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

Comparing libeio/eio.c (file contents):
Revision 1.26 by root, Sun Oct 12 22:30:33 2008 UTC vs.
Revision 1.139 by root, Thu Jun 25 18:14:19 2015 UTC

1/* 1/*
2 * libeio implementation 2 * libeio implementation
3 * 3 *
4 * Copyright (c) 2007,2008 Marc Alexander Lehmann <libeio@schmorp.de> 4 * Copyright (c) 2007,2008,2009,2010,2011,2012,2013 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"
45#include "ecb.h"
46
47#ifdef EIO_STACKSIZE
48# define X_STACKSIZE EIO_STACKSIZE
49#endif
41#include "xthread.h" 50#include "xthread.h"
42 51
43#include <errno.h> 52#include <errno.h>
44#include <stddef.h> 53#include <stddef.h>
45#include <stdlib.h> 54#include <stdlib.h>
… …
49#include <sys/stat.h> 58#include <sys/stat.h>
50#include <limits.h> 59#include <limits.h>
51#include <fcntl.h> 60#include <fcntl.h>
52#include <assert.h> 61#include <assert.h>
53 62
63/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
64/* intptr_t only comes from stdint.h, says idiot openbsd coder */
65#if HAVE_STDINT_H
66# include <stdint.h>
67#endif
68
69#ifndef ECANCELED
70# define ECANCELED EDOM
71#endif
72#ifndef ELOOP
73# define ELOOP EDOM
74#endif
75
76#if !defined(ENOTSOCK) && defined(WSAENOTSOCK)
77# define ENOTSOCK WSAENOTSOCK
78#endif
79
80static void eio_destroy (eio_req *req);
81
54#ifndef EIO_FINISH 82#ifndef EIO_FINISH
55# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0 83# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
56#endif 84#endif
57 85
58#ifndef EIO_DESTROY 86#ifndef EIO_DESTROY
… …
61 89
62#ifndef EIO_FEED 90#ifndef EIO_FEED
63# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0) 91# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)
64#endif 92#endif
65 93
94#ifndef EIO_FD_TO_WIN32_HANDLE
95# define EIO_FD_TO_WIN32_HANDLE(fd) _get_osfhandle (fd)
96#endif
97#ifndef EIO_WIN32_HANDLE_TO_FD
98# define EIO_WIN32_HANDLE_TO_FD(handle) _open_osfhandle (handle, 0)
99#endif
100
101#define EIO_ERRNO(errval,retval) ((errno = errval), retval)
102
103#define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1)
104
66#ifdef _WIN32 105#ifdef _WIN32
67 106
68 /*doh*/ 107 #undef PAGESIZE
108 #define PAGESIZE 4096 /* GetSystemInfo? */
109
110 /* TODO: look at how perl does stat (non-sloppy), unlink (ro-files), utime, link */
111
112 #ifdef EIO_STRUCT_STATI64
113 /* look at perl's non-sloppy stat */
114 #define stat(path,buf) _stati64 (path,buf)
115 #define fstat(fd,buf) _fstati64 (fd,buf)
116 #endif
117 #define lstat(path,buf) stat (path,buf)
118 #define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
119 #define mkdir(path,mode) _mkdir (path)
120 #define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
121
122 #define chmod(path,mode) _chmod (path, mode)
123 #define dup(fd) _dup (fd)
124 #define dup2(fd1,fd2) _dup2 (fd1, fd2)
125 #define pipe(fds) _pipe (fds, 4096, O_BINARY)
126
127 #define fchmod(fd,mode) EIO_ENOSYS ()
128 #define chown(path,uid,gid) EIO_ENOSYS ()
129 #define fchown(fd,uid,gid) EIO_ENOSYS ()
130 #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
131 #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
132 #define mknod(path,mode,dev) EIO_ENOSYS ()
133 #define sync() EIO_ENOSYS ()
134 #define readlink(path,buf,s) EIO_ENOSYS ()
135 #define statvfs(path,buf) EIO_ENOSYS ()
136 #define fstatvfs(fd,buf) EIO_ENOSYS ()
137
138 #define pread(fd,buf,count,offset) eio__pread (fd, buf, count, offset)
139 #define pwrite(fd,buf,count,offset) eio__pwrite (fd, buf, count, offset)
140
141 #if __GNUC__
142 typedef long long eio_off_t; /* signed for compatibility to msvc */
143 #else
144 typedef __int64 eio_off_t; /* unsigned not supported by msvc */
145 #endif
146
147 static eio_ssize_t
148 eio__pread (int fd, void *buf, eio_ssize_t count, eio_off_t offset)
149 {
150 OVERLAPPED o = { 0 };
151 DWORD got;
152
153 o.Offset = offset;
154 o.OffsetHigh = offset >> 32;
155
156 return ReadFile ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd), buf, count, &got, &o)
157 ? got : -1;
158 }
159
160 static eio_ssize_t
161 eio__pwrite (int fd, void *buf, eio_ssize_t count, eio_off_t offset)
162 {
163 OVERLAPPED o = { 0 };
164 DWORD got;
165
166 o.Offset = offset;
167 o.OffsetHigh = offset >> 32;
168
169 return WriteFile ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd), buf, count, &got, &o)
170 ? got : -1;
171 }
172
173 /* rename() uses MoveFile, which fails to overwrite */
174 #define rename(old,neu) eio__rename (old, neu)
175
176 static int
177 eio__rename (const char *old, const char *neu)
178 {
179 if (MoveFileEx (old, neu, MOVEFILE_REPLACE_EXISTING))
180 return 0;
181
182 /* should steal _dosmaperr */
183 switch (GetLastError ())
184 {
185 case ERROR_FILE_NOT_FOUND:
186 case ERROR_PATH_NOT_FOUND:
187 case ERROR_INVALID_DRIVE:
188 case ERROR_NO_MORE_FILES:
189 case ERROR_BAD_NETPATH:
190 case ERROR_BAD_NET_NAME:
191 case ERROR_BAD_PATHNAME:
192 case ERROR_FILENAME_EXCED_RANGE:
193 errno = ENOENT;
194 break;
195
196 default:
197 errno = EACCES;
198 break;
199 }
200
201 return -1;
202 }
203
204 /* we could even stat and see if it exists */
205 static int
206 symlink (const char *old, const char *neu)
207 {
208 #if WINVER >= 0x0600
209 if (CreateSymbolicLink (neu, old, 1))
210 return 0;
211
212 if (CreateSymbolicLink (neu, old, 0))
213 return 0;
214 #endif
215
216 return EIO_ERRNO (ENOENT, -1);
217 }
218
219 /* POSIX API only, causing trouble for win32 apps */
220 #define CreateHardLink(neu,old,flags) 0 /* not really creating hardlink, still using relative paths? */
221 #define CreateSymbolicLink(neu,old,flags) 0 /* vista+ only */
222
223 struct statvfs
224 {
225 int dummy;
226 };
227
228 #define DT_DIR EIO_DT_DIR
229 #define DT_REG EIO_DT_REG
230 #define D_NAME(entp) entp.cFileName
231 #define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
69 232
70#else 233#else
71 234
72# include "config.h"
73# include <sys/time.h> 235 #include <sys/time.h>
74# include <sys/select.h> 236 #include <sys/select.h>
75# include <unistd.h> 237 #include <unistd.h>
238 #include <signal.h>
239 #include <dirent.h>
240
241 #ifdef ANDROID
242 #include <sys/vfs.h>
243 #define statvfs statfs
244 #define fstatvfs fstatfs
245 #include <asm/page.h> /* supposedly limits.h does #define PAGESIZE PAGESIZE */
246 #else
247 #include <sys/statvfs.h>
248 #endif
249
250 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
251 #include <sys/mman.h>
252 #endif
253
254 #define D_NAME(entp) entp->d_name
255
256 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
257 #if __FreeBSD__ || __NetBSD__ || __OpenBSD__
258 #define _DIRENT_HAVE_D_TYPE /* sigh */
259 #define D_INO(de) (de)->d_fileno
260 #define D_NAMLEN(de) (de)->d_namlen
261 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
262 #define D_INO(de) (de)->d_ino
263 #endif
264
265 #ifdef _D_EXACT_NAMLEN
266 #undef D_NAMLEN
267 #define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
268 #endif
269
270 #ifdef _DIRENT_HAVE_D_TYPE
271 #define D_TYPE(de) (de)->d_type
272 #endif
273
274 #ifndef EIO_STRUCT_DIRENT
275 #define EIO_STRUCT_DIRENT struct dirent
276 #endif
277
278#endif
279
280#if HAVE_UTIMES
76# include <utime.h> 281# include <utime.h>
77# include <signal.h>
78# include <dirent.h>
79
80# ifndef EIO_STRUCT_DIRENT
81# define EIO_STRUCT_DIRENT struct dirent
82# endif 282#endif
83 283
284#if HAVE_SYS_SYSCALL_H
285# include <sys/syscall.h>
286#endif
287
288#if HAVE_SYS_PRCTL_H
289# include <sys/prctl.h>
84#endif 290#endif
85 291
86#if HAVE_SENDFILE 292#if HAVE_SENDFILE
87# if __linux 293# if __linux
88# include <sys/sendfile.h> 294# include <sys/sendfile.h>
89# elif __freebsd 295# elif __FreeBSD__ || defined __APPLE__
90# include <sys/socket.h> 296# include <sys/socket.h>
91# include <sys/uio.h> 297# include <sys/uio.h>
92# elif __hpux 298# elif __hpux
93# include <sys/socket.h> 299# include <sys/socket.h>
94# elif __solaris /* not yet */ 300# elif __solaris
95# include <sys/sendfile.h> 301# include <sys/sendfile.h>
96# else 302# else
97# error sendfile support requested but not available 303# error sendfile support requested but not available
98# endif 304# endif
99#endif 305#endif
100 306
101/* number of seconds after which an idle threads exit */ 307#ifndef D_TYPE
102#define IDLE_TIMEOUT 10 308# define D_TYPE(de) 0
309#endif
310#ifndef D_INO
311# define D_INO(de) 0
312#endif
313#ifndef D_NAMLEN
314# define D_NAMLEN(entp) strlen (D_NAME (entp))
315#endif
103 316
104/* used for struct dirent, AIX doesn't provide it */ 317/* used for struct dirent, AIX doesn't provide it */
105#ifndef NAME_MAX 318#ifndef NAME_MAX
106# define NAME_MAX 4096 319# define NAME_MAX 4096
107#endif 320#endif
108 321
322/* used for readlink etc. */
323#ifndef PATH_MAX
324# define PATH_MAX 4096
325#endif
326
109/* buffer size for various temporary buffers */ 327/* buffer size for various temporary buffers */
110#define EIO_BUFSIZE 65536 328#define EIO_BUFSIZE 65536
111 329
112#define dBUF \ 330#define dBUF \
113 char *eio_buf; \
114 ETP_WORKER_LOCK (self); \
115 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 331 char *eio_buf = malloc (EIO_BUFSIZE); \
116 ETP_WORKER_UNLOCK (self); \
117 errno = ENOMEM; \ 332 errno = ENOMEM; \
118 if (!eio_buf) \ 333 if (!eio_buf) \
119 return -1; 334 return -1
120 335
121#define EIO_TICKS ((1000000 + 1023) >> 10) 336#define FUBd \
337 free (eio_buf)
122 338
123/*****************************************************************************/ 339/*****************************************************************************/
124 340
125#if __GNUC__ >= 3 341struct etp_tmpbuf;
126# define expect(expr,value) __builtin_expect ((expr),(value)) 342
343#if _POSIX_VERSION >= 200809L
344 #define HAVE_AT 1
345 #define WD2FD(wd) ((wd) ? (wd)->fd : AT_FDCWD)
346 #ifndef O_SEARCH
347 #define O_SEARCH O_RDONLY
348 #endif
127#else 349#else
128# define expect(expr,value) (expr) 350 #define HAVE_AT 0
351 static const char *wd_expand (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path);
129#endif 352#endif
130 353
131#define expect_false(expr) expect ((expr) != 0, 0) 354struct eio_pwd
132#define expect_true(expr) expect ((expr) != 0, 1) 355{
356#if HAVE_AT
357 int fd;
