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

Comparing libeio/eio.c (file contents):
Revision 1.54 by root, Tue Feb 23 14:32:45 2010 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) 2007,2008,2009,2010 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"
41 46
42#ifdef EIO_STACKSIZE 47#ifdef EIO_STACKSIZE
43# define XTHREAD_STACKSIZE EIO_STACKSIZE 48# define X_STACKSIZE EIO_STACKSIZE
44#endif 49#endif
45#include "xthread.h" 50#include "xthread.h"
46 51
47#include <errno.h> 52#include <errno.h>
48#include <stddef.h> 53#include <stddef.h>
49#include <stdlib.h> 54#include <stdlib.h>
50#include <string.h> 55#include <string.h>
51#include <errno.h> 56#include <errno.h>
52#include <sys/types.h> 57#include <sys/types.h>
53#include <sys/stat.h> 58#include <sys/stat.h>
54#include <sys/statvfs.h>
55#include <limits.h> 59#include <limits.h>
56#include <fcntl.h> 60#include <fcntl.h>
57#include <assert.h> 61#include <assert.h>
58 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
59#ifndef EIO_FINISH 82#ifndef EIO_FINISH
60# 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
61#endif 84#endif
62 85
63#ifndef EIO_DESTROY 86#ifndef EIO_DESTROY
… …
66 89
67#ifndef EIO_FEED 90#ifndef EIO_FEED
68# 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)
69#endif 92#endif
70 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
71#ifdef _WIN32 105#ifdef _WIN32
72 106
73 /*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)
232
74#else 233#else
75 234
76# include "config.h"
77# include <sys/time.h> 235 #include <sys/time.h>
78# include <sys/select.h> 236 #include <sys/select.h>
79# include <sys/mman.h>
80# 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
81# include <utime.h> 281# 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 282#endif
93 283
94#ifdef _D_EXACT_NAMLEN 284#if HAVE_SYS_SYSCALL_H
95# undef D_NAMLEN 285# 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 286#endif
102 287
103# ifndef EIO_STRUCT_DIRENT 288#if HAVE_SYS_PRCTL_H
104# define EIO_STRUCT_DIRENT struct dirent 289# include <sys/prctl.h>
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>
… …
126#endif 309#endif
127#ifndef D_INO 310#ifndef D_INO
128# define D_INO(de) 0 311# define D_INO(de) 0
129#endif 312#endif
130#ifndef D_NAMLEN 313#ifndef D_NAMLEN
131# define D_NAMLEN(de) strlen ((de)->d_name) 314# define D_NAMLEN(entp) strlen (D_NAME (entp))
132#endif 315#endif
133
134/* number of seconds after which an idle threads exit */
135#define IDLE_TIMEOUT 10
136 316
137/* used for struct dirent, AIX doesn't provide it */ 317/* used for struct dirent, AIX doesn't provide it */
138#ifndef NAME_MAX 318#ifndef NAME_MAX
139# define NAME_MAX 4096 319# define NAME_MAX 4096
140#endif 320#endif
… …
145#endif 325#endif
146 326
147/* buffer size for various temporary buffers */ 327/* buffer size for various temporary buffers */
148#define EIO_BUFSIZE 65536 328#define EIO_BUFSIZE 65536
149 329
150#define dBUF \ 330#define dBUF \
151 char *eio_buf; \
152 ETP_WORKER_LOCK (self); \
153 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 331 char *eio_buf = malloc (EIO_BUFSIZE); \
154 ETP_WORKER_UNLOCK (self); \
155 errno = ENOMEM; \ 332 errno = ENOMEM; \
156 if (!eio_buf) \ 333 if (!eio_buf) \
157 return -1; 334 return -1
158 335
159#define EIO_TICKS ((1000000 + 1023) >> 10) 336#define FUBd \
337 free (eio_buf)
160 338
161/*****************************************************************************/ 339/*****************************************************************************/
162 340
163#if __GNUC__ >= 3 341struct etp_tmpbuf;
164# 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
165#else 349#else
166# 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);
167#endif 352#endif
168 353
169#define expect_false(expr) expect ((expr) != 0, 0) 354struct eio_pwd
170#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};
171 362
172/*****************************************************************************/ 363/*****************************************************************************/
173 364
174#define ETP_PRI_MIN EIO_PRI_MIN 365#define ETP_PRI_MIN EIO_PRI_MIN
175#define ETP_PRI_MAX EIO_PRI_MAX 366#define ETP_PRI_MAX EIO_PRI_MAX
176 367
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
177struct etp_worker; 378struct etp_worker;
178
179#define ETP_REQ eio_req 379#define ETP_REQ eio_req
180#define ETP_DESTROY(req) eio_destroy (req) 380#define ETP_DESTROY(req) eio_destroy (req)
181static int eio_finish (eio_req *req); 381static int eio_finish (eio_req *req);
182#define ETP_FINISH(req) eio_finish (req) 382#define ETP_FINISH(req) eio_finish (req)
183static void eio_execute (struct etp_worker *self, eio_req *req); 383static void eio_execute (struct etp_worker *self, eio_req *req);
184#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) 384#define ETP_EXECUTE(wrk,req) eio_execute (wrk, req)
185 385
186#define ETP_WORKER_CLEAR(req) \ 386#include "etp.c"
187 if (wrk->dbuf) \
188 { \
189 free (wrk->dbuf); \
190 wrk->dbuf = 0; \
191 } \
192 \
193 if (wrk->dirp) \
194 { \
195 closedir (wrk->dirp); \
196 wrk->dirp = 0; \
197 }
198 387
199#define ETP_WORKER_COMMON \ 388static struct etp_pool eio_pool;
200 void *dbuf; \ 389#define EIO_POOL (&eio_pool)
201 DIR *dirp;
202 390
203/*****************************************************************************/ 391/*****************************************************************************/
204 392
205#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
206
207/* calculate time difference in ~1/EIO_TICKS of a second */
208static int tvdiff (struct timeval *tv1, struct timeval *tv2)
209{
210 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
211 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
212}
213
214static unsigned int started, idle, wanted = 4;
215
216static void (*want_poll_cb) (void);
217static void (*done_poll_cb) (void);
218
219static unsigned int max_poll_time; /* reslock */
220static unsigned int max_poll_reqs; /* reslock */
221
222static volatile unsigned int nreqs; /* reqlock */
223static volatile unsigned int nready; /* reqlock */
224static volatile unsigned int npending; /* reqlock */
225static volatile unsigned int max_idle = 4;
226
227static mutex_t wrklock = X_MUTEX_INIT;
228static mutex_t reslock = X_MUTEX_INIT;
229static mutex_t reqlock = X_MUTEX_INIT;
230static cond_t reqwait = X_COND_INIT;
231
232#if !HAVE_PREADWRITE
233/*
234 * make our pread/pwrite emulation safe against themselves, but not against
235 * normal read/write by using a mutex. slows down execution a lot,
236 * but that's your problem, not mine.
237 */
238static mutex_t preadwritelock = X_MUTEX_INIT;
239#endif
240
241typedef struct etp_worker
242{
243 /* locked by wrklock */
244 struct etp_worker *prev, *next;
245
246 thread_t tid;
247
248 /* locked by reslock, reqlock or wrklock */
249 ETP_REQ *req; /* currently processed request */
250
251 ETP_WORKER_COMMON
252} etp_worker;
253
254static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */
255
256#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
257#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
258
259/* worker threads management */
260
261static void etp_worker_clear (etp_worker *wrk)
262{
263 ETP_WORKER_CLEAR (wrk);
264}
265
266static void etp_worker_free (etp_worker *wrk)
267{
268 wrk->next->prev = wrk->prev;
269 wrk->prev->next = wrk->next;
270
271 free (wrk);
272}
273
274static unsigned int etp_nreqs (void)
275{
276 int retval;
277 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
278 retval = nreqs;
279 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
280 return retval;
281}
282
283static unsigned int etp_nready (void)
284{
285 unsigned int retval;
286
287 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
288 retval = nready;
289 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
290
291 return retval;
292}
293
294static unsigned int etp_npending (void)
295{
296 unsigned int retval;
297
298 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
299 retval = npending;
300 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
301
302 return retval;
303}
304
305static unsigned int etp_nthreads (void)
306{
307 unsigned int retval;
308
309 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
310 retval = started;
311 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
312
313 return retval;
314}
315
316/*
317 * a somewhat faster data structure might be nice, but
318 * with 8 priorities this actually needs <20 insns
319 * per shift, the most expensive operation.
