#else
2049 {
2050 struct timeval tv;
2051
2052 tv.tv_sec = req->nv1;
2053 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
2054
2055 req->result = select (0, 0, 0, 0, &tv);
2056 }
2057#endif
2058 break;
2059
2060#if 0
1665 case EIO_GROUP: 2061 case EIO_GROUP:
1666 abort (); /* handled in eio_request */ 2062 abort (); /* handled in eio_request */
2063#endif
1667 2064
1668 case EIO_NOP: 2065 case EIO_NOP:
1669 req->result = 0; 2066 req->result = 0;
1670 break; 2067 break;
1671 2068
1672 case EIO_CUSTOM: 2069 case EIO_CUSTOM:
1673 ((void (*)(eio_req *))req->feed) (req); 2070 req->feed (req);
1674 break; 2071 break;
1675 2072
1676 default: 2073 default:
1677 req->result = -1; 2074 req->result = EIO_ENOSYS ();
1678 break; 2075 break;
1679 } 2076 }
1680 2077
2078alloc_fail:
1681 req->errorno = errno; 2079 req->errorno = errno;
1682} 2080}
1683 2081
1684#ifndef EIO_NO_WRAPPERS 2082#ifndef EIO_NO_WRAPPERS
1685 2083
2084eio_req *eio_wd_open (const char *path, int pri, eio_cb cb, void *data)
2085{
2086 REQ (EIO_WD_OPEN); PATH; SEND;
2087}
2088
2089eio_req *eio_wd_close (eio_wd wd, int pri, eio_cb cb, void *data)
2090{
2091 REQ (EIO_WD_CLOSE); req->wd = wd; SEND;
2092}
2093
1686eio_req *eio_nop (int pri, eio_cb cb, void *data) 2094eio_req *eio_nop (int pri, eio_cb cb, void *data)
1687{ 2095{
1688 REQ (EIO_NOP); SEND; 2096 REQ (EIO_NOP); SEND;
1689} 2097}
1690 2098
1706eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2114eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1707{ 2115{
1708 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2116 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1709} 2117}
1710 2118
2119eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2120{
2121 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2122}
2123
2124eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
2125{
2126 REQ (EIO_SYNCFS); req->int1 = fd; SEND;
2127}
2128
2129eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2130{
2131 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2132}
2133
1711eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2134eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1712{ 2135{
1713 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2136 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1714} 2137}
1715 2138
1716eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) 2139eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
1717{ 2140{
1718 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; 2141 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
1719} 2142}
1720 2143
1721eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) 2144eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
1722{ 2145{
1723 REQ (EIO_FDATASYNC); req->int1 = fd; SEND; 2146 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2147}
2148
2149eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2150{
2151 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
1724} 2152}
1725 2153
1726eio_req *eio_close (int fd, int pri, eio_cb cb, void *data) 2154eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
1727{ 2155{
1728 REQ (EIO_CLOSE); req->int1 = fd; SEND; 2156 REQ (EIO_CLOSE); req->int1 = fd; SEND;
1731eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data) 2159eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
1732{ 2160{
1733 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND; 2161 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
1734} 2162}
1735 2163
2164eio_req *eio_seek (int fd, off_t offset, int whence, int pri, eio_cb cb, void *data)
2165{
2166 REQ (EIO_SEEK); req->int1 = fd; req->offs = offset; req->int2 = whence; SEND;
2167}
2168
1736eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data) 2169eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1737{ 2170{
1738 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND; 2171 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1739} 2172}
1740 2173
1741eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data) 2174eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1742{ 2175{
1743 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND; 2176 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1744} 2177}
1745 2178
2179eio_req *eio_fcntl (int fd, int cmd, void *arg, int pri, eio_cb cb, void *data)
2180{
2181 REQ (EIO_IOCTL); req->int1 = fd; req->int2 = cmd; req->ptr2 = arg; SEND;
2182}
2183
2184eio_req *eio_ioctl (int fd, unsigned long request, void *buf, int pri, eio_cb cb, void *data)
2185{
2186 REQ (EIO_IOCTL); req->int1 = fd; req->int2 = request; req->ptr2 = buf; SEND;
2187}
2188
1746eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) 2189eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1747{ 2190{
1748 REQ (EIO_FSTAT); req->int1 = fd; SEND; 2191 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1749} 2192}
1750 2193
1766eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) 2209eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1767{ 2210{
1768 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2211 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1769} 2212}
1770 2213