358#endif
359 int len;
360 char str[1]; /* actually, a 0-terminated canonical path */
361};
133 362
134/*****************************************************************************/ 363/*****************************************************************************/
135 364
136#define ETP_PRI_MIN EIO_PRI_MIN 365#define ETP_PRI_MIN EIO_PRI_MIN
137#define ETP_PRI_MAX EIO_PRI_MAX 366#define ETP_PRI_MAX EIO_PRI_MAX
138 367
368#define ETP_TYPE_QUIT -1
369#define ETP_TYPE_GROUP EIO_GROUP
370
371static void eio_nop_callback (void) { }
372static void (*eio_want_poll_cb)(void) = eio_nop_callback;
373static void (*eio_done_poll_cb)(void) = eio_nop_callback;
374
375#define ETP_WANT_POLL(pool) eio_want_poll_cb ()
376#define ETP_DONE_POLL(pool) eio_done_poll_cb ()
377
139struct etp_worker; 378struct etp_worker;
140
141#define ETP_REQ eio_req 379#define ETP_REQ eio_req
142#define ETP_DESTROY(req) eio_destroy (req) 380#define ETP_DESTROY(req) eio_destroy (req)
143static int eio_finish (eio_req *req); 381static int eio_finish (eio_req *req);
144#define ETP_FINISH(req) eio_finish (req) 382#define ETP_FINISH(req) eio_finish (req)
145static void eio_execute (struct etp_worker *self, eio_req *req); 383static void eio_execute (struct etp_worker *self, eio_req *req);
146#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) 384#define ETP_EXECUTE(wrk,req) eio_execute (wrk, req)
147 385
148#define ETP_WORKER_CLEAR(req) \ 386#include "etp.c"
149 if (wrk->dbuf) \ 387
150 { \ 388static struct etp_pool eio_pool;
151 free (wrk->dbuf); \ 389#define EIO_POOL (&eio_pool)
152 wrk->dbuf = 0; \
153 } \
154 \
155 if (wrk->dirp) \
156 { \
157 closedir (wrk->dirp); \
158 wrk->dirp = 0; \
159 }
160#define ETP_WORKER_COMMON \
161 void *dbuf; \
162 DIR *dirp;
163 390
164/*****************************************************************************/ 391/*****************************************************************************/
165 392
166#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
167
168/* calculcate time difference in ~1/EIO_TICKS of a second */
169static int tvdiff (struct timeval *tv1, struct timeval *tv2)
170{
171 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
172 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
173}
174
175static unsigned int started, idle, wanted = 4;
176
177static void (*want_poll_cb) (void);
178static void (*done_poll_cb) (void);
179
180static unsigned int max_poll_time; /* reslock */
181static unsigned int max_poll_reqs; /* reslock */
182
183static volatile unsigned int nreqs; /* reqlock */
184static volatile unsigned int nready; /* reqlock */
185static volatile unsigned int npending; /* reqlock */
186static volatile unsigned int max_idle = 4;
187
188static mutex_t wrklock = X_MUTEX_INIT;
189static mutex_t reslock = X_MUTEX_INIT;
190static mutex_t reqlock = X_MUTEX_INIT;
191static cond_t reqwait = X_COND_INIT;
192
193#if !HAVE_PREADWRITE
194/*
195 * make our pread/pwrite emulation safe against themselves, but not against
196 * normal read/write by using a mutex. slows down execution a lot,
197 * but that's your problem, not mine.
198 */
199static mutex_t preadwritelock = X_MUTEX_INIT;
200#endif
201
202typedef struct etp_worker
203{
204 /* locked by wrklock */
205 struct etp_worker *prev, *next;
206
207 thread_t tid;
208
209 /* locked by reslock, reqlock or wrklock */
210 ETP_REQ *req; /* currently processed request */
211
212 ETP_WORKER_COMMON
213} etp_worker;
214
215static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */
216
217#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
218#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
219
220/* worker threads management */
221
222static void etp_worker_clear (etp_worker *wrk)
223{
224 ETP_WORKER_CLEAR (wrk);
225}
226
227static void etp_worker_free (etp_worker *wrk)
228{
229 wrk->next->prev = wrk->prev;
230 wrk->prev->next = wrk->next;
231
232 free (wrk);
233}
234
235static unsigned int etp_nreqs (void)
236{
237 int retval;
238 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
239 retval = nreqs;
240 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
241 return retval;
242}
243
244static unsigned int etp_nready (void)
245{
246 unsigned int retval;
247
248 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
249 retval = nready;
250 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
251
252 return retval;
253}
254
255static unsigned int etp_npending (void)
256{
257 unsigned int retval;
258
259 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
260 retval = npending;
261 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
262
263 return retval;
264}
265
266static unsigned int etp_nthreads (void)
267{
268 unsigned int retval;
269
270 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
271 retval = started;
272 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
273
274 return retval;
275}
276
277/*
278 * a somewhat faster data structure might be nice, but
279 * with 8 priorities this actually needs <20 insns
280 * per shift, the most expensive operation.
281 */
282typedef struct {
283 ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */
284 int size;
285} etp_reqq;
286
287static etp_reqq req_queue;
288static etp_reqq res_queue;
289
290static int reqq_push (etp_reqq *q, ETP_REQ *req)
291{
292 int pri = req->pri;
293 req->next = 0;
294
295 if (q->qe[pri])
296 {
297 q->qe[pri]->next = req;
298 q->qe[pri] = req;
299 }
300 else
301 q->qe[pri] = q->qs[pri] = req;
302
303 return q->size++;
304}
305
306static ETP_REQ *reqq_shift (etp_reqq *q)
307{
308 int pri;
309
310 if (!q->size)
311 return 0;
312
313 --q->size;
314
315 for (pri = ETP_NUM_PRI; pri--; )
316 {
317 eio_req *req = q->qs[pri];
318
319 if (req)
320 {
321 if (!(q->qs[pri] = (eio_req *)req->next))
322 q->qe[pri] = 0;
323
324 return req;
325 }
326 }
327
328 abort ();
329}
330
331static void etp_atfork_prepare (void)
332{
333 X_LOCK (wrklock);
334 X_LOCK (reqlock);
335 X_LOCK (reslock);
336#if !HAVE_PREADWRITE
337 X_LOCK (preadwritelock);
338#endif
339}
340
341static void etp_atfork_parent (void)
342{
343#if !HAVE_PREADWRITE
344 X_UNLOCK (preadwritelock);
345#endif
346 X_UNLOCK (reslock);
347 X_UNLOCK (reqlock);
348 X_UNLOCK (wrklock);
349}
350
351static void etp_atfork_child (void)
352{
353 ETP_REQ *prv;
354
355 while ((prv = reqq_shift (&req_queue)))
356 ETP_DESTROY (prv);
357
358 while ((prv = reqq_shift (&res_queue)))
359 ETP_DESTROY (prv);
360
361 while (wrk_first.next != &wrk_first)
362 {
363 etp_worker *wrk = wrk_first.next;
364
365 if (wrk->req)
366 ETP_DESTROY (wrk->req);
367
368 etp_worker_clear (wrk);
369 etp_worker_free (wrk);
370 }
371
372 started = 0;
373 idle = 0;
374 nreqs = 0;
375 nready = 0;
376 npending = 0;
377
378 etp_atfork_parent ();
379}
380
381static void 393static void
382etp_once_init (void)
383{
384 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
385}
386
387static int
388etp_init (void (*want_poll)(void), void (*done_poll)(void))
389{
390 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
391
392 pthread_once (&doinit, etp_once_init);
393
394 want_poll_cb = want_poll;
395 done_poll_cb = done_poll;
396
397 return 0;
398}
399
400X_THREAD_PROC (etp_proc);
401
402static void etp_start_thread (void)
403{
404 etp_worker *wrk = calloc (1, sizeof (etp_worker));
405
406 /*TODO*/
407 assert (("unable to allocate worker thread data", wrk));
408
409 X_LOCK (wrklock);
410
411 if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
412 {
413 wrk->prev = &wrk_first;
414 wrk->next = wrk_first.next;
415 wrk_first.next->prev = wrk;
416 wrk_first.next = wrk;
417 ++started;
418 }
419 else
420 free (wrk);
421
422 X_UNLOCK (wrklock);
423}
424
425static void etp_maybe_start_thread (void)
426{
427 if (expect_true (etp_nthreads () >= wanted))
428 return;
429
430 /* todo: maybe use idle here, but might be less exact */
431 if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
432 return;
433
434 etp_start_thread ();
435}
436
437static void etp_end_thread (void)
438{
439 eio_req *req = calloc (1, sizeof (eio_req));
440
441 req->type = -1;
442 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
443
444 X_LOCK (reqlock);
445 reqq_push (&req_queue, req);
446 X_COND_SIGNAL (reqwait);
447 X_UNLOCK (reqlock);
448
449 X_LOCK (wrklock);
450 --started;
451 X_UNLOCK (wrklock);
452}
453
454static int etp_poll (void)
455{
456 unsigned int maxreqs;
457 unsigned int maxtime;
458 struct timeval tv_start, tv_now;
459
460 X_LOCK (reslock);
461 maxreqs = max_poll_reqs;
462 maxtime = max_poll_time;
463 X_UNLOCK (reslock);
464
465 if (maxtime)
466 gettimeofday (&tv_start, 0);
467
468 for (;;)
469 {
470 ETP_REQ *req;
471
472 etp_maybe_start_thread ();
473
474 X_LOCK (reslock);
475 req = reqq_shift (&res_queue);
476
477 if (req)
478 {
479 --npending;
480
481 if (!res_queue.size && done_poll_cb)
482 done_poll_cb ();
483 }
484
485 X_UNLOCK (reslock);
486
487 if (!req)
488 return 0;
489
490 X_LOCK (reqlock);
491 --nreqs;
492 X_UNLOCK (reqlock);
493
494 if (expect_false (req->type == EIO_GROUP && req->size))
495 {
496 req->int1 = 1; /* mark request as delayed */
497 continue;
498 }
499 else
500 {
501 int res = ETP_FINISH (req);
502 if (expect_false (res))
503 return res;
504 }
505
506 if (expect_false (maxreqs && !--maxreqs))
507 break;
508
509 if (maxtime)
510 {
511 gettimeofday (&tv_now, 0);
512
513 if (tvdiff (&tv_start, &tv_now) >= maxtime)
514 break;
515 }
516 }
517
518 errno = EAGAIN;
519 return -1;
520}
521
522static void etp_cancel (ETP_REQ *req)
523{
524 X_LOCK (wrklock);
525 req->flags |= EIO_FLAG_CANCELLED;
526 X_UNLOCK (wrklock);
527
528 eio_grp_cancel (req);
529}
530
531static void etp_submit (ETP_REQ *req)
532{
533 req->pri -= ETP_PRI_MIN;
534
535 if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
536 if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
537
538 if (expect_false (req->type == EIO_GROUP))
539 {
540 /* I hope this is worth it :/ */
541 X_LOCK (reqlock);
542 ++nreqs;
543 X_UNLOCK (reqlock);
544
545 X_LOCK (reslock);
546
547 ++npending;
548
549 if (!reqq_push (&res_queue, req) && want_poll_cb)
550 want_poll_cb ();
551
552 X_UNLOCK (reslock);
553 }
554 else
555 {
556 X_LOCK (reqlock);
557 ++nreqs;
558 ++nready;
559 reqq_push (&req_queue, req);
560 X_COND_SIGNAL (reqwait);
561 X_UNLOCK (reqlock);
562
563 etp_maybe_start_thread ();
564 }
565}
566
567static void etp_set_max_poll_time (double nseconds)
568{
569 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
570 max_poll_time = nseconds;
571 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
572}
573
574static void etp_set_max_poll_reqs (unsigned int maxreqs)
575{
576 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
577 max_poll_reqs = maxreqs;
578 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
579}
580
581static void etp_set_max_idle (unsigned int nthreads)
582{
583 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
584 max_idle = nthreads <= 0 ? 1 : nthreads;
585 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
586}
587
588static void etp_set_min_parallel (unsigned int nthreads)
589{
590 if (wanted < nthreads)
591 wanted = nthreads;
592}
593
594static void etp_set_max_parallel (unsigned int nthreads)
595{
596 if (wanted > nthreads)
597 wanted = nthreads;
598
599 while (started > wanted)
600 etp_end_thread ();
601}
602
603/*****************************************************************************/
604
605static void grp_try_feed (eio_req *grp) 394grp_try_feed (eio_req *grp)
606{ 395{
607 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 396 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
608 { 397 {
609 grp->flags &= ~EIO_FLAG_GROUPADD; 398 grp->flags &= ~ETP_FLAG_GROUPADD;
610 399
611 EIO_FEED (grp); 400 EIO_FEED (grp);
612 401
613 /* stop if no progress has been made */ 402 /* stop if no progress has been made */
614 if (!(grp->flags & EIO_FLAG_GROUPADD)) 403 if (!(grp->flags & ETP_FLAG_GROUPADD))
615 { 404 {
616 grp->feed = 0; 405 grp->feed = 0;
617 break; 406 break;
618 } 407 }
619 } 408 }
620} 409}
621 410
411static int
622static int grp_dec (eio_req *grp) 412grp_dec (eio_req *grp)
623{ 413{
624 --grp->size; 414 --grp->size;
625 415
626 /* call feeder, if applicable */ 416 /* call feeder, if applicable */
627 grp_try_feed (grp); 417 grp_try_feed (grp);
628 418
629 /* finish, if done */ 419 /* finish, if done */
630 if (!grp->size && grp->int1) 420 if (!grp->size && grp->flags & ETP_FLAG_DELAYED)
631 return eio_finish (grp); 421 return eio_finish (grp);
632 else 422 else
633 return 0; 423 return 0;
634} 424}
635 425
426static void
636void eio_destroy (eio_req *req) 427eio_destroy (eio_req *req)
637{ 428{
638 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); 429 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
639 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); 430 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
640 431
641 EIO_DESTROY (req); 432 EIO_DESTROY (req);
642} 433}
643 434
435static int
644static int eio_finish (eio_req *req) 436eio_finish (eio_req *req)
645{ 437{
646 int res = EIO_FINISH (req); 438 int res = EIO_FINISH (req);
647 439
648 if (req->grp) 440 if (req->grp)
649 { 441 {
… …
657 if (grp->grp_first == req) 449 if (grp->grp_first == req)
658 grp->grp_first = req->grp_next; 450 grp->grp_first = req->grp_next;
659 451
660 res2 = grp_dec (grp); 452 res2 = grp_dec (grp);
661 453
662 if (!res && res2) 454 if (!res)
663 res = res2; 455 res = res2;
664 } 456 }
665 457
666 eio_destroy (req); 458 eio_destroy (req);
667 459
668 return res; 460 return res;
669} 461}
670 462
463void
671void eio_grp_cancel (eio_req *grp) 464eio_grp_cancel (eio_req *grp)
672{ 465{
673 for (grp = grp->grp_first; grp; grp = grp->grp_next) 466 etp_grp_cancel (EIO_POOL, grp);
674 eio_cancel (grp);
675} 467}
676 468
469void
677void eio_cancel (eio_req *req) 470eio_cancel (eio_req *req)
678{ 471{
679 etp_cancel (req); 472 etp_cancel (EIO_POOL, req);
680} 473}
681 474
475void
682void eio_submit (eio_req *req) 476eio_submit (eio_req *req)
683{ 477{
684 etp_submit (req); 478 etp_submit (EIO_POOL, req);
685} 479}
686 480
687unsigned int eio_nreqs (void) 481unsigned int
482eio_nreqs (void)
688{ 483{
689 return etp_nreqs (); 484 return etp_nreqs (EIO_POOL);
690} 485}
691 486
692unsigned int eio_nready (void) 487unsigned int
488eio_nready (void)
693{ 489{
694 return etp_nready (); 490 return etp_nready (EIO_POOL);
695} 491}
696 492
697unsigned int eio_npending (void) 493unsigned int
494eio_npending (void)
698{ 495{
699 return etp_npending (); 496 return etp_npending (EIO_POOL);
700} 497}
701 498
702unsigned int eio_nthreads (void) 499unsigned int ecb_cold
500eio_nthreads (void)
703{ 501{
704 return etp_nthreads (); 502 return etp_nthreads (EIO_POOL);
705} 503}
706 504
505void ecb_cold
707void eio_set_max_poll_time (double nseconds) 506eio_set_max_poll_time (double nseconds)
708{ 507{
709 etp_set_max_poll_time (nseconds); 508 etp_set_max_poll_time (EIO_POOL, nseconds);
710} 509}
711 510
511void ecb_cold
712void eio_set_max_poll_reqs (unsigned int maxreqs) 512eio_set_max_poll_reqs (unsigned int maxreqs)
713{ 513{
714 etp_set_max_poll_reqs (maxreqs); 514 etp_set_max_poll_reqs (EIO_POOL, maxreqs);
715} 515}
716 516
517void ecb_cold
717void eio_set_max_idle (unsigned int nthreads) 518eio_set_max_idle (unsigned int nthreads)
718{ 519{
719 etp_set_max_idle (nthreads); 520 etp_set_max_idle (EIO_POOL, nthreads);
720} 521}
721 522
523void ecb_cold
524eio_set_idle_timeout (unsigned int seconds)
525{
526 etp_set_idle_timeout (EIO_POOL, seconds);
527}
528
529void ecb_cold
722void eio_set_min_parallel (unsigned int nthreads) 530eio_set_min_parallel (unsigned int nthreads)
723{ 531{
724 etp_set_min_parallel (nthreads); 532 etp_set_min_parallel (EIO_POOL, nthreads);
725} 533}
726 534
535void ecb_cold
727void eio_set_max_parallel (unsigned int nthreads) 536eio_set_max_parallel (unsigned int nthreads)
728{ 537{
729 etp_set_max_parallel (nthreads); 538 etp_set_max_parallel (EIO_POOL, nthreads);
730} 539}
731 540
732int eio_poll (void) 541int eio_poll (void)
733{ 542{
734 return etp_poll (); 543 return etp_poll (EIO_POOL);
735} 544}
736 545
737/*****************************************************************************/ 546/*****************************************************************************/
738/* work around various missing functions */ 547/* work around various missing functions */
739 548
740#if !HAVE_PREADWRITE
741# define pread eio__pread
742# define pwrite eio__pwrite
743
744static ssize_t
745eio__pread (int fd, void *buf, size_t count, off_t offset)
746{
747 ssize_t res;
748 off_t ooffset;
749
750 X_LOCK (preadwritelock);
751 ooffset = lseek (fd, 0, SEEK_CUR);
752 lseek (fd, offset, SEEK_SET);
753 res = read (fd, buf, count);
754 lseek (fd, ooffset, SEEK_SET);
755 X_UNLOCK (preadwritelock);
756
757 return res;
758}
759
760static ssize_t
761eio__pwrite (int fd, void *buf, size_t count, off_t offset)
762{
763 ssize_t res;
764 off_t ooffset;
765
766 X_LOCK (preadwritelock);
767 ooffset = lseek (fd, 0, SEEK_CUR);
768 lseek (fd, offset, SEEK_SET);
769 res = write (fd, buf, count);
770 lseek (fd, offset, SEEK_SET);
771 X_UNLOCK (preadwritelock);
772
773 return res;
774}
775#endif
776
777#ifndef HAVE_FUTIMES 549#ifndef HAVE_UTIMES
778 550
551# undef utimes
779# define utimes(path,times) eio__utimes (path, times) 552# define utimes(path,times) eio__utimes (path, times)
780# define futimes(fd,times) eio__futimes (fd, times)
781 553
782static int 554static int
783eio__utimes (const char *filename, const struct timeval times[2]) 555eio__utimes (const char *filename, const struct timeval times[2])
784{ 556{
785 if (times) 557 if (times)
… …
793 } 565 }
794 else 566 else
795 return utime (filename, 0); 567 return utime (filename, 0);
796} 568}
797 569
570#endif
571
572#ifndef HAVE_FUTIMES
573
574# undef futimes
575# define futimes(fd,times) eio__futimes (fd, times)
576
577static int
798static int eio__futimes (int fd, const struct timeval tv[2]) 578eio__futimes (int fd, const struct timeval tv[2])
799{ 579{
800 errno = ENOSYS; 580 errno = ENOSYS;
801 return -1; 581 return -1;
802} 582}
803 583
804#endif 584#endif
805 585
806#if !HAVE_FDATASYNC 586#if !HAVE_FDATASYNC
587# undef fdatasync
807# define fdatasync fsync 588# define fdatasync(fd) fsync (fd)
808#endif 589#endif
590
591static int
592eio__syncfs (int fd)
593{
594 int res;
595
596#if HAVE_SYS_SYNCFS
597 res = (int)syscall (__NR_syncfs, (int)(fd));
598#else
599 res = EIO_ENOSYS ();
600#endif
601
602 if (res < 0 && errno == ENOSYS && fd >= 0)
603 sync ();
604
605 return res;
606}
607
608/* sync_file_range always needs emulation */
609static int
610eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
611{
612#if HAVE_SYNC_FILE_RANGE
613 int res;
614
615 if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
616 || EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
617 || EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
618 {
619 flags = 0
620 | (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
621 | (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
622 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
623 }
624
625 res = sync_file_range (fd, offset, nbytes, flags);
626
627 if (!res || errno != ENOSYS)
628 return res;
629#endif
630
631 /* even though we could play tricks with the flags, it's better to always
632 * call fdatasync, as that matches the expectation of its users best */
633 return fdatasync (fd);
634}
635
636static int
637eio__fallocate (int fd, int mode, off_t offset, size_t len)
638{
639#if HAVE_LINUX_FALLOCATE
640 return fallocate (fd, mode, offset, len);
641#else
642 return EIO_ENOSYS ();
643#endif
644}
809 645
810#if !HAVE_READAHEAD 646#if !HAVE_READAHEAD
647# undef readahead
811# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 648# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
812 649
813static ssize_t 650static eio_ssize_t
814eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) 651eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
815{ 652{
816 size_t todo = count; 653 size_t todo = count;
817 dBUF; 654 dBUF;
818 655
… …
823 pread (fd, eio_buf, len, offset); 660 pread (fd, eio_buf, len, offset);
824 offset += len; 661 offset += len;
825 todo -= len; 662 todo -= len;
826 } 663 }
827 664
828 errno = 0; 665 FUBd;
666
667 /* linux's readahead basically only fails for EBADF or EINVAL (not mmappable) */
668 /* but not for e.g. EIO or eof, so we also never fail */
829 return count; 669 return 0;
830} 670}
831 671
832#endif 672#endif
833 673
834/* sendfile always needs emulation */ 674/* sendfile always needs emulation */
835static ssize_t 675static eio_ssize_t
836eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) 676eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
837{ 677{
678 eio_ssize_t written = 0;
838 ssize_t res; 679 eio_ssize_t res;
839 680
840 if (!count) 681 if (!count)
841 return 0; 682 return 0;
842 683
684 for (;;)
685 {
686#ifdef __APPLE__
687# undef HAVE_SENDFILE /* broken, as everything on os x */
688#endif
843#if HAVE_SENDFILE 689#if HAVE_SENDFILE
844# if __linux 690# if __linux
691 off_t soffset = offset;
845 res = sendfile (ofd, ifd, &offset, count); 692 res = sendfile (ofd, ifd, &soffset, count);
846 693
847# elif __freebsd 694# elif __FreeBSD__
848 /* 695 /*
849 * Of course, the freebsd sendfile is a dire hack with no thoughts 696 * Of course, the freebsd sendfile is a dire hack with no thoughts
850 * wasted on making it similar to other I/O functions. 697 * wasted on making it similar to other I/O functions.
851 */ 698 */
852 {
853 off_t sbytes; 699 off_t sbytes;
854 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 700 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
855 701
856 if (res < 0 && sbytes) 702 #if 0 /* according to the manpage, this is correct, but broken behaviour */
857 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 703 /* freebsd' sendfile will return 0 on success */
704 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
705 /* not on e.g. EIO or EPIPE - sounds broken */
706 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
858 res = sbytes; 707 res = sbytes;
859 } 708 #endif
709
710 /* according to source inspection, this is correct, and useful behaviour */
711 if (sbytes)
712 res = sbytes;
713
714# elif defined __APPLE__
715 off_t sbytes = count;
716 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
717
718 /* according to the manpage, sbytes is always valid */
719 if (sbytes)
720 res = sbytes;
860 721
861# elif __hpux 722# elif __hpux
862 res = sendfile (ofd, ifd, offset, count, 0, 0); 723 res = sendfile (ofd, ifd, offset, count, 0, 0);
863 724
864# elif __solaris 725# elif __solaris
865 {
866 struct sendfilevec vec; 726 struct sendfilevec vec;
867 size_t sbytes; 727 size_t sbytes;
868 728
869 vec.sfv_fd = ifd; 729 vec.sfv_fd = ifd;
870 vec.sfv_flag = 0; 730 vec.sfv_flag = 0;
871 vec.sfv_off = offset; 731 vec.sfv_off = offset;
872 vec.sfv_len = count; 732 vec.sfv_len = count;
873 733
874 res = sendfilev (ofd, &vec, 1, &sbytes); 734 res = sendfilev (ofd, &vec, 1, &sbytes);
875 735
876 if (res < 0 && sbytes) 736 if (res < 0 && sbytes)
877 res = sbytes; 737 res = sbytes;
878 }
879 738
880# endif 739# endif
740
741#elif defined (_WIN32) && 0
742 /* does not work, just for documentation of what would need to be done */
743 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
744 /* libeio guarantees that the file offset does not change, and windows */
745 /* has no way to get an independent handle to the same file description */
746 HANDLE h = TO_SOCKET (ifd);
747 SetFilePointer (h, offset, 0, FILE_BEGIN);
748 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
749
881#else 750#else
882 res = -1; 751 res = EIO_ENOSYS ();
883 errno = ENOSYS;
884#endif 752#endif
885 753
754 /* we assume sendfile can copy at least 128mb in one go */
755 if (res <= 128 * 1024 * 1024)
756 {
757 if (res > 0)
758 written += res;
759
760 if (written)
761 return written;
762
763 break;
764 }
765 else
766 {
767 /* if we requested more, then probably the kernel was lazy */
768 written += res;
769 offset += res;
770 count -= res;
771
772 if (!count)
773 return written;
774 }
775 }
776
886 if (res < 0 777 if (res < 0
887 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 778 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
779 /* BSDs */
780#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
781 || errno == ENOTSUP
782#endif
783#ifdef EOPNOTSUPP /* windows */
784 || errno == EOPNOTSUPP /* BSDs */
785#endif
888#if __solaris 786#if __solaris
889 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 787 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
890#endif 788#endif
891 ) 789 )
892 ) 790 )
… …
896 794
897 res = 0; 795 res = 0;
898 796
899 while (count) 797 while (count)
900 { 798 {
901 ssize_t cnt; 799 eio_ssize_t cnt;
902 800
903 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 801 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
904 802
905 if (cnt <= 0) 803 if (cnt <= 0)
906 { 804 {
… …
918 816
919 offset += cnt; 817 offset += cnt;
920 res += cnt; 818 res += cnt;
921 count -= cnt; 819 count -= cnt;
922 } 820 }
821
822 FUBd;
923 } 823 }
924 824
925 return res; 825 return res;
826}
827
828#ifdef PAGESIZE
829# define eio_pagesize() PAGESIZE
830#else
831static intptr_t
832eio_pagesize (void)
833{
834 static intptr_t page;
835
836 if (!page)
837 page = sysconf (_SC_PAGESIZE);
838
839 return page;
840}
841#endif
842
843static void
844eio_page_align (void **addr, size_t *length)
845{
846 intptr_t mask = eio_pagesize () - 1;
847
848 /* round down addr */
849 intptr_t adj = mask & (intptr_t)*addr;
850
851 *addr = (void *)((intptr_t)*addr - adj);
852 *length += adj;
853
854 /* round up length */
855 *length = (*length + mask) & ~mask;
856}
857
858#if !_POSIX_MEMLOCK
859# define eio__mlockall(a) EIO_ENOSYS ()
860#else
861
862static int
863eio__mlockall (int flags)
864{
865 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
866 extern int mallopt (int, int);
867 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
868 #endif
869
870 if (EIO_MCL_CURRENT != MCL_CURRENT
871 || EIO_MCL_FUTURE != MCL_FUTURE)
872 {
873 flags = 0
874 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
875 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
876 }
877
878 return mlockall (flags);
879}
880#endif
881
882#if !_POSIX_MEMLOCK_RANGE
883# define eio__mlock(a,b) EIO_ENOSYS ()
884#else
885
886static int
887eio__mlock (void *addr, size_t length)
888{
889 eio_page_align (&addr, &length);
890
891 return mlock (addr, length);
892}
893
894#endif
895
896#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
897# define eio__msync(a,b,c) EIO_ENOSYS ()
898#else
899
900static int
901eio__msync (void *mem, size_t len, int flags)
902{
903 eio_page_align (&mem, &len);
904
905 if (EIO_MS_ASYNC != MS_SYNC
906 || EIO_MS_INVALIDATE != MS_INVALIDATE
907 || EIO_MS_SYNC != MS_SYNC)
908 {
909 flags = 0
910 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
911 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
912 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
913 }
914
915 return msync (mem, len, flags);
916}
917
918#endif
919
920static int
921eio__mtouch (eio_req *req)
922{
923 void *mem = req->ptr2;
924 size_t len = req->size;
925 int flags = req->int1;
926
927 eio_page_align (&mem, &len);
928
929 {
930 intptr_t addr = (intptr_t)mem;
931 intptr_t end = addr + len;
932 intptr_t page = eio_pagesize ();
933
934 if (addr < end)
935 if (flags & EIO_MT_MODIFY) /* modify */
936 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
937 else
938 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
939 }
940
941 return 0;
942}
943
944/*****************************************************************************/
945/* requests implemented outside eio_execute, because they are so large */
946
947static void
948eio__lseek (eio_req *req)
949{
950 /* this usually gets optimised away completely, or your compiler sucks, */
951 /* or the whence constants really are not 0, 1, 2 */
952 int whence = req->int2 == EIO_SEEK_SET ? SEEK_SET
953 : req->int2 == EIO_SEEK_CUR ? SEEK_CUR
954 : req->int2 == EIO_SEEK_END ? SEEK_END
955 : req->int2;
956
957 req->offs = lseek (req->int1, req->offs, whence);
958 req->result = req->offs == (off_t)-1 ? -1 : 0;
959}
960
961/* result will always end up in tmpbuf, there is always space for adding a 0-byte */
962static int
963eio__realpath (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
964{
965 char *res;
966 const char *rel = path;
967 char *tmp1, *tmp2;
968#if SYMLOOP_MAX > 32
969 int symlinks = SYMLOOP_MAX;
970#else
971 int symlinks = 32;
972#endif
973
974 errno = EINVAL;
975 if (!rel)
976 return -1;
977
978 errno = ENOENT;
979 if (!*rel)
980 return -1;
981
982 res = etp_tmpbuf_get (tmpbuf, PATH_MAX * 3);
983#ifdef _WIN32
984 if (_access (rel, 4) != 0)
985 return -1;
986
987 symlinks = GetFullPathName (rel, PATH_MAX * 3, res, 0);
988
989 errno = ENAMETOOLONG;
990 if (symlinks >= PATH_MAX * 3)
991 return -1;
992
993 errno = EIO;
994 if (symlinks <= 0)
995 return -1;
996
997 return symlinks;
998
999#else
1000 tmp1 = res + PATH_MAX;
1001 tmp2 = tmp1 + PATH_MAX;
1002
1003#if 0 /* disabled, the musl way to do things is just too racy */
1004#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1005 /* on linux we may be able to ask the kernel */
1006 {
1007 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1008
1009 if (fd >= 0)
1010 {
1011 sprintf (tmp1, "/proc/self/fd/%d", fd);
1012 req->result = readlink (tmp1, res, PATH_MAX);
1013 /* here we should probably stat the open file and the disk file, to make sure they still match */
1014 close (fd);
1015
1016 if (req->result > 0)
1017 goto done;
1018 }
1019 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1020 return -1;
1021 }
1022#endif
1023#endif
1024
1025 if (*rel != '/')
1026 {
1027 int len;
1028
1029 errno = ENOENT;
1030 if (wd == EIO_INVALID_WD)
1031 return -1;
1032
1033 if (wd == EIO_CWD)
1034 {
1035 if (!getcwd (res, PATH_MAX))
1036 return -1;
1037
1038 len = strlen (res);
1039 }
1040 else
1041 memcpy (res, wd->str, len = wd->len);
1042
1043 if (res [1]) /* only use if not / */
1044 res += len;
1045 }
1046
1047 while (*rel)
1048 {
1049 eio_ssize_t len, linklen;
1050 const char *beg = rel;
1051
1052 while (*rel && *rel != '/')
1053 ++rel;
1054
1055 len = rel - beg;
1056
1057 if (!len) /* skip slashes */
1058 {
1059 ++rel;
1060 continue;
1061 }
1062
1063 if (beg [0] == '.')
1064 {
1065 if (len == 1)
1066 continue; /* . - nop */
1067
1068 if (beg [1] == '.' && len == 2)
1069 {
1070 /* .. - back up one component, if possible */
1071
1072 while (res != tmpbuf->ptr)
1073 if (*--res == '/')
1074 break;
1075
1076 continue;
1077 }
1078 }
1079
1080 errno = ENAMETOOLONG;
1081 if (res + 1 + len + 1 >= tmp1)
1082 return -1;
1083
1084 /* copy one component */
1085 *res = '/';
1086 memcpy (res + 1, beg, len);
1087
1088 /* zero-terminate, for readlink */
1089 res [len + 1] = 0;
1090
1091 /* now check if it's a symlink */
1092 linklen = readlink (tmpbuf->ptr, tmp1, PATH_MAX);
1093
1094 if (linklen < 0)
1095 {
1096 if (errno != EINVAL)
1097 return -1;
1098
1099 /* it's a normal directory. hopefully */
1100 res += len + 1;
1101 }
1102 else
1103 {
1104 /* yay, it was a symlink - build new path in tmp2 */
1105 int rellen = strlen (rel);
1106
1107 errno = ENAMETOOLONG;
1108 if (linklen + 1 + rellen >= PATH_MAX)
1109 return -1;
1110
1111 errno = ELOOP;
1112 if (!--symlinks)
1113 return -1;
1114
1115 if (*tmp1 == '/')
1116 res = tmpbuf->ptr; /* symlink resolves to an absolute path */
1117
1118 /* we need to be careful, as rel might point into tmp2 already */
1119 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1120 tmp2 [linklen] = '/';
1121 memcpy (tmp2, tmp1, linklen);
1122
1123 rel = tmp2;
1124 }
1125 }
1126
1127 /* special case for the lone root path */
1128 if (res == tmpbuf->ptr)
1129 *res++ = '/';
1130
1131 return res - (char *)tmpbuf->ptr;
1132#endif
1133}
1134
1135static signed char
1136eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1137{
1138 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1139 : a->inode < b->inode ? -1
1140 : a->inode > b->inode ? 1
1141 : 0;
1142}
1143
1144#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1145
1146#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1147#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1148
1149static void
1150eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1151{
1152 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1153 unsigned char *bit = bits;
1154
1155 assert (CHAR_BIT == 8);
1156 assert (sizeof (eio_dirent) * 8 < 256);
1157 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1158 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1159
1160 if (size <= EIO_SORT_FAST)
1161 return;
1162
1163 /* first prepare an array of bits to test in our radix sort */
1164 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1165 /* inode_bits must contain all inodes ORed together */
1166 /* which is used to skip bits that are 0 everywhere, which is very common */
1167 {
1168 eio_ino_t endianness;
1169 int i, j;
1170
1171 /* we store the byte offset of byte n into byte n of "endianness" */
1172 for (i = 0; i < sizeof (eio_ino_t); ++i)
1173 ((unsigned char *)&endianness)[i] = i;
1174
1175 *bit++ = 0;
1176
1177 for (i = 0; i < sizeof (eio_ino_t); ++i)
1178 {
1179 /* shifting off the byte offsets out of "endianness" */
1180 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1181 endianness >>= 8;
1182
1183 for (j = 0; j < 8; ++j)
1184 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1185 *bit++ = offs + j;
1186 }
1187
1188 for (j = 0; j < 8; ++j)
1189 if (score_bits & (1 << j))
1190 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1191 }
1192
1193 /* now actually do the sorting (a variant of MSD radix sort) */
1194 {
1195 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1196 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1197 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1198 int stk_idx = 0;
1199
1200 base_stk [stk_idx] = dents;
1201 end_stk [stk_idx] = dents + size;
1202 bit_stk [stk_idx] = bit - 1;
1203
1204 do
1205 {
1206 base = base_stk [stk_idx];
1207 end = end_stk [stk_idx];
1208 bit = bit_stk [stk_idx];
1209
1210 for (;;)
1211 {
1212 unsigned char O = *bit >> 3;
1213 unsigned char M = 1 << (*bit & 7);
1214
1215 eio_dirent *a = base;
1216 eio_dirent *b = end;
1217
1218 if (b - a < EIO_SORT_CUTOFF)
1219 break;
1220
1221 /* now bit-partition the array on the bit */
1222 /* this ugly asymmetric loop seems to perform much better than typical */
1223 /* partition algos found in the literature */
1224 do
1225 if (!(((unsigned char *)a)[O] & M))
1226 ++a;
1227 else if (!(((unsigned char *)--b)[O] & M))
1228 {
1229 eio_dirent tmp = *a; *a = *b; *b = tmp;
1230 ++a;
1231 }
1232 while (b > a);
1233
1234 /* next bit, or stop, if no bits left in this path */
1235 if (!*--bit)
1236 break;
1237
1238 base_stk [stk_idx] = a;
1239 end_stk [stk_idx] = end;
1240 bit_stk [stk_idx] = bit;
1241 ++stk_idx;
1242
1243 end = a;
1244 }
1245 }
1246 while (stk_idx--);
1247 }
1248}
1249
1250static void
1251eio_dent_insertion_sort (eio_dirent *dents, int size)
1252{
1253 /* first move the smallest element to the front, to act as a sentinel */
1254 {
1255 int i;
1256 eio_dirent *min = dents;
1257
1258 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1259 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1260 if (EIO_DENT_CMP (dents [i], <, *min))
1261 min = &dents [i];
1262
1263 /* swap elements 0 and j (minimum) */
1264 {
1265 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1266 }
1267 }
1268
1269 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1270 {
1271 eio_dirent *i, *j;
1272
1273 for (i = dents + 1; i < dents + size; ++i)
1274 {
1275 eio_dirent value = *i;
1276
1277 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1278 j [1] = j [0];
1279
1280 j [1] = value;
1281 }
1282 }
1283}
1284
1285static void
1286eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1287{
1288 if (size <= 1)
1289 return; /* our insertion sort relies on size > 0 */
1290
1291 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1292 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1293 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1294
1295 /* use an insertion sort at the end, or for small arrays, */
1296 /* as insertion sort is more efficient for small partitions */
1297 eio_dent_insertion_sort (dents, size);
926} 1298}
927 1299
928/* read a full directory */ 1300/* read a full directory */
929static void 1301static void
930eio__scandir (eio_req *req, etp_worker *self) 1302eio__scandir (eio_req *req, etp_worker *self)
931{ 1303{
1304 char *name, *names;
1305 int namesalloc = 4096 - sizeof (void *) * 4;
1306 int namesoffs = 0;
1307 int flags = req->int1;
1308 eio_dirent *dents = 0;
1309 int dentalloc = 128;
1310 int dentoffs = 0;
1311 eio_ino_t inode_bits = 0;
1312#ifdef _WIN32
1313 HANDLE dirp;
1314 WIN32_FIND_DATA entp;
1315#else
932 DIR *dirp; 1316 DIR *dirp;
933 EIO_STRUCT_DIRENT *entp; 1317 EIO_STRUCT_DIRENT *entp;
934 char *name, *names; 1318#endif
935 int memlen = 4096;
936 int memofs = 0;
937 int res = 0;
938 1319
939 X_LOCK (wrklock); 1320 req->result = -1;
940 /* the corresponding closedir is in ETP_WORKER_CLEAR */
941 self->dirp = dirp = opendir (req->ptr1);
942 req->flags |= EIO_FLAG_PTR2_FREE;
943 req->ptr2 = names = malloc (memlen);
944 X_UNLOCK (wrklock);
945 1321
946 if (dirp && names) 1322 if (!(flags & EIO_READDIR_DENTS))
947 for (;;) 1323 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1324
1325#ifdef _WIN32
1326 {
1327 int len = strlen ((const char *)req->ptr1);
1328 char *path = malloc (MAX_PATH);
1329 const char *fmt;
1330 const char *reqpath = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1331
1332 if (!len)
1333 fmt = "./*";
1334 else if (reqpath[len - 1] == '/' || reqpath[len - 1] == '\\')
1335 fmt = "%s*";
1336 else
1337 fmt = "%s/*";
1338
1339 _snprintf (path, MAX_PATH, fmt, reqpath);
1340 dirp = FindFirstFile (path, &entp);
1341 free (path);
1342
1343 if (dirp == INVALID_HANDLE_VALUE)
1344 {
1345 /* should steal _dosmaperr */
1346 switch (GetLastError ())
1347 {
1348 case ERROR_FILE_NOT_FOUND:
1349 req->result = 0;
1350 break;
1351
1352 case ERROR_INVALID_NAME:
1353 case ERROR_PATH_NOT_FOUND:
1354 case ERROR_NO_MORE_FILES:
1355 errno = ENOENT;
1356 break;
1357
1358 case ERROR_NOT_ENOUGH_MEMORY:
1359 errno = ENOMEM;
1360 break;
1361
1362 default:
1363 errno = EINVAL;
1364 break;
1365 }
1366
1367 return;
1368 }
1369 }
1370#else
1371 #if HAVE_AT
1372 if (req->wd)
948 { 1373 {
1374 int fd = openat (WD2FD (req->wd), req->ptr1, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1375
1376 if (fd < 0)
1377 return;
1378
1379 dirp = fdopendir (fd);
1380
1381 if (!dirp)
1382 close (fd);
1383 }
1384 else
1385 dirp = opendir (req->ptr1);
1386 #else
1387 dirp = opendir (wd_expand (&self->tmpbuf, req->wd, req->ptr1));
1388 #endif
1389
1390 if (!dirp)
1391 return;
1392#endif
1393
1394 if (req->flags & EIO_FLAG_PTR1_FREE)
1395 free (req->ptr1);
1396
1397 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1398 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1399 req->ptr2 = names = malloc (namesalloc);
1400
1401 if (!names || (flags && !dents))
1402 return;
1403
1404 for (;;)
1405 {
1406 int done;
1407
1408#ifdef _WIN32
1409 done = !dirp;
1410#else
949 errno = 0; 1411 errno = 0;
950 entp = readdir (dirp); 1412 entp = readdir (dirp);
1413 done = !entp;
1414#endif
951 1415
1416 if (done)
1417 {
1418#ifndef _WIN32
1419 int old_errno = errno;
1420 closedir (dirp);
1421 errno = old_errno;
1422
952 if (!entp) 1423 if (errno)
1424 break;
1425#endif
1426
1427 /* sort etc. */
1428 req->int1 = flags;
1429 req->result = dentoffs;
1430
1431 if (flags & EIO_READDIR_STAT_ORDER)
1432 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1433 else if (flags & EIO_READDIR_DIRS_FIRST)
1434 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1435 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1436 else
1437 {
1438 /* in this case, all is known, and we just put dirs first and sort them */
1439 eio_dirent *oth = dents + dentoffs;
1440 eio_dirent *dir = dents;
1441
1442 /* now partition dirs to the front, and non-dirs to the back */
1443 /* by walking from both sides and swapping if necessary */
1444 while (oth > dir)
1445 {
1446 if (dir->type == EIO_DT_DIR)
1447 ++dir;
1448 else if ((--oth)->type == EIO_DT_DIR)
1449 {
1450 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1451
1452 ++dir;
1453 }
1454 }
1455
1456 /* now sort the dirs only (dirs all have the same score) */
1457 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1458 }
1459
953 break; 1460 break;
1461 }
954 1462
955 name = entp->d_name; 1463 /* now add the entry to our list(s) */
1464 name = D_NAME (entp);
956 1465
1466 /* skip . and .. entries */
957 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1467 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
958 { 1468 {
959 int len = strlen (name) + 1; 1469 int len = D_NAMLEN (entp) + 1;
960 1470
961 res++; 1471 while (ecb_expect_false (namesoffs + len > namesalloc))
962
963 while (memofs + len > memlen)
964 { 1472 {
965 memlen *= 2; 1473 namesalloc *= 2;
966 X_LOCK (wrklock);
967 req->ptr2 = names = realloc (names, memlen); 1474 req->ptr2 = names = realloc (names, namesalloc);
968 X_UNLOCK (wrklock);
969 1475
970 if (!names) 1476 if (!names)
1477 break;
1478 }
1479
1480 memcpy (names + namesoffs, name, len);
1481
1482 if (dents)
1483 {
1484 struct eio_dirent *ent;
1485
1486 if (ecb_expect_false (dentoffs == dentalloc))
1487 {
1488 dentalloc *= 2;
1489 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1490
1491 if (!dents)
971 break; 1492 break;
972 } 1493 }
973 1494
974 memcpy (names + memofs, name, len); 1495 ent = dents + dentoffs;
975 memofs += len; 1496
1497 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1498 ent->namelen = len - 1;
1499 ent->inode = D_INO (entp);
1500
1501 inode_bits |= ent->inode;
1502
1503 switch (D_TYPE (entp))
1504 {
1505 default:
1506 ent->type = EIO_DT_UNKNOWN;
1507 flags |= EIO_READDIR_FOUND_UNKNOWN;
1508 break;
1509
1510 #ifdef DT_FIFO
1511 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1512 #endif
1513 #ifdef DT_CHR
1514 case DT_CHR: ent->type = EIO_DT_CHR; break;
1515 #endif
1516 #ifdef DT_MPC
1517 case DT_MPC: ent->type = EIO_DT_MPC; break;
1518 #endif
1519 #ifdef DT_DIR
1520 case DT_DIR: ent->type = EIO_DT_DIR; break;
1521 #endif
1522 #ifdef DT_NAM
1523 case DT_NAM: ent->type = EIO_DT_NAM; break;
1524 #endif
1525 #ifdef DT_BLK
1526 case DT_BLK: ent->type = EIO_DT_BLK; break;
1527 #endif
1528 #ifdef DT_MPB
1529 case DT_MPB: ent->type = EIO_DT_MPB; break;
1530 #endif
1531 #ifdef DT_REG
1532 case DT_REG: ent->type = EIO_DT_REG; break;
1533 #endif
1534 #ifdef DT_NWK
1535 case DT_NWK: ent->type = EIO_DT_NWK; break;
1536 #endif
1537 #ifdef DT_CMP
1538 case DT_CMP: ent->type = EIO_DT_CMP; break;
1539 #endif
1540 #ifdef DT_LNK
1541 case DT_LNK: ent->type = EIO_DT_LNK; break;
1542 #endif
1543 #ifdef DT_SOCK
1544 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1545 #endif
1546 #ifdef DT_DOOR
1547 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1548 #endif
1549 #ifdef DT_WHT
1550 case DT_WHT: ent->type = EIO_DT_WHT; break;
1551 #endif
1552 }
1553
1554 ent->score = 7;
1555
1556 if (flags & EIO_READDIR_DIRS_FIRST)
1557 {
1558 if (ent->type == EIO_DT_UNKNOWN)
1559 {
1560 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1561 ent->score = 1;
1562 else if (!strchr (name, '.')) /* absence of dots indicate likely dirs */
1563 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1564 }
1565 else if (ent->type == EIO_DT_DIR)
1566 ent->score = 0;
1567 }
976 } 1568 }
1569
1570 namesoffs += len;
1571 ++dentoffs;
977 } 1572 }
978 1573
979 if (errno) 1574 if (EIO_CANCELLED (req))
980 res = -1; 1575 {
1576 errno = ECANCELED;
1577 break;
1578 }
1579
1580#ifdef _WIN32
1581 if (!FindNextFile (dirp, &entp))
1582 {
1583 FindClose (dirp);
1584 dirp = 0;
1585 }
1586#endif
1587 }
1588}
1589
1590/*****************************************************************************/
1591/* working directory stuff */
1592/* various deficiencies in the posix 2008 api force us to */
1593/* keep the absolute path in string form at all times */
1594/* fuck yeah. */
1595
1596#if !HAVE_AT
1597
1598/* a bit like realpath, but usually faster because it doesn'T have to return */
1599/* an absolute or canonical path */
1600static const char *
1601wd_expand (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
1602{
1603 if (!wd || *path == '/')
1604 return path;
1605
1606 if (path [0] == '.' && !path [1])
1607 return wd->str;
1608
981 1609 {
982 req->result = res; 1610 int l1 = wd->len;
983} 1611 int l2 = strlen (path);
984 1612
985#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) 1613 char *res = etp_tmpbuf_get (tmpbuf, l1 + l2 + 2);
986# define msync(a,b,c) ENOSYS
987#endif
988 1614
989int 1615 memcpy (res, wd->str, l1);
990eio__mtouch (void *mem, size_t len, int flags) 1616 res [l1] = '/';
991{ 1617 memcpy (res + l1 + 1, path, l2 + 1);
992 intptr_t addr = (intptr_t)mem;
993 intptr_t end = addr + len;
994#ifdef PAGESIZE
995 const intptr_t page = PAGESIZE;
996#else
997 static intptr_t page;
998 1618
999 if (!page) 1619 return res;
1000 page = sysconf (_SC_PAGESIZE); 1620 }
1001#endif 1621}
1002 1622
1003 addr &= ~(page - 1); /* assume page size is always a power of two */ 1623#endif
1004 1624
1005 if (addr < end) 1625static eio_wd
1006 if (flags) /* modify */ 1626eio__wd_open_sync (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
1007 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len); 1627{
1008 else 1628 int fd;
1009 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len); 1629 eio_wd res;
1630 int len = eio__realpath (tmpbuf, wd, path);
1010 1631
1632 if (len < 0)
1633 return EIO_INVALID_WD;
1634
1635#if HAVE_AT
1636 fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1637
1638 if (fd < 0)
1639 return EIO_INVALID_WD;
1640#endif
1641
1642 res = malloc (sizeof (*res) + len); /* one extra 0-byte */
1643
1644#if HAVE_AT
1645 res->fd = fd;
1646#endif
1647
1648 res->len = len;
1649 memcpy (res->str, tmpbuf->ptr, len);
1650 res->str [len] = 0;
1651
1652 return res;
1653}
1654
1655eio_wd
1656eio_wd_open_sync (eio_wd wd, const char *path)
1657{
1658 struct etp_tmpbuf tmpbuf = { };
1659 wd = eio__wd_open_sync (&tmpbuf, wd, path);
1660 free (tmpbuf.ptr);
1661
1011 return 0; 1662 return wd;
1012} 1663}
1664
1665void
1666eio_wd_close_sync (eio_wd wd)
1667{
1668 if (wd != EIO_INVALID_WD && wd != EIO_CWD)
1669 {
1670 #if HAVE_AT
1671 close (wd->fd);
1672 #endif
1673 free (wd);
1674 }
1675}
1676
1677#if HAVE_AT
1678
1679/* they forgot these */
1680
1681static int
1682eio__truncateat (int dirfd, const char *path, off_t length)
1683{
1684 int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
1685 int res;
1686
1687 if (fd < 0)
1688 return fd;
1689
1690 res = ftruncate (fd, length);
1691 close (fd);
1692 return res;
1693}
1694
1695static int
1696eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
1697{
1698 int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
1699 int res;
1700
1701 if (fd < 0)
1702 return fd;
1703
1704 res = fstatvfs (fd, buf);
1705 close (fd);
1706 return res;
1707
1708}
1709
1710#endif
1013 1711
1014/*****************************************************************************/ 1712/*****************************************************************************/
1015 1713
1016#define ALLOC(len) \ 1714#define ALLOC(len) \
1017 if (!req->ptr2) \ 1715 if (!req->ptr2) \
1018 { \ 1716 { \
1019 X_LOCK (wrklock); \ 1717 X_LOCK (EIO_POOL->wrklock); \
1020 req->flags |= EIO_FLAG_PTR2_FREE; \ 1718 req->flags |= EIO_FLAG_PTR2_FREE; \
1021 X_UNLOCK (wrklock); \ 1719 X_UNLOCK (EIO_POOL->wrklock); \
1022 req->ptr2 = malloc (len); \ 1720 req->ptr2 = malloc (len); \
1023 if (!req->ptr2) \ 1721 if (!req->ptr2) \
1024 { \ 1722 { \
1025 errno = ENOMEM; \ 1723 errno = ENOMEM; \
1026 req->result = -1; \ 1724 req->result = -1; \
1027 break; \ 1725 break; \
1028 } \ 1726 } \
1029 } 1727 }
1030 1728
1031X_THREAD_PROC (etp_proc)
1032{
1033 ETP_REQ *req;
1034 struct timespec ts;
1035 etp_worker *self = (etp_worker *)thr_arg;
1036
1037 /* try to distribute timeouts somewhat randomly */
1038 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1039
1040 for (;;)
1041 {
1042 X_LOCK (reqlock);
1043
1044 for (;;)
1045 {
1046 self->req = req = reqq_shift (&req_queue);
1047
1048 if (req)
1049 break;
1050
1051 ++idle;
1052
1053 ts.tv_sec = time (0) + IDLE_TIMEOUT;
1054 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
1055 {
1056 if (idle > max_idle)
1057 {
1058 --idle;
1059 X_UNLOCK (reqlock);
1060 X_LOCK (wrklock);
1061 --started;
1062 X_UNLOCK (wrklock);
1063 goto quit;
1064 }
1065
1066 /* we are allowed to idle, so do so without any timeout */
1067 X_COND_WAIT (reqwait, reqlock);
1068 }
1069
1070 --idle;
1071 }
1072
1073 --nready;
1074
1075 X_UNLOCK (reqlock);
1076
1077 if (req->type < 0)
1078 goto quit;
1079
1080 if (!EIO_CANCELLED (req))
1081 ETP_EXECUTE (self, req);
1082
1083 X_LOCK (reslock);
1084
1085 ++npending;
1086
1087 if (!reqq_push (&res_queue, req) && want_poll_cb)
1088 want_poll_cb ();
1089
1090 self->req = 0;
1091 etp_worker_clear (self);
1092
1093 X_UNLOCK (reslock);
1094 }
1095
1096quit:
1097 X_LOCK (wrklock);
1098 etp_worker_free (self);
1099 X_UNLOCK (wrklock);
1100
1101 return 0;
1102}
1103
1104/*****************************************************************************/ 1729/*****************************************************************************/
1105 1730
1731int ecb_cold
1106int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1732eio_init (void (*want_poll)(void), void (*done_poll)(void))
1107{ 1733{
1108 return etp_init (want_poll, done_poll); 1734 eio_want_poll_cb = want_poll;
1109} 1735 eio_done_poll_cb = done_poll;
1110 1736
1737 return etp_init (EIO_POOL, 0, 0, 0);
1738}
1739
1740ecb_inline void
1111static void eio_api_destroy (eio_req *req) 1741eio_api_destroy (eio_req *req)
1112{ 1742{
1113 free (req); 1743 free (req);
1114} 1744}
1115 1745
1116#define REQ(rtype) \ 1746#define REQ(rtype) \
1117 eio_req *req; \ 1747 eio_req *req; \
1118 \ 1748 \
1119 req = (eio_req *)calloc (1, sizeof *req); \ 1749 req = (eio_req *)calloc (1, sizeof *req); \
1120 if (!req) \ 1750 if (!req) \
1121 return 0; \ 1751 return 0; \
… …
1135 { \ 1765 { \
1136 eio_api_destroy (req); \ 1766 eio_api_destroy (req); \
1137 return 0; \ 1767 return 0; \
1138 } 1768 }
1139 1769
1770#define SINGLEDOT(ptr) (0[(char *)(ptr)] == '.' && !1[(char *)(ptr)])
1771
1772static void
1140static void eio_execute (etp_worker *self, eio_req *req) 1773eio_execute (etp_worker *self, eio_req *req)
1141{ 1774{
1142 errno = 0; 1775#if HAVE_AT
1776 int dirfd;
1777#else
1778 const char *path;
1779#endif
1780
1781 if (ecb_expect_false (EIO_CANCELLED (req)))
1782 {
1783 req->result = -1;
1784 req->errorno = ECANCELED;
1785 return;
1786 }
1787
1788 if (ecb_expect_false (req->wd == EIO_INVALID_WD))
1789 {
1790 req->result = -1;
1791 req->errorno = ENOENT;
1792 return;
1793 }
1794
1795 if (req->type >= EIO_OPEN)
1796 {
1797 #if HAVE_AT
1798 dirfd = WD2FD (req->wd);
1799 #else
1800 path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1801 #endif
1802 }
1143 1803
1144 switch (req->type) 1804 switch (req->type)
1145 { 1805 {
1806 case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
1807 req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
1808 break;
1809 case EIO_WD_CLOSE: req->result = 0;
1810 eio_wd_close_sync (req->wd); break;
1811
1812 case EIO_SEEK: eio__lseek (req); break;
1146 case EIO_READ: ALLOC (req->size); 1813 case EIO_READ: ALLOC (req->size);
1147 req->result = req->offs >= 0 1814 req->result = req->offs >= 0
1148 ? pread (req->int1, req->ptr2, req->size, req->offs) 1815 ? pread (req->int1, req->ptr2, req->size, req->offs)
1149 : read (req->int1, req->ptr2, req->size); break; 1816 : read (req->int1, req->ptr2, req->size); break;
1150 case EIO_WRITE: req->result = req->offs >= 0 1817 case EIO_WRITE: req->result = req->offs >= 0
1151 ? pwrite (req->int1, req->ptr2, req->size, req->offs) 1818 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1152 : write (req->int1, req->ptr2, req->size); break; 1819 : write (req->int1, req->ptr2, req->size); break;
1153 1820
1154 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break; 1821 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1155 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break; 1822 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
1823
1824#if HAVE_AT
1156 1825
1157 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1826 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1158 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 1827 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break;
1159 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1828 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1160 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 1829 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, AT_SYMLINK_NOFOLLOW); break;
1161 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1162 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1163
1164 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; 1830 case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break;
1165 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1166 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break; 1831 case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break;
1167 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1168 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break; 1832 case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break;
1169 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1170
1171 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break; 1833 case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break;
1172 case EIO_CLOSE: req->result = close (req->int1); break; 1834
1173 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1174 case EIO_UNLINK: req->result = unlink (req->ptr1); break; 1835 case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
1175 case EIO_RMDIR: req->result = rmdir (req->ptr1); break; 1836 case EIO_RMDIR: /* complications arise because "." cannot be removed, so we might have to expand */
1837 req->result = req->wd && SINGLEDOT (req->ptr1)
1838 ? rmdir (req->wd->str)
1839 : unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break;
1176 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; 1840 case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); break;
1177 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; 1841 case EIO_RENAME: /* complications arise because "." cannot be renamed, so we might have to expand */
1842 req->result = req->wd && SINGLEDOT (req->ptr1)
1843 ? rename (req->wd->str, req->ptr2)
1844 : renameat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2); break;
1178 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; 1845 case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2, 0); break;
1179 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; 1846 case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
1180 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break; 1847 case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1181
1182 case EIO_READLINK: ALLOC (NAME_MAX); 1848 case EIO_READLINK: ALLOC (PATH_MAX);
1183 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; 1849 req->result = readlinkat (dirfd, req->ptr1, req->ptr2, PATH_MAX); break;
1184 1850 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1185 case EIO_SYNC: req->result = 0; sync (); break; 1851 req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1186 case EIO_FSYNC: req->result = fsync (req->int1); break;
1187 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1188 case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break;
1189 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;
1190
1191 case EIO_READDIR: eio__scandir (req, self); break;
1192
1193 case EIO_BUSY: 1852 case EIO_UTIME:
1194#ifdef _WIN32 1853 case EIO_FUTIME:
1195 Sleep (req->nv1 * 1000.); 1854 {
1855 struct timespec ts[2];
1856 struct timespec *times;
1857
1858 if (req->nv1 != -1. || req->nv2 != -1.)
1859 {
1860 ts[0].tv_sec = req->nv1;
1861 ts[0].tv_nsec = (req->nv1 - ts[0].tv_sec) * 1e9;
1862 ts[1].tv_sec = req->nv2;
1863 ts[1].tv_nsec = (req->nv2 - ts[1].tv_sec) * 1e9;
1864
1865 times = ts;
1866 }
1867 else
1868 times = 0;
1869
1870 req->result = req->type == EIO_FUTIME
1871 ? futimens (req->int1, times)
1872 : utimensat (dirfd, req->ptr1, times, 0);
1873 }
1874 break;
1875
1196#else 1876#else
1197 {
1198 struct timeval tv;
1199 1877
1200 tv.tv_sec = req->nv1; 1878 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1201 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; 1879 req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
1880 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1881 req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
1882 case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
1883 case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
1884 case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
1885 case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
1202 1886
1203 req->result = select (0, 0, 0, 0, &tv); 1887 case EIO_UNLINK: req->result = unlink (path ); break;
1204 } 1888 case EIO_RMDIR: req->result = rmdir (path ); break;
1205#endif 1889 case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
1206 break; 1890 case EIO_RENAME: req->result = rename (path , req->ptr2); break;
1891 case EIO_LINK: req->result = link (path , req->ptr2); break;
1892 case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
1893 case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
1894 case EIO_READLINK: ALLOC (PATH_MAX);
1895 req->result = readlink (path, req->ptr2, PATH_MAX); break;
1896 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1897 req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
1207 1898
1208 case EIO_UTIME: 1899 case EIO_UTIME:
1209 case EIO_FUTIME: 1900 case EIO_FUTIME:
1210 { 1901 {
1211 struct timeval tv[2]; 1902 struct timeval tv[2];
1212 struct timeval *times; 1903 struct timeval *times;
1213 1904
1214 if (req->nv1 != -1. || req->nv2 != -1.) 1905 if (req->nv1 != -1. || req->nv2 != -1.)
1215 { 1906 {
1216 tv[0].tv_sec = req->nv1; 1907 tv[0].tv_sec = req->nv1;
1217 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.; 1908 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
1218 tv[1].tv_sec = req->nv2; 1909 tv[1].tv_sec = req->nv2;
1219 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.; 1910 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
1220 1911
1221 times = tv; 1912 times = tv;
1222 } 1913 }
1223 else 1914 else
1224 times = 0; 1915 times = 0;
1225
1226 1916
1227 req->result = req->type == EIO_FUTIME 1917 req->result = req->type == EIO_FUTIME
1228 ? futimes (req->int1, times) 1918 ? futimes (req->int1, times)
1229 : utimes (req->ptr1, times); 1919 : utimes (req->ptr1, times);
1230 } 1920 }
1231 break; 1921 break;
1232 1922
1923#endif
1924
1925 case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
1926 {
1927 ALLOC (req->result);
1928 memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
1929 }
1930 break;
1931
1932 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1933 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1934
1935 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1936 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1937
1938 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1939 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1940 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1941
1942 case EIO_CLOSE: req->result = close (req->int1); break;
1943 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1944 case EIO_SYNC: req->result = 0; sync (); break;
1945 case EIO_FSYNC: req->result = fsync (req->int1); break;
1946 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1947 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
1948 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1949 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1950 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1951 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1952 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1953 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
1954
1955 case EIO_READDIR: eio__scandir (req, self); break;
1956
1957 case EIO_BUSY:
1958#ifdef _WIN32
1959 Sleep (req->nv1 * 1e3);
1960#else
1961 {
1962 struct timeval tv;
1963
1964 tv.tv_sec = req->nv1;
1965 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1966
1967 req->result = select (0, 0, 0, 0, &tv);
1968 }
1969#endif
1970 break;
1971
1972#if 0
1233 case EIO_GROUP: 1973 case EIO_GROUP:
1234 abort (); /* handled in eio_request */ 1974 abort (); /* handled in eio_request */
1975#endif
1235 1976
1236 case EIO_NOP: 1977 case EIO_NOP:
1237 req->result = 0; 1978 req->result = 0;
1238 break; 1979 break;
1239 1980
1240 case EIO_CUSTOM: 1981 case EIO_CUSTOM:
1241 ((void (*)(eio_req *))req->feed) (req); 1982 req->feed (req);
1242 break; 1983 break;
1243 1984
1244 default: 1985 default:
1245 req->result = -1; 1986 req->result = EIO_ENOSYS ();
1246 break; 1987 break;
1247 } 1988 }
1248 1989
1249 req->errorno = errno; 1990 req->errorno = errno;
1250} 1991}
1251 1992
1252#ifndef EIO_NO_WRAPPERS 1993#ifndef EIO_NO_WRAPPERS
1253 1994
1995eio_req *eio_wd_open (const char *path, int pri, eio_cb cb, void *data)
1996{
1997 REQ (EIO_WD_OPEN); PATH; SEND;
1998}
1999
2000eio_req *eio_wd_close (eio_wd wd, int pri, eio_cb cb, void *data)
2001{
2002 REQ (EIO_WD_CLOSE); req->wd = wd; SEND;
2003}
2004
1254eio_req *eio_nop (int pri, eio_cb cb, void *data) 2005eio_req *eio_nop (int pri, eio_cb cb, void *data)
1255{ 2006{
1256 REQ (EIO_NOP); SEND; 2007 REQ (EIO_NOP); SEND;
1257} 2008}
1258 2009
… …
1274eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2025eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1275{ 2026{
1276 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2027 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1277} 2028}
1278 2029
2030eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2031{
2032 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2033}
2034
2035eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
2036{
2037 REQ (EIO_SYNCFS); req->int1 = fd; SEND;
2038}
2039
2040eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2041{
2042 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2043}
2044
1279eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2045eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1280{ 2046{
1281 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2047 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1282} 2048}
1283 2049
2050eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
2051{
2052 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
2053}
2054
1284eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) 2055eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
1285{ 2056{
1286 REQ (EIO_FDATASYNC); req->int1 = fd; SEND; 2057 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2058}
2059
2060eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2061{
2062 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
1287} 2063}
1288 2064
1289eio_req *eio_close (int fd, int pri, eio_cb cb, void *data) 2065eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
1290{ 2066{
1291 REQ (EIO_CLOSE); req->int1 = fd; SEND; 2067 REQ (EIO_CLOSE); req->int1 = fd; SEND;
… …
1294eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data) 2070eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
1295{ 2071{
1296 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND; 2072 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
1297} 2073}
1298 2074
2075eio_req *eio_seek (int fd, off_t offset, int whence, int pri, eio_cb cb, void *data)
2076{
2077 REQ (EIO_SEEK); req->int1 = fd; req->offs = offset; req->int2 = whence; SEND;
2078}
2079
1299eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data) 2080eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1300{ 2081{
1301 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND; 2082 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1302} 2083}
1303 2084
… …
1309eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) 2090eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1310{ 2091{
1311 REQ (EIO_FSTAT); req->int1 = fd; SEND; 2092 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1312} 2093}
1313 2094
2095eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
2096{
2097 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
2098}
2099
1314eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) 2100eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1315{ 2101{
1316 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; 2102 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1317} 2103}
1318 2104
… …
1324eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) 2110eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1325{ 2111{
1326 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2112 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1327} 2113}
1328 2114
1329eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2115eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1330{ 2116{
1331 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2117 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1332} 2118}
1333 2119
1334eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) 2120eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
… …
1354eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data) 2140eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1355{ 2141{
1356 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2142 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1357} 2143}
1358 2144
1359eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2145eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1360{ 2146{
1361 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2147 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1362} 2148}
1363 2149
1364eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) 2150eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
… …
1380eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data) 2166eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1381{ 2167{
1382 return eio__1path (EIO_READLINK, path, pri, cb, data); 2168 return eio__1path (EIO_READLINK, path, pri, cb, data);
1383} 2169}
1384 2170
2171eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2172{
2173 return eio__1path (EIO_REALPATH, path, pri, cb, data);
2174}
2175
1385eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) 2176eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1386{ 2177{
1387 return eio__1path (EIO_STAT, path, pri, cb, data); 2178 return eio__1path (EIO_STAT, path, pri, cb, data);
1388} 2179}
1389 2180
1390eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) 2181eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1391{ 2182{
1392 return eio__1path (EIO_LSTAT, path, pri, cb, data); 2183 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1393} 2184}
1394 2185
2186eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
2187{
2188 return eio__1path (EIO_STATVFS, path, pri, cb, data);
2189}
2190
1395eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) 2191eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1396{ 2192{
1397 return eio__1path (EIO_UNLINK, path, pri, cb, data); 2193 return eio__1path (EIO_UNLINK, path, pri, cb, data);
1398} 2194}
1399 2195
1400eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data) 2196eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1401{ 2197{
1402 return eio__1path (EIO_RMDIR, path, pri, cb, data); 2198 return eio__1path (EIO_RMDIR, path, pri, cb, data);
1403} 2199}
1404 2200
1405eio_req *eio_readdir (const char *path, int pri, eio_cb cb, void *data) 2201eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1406{ 2202{
1407 return eio__1path (EIO_READDIR, path, pri, cb, data); 2203 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1408} 2204}
1409 2205
1410eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) 2206eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1411{ 2207{
1412 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; 2208 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1413} 2209}
1414 2210
1415static eio_req * 2211static eio_req *
1416eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2212eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1417{ 2213{
… …
1441eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2237eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1442{ 2238{
1443 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2239 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1444} 2240}
1445 2241
1446eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) 2242eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
1447{ 2243{
1448 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND; 2244 REQ (EIO_CUSTOM); req->feed = execute; SEND;
1449} 2245}
1450 2246
1451#endif 2247#endif
1452 2248
1453eio_req *eio_grp (eio_cb cb, void *data) 2249eio_req *eio_grp (eio_cb cb, void *data)
… …
1462#undef SEND 2258#undef SEND
1463 2259
1464/*****************************************************************************/ 2260/*****************************************************************************/
1465/* grp functions */ 2261/* grp functions */
1466 2262
2263void
1467void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2264eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1468{ 2265{
1469 grp->int2 = limit; 2266 grp->int2 = limit;
1470 grp->feed = feed; 2267 grp->feed = feed;
1471 2268
1472 grp_try_feed (grp); 2269 grp_try_feed (grp);
1473} 2270}
1474 2271
2272void
1475void eio_grp_limit (eio_req *grp, int limit) 2273eio_grp_limit (eio_req *grp, int limit)
1476{ 2274{
1477 grp->int2 = limit; 2275 grp->int2 = limit;
1478 2276
1479 grp_try_feed (grp); 2277 grp_try_feed (grp);
1480} 2278}
1481 2279
2280void
1482void eio_grp_add (eio_req *grp, eio_req *req) 2281eio_grp_add (eio_req *grp, eio_req *req)
1483{ 2282{
1484 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2283 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1485 2284
1486 grp->flags |= EIO_FLAG_GROUPADD; 2285 grp->flags |= ETP_FLAG_GROUPADD;
1487 2286
1488 ++grp->size; 2287 ++grp->size;
1489 req->grp = grp; 2288 req->grp = grp;
1490 2289
1491 req->grp_prev = 0; 2290 req->grp_prev = 0;
… …
1498} 2297}
1499 2298
1500/*****************************************************************************/ 2299/*****************************************************************************/
1501/* misc garbage */ 2300/* misc garbage */
1502 2301
2302eio_ssize_t
1503ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2303eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1504{ 2304{
1505 etp_worker wrk;
1506
1507 wrk.dbuf = 0;
1508
1509 eio__sendfile (ofd, ifd, offset, count, &wrk); 2305 return eio__sendfile (ofd, ifd, offset, count);
1510
1511 if (wrk.dbuf)
1512 free (wrk.dbuf);
1513} 2306}
1514 2307

Diff Legend

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