320 */
321typedef struct {
322 ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */
323 int size;
324} etp_reqq;
325
326static etp_reqq req_queue;
327static etp_reqq res_queue;
328
329static int reqq_push (etp_reqq *q, ETP_REQ *req)
330{
331 int pri = req->pri;
332 req->next = 0;
333
334 if (q->qe[pri])
335 {
336 q->qe[pri]->next = req;
337 q->qe[pri] = req;
338 }
339 else
340 q->qe[pri] = q->qs[pri] = req;
341
342 return q->size++;
343}
344
345static ETP_REQ *reqq_shift (etp_reqq *q)
346{
347 int pri;
348
349 if (!q->size)
350 return 0;
351
352 --q->size;
353
354 for (pri = ETP_NUM_PRI; pri--; )
355 {
356 eio_req *req = q->qs[pri];
357
358 if (req)
359 {
360 if (!(q->qs[pri] = (eio_req *)req->next))
361 q->qe[pri] = 0;
362
363 return req;
364 }
365 }
366
367 abort ();
368}
369
370static void etp_atfork_prepare (void)
371{
372 X_LOCK (wrklock);
373 X_LOCK (reqlock);
374 X_LOCK (reslock);
375#if !HAVE_PREADWRITE
376 X_LOCK (preadwritelock);
377#endif
378}
379
380static void etp_atfork_parent (void)
381{
382#if !HAVE_PREADWRITE
383 X_UNLOCK (preadwritelock);
384#endif
385 X_UNLOCK (reslock);
386 X_UNLOCK (reqlock);
387 X_UNLOCK (wrklock);
388}
389
390static void etp_atfork_child (void)
391{
392 ETP_REQ *prv;
393
394 while ((prv = reqq_shift (&req_queue)))
395 ETP_DESTROY (prv);
396
397 while ((prv = reqq_shift (&res_queue)))
398 ETP_DESTROY (prv);
399
400 while (wrk_first.next != &wrk_first)
401 {
402 etp_worker *wrk = wrk_first.next;
403
404 if (wrk->req)
405 ETP_DESTROY (wrk->req);
406
407 etp_worker_clear (wrk);
408 etp_worker_free (wrk);
409 }
410
411 started = 0;
412 idle = 0;
413 nreqs = 0;
414 nready = 0;
415 npending = 0;
416
417 etp_atfork_parent ();
418}
419
420static void 393static void
421etp_once_init (void)
422{
423 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
424}
425
426static int
427etp_init (void (*want_poll)(void), void (*done_poll)(void))
428{
429 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
430
431 pthread_once (&doinit, etp_once_init);
432
433 want_poll_cb = want_poll;
434 done_poll_cb = done_poll;
435
436 return 0;
437}
438
439X_THREAD_PROC (etp_proc);
440
441static void etp_start_thread (void)
442{
443 etp_worker *wrk = calloc (1, sizeof (etp_worker));
444
445 /*TODO*/
446 assert (("unable to allocate worker thread data", wrk));
447
448 X_LOCK (wrklock);
449
450 if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
451 {
452 wrk->prev = &wrk_first;
453 wrk->next = wrk_first.next;
454 wrk_first.next->prev = wrk;
455 wrk_first.next = wrk;
456 ++started;
457 }
458 else
459 free (wrk);
460
461 X_UNLOCK (wrklock);
462}
463
464static void etp_maybe_start_thread (void)
465{
466 if (expect_true (etp_nthreads () >= wanted))
467 return;
468
469 /* todo: maybe use idle here, but might be less exact */
470 if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
471 return;
472
473 etp_start_thread ();
474}
475
476static void etp_end_thread (void)
477{
478 eio_req *req = calloc (1, sizeof (eio_req));
479
480 req->type = -1;
481 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
482
483 X_LOCK (reqlock);
484 reqq_push (&req_queue, req);
485 X_COND_SIGNAL (reqwait);
486 X_UNLOCK (reqlock);
487
488 X_LOCK (wrklock);
489 --started;
490 X_UNLOCK (wrklock);
491}
492
493static int etp_poll (void)
494{
495 unsigned int maxreqs;
496 unsigned int maxtime;
497 struct timeval tv_start, tv_now;
498
499 X_LOCK (reslock);
500 maxreqs = max_poll_reqs;
501 maxtime = max_poll_time;
502 X_UNLOCK (reslock);
503
504 if (maxtime)
505 gettimeofday (&tv_start, 0);
506
507 for (;;)
508 {
509 ETP_REQ *req;
510
511 etp_maybe_start_thread ();
512
513 X_LOCK (reslock);
514 req = reqq_shift (&res_queue);
515
516 if (req)
517 {
518 --npending;
519
520 if (!res_queue.size && done_poll_cb)
521 done_poll_cb ();
522 }
523
524 X_UNLOCK (reslock);
525
526 if (!req)
527 return 0;
528
529 X_LOCK (reqlock);
530 --nreqs;
531 X_UNLOCK (reqlock);
532
533 if (expect_false (req->type == EIO_GROUP && req->size))
534 {
535 req->int1 = 1; /* mark request as delayed */
536 continue;
537 }
538 else
539 {
540 int res = ETP_FINISH (req);
541 if (expect_false (res))
542 return res;
543 }
544
545 if (expect_false (maxreqs && !--maxreqs))
546 break;
547
548 if (maxtime)
549 {
550 gettimeofday (&tv_now, 0);
551
552 if (tvdiff (&tv_start, &tv_now) >= maxtime)
553 break;
554 }
555 }
556
557 errno = EAGAIN;
558 return -1;
559}
560
561static void etp_cancel (ETP_REQ *req)
562{
563 X_LOCK (wrklock);
564 req->flags |= EIO_FLAG_CANCELLED;
565 X_UNLOCK (wrklock);
566
567 eio_grp_cancel (req);
568}
569
570static void etp_submit (ETP_REQ *req)
571{
572 req->pri -= ETP_PRI_MIN;
573
574 if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
575 if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
576
577 if (expect_false (req->type == EIO_GROUP))
578 {
579 /* I hope this is worth it :/ */
580 X_LOCK (reqlock);
581 ++nreqs;
582 X_UNLOCK (reqlock);
583
584 X_LOCK (reslock);
585
586 ++npending;
587
588 if (!reqq_push (&res_queue, req) && want_poll_cb)
589 want_poll_cb ();
590
591 X_UNLOCK (reslock);
592 }
593 else
594 {
595 X_LOCK (reqlock);
596 ++nreqs;
597 ++nready;
598 reqq_push (&req_queue, req);
599 X_COND_SIGNAL (reqwait);
600 X_UNLOCK (reqlock);
601
602 etp_maybe_start_thread ();
603 }
604}
605
606static void etp_set_max_poll_time (double nseconds)
607{
608 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
609 max_poll_time = nseconds * EIO_TICKS;
610 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
611}
612
613static void etp_set_max_poll_reqs (unsigned int maxreqs)
614{
615 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
616 max_poll_reqs = maxreqs;
617 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
618}
619
620static void etp_set_max_idle (unsigned int nthreads)
621{
622 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
623 max_idle = nthreads <= 0 ? 1 : nthreads;
624 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
625}
626
627static void etp_set_min_parallel (unsigned int nthreads)
628{
629 if (wanted < nthreads)
630 wanted = nthreads;
631}
632
633static void etp_set_max_parallel (unsigned int nthreads)
634{
635 if (wanted > nthreads)
636 wanted = nthreads;
637
638 while (started > wanted)
639 etp_end_thread ();
640}
641
642/*****************************************************************************/
643
644static void grp_try_feed (eio_req *grp) 394grp_try_feed (eio_req *grp)
645{ 395{
646 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 396 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
647 { 397 {
648 grp->flags &= ~EIO_FLAG_GROUPADD; 398 grp->flags &= ~ETP_FLAG_GROUPADD;
649 399
650 EIO_FEED (grp); 400 EIO_FEED (grp);
651 401
652 /* stop if no progress has been made */ 402 /* stop if no progress has been made */
653 if (!(grp->flags & EIO_FLAG_GROUPADD)) 403 if (!(grp->flags & ETP_FLAG_GROUPADD))
654 { 404 {
655 grp->feed = 0; 405 grp->feed = 0;
656 break; 406 break;
657 } 407 }
658 } 408 }
659} 409}
660 410
411static int
661static int grp_dec (eio_req *grp) 412grp_dec (eio_req *grp)
662{ 413{
663 --grp->size; 414 --grp->size;
664 415
665 /* call feeder, if applicable */ 416 /* call feeder, if applicable */
666 grp_try_feed (grp); 417 grp_try_feed (grp);
667 418
668 /* finish, if done */ 419 /* finish, if done */
669 if (!grp->size && grp->int1) 420 if (!grp->size && grp->flags & ETP_FLAG_DELAYED)
670 return eio_finish (grp); 421 return eio_finish (grp);
671 else 422 else
672 return 0; 423 return 0;
673} 424}
674 425
426static void
675void eio_destroy (eio_req *req) 427eio_destroy (eio_req *req)
676{ 428{
677 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); 429 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
678 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); 430 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
679 431
680 EIO_DESTROY (req); 432 EIO_DESTROY (req);
681} 433}
682 434
435static int
683static int eio_finish (eio_req *req) 436eio_finish (eio_req *req)
684{ 437{
685 int res = EIO_FINISH (req); 438 int res = EIO_FINISH (req);
686 439
687 if (req->grp) 440 if (req->grp)
688 { 441 {
… …
696 if (grp->grp_first == req) 449 if (grp->grp_first == req)
697 grp->grp_first = req->grp_next; 450 grp->grp_first = req->grp_next;
698 451
699 res2 = grp_dec (grp); 452 res2 = grp_dec (grp);
700 453
701 if (!res && res2) 454 if (!res)
702 res = res2; 455 res = res2;
703 } 456 }
704 457
705 eio_destroy (req); 458 eio_destroy (req);
706 459
707 return res; 460 return res;
708} 461}
709 462
463void
710void eio_grp_cancel (eio_req *grp) 464eio_grp_cancel (eio_req *grp)
711{ 465{
712 for (grp = grp->grp_first; grp; grp = grp->grp_next) 466 etp_grp_cancel (EIO_POOL, grp);
713 eio_cancel (grp);
714} 467}
715 468
469void
716void eio_cancel (eio_req *req) 470eio_cancel (eio_req *req)
717{ 471{
718 etp_cancel (req); 472 etp_cancel (EIO_POOL, req);
719} 473}
720 474
475void
721void eio_submit (eio_req *req) 476eio_submit (eio_req *req)
722{ 477{
723 etp_submit (req); 478 etp_submit (EIO_POOL, req);
724} 479}
725 480
726unsigned int eio_nreqs (void) 481unsigned int
482eio_nreqs (void)
727{ 483{
728 return etp_nreqs (); 484 return etp_nreqs (EIO_POOL);
729} 485}
730 486
731unsigned int eio_nready (void) 487unsigned int
488eio_nready (void)
732{ 489{
733 return etp_nready (); 490 return etp_nready (EIO_POOL);
734} 491}
735 492
736unsigned int eio_npending (void) 493unsigned int
494eio_npending (void)
737{ 495{
738 return etp_npending (); 496 return etp_npending (EIO_POOL);
739} 497}
740 498
741unsigned int eio_nthreads (void) 499unsigned int ecb_cold
500eio_nthreads (void)
742{ 501{
743 return etp_nthreads (); 502 return etp_nthreads (EIO_POOL);
744} 503}
745 504
505void ecb_cold
746void eio_set_max_poll_time (double nseconds) 506eio_set_max_poll_time (double nseconds)
747{ 507{
748 etp_set_max_poll_time (nseconds); 508 etp_set_max_poll_time (EIO_POOL, nseconds);
749} 509}
750 510
511void ecb_cold
751void eio_set_max_poll_reqs (unsigned int maxreqs) 512eio_set_max_poll_reqs (unsigned int maxreqs)
752{ 513{
753 etp_set_max_poll_reqs (maxreqs); 514 etp_set_max_poll_reqs (EIO_POOL, maxreqs);
754} 515}
755 516
517void ecb_cold
756void eio_set_max_idle (unsigned int nthreads) 518eio_set_max_idle (unsigned int nthreads)
757{ 519{
758 etp_set_max_idle (nthreads); 520 etp_set_max_idle (EIO_POOL, nthreads);
759} 521}
760 522
523void ecb_cold
524eio_set_idle_timeout (unsigned int seconds)
525{
526 etp_set_idle_timeout (EIO_POOL, seconds);
527}
528
529void ecb_cold
761void eio_set_min_parallel (unsigned int nthreads) 530eio_set_min_parallel (unsigned int nthreads)
762{ 531{
763 etp_set_min_parallel (nthreads); 532 etp_set_min_parallel (EIO_POOL, nthreads);
764} 533}
765 534
535void ecb_cold
766void eio_set_max_parallel (unsigned int nthreads) 536eio_set_max_parallel (unsigned int nthreads)
767{ 537{
768 etp_set_max_parallel (nthreads); 538 etp_set_max_parallel (EIO_POOL, nthreads);
769} 539}
770 540
771int eio_poll (void) 541int eio_poll (void)
772{ 542{
773 return etp_poll (); 543 return etp_poll (EIO_POOL);
774} 544}
775 545
776/*****************************************************************************/ 546/*****************************************************************************/
777/* work around various missing functions */ 547/* work around various missing functions */
778 548
779#if !HAVE_PREADWRITE
780# undef pread
781# undef pwrite
782# define pread eio__pread
783# define pwrite eio__pwrite
784
785static ssize_t
786eio__pread (int fd, void *buf, size_t count, off_t offset)
787{
788 ssize_t res;
789 off_t ooffset;
790
791 X_LOCK (preadwritelock);
792 ooffset = lseek (fd, 0, SEEK_CUR);
793 lseek (fd, offset, SEEK_SET);
794 res = read (fd, buf, count);
795 lseek (fd, ooffset, SEEK_SET);
796 X_UNLOCK (preadwritelock);
797
798 return res;
799}
800
801static ssize_t
802eio__pwrite (int fd, void *buf, size_t count, off_t offset)
803{
804 ssize_t res;
805 off_t ooffset;
806
807 X_LOCK (preadwritelock);
808 ooffset = lseek (fd, 0, SEEK_CUR);
809 lseek (fd, offset, SEEK_SET);
810 res = write (fd, buf, count);
811 lseek (fd, ooffset, SEEK_SET);
812 X_UNLOCK (preadwritelock);
813
814 return res;
815}
816#endif
817
818#ifndef HAVE_FUTIMES 549#ifndef HAVE_UTIMES
819 550
820# undef utimes 551# undef utimes
821# undef futimes
822# define utimes(path,times) eio__utimes (path, times) 552# define utimes(path,times) eio__utimes (path, times)
823# define futimes(fd,times) eio__futimes (fd, times)
824 553
825static int 554static int
826eio__utimes (const char *filename, const struct timeval times[2]) 555eio__utimes (const char *filename, const struct timeval times[2])
827{ 556{
828 if (times) 557 if (times)
… …
836 } 565 }
837 else 566 else
838 return utime (filename, 0); 567 return utime (filename, 0);
839} 568}
840 569
570#endif
571
572#ifndef HAVE_FUTIMES
573
574# undef futimes
575# define futimes(fd,times) eio__futimes (fd, times)
576
577static int
841static int eio__futimes (int fd, const struct timeval tv[2]) 578eio__futimes (int fd, const struct timeval tv[2])
842{ 579{
843 errno = ENOSYS; 580 errno = ENOSYS;
844 return -1; 581 return -1;
845} 582}
846 583
… …
849#if !HAVE_FDATASYNC 586#if !HAVE_FDATASYNC
850# undef fdatasync 587# undef fdatasync
851# define fdatasync(fd) fsync (fd) 588# define fdatasync(fd) fsync (fd)
852#endif 589#endif
853 590
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
854/* sync_file_range always needs emulation */ 608/* sync_file_range always needs emulation */
855int 609static int
856eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags) 610eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
857{ 611{
858#if HAVE_SYNC_FILE_RANGE 612#if HAVE_SYNC_FILE_RANGE
859 int res; 613 int res;
860 614
… …
877 /* even though we could play tricks with the flags, it's better to always 631 /* even though we could play tricks with the flags, it's better to always
878 * call fdatasync, as that matches the expectation of its users best */ 632 * call fdatasync, as that matches the expectation of its users best */
879 return fdatasync (fd); 633 return fdatasync (fd);
880} 634}
881 635
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}
645
882#if !HAVE_READAHEAD 646#if !HAVE_READAHEAD
883# undef readahead 647# undef readahead
884# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 648# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
885 649
886static ssize_t 650static eio_ssize_t
887eio__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)
888{ 652{
889 size_t todo = count; 653 size_t todo = count;
890 dBUF; 654 dBUF;
891 655
… …
896 pread (fd, eio_buf, len, offset); 660 pread (fd, eio_buf, len, offset);
897 offset += len; 661 offset += len;
898 todo -= len; 662 todo -= len;
899 } 663 }
900 664
901 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 */
902 return count; 669 return 0;
903} 670}
904 671
905#endif 672#endif
906 673
907/* sendfile always needs emulation */ 674/* sendfile always needs emulation */
908static ssize_t 675static eio_ssize_t
909eio__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)
910{ 677{
678 eio_ssize_t written = 0;
911 ssize_t res; 679 eio_ssize_t res;
912 680
913 if (!count) 681 if (!count)
914 return 0; 682 return 0;
915 683
684 for (;;)
685 {
686#ifdef __APPLE__
687# undef HAVE_SENDFILE /* broken, as everything on os x */
688#endif
916#if HAVE_SENDFILE 689#if HAVE_SENDFILE
917# if __linux 690# if __linux
691 off_t soffset = offset;
918 res = sendfile (ofd, ifd, &offset, count); 692 res = sendfile (ofd, ifd, &soffset, count);
919 693
920# elif __FreeBSD__ 694# elif __FreeBSD__
921 /* 695 /*
922 * 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
923 * wasted on making it similar to other I/O functions. 697 * wasted on making it similar to other I/O functions.
924 */ 698 */
925 {
926 off_t sbytes; 699 off_t sbytes;
927 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 700 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
928 701
929 #if 0 /* according to the manpage, this is correct, but broken behaviour */ 702 #if 0 /* according to the manpage, this is correct, but broken behaviour */
930 /* freebsd' sendfile will return 0 on success */ 703 /* freebsd' sendfile will return 0 on success */
931 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */ 704 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
932 /* not on e.g. EIO or EPIPE - sounds broken */ 705 /* not on e.g. EIO or EPIPE - sounds broken */
933 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0) 706 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
934 res = sbytes; 707 res = sbytes;
935 #endif 708 #endif
936 709
937 /* according to source inspection, this is correct, and useful behaviour */ 710 /* according to source inspection, this is correct, and useful behaviour */
938 if (sbytes) 711 if (sbytes)
939 res = sbytes; 712 res = sbytes;
940 }
941 713
942# elif defined (__APPLE__) 714# elif defined __APPLE__
943
944 {
945 off_t sbytes = count; 715 off_t sbytes = count;
946 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0); 716 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
947 717
948 /* according to the manpage, sbytes is always valid */ 718 /* according to the manpage, sbytes is always valid */
949 if (sbytes) 719 if (sbytes)
950 res = sbytes; 720 res = sbytes;
951 }
952 721
953# elif __hpux 722# elif __hpux
954 res = sendfile (ofd, ifd, offset, count, 0, 0); 723 res = sendfile (ofd, ifd, offset, count, 0, 0);
955 724
956# elif __solaris 725# elif __solaris
957 {
958 struct sendfilevec vec; 726 struct sendfilevec vec;
959 size_t sbytes; 727 size_t sbytes;
960 728
961 vec.sfv_fd = ifd; 729 vec.sfv_fd = ifd;
962 vec.sfv_flag = 0; 730 vec.sfv_flag = 0;
963 vec.sfv_off = offset; 731 vec.sfv_off = offset;
964 vec.sfv_len = count; 732 vec.sfv_len = count;
965 733
966 res = sendfilev (ofd, &vec, 1, &sbytes); 734 res = sendfilev (ofd, &vec, 1, &sbytes);
967 735
968 if (res < 0 && sbytes) 736 if (res < 0 && sbytes)
969 res = sbytes; 737 res = sbytes;
970 }
971 738
972# endif 739# endif
973 740
974#elif defined (_WIN32) 741#elif defined (_WIN32) && 0
975
976 /* does not work, just for documentation of what would need to be done */ 742 /* does not work, just for documentation of what would need to be done */
977 { 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 */
978 HANDLE h = TO_SOCKET (ifd); 746 HANDLE h = TO_SOCKET (ifd);
979 SetFilePointer (h, offset, 0, FILE_BEGIN); 747 SetFilePointer (h, offset, 0, FILE_BEGIN);
980 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0); 748 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
981 }
982 749
983#else 750#else
984 res = -1; 751 res = EIO_ENOSYS ();
985 errno = ENOSYS;
986#endif 752#endif
987 753
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
988 if (res < 0 777 if (res < 0
989 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 778 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
990 /* BSDs */ 779 /* BSDs */
991#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */ 780#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
992 || errno == ENOTSUP 781 || errno == ENOTSUP
993#endif 782#endif
783#ifdef EOPNOTSUPP /* windows */
994 || errno == EOPNOTSUPP /* BSDs */ 784 || errno == EOPNOTSUPP /* BSDs */
785#endif
995#if __solaris 786#if __solaris
996 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 787 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
997#endif 788#endif
998 ) 789 )
999 ) 790 )
… …
1003 794
1004 res = 0; 795 res = 0;
1005 796
1006 while (count) 797 while (count)
1007 { 798 {
1008 ssize_t cnt; 799 eio_ssize_t cnt;
1009 800
1010 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);
1011 802
1012 if (cnt <= 0) 803 if (cnt <= 0)
1013 { 804 {
… …
1025 816
1026 offset += cnt; 817 offset += cnt;
1027 res += cnt; 818 res += cnt;
1028 count -= cnt; 819 count -= cnt;
1029 } 820 }
821
822 FUBd;
1030 } 823 }
1031 824
1032 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
1033} 1133}
1034 1134
1035static signed char 1135static signed char
1036eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) 1136eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1037{ 1137{
1038 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */ 1138 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1039 : a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0; 1139 : a->inode < b->inode ? -1
1140 : a->inode > b->inode ? 1
1141 : 0;
1040} 1142}
1041 1143
1042#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0 1144#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1043 1145
1044#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */ 1146#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1045#define EIO_SORT_FAST 60 /* when to only use insertion sort */ 1147#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1046 1148
1047static void 1149static void
1048eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1150eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1049{ 1151{
1050 unsigned char bits [9 + sizeof (ino_t) * 8]; 1152 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1051 unsigned char *bit = bits; 1153 unsigned char *bit = bits;
1052 1154
1053 assert (CHAR_BIT == 8); 1155 assert (CHAR_BIT == 8);
1054 assert (sizeof (eio_dirent) * 8 < 256); 1156 assert (sizeof (eio_dirent) * 8 < 256);
1055 assert (offsetof (eio_dirent, inode)); /* we use 0 as sentinel */ 1157 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1056 assert (offsetof (eio_dirent, score)); /* we use 0 as sentinel */ 1158 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1057 1159
1058 if (size <= EIO_SORT_FAST) 1160 if (size <= EIO_SORT_FAST)
1059 return; 1161 return;
1060 1162
1061 /* first prepare an array of bits to test in our radix sort */ 1163 /* first prepare an array of bits to test in our radix sort */
1062 /* try to take endianness into account, as well as differences in ino_t sizes */ 1164 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1063 /* inode_bits must contain all inodes ORed together */ 1165 /* inode_bits must contain all inodes ORed together */
1064 /* which is used to skip bits that are 0 everywhere, which is very common */ 1166 /* which is used to skip bits that are 0 everywhere, which is very common */
1065 { 1167 {
1066 ino_t endianness; 1168 eio_ino_t endianness;
1067 int i, j; 1169 int i, j;
1068 1170
1069 /* we store the byte offset of byte n into byte n of "endianness" */ 1171 /* we store the byte offset of byte n into byte n of "endianness" */
1070 for (i = 0; i < sizeof (ino_t); ++i) 1172 for (i = 0; i < sizeof (eio_ino_t); ++i)
1071 ((unsigned char *)&endianness)[i] = i; 1173 ((unsigned char *)&endianness)[i] = i;
1072 1174
1073 *bit++ = 0; 1175 *bit++ = 0;
1074 1176
1075 for (i = 0; i < sizeof (ino_t); ++i) 1177 for (i = 0; i < sizeof (eio_ino_t); ++i)
1076 { 1178 {
1077 /* shifting off the byte offsets out of "endianness" */ 1179 /* shifting off the byte offsets out of "endianness" */
1078 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8; 1180 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1079 endianness >>= 8; 1181 endianness >>= 8;
1080 1182
1081 for (j = 0; j < 8; ++j) 1183 for (j = 0; j < 8; ++j)
1082 if (inode_bits & (((ino_t)1) << (i * 8 + j))) 1184 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1083 *bit++ = offs + j; 1185 *bit++ = offs + j;
1084 } 1186 }
1085 1187
1086 for (j = 0; j < 8; ++j) 1188 for (j = 0; j < 8; ++j)
1087 if (score_bits & (1 << j)) 1189 if (score_bits & (1 << j))
1088 *bit++ = offsetof (eio_dirent, score) * 8 + j; 1190 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1089 } 1191 }
1090 1192
1091 /* now actually do the sorting (a variant of MSD radix sort) */ 1193 /* now actually do the sorting (a variant of MSD radix sort) */
1092 { 1194 {
1093 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base; 1195 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1094 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end; 1196 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1095 unsigned char *bit_stk [9 + sizeof (ino_t) * 8]; 1197 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1096 int stk_idx = 0; 1198 int stk_idx = 0;
1097 1199
1098 base_stk [stk_idx] = dents; 1200 base_stk [stk_idx] = dents;
1099 end_stk [stk_idx] = dents + size; 1201 end_stk [stk_idx] = dents + size;
1100 bit_stk [stk_idx] = bit - 1; 1202 bit_stk [stk_idx] = bit - 1;
… …
1179 } 1281 }
1180 } 1282 }
1181} 1283}
1182 1284
1183static void 1285static void
1184eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1286eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1185{ 1287{
1186 if (size <= 1) 1288 if (size <= 1)
1187 return; /* our insertion sort relies on size > 0 */ 1289 return; /* our insertion sort relies on size > 0 */
1188 1290
1189 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */ 1291 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
… …
1197 1299
1198/* read a full directory */ 1300/* read a full directory */
1199static void 1301static void
1200eio__scandir (eio_req *req, etp_worker *self) 1302eio__scandir (eio_req *req, etp_worker *self)
1201{ 1303{
1202 DIR *dirp;
1203 EIO_STRUCT_DIRENT *entp;
1204 char *name, *names; 1304 char *name, *names;
1205 int namesalloc = 4096; 1305 int namesalloc = 4096 - sizeof (void *) * 4;
1206 int namesoffs = 0; 1306 int namesoffs = 0;
1207 int flags = req->int1; 1307 int flags = req->int1;
1208 eio_dirent *dents = 0; 1308 eio_dirent *dents = 0;
1209 int dentalloc = 128; 1309 int dentalloc = 128;
1210 int dentoffs = 0; 1310 int dentoffs = 0;
1211 ino_t inode_bits = 0; 1311 eio_ino_t inode_bits = 0;
1312#ifdef _WIN32
1313 HANDLE dirp;
1314 WIN32_FIND_DATA entp;
1315#else
1316 DIR *dirp;
1317 EIO_STRUCT_DIRENT *entp;
1318#endif
1212 1319
1213 req->result = -1; 1320 req->result = -1;
1214 1321
1215 if (!(flags & EIO_READDIR_DENTS)) 1322 if (!(flags & EIO_READDIR_DENTS))
1216 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); 1323 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1217 1324
1218 X_LOCK (wrklock); 1325#ifdef _WIN32
1219 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 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)
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
1220 self->dirp = dirp = opendir (req->ptr1); 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
1221 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1397 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1222 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; 1398 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1223 req->ptr2 = names = malloc (namesalloc); 1399 req->ptr2 = names = malloc (namesalloc);
1224 X_UNLOCK (wrklock);
1225 1400
1226 if (dirp && names && (!flags || dents)) 1401 if (!names || (flags && !dents))
1402 return;
1403
1227 for (;;) 1404 for (;;)
1228 { 1405 {
1406 int done;
1407
1408#ifdef _WIN32
1409 done = !dirp;
1410#else
1229 errno = 0; 1411 errno = 0;
1230 entp = readdir (dirp); 1412 entp = readdir (dirp);
1413 done = !entp;
1414#endif
1231 1415
1232 if (!entp) 1416 if (done)
1233 { 1417 {
1418#ifndef _WIN32
1419 int old_errno = errno;
1420 closedir (dirp);
1421 errno = old_errno;
1422
1234 if (errno) 1423 if (errno)
1235 break; 1424 break;
1425#endif
1236 1426
1237 /* sort etc. */ 1427 /* sort etc. */
1238 req->int1 = flags; 1428 req->int1 = flags;
1239 req->result = dentoffs; 1429 req->result = dentoffs;
1240 1430
1241 if (flags & EIO_READDIR_STAT_ORDER) 1431 if (flags & EIO_READDIR_STAT_ORDER)
1242 eio_dent_sort (dents, dentoffs, 0, inode_bits); /* sort by inode exclusively */ 1432 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1243 else if (flags & EIO_READDIR_DIRS_FIRST) 1433 else if (flags & EIO_READDIR_DIRS_FIRST)
1244 if (flags & EIO_READDIR_FOUND_UNKNOWN) 1434 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1245 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */ 1435 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1246 else 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
1460 break;
1461 }
1462
1463 /* now add the entry to our list(s) */
1464 name = D_NAME (entp);
1465
1466 /* skip . and .. entries */
1467 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1468 {
1469 int len = D_NAMLEN (entp) + 1;
1470
1471 while (ecb_expect_false (namesoffs + len > namesalloc))
1472 {
1473 namesalloc *= 2;
1474 req->ptr2 = names = realloc (names, namesalloc);
1475
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))
1247 { 1487 {
1248 /* in this case, all is known, and we just put dirs first and sort them */ 1488 dentalloc *= 2;
1489 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1490
1491 if (!dents)
1492 break;
1493 }
1494
1249 eio_dirent *oth = dents + dentoffs; 1495 ent = dents + dentoffs;
1250 eio_dirent *dir = dents;
1251 1496
1252 /* now partition dirs to the front, and non-dirs to the back */ 1497 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1253 /* by walking from both sides and swapping if necessary */ 1498 ent->namelen = len - 1;
1254 /* also clear score, so it doesn't influence sorting */ 1499 ent->inode = D_INO (entp);
1255 while (oth > dir) 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)
1256 { 1559 {
1257 if (dir->type == EIO_DT_DIR) 1560 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1258 ++dir; 1561 ent->score = 1;
1259 else if ((--oth)->type == EIO_DT_DIR) 1562 else if (!strchr (name, '.')) /* absence of dots indicate likely dirs */
1260 { 1563 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1261 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1262
1263 ++dir;
1264 }
1265 } 1564 }
1266 1565 else if (ent->type == EIO_DT_DIR)
1267 /* now sort the dirs only */ 1566 ent->score = 0;
1268 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1269 } 1567 }
1270
1271 break;
1272 }
1273
1274 /* now add the entry to our list(s) */
1275 name = entp->d_name;
1276
1277 /* skip . and .. entries */
1278 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1279 {
1280 int len = D_NAMLEN (entp) + 1;
1281
1282 while (expect_false (namesoffs + len > namesalloc))
1283 {
1284 namesalloc *= 2;
1285 X_LOCK (wrklock);
1286 req->ptr2 = names = realloc (names, namesalloc);
1287 X_UNLOCK (wrklock);
1288
1289 if (!names)
1290 break;
1291 } 1568 }
1292 1569
1293 memcpy (names + namesoffs, name, len);
1294
1295 if (dents)
1296 {
1297 struct eio_dirent *ent;
1298
1299 if (expect_false (dentoffs == dentalloc))
1300 {
1301 dentalloc *= 2;
1302 X_LOCK (wrklock);
1303 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1304 X_UNLOCK (wrklock);
1305
1306 if (!dents)
1307 break;
1308 }
1309
1310 ent = dents + dentoffs;
1311
1312 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1313 ent->namelen = len - 1;
1314 ent->inode = D_INO (entp);
1315
1316 inode_bits |= ent->inode;
1317
1318 switch (D_TYPE (entp))
1319 {
1320 default:
1321 ent->type = EIO_DT_UNKNOWN;
1322 flags |= EIO_READDIR_FOUND_UNKNOWN;
1323 break;
1324
1325 #ifdef DT_FIFO
1326 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1327 #endif
1328 #ifdef DT_CHR
1329 case DT_CHR: ent->type = EIO_DT_CHR; break;
1330 #endif
1331 #ifdef DT_MPC
1332 case DT_MPC: ent->type = EIO_DT_MPC; break;
1333 #endif
1334 #ifdef DT_DIR
1335 case DT_DIR: ent->type = EIO_DT_DIR; break;
1336 #endif
1337 #ifdef DT_NAM
1338 case DT_NAM: ent->type = EIO_DT_NAM; break;
1339 #endif
1340 #ifdef DT_BLK
1341 case DT_BLK: ent->type = EIO_DT_BLK; break;
1342 #endif
1343 #ifdef DT_MPB
1344 case DT_MPB: ent->type = EIO_DT_MPB; break;
1345 #endif
1346 #ifdef DT_REG
1347 case DT_REG: ent->type = EIO_DT_REG; break;
1348 #endif
1349 #ifdef DT_NWK
1350 case DT_NWK: ent->type = EIO_DT_NWK; break;
1351 #endif
1352 #ifdef DT_CMP
1353 case DT_CMP: ent->type = EIO_DT_CMP; break;
1354 #endif
1355 #ifdef DT_LNK
1356 case DT_LNK: ent->type = EIO_DT_LNK; break;
1357 #endif
1358 #ifdef DT_SOCK
1359 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1360 #endif
1361 #ifdef DT_DOOR
1362 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1363 #endif
1364 #ifdef DT_WHT
1365 case DT_WHT: ent->type = EIO_DT_WHT; break;
1366 #endif
1367 }
1368
1369 ent->score = 7;
1370
1371 if (flags & EIO_READDIR_DIRS_FIRST)
1372 {
1373 if (ent->type == EIO_DT_UNKNOWN)
1374 {
1375 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1376 ent->score = 1;
1377 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1378 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1379 }
1380 else if (ent->type == EIO_DT_DIR)
1381 ent->score = 0;
1382 }
1383 }
1384
1385 namesoffs += len; 1570 namesoffs += len;
1386 ++dentoffs; 1571 ++dentoffs;
1387 } 1572 }
1388 1573
1389 if (EIO_CANCELLED (req)) 1574 if (EIO_CANCELLED (req))
1390 { 1575 {
1391 errno = ECANCELED; 1576 errno = ECANCELED;
1392 break; 1577 break;
1393 } 1578 }
1579
1580#ifdef _WIN32
1581 if (!FindNextFile (dirp, &entp))
1582 {
1583 FindClose (dirp);
1584 dirp = 0;
1394 } 1585 }
1586#endif
1587 }
1395} 1588}
1396 1589
1397#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) 1590/*****************************************************************************/
1398# define eio__msync(a,b,c) ((errno = ENOSYS), -1) 1591/* working directory stuff */
1399#else 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. */
1400 1595
1401int 1596#if !HAVE_AT
1402eio__msync (void *mem, size_t len, int flags) 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)
1403{ 1602{
1404 if (EIO_MS_ASYNC != MS_SYNC 1603 if (!wd || *path == '/')
1405 || EIO_MS_INVALIDATE != MS_INVALIDATE 1604 return path;
1406 || EIO_MS_SYNC != MS_SYNC) 1605
1606 if (path [0] == '.' && !path [1])
1607 return wd->str;
1608
1609 {
1610 int l1 = wd->len;
1611 int l2 = strlen (path);
1612
1613 char *res = etp_tmpbuf_get (tmpbuf, l1 + l2 + 2);
1614
1615 memcpy (res, wd->str, l1);
1616 res [l1] = '/';
1617 memcpy (res + l1 + 1, path, l2 + 1);
1618
1619 return res;
1620 }
1621}
1622
1623#endif
1624
1625static eio_wd
1626eio__wd_open_sync (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
1627{
1628 int fd;
1629 eio_wd res;
1630 int len = eio__realpath (tmpbuf, wd, path);
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
1662 return wd;
1663}
1664
1665void
1666eio_wd_close_sync (eio_wd wd)
1667{
1668 if (wd != EIO_INVALID_WD && wd != EIO_CWD)
1407 { 1669 {
1408 flags = 0 1670 #if HAVE_AT
1409 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0) 1671 close (wd->fd);
1410 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0) 1672 #endif
1411 | (flags & EIO_MS_SYNC ? MS_SYNC : 0); 1673 free (wd);
1412 } 1674 }
1413
1414 return msync (mem, len, flags);
1415} 1675}
1416 1676
1417#endif 1677#if HAVE_AT
1418 1678
1419int 1679/* they forgot these */
1420eio__mtouch (void *mem, size_t len, int flags)
1421{
1422 intptr_t addr = (intptr_t)mem;
1423 intptr_t end = addr + len;
1424#ifdef PAGESIZE
1425 const intptr_t page = PAGESIZE;
1426#else
1427 static intptr_t page;
1428 1680
1429 if (!page) 1681static int
1430 page = sysconf (_SC_PAGESIZE); 1682eio__truncateat (int dirfd, const char *path, off_t length)
1431#endif 1683{
1684 int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
1685 int res;
1432 1686
1433 /* round down to start of page, although this is probably useless */ 1687 if (fd < 0)
1434 addr &= ~(page - 1); /* assume page size is always a power of two */ 1688 return fd;
1435 1689
1436 if (addr < end) 1690 res = ftruncate (fd, length);
1437 if (flags & EIO_MT_MODIFY) /* modify */ 1691 close (fd);
1438 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
1439 else
1440 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
1441
1442 return 0; 1692 return res;
1443} 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
1444 1711
1445/*****************************************************************************/ 1712/*****************************************************************************/
1446 1713
1447#define ALLOC(len) \ 1714#define ALLOC(len) \
1448 if (!req->ptr2) \ 1715 if (!req->ptr2) \
1449 { \ 1716 { \
1450 X_LOCK (wrklock); \ 1717 X_LOCK (EIO_POOL->wrklock); \
1451 req->flags |= EIO_FLAG_PTR2_FREE; \ 1718 req->flags |= EIO_FLAG_PTR2_FREE; \
1452 X_UNLOCK (wrklock); \ 1719 X_UNLOCK (EIO_POOL->wrklock); \
1453 req->ptr2 = malloc (len); \ 1720 req->ptr2 = malloc (len); \
1454 if (!req->ptr2) \ 1721 if (!req->ptr2) \
1455 { \ 1722 { \
1456 errno = ENOMEM; \ 1723 errno = ENOMEM; \
1457 req->result = -1; \ 1724 req->result = -1; \
1458 break; \ 1725 break; \
1459 } \ 1726 } \
1460 } 1727 }
1461 1728
1462X_THREAD_PROC (etp_proc)
1463{
1464 ETP_REQ *req;
1465 struct timespec ts;
1466 etp_worker *self = (etp_worker *)thr_arg;
1467
1468 /* try to distribute timeouts somewhat randomly */
1469 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1470
1471 for (;;)
1472 {
1473 X_LOCK (reqlock);
1474
1475 for (;;)
1476 {
1477 self->req = req = reqq_shift (&req_queue);
1478
1479 if (req)
1480 break;
1481
1482 ++idle;
1483
1484 ts.tv_sec = time (0) + IDLE_TIMEOUT;
1485 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
1486 {
1487 if (idle > max_idle)
1488 {
1489 --idle;
1490 X_UNLOCK (reqlock);
1491 X_LOCK (wrklock);
1492 --started;
1493 X_UNLOCK (wrklock);
1494 goto quit;
1495 }
1496
1497 /* we are allowed to idle, so do so without any timeout */
1498 X_COND_WAIT (reqwait, reqlock);
1499 }
1500
1501 --idle;
1502 }
1503
1504 --nready;
1505
1506 X_UNLOCK (reqlock);
1507
1508 if (req->type < 0)
1509 goto quit;
1510
1511 if (!EIO_CANCELLED (req))
1512 ETP_EXECUTE (self, req);
1513
1514 X_LOCK (reslock);
1515
1516 ++npending;
1517
1518 if (!reqq_push (&res_queue, req) && want_poll_cb)
1519 want_poll_cb ();
1520
1521 self->req = 0;
1522 etp_worker_clear (self);
1523
1524 X_UNLOCK (reslock);
1525 }
1526
1527quit:
1528 X_LOCK (wrklock);
1529 etp_worker_free (self);
1530 X_UNLOCK (wrklock);
1531
1532 return 0;
1533}
1534
1535/*****************************************************************************/ 1729/*****************************************************************************/
1536 1730
1731int ecb_cold
1537int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1732eio_init (void (*want_poll)(void), void (*done_poll)(void))
1538{ 1733{
1539 return etp_init (want_poll, done_poll); 1734 eio_want_poll_cb = want_poll;
1540} 1735 eio_done_poll_cb = done_poll;
1541 1736
1737 return etp_init (EIO_POOL, 0, 0, 0);
1738}
1739
1740ecb_inline void
1542static void eio_api_destroy (eio_req *req) 1741eio_api_destroy (eio_req *req)
1543{ 1742{
1544 free (req); 1743 free (req);
1545} 1744}
1546 1745
1547#define REQ(rtype) \ 1746#define REQ(rtype) \
1548 eio_req *req; \ 1747 eio_req *req; \
1549 \ 1748 \
1550 req = (eio_req *)calloc (1, sizeof *req); \ 1749 req = (eio_req *)calloc (1, sizeof *req); \
1551 if (!req) \ 1750 if (!req) \
1552 return 0; \ 1751 return 0; \
… …
1566 { \ 1765 { \
1567 eio_api_destroy (req); \ 1766 eio_api_destroy (req); \
1568 return 0; \ 1767 return 0; \
1569 } 1768 }
1570 1769
1770#define SINGLEDOT(ptr) (0[(char *)(ptr)] == '.' && !1[(char *)(ptr)])
1771
1772static void
1571static void eio_execute (etp_worker *self, eio_req *req) 1773eio_execute (etp_worker *self, eio_req *req)
1572{ 1774{
1573 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 }
1574 1803
1575 switch (req->type) 1804 switch (req->type)
1576 { 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;
1577 case EIO_READ: ALLOC (req->size); 1813 case EIO_READ: ALLOC (req->size);
1578 req->result = req->offs >= 0 1814 req->result = req->offs >= 0
1579 ? pread (req->int1, req->ptr2, req->size, req->offs) 1815 ? pread (req->int1, req->ptr2, req->size, req->offs)
1580 : read (req->int1, req->ptr2, req->size); break; 1816 : read (req->int1, req->ptr2, req->size); break;
1581 case EIO_WRITE: req->result = req->offs >= 0 1817 case EIO_WRITE: req->result = req->offs >= 0
1582 ? pwrite (req->int1, req->ptr2, req->size, req->offs) 1818 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1583 : write (req->int1, req->ptr2, req->size); break; 1819 : write (req->int1, req->ptr2, req->size); break;
1584 1820
1585 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;
1586 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
1587 1825
1588 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1826 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1589 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;
1590 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1828 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1591 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;
1592 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1830 case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break;
1593 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; 1831 case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break;
1832 case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break;
1833 case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break;
1594 1834
1835 case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); 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;
1840 case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); 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;
1845 case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2, 0); break;
1846 case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
1847 case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1848 case EIO_READLINK: ALLOC (PATH_MAX);
1849 req->result = readlinkat (dirfd, req->ptr1, req->ptr2, PATH_MAX); break;
1595 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS)); 1850 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1596 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break; 1851 req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1852 case EIO_UTIME:
1853 case EIO_FUTIME:
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
1876#else
1877
1597 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS)); 1878 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1598 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break; 1879 req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
1599 1880 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1881 req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
1600 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; 1882 case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
1601 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1602 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break; 1883 case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
1603 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1604 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break; 1884 case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
1605 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1606
1607 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break; 1885 case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
1608 case EIO_CLOSE: req->result = close (req->int1); break; 1886
1609 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1610 case EIO_UNLINK: req->result = unlink (req->ptr1); break; 1887 case EIO_UNLINK: req->result = unlink (path ); break;
1611 case EIO_RMDIR: req->result = rmdir (req->ptr1); break; 1888 case EIO_RMDIR: req->result = rmdir (path ); break;
1612 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; 1889 case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
1613 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; 1890 case EIO_RENAME: req->result = rename (path , req->ptr2); break;
1614 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; 1891 case EIO_LINK: req->result = link (path , req->ptr2); break;
1615 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; 1892 case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
1616 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break; 1893 case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
1617
1618 case EIO_READLINK: ALLOC (PATH_MAX); 1894 case EIO_READLINK: ALLOC (PATH_MAX);
1619 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break; 1895 req->result = readlink (path, req->ptr2, PATH_MAX); break;
1620 1896 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1621 case EIO_SYNC: req->result = 0; sync (); break; 1897 req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
1622 case EIO_FSYNC: req->result = fsync (req->int1); break;
1623 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1624 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1625 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;
1626 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1627
1628 case EIO_READDIR: eio__scandir (req, self); break;
1629
1630 case EIO_BUSY:
1631#ifdef _WIN32
1632 Sleep (req->nv1 * 1000.);
1633#else
1634 {
1635 struct timeval tv;
1636
1637 tv.tv_sec = req->nv1;
1638 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;
1639
1640 req->result = select (0, 0, 0, 0, &tv);
1641 }
1642#endif
1643 break;
1644 1898
1645 case EIO_UTIME: 1899 case EIO_UTIME:
1646 case EIO_FUTIME: 1900 case EIO_FUTIME:
1647 { 1901 {
1648 struct timeval tv[2]; 1902 struct timeval tv[2];
1649 struct timeval *times; 1903 struct timeval *times;
1650 1904
1651 if (req->nv1 != -1. || req->nv2 != -1.) 1905 if (req->nv1 != -1. || req->nv2 != -1.)
1652 { 1906 {
1653 tv[0].tv_sec = req->nv1; 1907 tv[0].tv_sec = req->nv1;
1654 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.; 1908 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
1655 tv[1].tv_sec = req->nv2; 1909 tv[1].tv_sec = req->nv2;
1656 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.; 1910 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
1657 1911
1658 times = tv; 1912 times = tv;
1659 } 1913 }
1660 else 1914 else
1661 times = 0; 1915 times = 0;
1662
1663 1916
1664 req->result = req->type == EIO_FUTIME 1917 req->result = req->type == EIO_FUTIME
1665 ? futimes (req->int1, times) 1918 ? futimes (req->int1, times)
1666 : utimes (req->ptr1, times); 1919 : utimes (req->ptr1, times);
1667 } 1920 }
1668 break; 1921 break;
1669 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
1670 case EIO_GROUP: 1973 case EIO_GROUP:
1671 abort (); /* handled in eio_request */ 1974 abort (); /* handled in eio_request */
1975#endif
1672 1976
1673 case EIO_NOP: 1977 case EIO_NOP:
1674 req->result = 0; 1978 req->result = 0;
1675 break; 1979 break;
1676 1980
1677 case EIO_CUSTOM: 1981 case EIO_CUSTOM:
1678 ((void (*)(eio_req *))req->feed) (req); 1982 req->feed (req);
1679 break; 1983 break;
1680 1984
1681 default: 1985 default:
1682 req->result = -1; 1986 req->result = EIO_ENOSYS ();
1683 break; 1987 break;
1684 } 1988 }
1685 1989
1686 req->errorno = errno; 1990 req->errorno = errno;
1687} 1991}
1688 1992
1689#ifndef EIO_NO_WRAPPERS 1993#ifndef EIO_NO_WRAPPERS
1690 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
1691eio_req *eio_nop (int pri, eio_cb cb, void *data) 2005eio_req *eio_nop (int pri, eio_cb cb, void *data)
1692{ 2006{
1693 REQ (EIO_NOP); SEND; 2007 REQ (EIO_NOP); SEND;
1694} 2008}
1695 2009
… …
1711eio_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)
1712{ 2026{
1713 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;
1714} 2028}
1715 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
1716eio_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)
1717{ 2046{
1718 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;
1719} 2048}
1720 2049
1721eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) 2050eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
1722{ 2051{
1723 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; 2052 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
1724} 2053}
1725 2054
1726eio_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)
1727{ 2056{
1728 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;
1729} 2063}
1730 2064
1731eio_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)
1732{ 2066{
1733 REQ (EIO_CLOSE); req->int1 = fd; SEND; 2067 REQ (EIO_CLOSE); req->int1 = fd; SEND;
… …
1736eio_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)
1737{ 2071{
1738 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;
1739} 2073}
1740 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
1741eio_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)
1742{ 2081{
1743 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;
1744} 2083}
1745 2084
… …
1771eio_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)
1772{ 2111{
1773 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2112 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1774} 2113}
1775 2114
1776eio_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)
1777{ 2116{
1778 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;
1779} 2118}
1780 2119
1781eio_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)
… …
1801eio_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)
1802{ 2141{
1803 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2142 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1804} 2143}
1805 2144
1806eio_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)
1807{ 2146{
1808 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;
1809} 2148}
1810 2149
1811eio_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)
… …
1827eio_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)
1828{ 2167{
1829 return eio__1path (EIO_READLINK, path, pri, cb, data); 2168 return eio__1path (EIO_READLINK, path, pri, cb, data);
1830} 2169}
1831 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
1832eio_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)
1833{ 2177{
1834 return eio__1path (EIO_STAT, path, pri, cb, data); 2178 return eio__1path (EIO_STAT, path, pri, cb, data);
1835} 2179}
1836 2180
… …
1859 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND; 2203 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1860} 2204}
1861 2205
1862eio_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)
1863{ 2207{
1864 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;
1865} 2209}
1866 2210
1867static eio_req * 2211static eio_req *
1868eio__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)
1869{ 2213{
… …
1893eio_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)
1894{ 2238{
1895 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2239 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1896} 2240}
1897 2241
1898eio_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)
1899{ 2243{
1900 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND; 2244 REQ (EIO_CUSTOM); req->feed = execute; SEND;
1901} 2245}
1902 2246
1903#endif 2247#endif
1904 2248
1905eio_req *eio_grp (eio_cb cb, void *data) 2249eio_req *eio_grp (eio_cb cb, void *data)
… …
1914#undef SEND 2258#undef SEND
1915 2259
1916/*****************************************************************************/ 2260/*****************************************************************************/
1917/* grp functions */ 2261/* grp functions */
1918 2262
2263void
1919void 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)
1920{ 2265{
1921 grp->int2 = limit; 2266 grp->int2 = limit;
1922 grp->feed = feed; 2267 grp->feed = feed;
1923 2268
1924 grp_try_feed (grp); 2269 grp_try_feed (grp);
1925} 2270}
1926 2271
2272void
1927void eio_grp_limit (eio_req *grp, int limit) 2273eio_grp_limit (eio_req *grp, int limit)
1928{ 2274{
1929 grp->int2 = limit; 2275 grp->int2 = limit;
1930 2276
1931 grp_try_feed (grp); 2277 grp_try_feed (grp);
1932} 2278}
1933 2279
2280void
1934void eio_grp_add (eio_req *grp, eio_req *req) 2281eio_grp_add (eio_req *grp, eio_req *req)
1935{ 2282{
1936 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));
1937 2284
1938 grp->flags |= EIO_FLAG_GROUPADD; 2285 grp->flags |= ETP_FLAG_GROUPADD;
1939 2286
1940 ++grp->size; 2287 ++grp->size;
1941 req->grp = grp; 2288 req->grp = grp;
1942 2289
1943 req->grp_prev = 0; 2290 req->grp_prev = 0;
… …
1950} 2297}
1951 2298
1952/*****************************************************************************/ 2299/*****************************************************************************/
1953/* misc garbage */ 2300/* misc garbage */
1954 2301
2302eio_ssize_t
1955ssize_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)
1956{ 2304{
1957 etp_worker wrk;
1958 ssize_t ret;
1959
1960 wrk.dbuf = 0;
1961
1962 ret = eio__sendfile (ofd, ifd, offset, count, &wrk); 2305 return eio__sendfile (ofd, ifd, offset, count);
1963
1964 if (wrk.dbuf)
1965 free (wrk.dbuf);
1966
1967 return ret;
1968} 2306}
1969 2307

Diff Legend

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