1771eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2214eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1772{ 2215{
1773 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2216 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1774} 2217}
1775 2218
1776eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) 2219eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
1796eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data) 2239eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1797{ 2240{
1798 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2241 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1799} 2242}
1800 2243
1801eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2244eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1802{ 2245{
1803 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2246 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1804} 2247}
1805 2248
1806eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) 2249eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
1822eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data) 2265eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1823{ 2266{
1824 return eio__1path (EIO_READLINK, path, pri, cb, data); 2267 return eio__1path (EIO_READLINK, path, pri, cb, data);
1825} 2268}
1826 2269
2270eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2271{
2272 return eio__1path (EIO_REALPATH, path, pri, cb, data);
2273}
2274
1827eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) 2275eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1828{ 2276{
1829 return eio__1path (EIO_STAT, path, pri, cb, data); 2277 return eio__1path (EIO_STAT, path, pri, cb, data);
1830} 2278}
1831 2279
1854 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND; 2302 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1855} 2303}
1856 2304
1857eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) 2305eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1858{ 2306{
1859 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; 2307 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1860} 2308}
1861 2309
1862static eio_req * 2310static eio_req *
1863eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2311eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1864{ 2312{
1888eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2336eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1889{ 2337{
1890 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2338 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1891} 2339}
1892 2340
2341eio_req *eio_slurp (const char *path, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
2342{
2343 REQ (EIO_SLURP); PATH; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
2344}
2345
1893eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) 2346eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
1894{ 2347{
1895 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND; 2348 REQ (EIO_CUSTOM); req->feed = execute; SEND;
1896} 2349}
1897 2350
1898#endif 2351#endif
1899 2352
1900eio_req *eio_grp (eio_cb cb, void *data) 2353eio_req *eio_grp (eio_cb cb, void *data)
1909#undef SEND 2362#undef SEND
1910 2363
1911/*****************************************************************************/ 2364/*****************************************************************************/
1912/* grp functions */ 2365/* grp functions */
1913 2366
2367void
1914void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2368eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1915{ 2369{
1916 grp->int2 = limit; 2370 grp->int2 = limit;
1917 grp->feed = feed; 2371 grp->feed = feed;
1918 2372
1919 grp_try_feed (grp); 2373 grp_try_feed (grp);
1920} 2374}
1921 2375
2376void
1922void eio_grp_limit (eio_req *grp, int limit) 2377eio_grp_limit (eio_req *grp, int limit)
1923{ 2378{
1924 grp->int2 = limit; 2379 grp->int2 = limit;
1925 2380
1926 grp_try_feed (grp); 2381 grp_try_feed (grp);
1927} 2382}
1928 2383
2384void
1929void eio_grp_add (eio_req *grp, eio_req *req) 2385eio_grp_add (eio_req *grp, eio_req *req)
1930{ 2386{
1931 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2387 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1932 2388
1933 grp->flags |= EIO_FLAG_GROUPADD; 2389 grp->flags |= ETP_FLAG_GROUPADD;
1934 2390
1935 ++grp->size; 2391 ++grp->size;
1936 req->grp = grp; 2392 req->grp = grp;
1937 2393
1938 req->grp_prev = 0; 2394 req->grp_prev = 0;
1945} 2401}
1946 2402
1947/*****************************************************************************/ 2403/*****************************************************************************/
1948/* misc garbage */ 2404/* misc garbage */
1949 2405
2406eio_ssize_t
1950ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2407eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1951{ 2408{
1952 etp_worker wrk;
1953 ssize_t ret;
1954
1955 wrk.dbuf = 0;
1956
1957 ret = eio__sendfile (ofd, ifd, offset, count, &wrk); 2409 return eio__sendfile (ofd, ifd, offset, count);
1958
1959 if (wrk.dbuf)
1960 free (wrk.dbuf);
1961
1962 return ret;
1963} 2410}
1964 2411

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines