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

Comparing libeio/eio.c (file contents):
Revision 1.53 by root, Sat Jan 9 10:03:09 2010 UTC vs.
Revision 1.132 by root, Sun Sep 7 14:50:00 2014 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
126 #define fchmod(fd,mode) EIO_ENOSYS ()
127 #define chown(path,uid,gid) EIO_ENOSYS ()
128 #define fchown(fd,uid,gid) EIO_ENOSYS ()
129 #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
130 #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
131 #define mknod(path,mode,dev) EIO_ENOSYS ()
132 #define sync() EIO_ENOSYS ()
133 #define readlink(path,buf,s) EIO_ENOSYS ()
134 #define statvfs(path,buf) EIO_ENOSYS ()
135 #define fstatvfs(fd,buf) EIO_ENOSYS ()
136
137 #define pread(fd,buf,count,offset) eio__pread (fd, buf, count, offset)
138 #define pwrite(fd,buf,count,offset) eio__pwrite (fd, buf, count, offset)
139
140 #if __GNUC__
141 typedef long long eio_off_t; /* signed for compatibility to msvc */
142 #else
143 typedef __int64 eio_off_t; /* unsigned not supported by msvc */
144 #endif
145
146 static eio_ssize_t
147 eio__pread (int fd, void *buf, eio_ssize_t count, eio_off_t offset)
148 {
149 OVERLAPPED o = { 0 };
150 DWORD got;
151
152 o.Offset = offset;
153 o.OffsetHigh = offset >> 32;
154
155 return ReadFile ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd), buf, count, &got, &o)
156 ? got : -1;
157 }
158
159 static eio_ssize_t
160 eio__pwrite (int fd, void *buf, eio_ssize_t count, eio_off_t offset)
161 {
162 OVERLAPPED o = { 0 };
163 DWORD got;
164
165 o.Offset = offset;
166 o.OffsetHigh = offset >> 32;
167
168 return WriteFile ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd), buf, count, &got, &o)
169 ? got : -1;
170 }
171
172 /* rename() uses MoveFile, which fails to overwrite */
173 #define rename(old,neu) eio__rename (old, neu)
174
175 static int
176 eio__rename (const char *old, const char *neu)
177 {
178 if (MoveFileEx (old, neu, MOVEFILE_REPLACE_EXISTING))
179 return 0;
180
181 /* should steal _dosmaperr */
182 switch (GetLastError ())
183 {
184 case ERROR_FILE_NOT_FOUND:
185 case ERROR_PATH_NOT_FOUND:
186 case ERROR_INVALID_DRIVE:
187 case ERROR_NO_MORE_FILES:
188 case ERROR_BAD_NETPATH:
189 case ERROR_BAD_NET_NAME:
190 case ERROR_BAD_PATHNAME:
191 case ERROR_FILENAME_EXCED_RANGE:
192 errno = ENOENT;
193 break;
194
195 default:
196 errno = EACCES;
197 break;
198 }
199
200 return -1;
201 }
202
203 /* we could even stat and see if it exists */
204 static int
205 symlink (const char *old, const char *neu)
206 {
207 #if WINVER >= 0x0600
208 if (CreateSymbolicLink (neu, old, 1))
209 return 0;
210
211 if (CreateSymbolicLink (neu, old, 0))
212 return 0;
213 #endif
214
215 return EIO_ERRNO (ENOENT, -1);
216 }
217
218 /* POSIX API only, causing trouble for win32 apps */
219 #define CreateHardLink(neu,old,flags) 0 /* not really creating hardlink, still using relative paths? */
220 #define CreateSymbolicLink(neu,old,flags) 0 /* vista+ only */
221
222 struct statvfs
223 {
224 int dummy;
225 };
226
227 #define DT_DIR EIO_DT_DIR
228 #define DT_REG EIO_DT_REG
229 #define D_NAME(entp) entp.cFileName
230 #define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
231
74#else 232#else
75 233
76# include "config.h"
77# include <sys/time.h> 234 #include <sys/time.h>
78# include <sys/select.h> 235 #include <sys/select.h>
79# include <sys/mman.h>
80# include <unistd.h> 236 #include <unistd.h>
237 #include <signal.h>
238 #include <dirent.h>
239
240 #ifdef ANDROID
241 #include <sys/vfs.h>
242 #define statvfs statfs
243 #define fstatvfs fstatfs
244 #include <asm/page.h> /* supposedly limits.h does #define PAGESIZE PAGESIZE */
245 #else
246 #include <sys/statvfs.h>
247 #endif
248
249 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
250 #include <sys/mman.h>
251 #endif
252
253 #define D_NAME(entp) entp->d_name
254
255 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
256 #if __FreeBSD__ || __NetBSD__ || __OpenBSD__
257 #define _DIRENT_HAVE_D_TYPE /* sigh */
258 #define D_INO(de) (de)->d_fileno
259 #define D_NAMLEN(de) (de)->d_namlen
260 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
261 #define D_INO(de) (de)->d_ino
262 #endif
263
264 #ifdef _D_EXACT_NAMLEN
265 #undef D_NAMLEN
266 #define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
267 #endif
268
269 #ifdef _DIRENT_HAVE_D_TYPE
270 #define D_TYPE(de) (de)->d_type
271 #endif
272
273 #ifndef EIO_STRUCT_DIRENT
274 #define EIO_STRUCT_DIRENT struct dirent
275 #endif
276
277#endif
278
279#if HAVE_UTIMES
81# include <utime.h> 280# 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 281#endif
93 282
94#ifdef _D_EXACT_NAMLEN 283#if HAVE_SYS_SYSCALL_H
95# undef D_NAMLEN 284# 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 285#endif
102 286
103# ifndef EIO_STRUCT_DIRENT 287#if HAVE_SYS_PRCTL_H
104# define EIO_STRUCT_DIRENT struct dirent 288# include <sys/prctl.h>
105# endif
106
107#endif 289#endif
108 290
109#if HAVE_SENDFILE 291#if HAVE_SENDFILE
110# if __linux 292# if __linux
111# include <sys/sendfile.h> 293# include <sys/sendfile.h>
… …
126#endif 308#endif
127#ifndef D_INO 309#ifndef D_INO
128# define D_INO(de) 0 310# define D_INO(de) 0
129#endif 311#endif
130#ifndef D_NAMLEN 312#ifndef D_NAMLEN
131# define D_NAMLEN(de) strlen ((de)->d_name) 313# define D_NAMLEN(entp) strlen (D_NAME (entp))
132#endif 314#endif
133
134/* number of seconds after which an idle threads exit */
135#define IDLE_TIMEOUT 10
136 315
137/* used for struct dirent, AIX doesn't provide it */ 316/* used for struct dirent, AIX doesn't provide it */
138#ifndef NAME_MAX 317#ifndef NAME_MAX
139# define NAME_MAX 4096 318# define NAME_MAX 4096
140#endif 319#endif
141 320
321/* used for readlink etc. */
322#ifndef PATH_MAX
323# define PATH_MAX 4096
324#endif
325
142/* buffer size for various temporary buffers */ 326/* buffer size for various temporary buffers */
143#define EIO_BUFSIZE 65536 327#define EIO_BUFSIZE 65536
144 328
145#define dBUF \ 329#define dBUF \
146 char *eio_buf; \
147 ETP_WORKER_LOCK (self); \
148 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 330 char *eio_buf = malloc (EIO_BUFSIZE); \
149 ETP_WORKER_UNLOCK (self); \
150 errno = ENOMEM; \ 331 errno = ENOMEM; \
151 if (!eio_buf) \ 332 if (!eio_buf) \
152 return -1; 333 return -1
153 334
154#define EIO_TICKS ((1000000 + 1023) >> 10) 335#define FUBd \
336 free (eio_buf)
155 337
156/*****************************************************************************/ 338/*****************************************************************************/
157 339
158#if __GNUC__ >= 3 340struct tmpbuf
159# define expect(expr,value) __builtin_expect ((expr),(value)) 341{
342 void *ptr;
343 int len;
344};
345
346static void *
347tmpbuf_get (struct tmpbuf *buf, int len)
348{
349 if (buf->len < len)
350 {
351 free (buf->ptr);
352 buf->ptr = malloc (buf->len = len);
353 }
354
355 return buf->ptr;
356}
357
358struct tmpbuf;
359
360#if _POSIX_VERSION >= 200809L
361 #define HAVE_AT 1
362 #define WD2FD(wd) ((wd) ? (wd)->fd : AT_FDCWD)
363 #ifndef O_SEARCH
364 #define O_SEARCH O_RDONLY
365 #endif
160#else 366#else
161# define expect(expr,value) (expr) 367 #define HAVE_AT 0
368 static const char *wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path);
162#endif 369#endif
163 370
164#define expect_false(expr) expect ((expr) != 0, 0) 371struct eio_pwd
165#define expect_true(expr) expect ((expr) != 0, 1) 372{
373#if HAVE_AT
374 int fd;
375#endif
376 int len;
377 char str[1]; /* actually, a 0-terminated canonical path */
378};
166 379
167/*****************************************************************************/ 380/*****************************************************************************/
168 381
169#define ETP_PRI_MIN EIO_PRI_MIN 382#define ETP_PRI_MIN EIO_PRI_MIN
170#define ETP_PRI_MAX EIO_PRI_MAX 383#define ETP_PRI_MAX EIO_PRI_MAX
384
385#define ETP_TYPE_QUIT -1
386#define ETP_TYPE_GROUP EIO_GROUP
171 387
172struct etp_worker; 388struct etp_worker;
173 389
174#define ETP_REQ eio_req 390#define ETP_REQ eio_req
175#define ETP_DESTROY(req) eio_destroy (req) 391#define ETP_DESTROY(req) eio_destroy (req)
176static int eio_finish (eio_req *req); 392static int eio_finish (eio_req *req);
177#define ETP_FINISH(req) eio_finish (req) 393#define ETP_FINISH(req) eio_finish (req)
178static void eio_execute (struct etp_worker *self, eio_req *req); 394static void eio_execute (struct etp_worker *self, eio_req *req);
179#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) 395#define ETP_EXECUTE(wrk,req) eio_execute (wrk, req)
180 396
181#define ETP_WORKER_CLEAR(req) \ 397#include "etp.c"
182 if (wrk->dbuf) \
183 { \
184 free (wrk->dbuf); \
185 wrk->dbuf = 0; \
186 } \
187 \
188 if (wrk->dirp) \
189 { \
190 closedir (wrk->dirp); \
191 wrk->dirp = 0; \
192 }
193
194#define ETP_WORKER_COMMON \
195 void *dbuf; \
196 DIR *dirp;
197 398
198/*****************************************************************************/ 399/*****************************************************************************/
199 400
200#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
201
202/* calculate time difference in ~1/EIO_TICKS of a second */
203static int tvdiff (struct timeval *tv1, struct timeval *tv2)
204{
205 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
206 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
207}
208
209static unsigned int started, idle, wanted = 4;
210
211static void (*want_poll_cb) (void);
212static void (*done_poll_cb) (void);
213
214static unsigned int max_poll_time; /* reslock */
215static unsigned int max_poll_reqs; /* reslock */
216
217static volatile unsigned int nreqs; /* reqlock */
218static volatile unsigned int nready; /* reqlock */
219static volatile unsigned int npending; /* reqlock */
220static volatile unsigned int max_idle = 4;
221
222static mutex_t wrklock = X_MUTEX_INIT;
223static mutex_t reslock = X_MUTEX_INIT;
224static mutex_t reqlock = X_MUTEX_INIT;
225static cond_t reqwait = X_COND_INIT;
226
227#if !HAVE_PREADWRITE
228/*
229 * make our pread/pwrite emulation safe against themselves, but not against
230 * normal read/write by using a mutex. slows down execution a lot,
231 * but that's your problem, not mine.
232 */
233static mutex_t preadwritelock = X_MUTEX_INIT;
234#endif
235
236typedef struct etp_worker
237{
238 /* locked by wrklock */
239 struct etp_worker *prev, *next;
240
241 thread_t tid;
242
243 /* locked by reslock, reqlock or wrklock */
244 ETP_REQ *req; /* currently processed request */
245
246 ETP_WORKER_COMMON
247} etp_worker;
248
249static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */
250
251#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
252#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
253
254/* worker threads management */
255
256static void etp_worker_clear (etp_worker *wrk)
257{
258 ETP_WORKER_CLEAR (wrk);
259}
260
261static void etp_worker_free (etp_worker *wrk)
262{
263 wrk->next->prev = wrk->prev;
264 wrk->prev->next = wrk->next;
265
266 free (wrk);
267}
268
269static unsigned int etp_nreqs (void)
270{
271 int retval;
272 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
273 retval = nreqs;
274 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
275 return retval;
276}
277
278static unsigned int etp_nready (void)
279{
280 unsigned int retval;
281
282 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
283 retval = nready;
284 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
285
286 return retval;
287}
288
289static unsigned int etp_npending (void)
290{
291 unsigned int retval;
292
293 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
294 retval = npending;
295 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
296
297 return retval;
298}
299
300static unsigned int etp_nthreads (void)
301{
302 unsigned int retval;
303
304 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
305 retval = started;
306 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
307
308 return retval;
309}
310
311/*
312 * a somewhat faster data structure might be nice, but
313 * with 8 priorities this actually needs <20 insns
314 * per shift, the most expensive operation.
315 */
316typedef struct {
317 ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */
318 int size;
319} etp_reqq;
320
321static etp_reqq req_queue;
322static etp_reqq res_queue;
323
324static int reqq_push (etp_reqq *q, ETP_REQ *req)
325{
326 int pri = req->pri;
327 req->next = 0;
328
329 if (q->qe[pri])
330 {
331 q->qe[pri]->next = req;
332 q->qe[pri] = req;
333 }
334 else
335 q->qe[pri] = q->qs[pri] = req;
336
337 return q->size++;
338}
339
340static ETP_REQ *reqq_shift (etp_reqq *q)
341{
342 int pri;
343
344 if (!q->size)
345 return 0;
346
347 --q->size;
348
349 for (pri = ETP_NUM_PRI; pri--; )
350 {
351 eio_req *req = q->qs[pri];
352
353 if (req)
354 {
355 if (!(q->qs[pri] = (eio_req *)req->next))
356 q->qe[pri] = 0;
357
358 return req;
359 }
360 }
361
362 abort ();
363}
364
365static void etp_atfork_prepare (void)
366{
367 X_LOCK (wrklock);
368 X_LOCK (reqlock);
369 X_LOCK (reslock);
370#if !HAVE_PREADWRITE
371 X_LOCK (preadwritelock);
372#endif
373}
374
375static void etp_atfork_parent (void)
376{
377#if !HAVE_PREADWRITE
378 X_UNLOCK (preadwritelock);
379#endif
380 X_UNLOCK (reslock);
381 X_UNLOCK (reqlock);
382 X_UNLOCK (wrklock);
383}
384
385static void etp_atfork_child (void)
386{
387 ETP_REQ *prv;
388
389 while ((prv = reqq_shift (&req_queue)))
390 ETP_DESTROY (prv);
391
392 while ((prv = reqq_shift (&res_queue)))
393 ETP_DESTROY (prv);
394
395 while (wrk_first.next != &wrk_first)
396 {
397 etp_worker *wrk = wrk_first.next;
398
399 if (wrk->req)
400 ETP_DESTROY (wrk->req);
401
402 etp_worker_clear (wrk);
403 etp_worker_free (wrk);
404 }
405
406 started = 0;
407 idle = 0;
408 nreqs = 0;
409 nready = 0;
410 npending = 0;
411
412 etp_atfork_parent ();
413}
414
415static void 401static void
416etp_once_init (void)
417{
418 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
419}
420
421static int
422etp_init (void (*want_poll)(void), void (*done_poll)(void))
423{
424 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
425
426 pthread_once (&doinit, etp_once_init);
427
428 want_poll_cb = want_poll;
429 done_poll_cb = done_poll;
430
431 return 0;
432}
433
434X_THREAD_PROC (etp_proc);
435
436static void etp_start_thread (void)
437{
438 etp_worker *wrk = calloc (1, sizeof (etp_worker));
439
440 /*TODO*/
441 assert (("unable to allocate worker thread data", wrk));
442
443 X_LOCK (wrklock);
444
445 if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
446 {
447 wrk->prev = &wrk_first;
448 wrk->next = wrk_first.next;
449 wrk_first.next->prev = wrk;
450 wrk_first.next = wrk;
451 ++started;
452 }
453 else
454 free (wrk);
455
456 X_UNLOCK (wrklock);
457}
458
459static void etp_maybe_start_thread (void)
460{
461 if (expect_true (etp_nthreads () >= wanted))
462 return;
463
464 /* todo: maybe use idle here, but might be less exact */
465 if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
466 return;
467
468 etp_start_thread ();
469}
470
471static void etp_end_thread (void)
472{
473 eio_req *req = calloc (1, sizeof (eio_req));
474
475 req->type = -1;
476 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
477
478 X_LOCK (reqlock);
479 reqq_push (&req_queue, req);
480 X_COND_SIGNAL (reqwait);
481 X_UNLOCK (reqlock);
482
483 X_LOCK (wrklock);
484 --started;
485 X_UNLOCK (wrklock);
486}
487
488static int etp_poll (void)
489{
490 unsigned int maxreqs;
491 unsigned int maxtime;
492 struct timeval tv_start, tv_now;
493
494 X_LOCK (reslock);
495 maxreqs = max_poll_reqs;
496 maxtime = max_poll_time;
497 X_UNLOCK (reslock);
498
499 if (maxtime)
500 gettimeofday (&tv_start, 0);
501
502 for (;;)
503 {
504 ETP_REQ *req;
505
506 etp_maybe_start_thread ();
507
508 X_LOCK (reslock);
509 req = reqq_shift (&res_queue);
510
511 if (req)
512 {
513 --npending;
514
515 if (!res_queue.size && done_poll_cb)
516 done_poll_cb ();
517 }
518
519 X_UNLOCK (reslock);
520
521 if (!req)
522 return 0;
523
524 X_LOCK (reqlock);
525 --nreqs;
526 X_UNLOCK (reqlock);
527
528 if (expect_false (req->type == EIO_GROUP && req->size))
529 {
530 req->int1 = 1; /* mark request as delayed */
531 continue;
532 }
533 else
534 {
535 int res = ETP_FINISH (req);
536 if (expect_false (res))
537 return res;
538 }
539
540 if (expect_false (maxreqs && !--maxreqs))
541 break;
542
543 if (maxtime)
544 {
545 gettimeofday (&tv_now, 0);
546
547 if (tvdiff (&tv_start, &tv_now) >= maxtime)
548 break;
549 }
550 }
551
552 errno = EAGAIN;
553 return -1;
554}
555
556static void etp_cancel (ETP_REQ *req)
557{
558 X_LOCK (wrklock);
559 req->flags |= EIO_FLAG_CANCELLED;
560 X_UNLOCK (wrklock);
561
562 eio_grp_cancel (req);
563}
564
565static void etp_submit (ETP_REQ *req)
566{
567 req->pri -= ETP_PRI_MIN;
568
569 if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
570 if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
571
572 if (expect_false (req->type == EIO_GROUP))
573 {
574 /* I hope this is worth it :/ */
575 X_LOCK (reqlock);
576 ++nreqs;
577 X_UNLOCK (reqlock);
578
579 X_LOCK (reslock);
580
581 ++npending;
582
583 if (!reqq_push (&res_queue, req) && want_poll_cb)
584 want_poll_cb ();
585
586 X_UNLOCK (reslock);
587 }
588 else
589 {
590 X_LOCK (reqlock);
591 ++nreqs;
592 ++nready;
593 reqq_push (&req_queue, req);
594 X_COND_SIGNAL (reqwait);
595 X_UNLOCK (reqlock);
596
597 etp_maybe_start_thread ();
598 }
599}
600
601static void etp_set_max_poll_time (double nseconds)
602{
603 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
604 max_poll_time = nseconds * EIO_TICKS;
605 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
606}
607
608static void etp_set_max_poll_reqs (unsigned int maxreqs)
609{
610 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
611 max_poll_reqs = maxreqs;
612 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
613}
614
615static void etp_set_max_idle (unsigned int nthreads)
616{
617 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
618 max_idle = nthreads <= 0 ? 1 : nthreads;
619 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
620}
621
622static void etp_set_min_parallel (unsigned int nthreads)
623{
624 if (wanted < nthreads)
625 wanted = nthreads;
626}
627
628static void etp_set_max_parallel (unsigned int nthreads)
629{
630 if (wanted > nthreads)
631 wanted = nthreads;
632
633 while (started > wanted)
634 etp_end_thread ();
635}
636
637/*****************************************************************************/
638
639static void grp_try_feed (eio_req *grp) 402grp_try_feed (eio_req *grp)
640{ 403{
641 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 404 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
642 { 405 {
643 grp->flags &= ~EIO_FLAG_GROUPADD; 406 grp->flags &= ~EIO_FLAG_GROUPADD;
644 407
… …
651 break; 414 break;
652 } 415 }
653 } 416 }
654} 417}
655 418
419static int
656static int grp_dec (eio_req *grp) 420grp_dec (eio_req *grp)
657{ 421{
658 --grp->size; 422 --grp->size;
659 423
660 /* call feeder, if applicable */ 424 /* call feeder, if applicable */
661 grp_try_feed (grp); 425 grp_try_feed (grp);
… …
665 return eio_finish (grp); 429 return eio_finish (grp);
666 else 430 else
667 return 0; 431 return 0;
668} 432}
669 433
434static void
670void eio_destroy (eio_req *req) 435eio_destroy (eio_req *req)
671{ 436{
672 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); 437 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
673 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); 438 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
674 439
675 EIO_DESTROY (req); 440 EIO_DESTROY (req);
676} 441}
677 442
443static int
678static int eio_finish (eio_req *req) 444eio_finish (eio_req *req)
679{ 445{
680 int res = EIO_FINISH (req); 446 int res = EIO_FINISH (req);
681 447
682 if (req->grp) 448 if (req->grp)
683 { 449 {
… …
691 if (grp->grp_first == req) 457 if (grp->grp_first == req)
692 grp->grp_first = req->grp_next; 458 grp->grp_first = req->grp_next;
693 459
694 res2 = grp_dec (grp); 460 res2 = grp_dec (grp);
695 461
696 if (!res && res2) 462 if (!res)
697 res = res2; 463 res = res2;
698 } 464 }
699 465
700 eio_destroy (req); 466 eio_destroy (req);
701 467
702 return res; 468 return res;
703} 469}
704 470
471void
705void eio_grp_cancel (eio_req *grp) 472eio_grp_cancel (eio_req *grp)
706{ 473{
707 for (grp = grp->grp_first; grp; grp = grp->grp_next)
708 eio_cancel (grp); 474 etp_grp_cancel (grp);
709} 475}
710 476
477void
711void eio_cancel (eio_req *req) 478eio_cancel (eio_req *req)
712{ 479{
713 etp_cancel (req); 480 etp_cancel (req);
714} 481}
715 482
483void
716void eio_submit (eio_req *req) 484eio_submit (eio_req *req)
717{ 485{
718 etp_submit (req); 486 etp_submit (req);
719} 487}
720 488
721unsigned int eio_nreqs (void) 489unsigned int
490eio_nreqs (void)
722{ 491{
723 return etp_nreqs (); 492 return etp_nreqs ();
724} 493}
725 494
726unsigned int eio_nready (void) 495unsigned int
496eio_nready (void)
727{ 497{
728 return etp_nready (); 498 return etp_nready ();
729} 499}
730 500
731unsigned int eio_npending (void) 501unsigned int
502eio_npending (void)
732{ 503{
733 return etp_npending (); 504 return etp_npending ();
734} 505}
735 506
736unsigned int eio_nthreads (void) 507unsigned int ecb_cold
508eio_nthreads (void)
737{ 509{
738 return etp_nthreads (); 510 return etp_nthreads ();
739} 511}
740 512
513void ecb_cold
741void eio_set_max_poll_time (double nseconds) 514eio_set_max_poll_time (double nseconds)
742{ 515{
743 etp_set_max_poll_time (nseconds); 516 etp_set_max_poll_time (nseconds);
744} 517}
745 518
519void ecb_cold
746void eio_set_max_poll_reqs (unsigned int maxreqs) 520eio_set_max_poll_reqs (unsigned int maxreqs)
747{ 521{
748 etp_set_max_poll_reqs (maxreqs); 522 etp_set_max_poll_reqs (maxreqs);
749} 523}
750 524
525void ecb_cold
751void eio_set_max_idle (unsigned int nthreads) 526eio_set_max_idle (unsigned int nthreads)
752{ 527{
753 etp_set_max_idle (nthreads); 528 etp_set_max_idle (nthreads);
754} 529}
755 530
531void ecb_cold
532eio_set_idle_timeout (unsigned int seconds)
533{
534 etp_set_idle_timeout (seconds);
535}
536
537void ecb_cold
756void eio_set_min_parallel (unsigned int nthreads) 538eio_set_min_parallel (unsigned int nthreads)
757{ 539{
758 etp_set_min_parallel (nthreads); 540 etp_set_min_parallel (nthreads);
759} 541}
760 542
543void ecb_cold
761void eio_set_max_parallel (unsigned int nthreads) 544eio_set_max_parallel (unsigned int nthreads)
762{ 545{
763 etp_set_max_parallel (nthreads); 546 etp_set_max_parallel (nthreads);
764} 547}
765 548
766int eio_poll (void) 549int eio_poll (void)
… …
769} 552}
770 553
771/*****************************************************************************/ 554/*****************************************************************************/
772/* work around various missing functions */ 555/* work around various missing functions */
773 556
774#if !HAVE_PREADWRITE
775# undef pread
776# undef pwrite
777# define pread eio__pread
778# define pwrite eio__pwrite
779
780static ssize_t
781eio__pread (int fd, void *buf, size_t count, off_t offset)
782{
783 ssize_t res;
784 off_t ooffset;
785
786 X_LOCK (preadwritelock);
787 ooffset = lseek (fd, 0, SEEK_CUR);
788 lseek (fd, offset, SEEK_SET);
789 res = read (fd, buf, count);
790 lseek (fd, ooffset, SEEK_SET);
791 X_UNLOCK (preadwritelock);
792
793 return res;
794}
795
796static ssize_t
797eio__pwrite (int fd, void *buf, size_t count, off_t offset)
798{
799 ssize_t res;
800 off_t ooffset;
801
802 X_LOCK (preadwritelock);
803 ooffset = lseek (fd, 0, SEEK_CUR);
804 lseek (fd, offset, SEEK_SET);
805 res = write (fd, buf, count);
806 lseek (fd, ooffset, SEEK_SET);
807 X_UNLOCK (preadwritelock);
808
809 return res;
810}
811#endif
812
813#ifndef HAVE_FUTIMES 557#ifndef HAVE_UTIMES
814 558
815# undef utimes 559# undef utimes
816# undef futimes
817# define utimes(path,times) eio__utimes (path, times) 560# define utimes(path,times) eio__utimes (path, times)
818# define futimes(fd,times) eio__futimes (fd, times)
819 561
820static int 562static int
821eio__utimes (const char *filename, const struct timeval times[2]) 563eio__utimes (const char *filename, const struct timeval times[2])
822{ 564{
823 if (times) 565 if (times)
… …
831 } 573 }
832 else 574 else
833 return utime (filename, 0); 575 return utime (filename, 0);
834} 576}
835 577
578#endif
579
580#ifndef HAVE_FUTIMES
581
582# undef futimes
583# define futimes(fd,times) eio__futimes (fd, times)
584
585static int
836static int eio__futimes (int fd, const struct timeval tv[2]) 586eio__futimes (int fd, const struct timeval tv[2])
837{ 587{
838 errno = ENOSYS; 588 errno = ENOSYS;
839 return -1; 589 return -1;
840} 590}
841 591
… …
844#if !HAVE_FDATASYNC 594#if !HAVE_FDATASYNC
845# undef fdatasync 595# undef fdatasync
846# define fdatasync(fd) fsync (fd) 596# define fdatasync(fd) fsync (fd)
847#endif 597#endif
848 598
599static int
600eio__syncfs (int fd)
601{
602 int res;
603
604#if HAVE_SYS_SYNCFS
605 res = (int)syscall (__NR_syncfs, (int)(fd));
606#else
607 res = EIO_ENOSYS ();
608#endif
609
610 if (res < 0 && errno == ENOSYS && fd >= 0)
611 sync ();
612
613 return res;
614}
615
849/* sync_file_range always needs emulation */ 616/* sync_file_range always needs emulation */
850int 617static int
851eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags) 618eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
852{ 619{
853#if HAVE_SYNC_FILE_RANGE 620#if HAVE_SYNC_FILE_RANGE
854 int res; 621 int res;
855 622
… …
872 /* even though we could play tricks with the flags, it's better to always 639 /* even though we could play tricks with the flags, it's better to always
873 * call fdatasync, as that matches the expectation of its users best */ 640 * call fdatasync, as that matches the expectation of its users best */
874 return fdatasync (fd); 641 return fdatasync (fd);
875} 642}
876 643
644static int
645eio__fallocate (int fd, int mode, off_t offset, size_t len)
646{
647#if HAVE_LINUX_FALLOCATE
648 return fallocate (fd, mode, offset, len);
649#else
650 return EIO_ENOSYS ();
651#endif
652}
653
877#if !HAVE_READAHEAD 654#if !HAVE_READAHEAD
878# undef readahead 655# undef readahead
879# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 656# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
880 657
881static ssize_t 658static eio_ssize_t
882eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) 659eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
883{ 660{
884 size_t todo = count; 661 size_t todo = count;
885 dBUF; 662 dBUF;
886 663
… …
891 pread (fd, eio_buf, len, offset); 668 pread (fd, eio_buf, len, offset);
892 offset += len; 669 offset += len;
893 todo -= len; 670 todo -= len;
894 } 671 }
895 672
896 errno = 0; 673 FUBd;
674
675 /* linux's readahead basically only fails for EBADF or EINVAL (not mmappable) */
676 /* but not for e.g. EIO or eof, so we also never fail */
897 return count; 677 return 0;
898} 678}
899 679
900#endif 680#endif
901 681
902/* sendfile always needs emulation */ 682/* sendfile always needs emulation */
903static ssize_t 683static eio_ssize_t
904eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) 684eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
905{ 685{
686 eio_ssize_t written = 0;
906 ssize_t res; 687 eio_ssize_t res;
907 688
908 if (!count) 689 if (!count)
909 return 0; 690 return 0;
910 691
692 for (;;)
693 {
694#ifdef __APPLE__
695# undef HAVE_SENDFILE /* broken, as everything on os x */
696#endif
911#if HAVE_SENDFILE 697#if HAVE_SENDFILE
912# if __linux 698# if __linux
699 off_t soffset = offset;
913 res = sendfile (ofd, ifd, &offset, count); 700 res = sendfile (ofd, ifd, &soffset, count);
914 701
915# elif __FreeBSD__ 702# elif __FreeBSD__
916 /* 703 /*
917 * Of course, the freebsd sendfile is a dire hack with no thoughts 704 * Of course, the freebsd sendfile is a dire hack with no thoughts
918 * wasted on making it similar to other I/O functions. 705 * wasted on making it similar to other I/O functions.
919 */ 706 */
920 {
921 off_t sbytes; 707 off_t sbytes;
922 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 708 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
923 709
924 #if 0 /* according to the manpage, this is correct, but broken behaviour */ 710 #if 0 /* according to the manpage, this is correct, but broken behaviour */
925 /* freebsd' sendfile will return 0 on success */ 711 /* freebsd' sendfile will return 0 on success */
926 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */ 712 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
927 /* not on e.g. EIO or EPIPE - sounds broken */ 713 /* not on e.g. EIO or EPIPE - sounds broken */
928 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0) 714 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
929 res = sbytes; 715 res = sbytes;
930 #endif 716 #endif
931 717
932 /* according to source inspection, this is correct, and useful behaviour */ 718 /* according to source inspection, this is correct, and useful behaviour */
933 if (sbytes) 719 if (sbytes)
934 res = sbytes; 720 res = sbytes;
935 }
936 721
937# elif defined (__APPLE__) 722# elif defined __APPLE__
938
939 {
940 off_t sbytes = count; 723 off_t sbytes = count;
941 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0); 724 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
942 725
943 /* according to the manpage, sbytes is always valid */ 726 /* according to the manpage, sbytes is always valid */
944 if (sbytes) 727 if (sbytes)
945 res = sbytes; 728 res = sbytes;
946 }
947 729
948# elif __hpux 730# elif __hpux
949 res = sendfile (ofd, ifd, offset, count, 0, 0); 731 res = sendfile (ofd, ifd, offset, count, 0, 0);
950 732
951# elif __solaris 733# elif __solaris
952 {
953 struct sendfilevec vec; 734 struct sendfilevec vec;
954 size_t sbytes; 735 size_t sbytes;
955 736
956 vec.sfv_fd = ifd; 737 vec.sfv_fd = ifd;
957 vec.sfv_flag = 0; 738 vec.sfv_flag = 0;
958 vec.sfv_off = offset; 739 vec.sfv_off = offset;
959 vec.sfv_len = count; 740 vec.sfv_len = count;
960 741
961 res = sendfilev (ofd, &vec, 1, &sbytes); 742 res = sendfilev (ofd, &vec, 1, &sbytes);
962 743
963 if (res < 0 && sbytes) 744 if (res < 0 && sbytes)
964 res = sbytes; 745 res = sbytes;
965 }
966 746
967# endif 747# endif
968 748
969#elif defined (_WIN32) 749#elif defined (_WIN32) && 0
970
971 /* does not work, just for documentation of what would need to be done */ 750 /* does not work, just for documentation of what would need to be done */
972 { 751 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
752 /* libeio guarantees that the file offset does not change, and windows */
753 /* has no way to get an independent handle to the same file description */
973 HANDLE h = TO_SOCKET (ifd); 754 HANDLE h = TO_SOCKET (ifd);
974 SetFilePointer (h, offset, 0, FILE_BEGIN); 755 SetFilePointer (h, offset, 0, FILE_BEGIN);
975 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0); 756 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
976 }
977 757
978#else 758#else
979 res = -1; 759 res = EIO_ENOSYS ();
980 errno = ENOSYS;
981#endif 760#endif
982 761
762 /* we assume sendfile can copy at least 128mb in one go */
763 if (res <= 128 * 1024 * 1024)
764 {
765 if (res > 0)
766 written += res;
767
768 if (written)
769 return written;
770
771 break;
772 }
773 else
774 {
775 /* if we requested more, then probably the kernel was lazy */
776 written += res;
777 offset += res;
778 count -= res;
779
780 if (!count)
781 return written;
782 }
783 }
784
983 if (res < 0 785 if (res < 0
984 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 786 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
985 /* BSDs */ 787 /* BSDs */
986#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */ 788#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
987 || errno == ENOTSUP 789 || errno == ENOTSUP
988#endif 790#endif
791#ifdef EOPNOTSUPP /* windows */
989 || errno == EOPNOTSUPP /* BSDs */ 792 || errno == EOPNOTSUPP /* BSDs */
793#endif
990#if __solaris 794#if __solaris
991 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 795 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
992#endif 796#endif
993 ) 797 )
994 ) 798 )
… …
998 802
999 res = 0; 803 res = 0;
1000 804
1001 while (count) 805 while (count)
1002 { 806 {
1003 ssize_t cnt; 807 eio_ssize_t cnt;
1004 808
1005 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 809 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
1006 810
1007 if (cnt <= 0) 811 if (cnt <= 0)
1008 { 812 {
… …
1020 824
1021 offset += cnt; 825 offset += cnt;
1022 res += cnt; 826 res += cnt;
1023 count -= cnt; 827 count -= cnt;
1024 } 828 }
829
830 FUBd;
1025 } 831 }
1026 832
1027 return res; 833 return res;
834}
835
836#ifdef PAGESIZE
837# define eio_pagesize() PAGESIZE
838#else
839static intptr_t
840eio_pagesize (void)
841{
842 static intptr_t page;
843
844 if (!page)
845 page = sysconf (_SC_PAGESIZE);
846
847 return page;
848}
849#endif
850
851static void
852eio_page_align (void **addr, size_t *length)
853{
854 intptr_t mask = eio_pagesize () - 1;
855
856 /* round down addr */
857 intptr_t adj = mask & (intptr_t)*addr;
858
859 *addr = (void *)((intptr_t)*addr - adj);
860 *length += adj;
861
862 /* round up length */
863 *length = (*length + mask) & ~mask;
864}
865
866#if !_POSIX_MEMLOCK
867# define eio__mlockall(a) EIO_ENOSYS ()
868#else
869
870static int
871eio__mlockall (int flags)
872{
873 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
874 extern int mallopt (int, int);
875 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
876 #endif
877
878 if (EIO_MCL_CURRENT != MCL_CURRENT
879 || EIO_MCL_FUTURE != MCL_FUTURE)
880 {
881 flags = 0
882 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
883 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
884 }
885
886 return mlockall (flags);
887}
888#endif
889
890#if !_POSIX_MEMLOCK_RANGE
891# define eio__mlock(a,b) EIO_ENOSYS ()
892#else
893
894static int
895eio__mlock (void *addr, size_t length)
896{
897 eio_page_align (&addr, &length);
898
899 return mlock (addr, length);
900}
901
902#endif
903
904#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
905# define eio__msync(a,b,c) EIO_ENOSYS ()
906#else
907
908static int
909eio__msync (void *mem, size_t len, int flags)
910{
911 eio_page_align (&mem, &len);
912
913 if (EIO_MS_ASYNC != MS_SYNC
914 || EIO_MS_INVALIDATE != MS_INVALIDATE
915 || EIO_MS_SYNC != MS_SYNC)
916 {
917 flags = 0
918 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
919 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
920 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
921 }
922
923 return msync (mem, len, flags);
924}
925
926#endif
927
928static int
929eio__mtouch (eio_req *req)
930{
931 void *mem = req->ptr2;
932 size_t len = req->size;
933 int flags = req->int1;
934
935 eio_page_align (&mem, &len);
936
937 {
938 intptr_t addr = (intptr_t)mem;
939 intptr_t end = addr + len;
940 intptr_t page = eio_pagesize ();
941
942 if (addr < end)
943 if (flags & EIO_MT_MODIFY) /* modify */
944 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
945 else
946 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
947 }
948
949 return 0;
950}
951
952/*****************************************************************************/
953/* requests implemented outside eio_execute, because they are so large */
954
955static void
956eio__lseek (eio_req *req)
957{
958 /* this usually gets optimised away completely, or your compiler sucks, */
959 /* or the whence constants really are not 0, 1, 2 */
960 int whence = req->int2 == EIO_SEEK_SET ? SEEK_SET
961 : req->int2 == EIO_SEEK_CUR ? SEEK_CUR
962 : req->int2 == EIO_SEEK_END ? SEEK_END
963 : req->int2;
964
965 req->offs = lseek (req->int1, req->offs, whence);
966 req->result = req->offs == (off_t)-1 ? -1 : 0;
967}
968
969/* result will always end up in tmpbuf, there is always space for adding a 0-byte */
970static int
971eio__realpath (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
972{
973 char *res;
974 const char *rel = path;
975 char *tmp1, *tmp2;
976#if SYMLOOP_MAX > 32
977 int symlinks = SYMLOOP_MAX;
978#else
979 int symlinks = 32;
980#endif
981
982 errno = EINVAL;
983 if (!rel)
984 return -1;
985
986 errno = ENOENT;
987 if (!*rel)
988 return -1;
989
990 res = tmpbuf_get (tmpbuf, PATH_MAX * 3);
991#ifdef _WIN32
992 if (_access (rel, 4) != 0)
993 return -1;
994
995 symlinks = GetFullPathName (rel, PATH_MAX * 3, res, 0);
996
997 errno = ENAMETOOLONG;
998 if (symlinks >= PATH_MAX * 3)
999 return -1;
1000
1001 errno = EIO;
1002 if (symlinks <= 0)
1003 return -1;
1004
1005 return symlinks;
1006
1007#else
1008 tmp1 = res + PATH_MAX;
1009 tmp2 = tmp1 + PATH_MAX;
1010
1011#if 0 /* disabled, the musl way to do things is just too racy */
1012#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1013 /* on linux we may be able to ask the kernel */
1014 {
1015 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1016
1017 if (fd >= 0)
1018 {
1019 sprintf (tmp1, "/proc/self/fd/%d", fd);
1020 req->result = readlink (tmp1, res, PATH_MAX);
1021 /* here we should probably stat the open file and the disk file, to make sure they still match */
1022 close (fd);
1023
1024 if (req->result > 0)
1025 goto done;
1026 }
1027 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1028 return -1;
1029 }
1030#endif
1031#endif
1032
1033 if (*rel != '/')
1034 {
1035 int len;
1036
1037 errno = ENOENT;
1038 if (wd == EIO_INVALID_WD)
1039 return -1;
1040
1041 if (wd == EIO_CWD)
1042 {
1043 if (!getcwd (res, PATH_MAX))
1044 return -1;
1045
1046 len = strlen (res);
1047 }
1048 else
1049 memcpy (res, wd->str, len = wd->len);
1050
1051 if (res [1]) /* only use if not / */
1052 res += len;
1053 }
1054
1055 while (*rel)
1056 {
1057 eio_ssize_t len, linklen;
1058 const char *beg = rel;
1059
1060 while (*rel && *rel != '/')
1061 ++rel;
1062
1063 len = rel - beg;
1064
1065 if (!len) /* skip slashes */
1066 {
1067 ++rel;
1068 continue;
1069 }
1070
1071 if (beg [0] == '.')
1072 {
1073 if (len == 1)
1074 continue; /* . - nop */
1075
1076 if (beg [1] == '.' && len == 2)
1077 {
1078 /* .. - back up one component, if possible */
1079
1080 while (res != tmpbuf->ptr)
1081 if (*--res == '/')
1082 break;
1083
1084 continue;
1085 }
1086 }
1087
1088 errno = ENAMETOOLONG;
1089 if (res + 1 + len + 1 >= tmp1)
1090 return -1;
1091
1092 /* copy one component */
1093 *res = '/';
1094 memcpy (res + 1, beg, len);
1095
1096 /* zero-terminate, for readlink */
1097 res [len + 1] = 0;
1098
1099 /* now check if it's a symlink */
1100 linklen = readlink (tmpbuf->ptr, tmp1, PATH_MAX);
1101
1102 if (linklen < 0)
1103 {
1104 if (errno != EINVAL)
1105 return -1;
1106
1107 /* it's a normal directory. hopefully */
1108 res += len + 1;
1109 }
1110 else
1111 {
1112 /* yay, it was a symlink - build new path in tmp2 */
1113 int rellen = strlen (rel);
1114
1115 errno = ENAMETOOLONG;
1116 if (linklen + 1 + rellen >= PATH_MAX)
1117 return -1;
1118
1119 errno = ELOOP;
1120 if (!--symlinks)
1121 return -1;
1122
1123 if (*tmp1 == '/')
1124 res = tmpbuf->ptr; /* symlink resolves to an absolute path */
1125
1126 /* we need to be careful, as rel might point into tmp2 already */
1127 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1128 tmp2 [linklen] = '/';
1129 memcpy (tmp2, tmp1, linklen);
1130
1131 rel = tmp2;
1132 }
1133 }
1134
1135 /* special case for the lone root path */
1136 if (res == tmpbuf->ptr)
1137 *res++ = '/';
1138
1139 return res - (char *)tmpbuf->ptr;
1140#endif
1028} 1141}
1029 1142
1030static signed char 1143static signed char
1031eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) 1144eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1032{ 1145{
1033 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */ 1146 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1034 : a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0; 1147 : a->inode < b->inode ? -1
1148 : a->inode > b->inode ? 1
1149 : 0;
1035} 1150}
1036 1151
1037#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0 1152#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1038 1153
1039#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */ 1154#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1040#define EIO_SORT_FAST 60 /* when to only use insertion sort */ 1155#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1041 1156
1042static void 1157static void
1043eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1158eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1044{ 1159{
1045 unsigned char bits [9 + sizeof (ino_t) * 8]; 1160 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1046 unsigned char *bit = bits; 1161 unsigned char *bit = bits;
1047 1162
1048 assert (CHAR_BIT == 8); 1163 assert (CHAR_BIT == 8);
1049 assert (sizeof (eio_dirent) * 8 < 256); 1164 assert (sizeof (eio_dirent) * 8 < 256);
1050 assert (offsetof (eio_dirent, inode)); /* we use 0 as sentinel */ 1165 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1051 assert (offsetof (eio_dirent, score)); /* we use 0 as sentinel */ 1166 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1052 1167
1053 if (size <= EIO_SORT_FAST) 1168 if (size <= EIO_SORT_FAST)
1054 return; 1169 return;
1055 1170
1056 /* first prepare an array of bits to test in our radix sort */ 1171 /* first prepare an array of bits to test in our radix sort */
1057 /* try to take endianness into account, as well as differences in ino_t sizes */ 1172 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1058 /* inode_bits must contain all inodes ORed together */ 1173 /* inode_bits must contain all inodes ORed together */
1059 /* which is used to skip bits that are 0 everywhere, which is very common */ 1174 /* which is used to skip bits that are 0 everywhere, which is very common */
1060 { 1175 {
1061 ino_t endianness; 1176 eio_ino_t endianness;
1062 int i, j; 1177 int i, j;
1063 1178
1064 /* we store the byte offset of byte n into byte n of "endianness" */ 1179 /* we store the byte offset of byte n into byte n of "endianness" */
1065 for (i = 0; i < sizeof (ino_t); ++i) 1180 for (i = 0; i < sizeof (eio_ino_t); ++i)
1066 ((unsigned char *)&endianness)[i] = i; 1181 ((unsigned char *)&endianness)[i] = i;
1067 1182
1068 *bit++ = 0; 1183 *bit++ = 0;
1069 1184
1070 for (i = 0; i < sizeof (ino_t); ++i) 1185 for (i = 0; i < sizeof (eio_ino_t); ++i)
1071 { 1186 {
1072 /* shifting off the byte offsets out of "endianness" */ 1187 /* shifting off the byte offsets out of "endianness" */
1073 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8; 1188 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1074 endianness >>= 8; 1189 endianness >>= 8;
1075 1190
1076 for (j = 0; j < 8; ++j) 1191 for (j = 0; j < 8; ++j)
1077 if (inode_bits & (((ino_t)1) << (i * 8 + j))) 1192 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1078 *bit++ = offs + j; 1193 *bit++ = offs + j;
1079 } 1194 }
1080 1195
1081 for (j = 0; j < 8; ++j) 1196 for (j = 0; j < 8; ++j)
1082 if (score_bits & (1 << j)) 1197 if (score_bits & (1 << j))
1083 *bit++ = offsetof (eio_dirent, score) * 8 + j; 1198 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1084 } 1199 }
1085 1200
1086 /* now actually do the sorting (a variant of MSD radix sort) */ 1201 /* now actually do the sorting (a variant of MSD radix sort) */
1087 { 1202 {
1088 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base; 1203 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1089 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end; 1204 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1090 unsigned char *bit_stk [9 + sizeof (ino_t) * 8]; 1205 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1091 int stk_idx = 0; 1206 int stk_idx = 0;
1092 1207
1093 base_stk [stk_idx] = dents; 1208 base_stk [stk_idx] = dents;
1094 end_stk [stk_idx] = dents + size; 1209 end_stk [stk_idx] = dents + size;
1095 bit_stk [stk_idx] = bit - 1; 1210 bit_stk [stk_idx] = bit - 1;
… …
1174 } 1289 }
1175 } 1290 }
1176} 1291}
1177 1292
1178static void 1293static void
1179eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1294eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1180{ 1295{
1181 if (size <= 1) 1296 if (size <= 1)
1182 return; /* our insertion sort relies on size > 0 */ 1297 return; /* our insertion sort relies on size > 0 */
1183 1298
1184 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */ 1299 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
… …
1192 1307
1193/* read a full directory */ 1308/* read a full directory */
1194static void 1309static void
1195eio__scandir (eio_req *req, etp_worker *self) 1310eio__scandir (eio_req *req, etp_worker *self)
1196{ 1311{
1197 DIR *dirp;
1198 EIO_STRUCT_DIRENT *entp;
1199 char *name, *names; 1312 char *name, *names;
1200 int namesalloc = 4096; 1313 int namesalloc = 4096 - sizeof (void *) * 4;
1201 int namesoffs = 0; 1314 int namesoffs = 0;
1202 int flags = req->int1; 1315 int flags = req->int1;
1203 eio_dirent *dents = 0; 1316 eio_dirent *dents = 0;
1204 int dentalloc = 128; 1317 int dentalloc = 128;
1205 int dentoffs = 0; 1318 int dentoffs = 0;
1206 ino_t inode_bits = 0; 1319 eio_ino_t inode_bits = 0;
1320#ifdef _WIN32
1321 HANDLE dirp;
1322 WIN32_FIND_DATA entp;
1323#else
1324 DIR *dirp;
1325 EIO_STRUCT_DIRENT *entp;
1326#endif
1207 1327
1208 req->result = -1; 1328 req->result = -1;
1209 1329
1210 if (!(flags & EIO_READDIR_DENTS)) 1330 if (!(flags & EIO_READDIR_DENTS))
1211 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); 1331 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1212 1332
1213 X_LOCK (wrklock); 1333#ifdef _WIN32
1214 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 1334 {
1335 int len = strlen ((const char *)req->ptr1);
1336 char *path = malloc (MAX_PATH);
1337 const char *fmt;
1338 const char *reqpath = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1339
1340 if (!len)
1341 fmt = "./*";
1342 else if (reqpath[len - 1] == '/' || reqpath[len - 1] == '\\')
1343 fmt = "%s*";
1344 else
1345 fmt = "%s/*";
1346
1347 _snprintf (path, MAX_PATH, fmt, reqpath);
1348 dirp = FindFirstFile (path, &entp);
1349 free (path);
1350
1351 if (dirp == INVALID_HANDLE_VALUE)
1352 {
1353 /* should steal _dosmaperr */
1354 switch (GetLastError ())
1355 {
1356 case ERROR_FILE_NOT_FOUND:
1357 req->result = 0;
1358 break;
1359
1360 case ERROR_INVALID_NAME:
1361 case ERROR_PATH_NOT_FOUND:
1362 case ERROR_NO_MORE_FILES:
1363 errno = ENOENT;
1364 break;
1365
1366 case ERROR_NOT_ENOUGH_MEMORY:
1367 errno = ENOMEM;
1368 break;
1369
1370 default:
1371 errno = EINVAL;
1372 break;
1373 }
1374
1375 return;
1376 }
1377 }
1378#else
1379 #if HAVE_AT
1380 if (req->wd)
1381 {
1382 int fd = openat (WD2FD (req->wd), req->ptr1, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1383
1384 if (fd < 0)
1385 return;
1386
1387 dirp = fdopendir (fd);
1388
1389 if (!dirp)
1390 close (fd);
1391 }
1392 else
1215 self->dirp = dirp = opendir (req->ptr1); 1393 dirp = opendir (req->ptr1);
1394 #else
1395 dirp = opendir (wd_expand (&self->tmpbuf, req->wd, req->ptr1));
1396 #endif
1397
1398 if (!dirp)
1399 return;
1400#endif
1401
1402 if (req->flags & EIO_FLAG_PTR1_FREE)
1403 free (req->ptr1);
1404
1216 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1405 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1217 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; 1406 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1218 req->ptr2 = names = malloc (namesalloc); 1407 req->ptr2 = names = malloc (namesalloc);
1219 X_UNLOCK (wrklock);
1220 1408
1221 if (dirp && names && (!flags || dents)) 1409 if (!names || (flags && !dents))
1410 return;
1411
1222 for (;;) 1412 for (;;)
1223 { 1413 {
1414 int done;
1415
1416#ifdef _WIN32
1417 done = !dirp;
1418#else
1224 errno = 0; 1419 errno = 0;
1225 entp = readdir (dirp); 1420 entp = readdir (dirp);
1421 done = !entp;
1422#endif
1226 1423
1227 if (!entp) 1424 if (done)
1228 { 1425 {
1426#ifndef _WIN32
1427 int old_errno = errno;
1428 closedir (dirp);
1429 errno = old_errno;
1430
1229 if (errno) 1431 if (errno)
1230 break; 1432 break;
1433#endif
1231 1434
1232 /* sort etc. */ 1435 /* sort etc. */
1233 req->int1 = flags; 1436 req->int1 = flags;
1234 req->result = dentoffs; 1437 req->result = dentoffs;
1235 1438
1236 if (flags & EIO_READDIR_STAT_ORDER) 1439 if (flags & EIO_READDIR_STAT_ORDER)
1237 eio_dent_sort (dents, dentoffs, 0, inode_bits); /* sort by inode exclusively */ 1440 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1238 else if (flags & EIO_READDIR_DIRS_FIRST) 1441 else if (flags & EIO_READDIR_DIRS_FIRST)
1239 if (flags & EIO_READDIR_FOUND_UNKNOWN) 1442 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1240 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */ 1443 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1241 else 1444 else
1445 {
1446 /* in this case, all is known, and we just put dirs first and sort them */
1447 eio_dirent *oth = dents + dentoffs;
1448 eio_dirent *dir = dents;
1449
1450 /* now partition dirs to the front, and non-dirs to the back */
1451 /* by walking from both sides and swapping if necessary */
1452 while (oth > dir)
1453 {
1454 if (dir->type == EIO_DT_DIR)
1455 ++dir;
1456 else if ((--oth)->type == EIO_DT_DIR)
1457 {
1458 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1459
1460 ++dir;
1461 }
1462 }
1463
1464 /* now sort the dirs only (dirs all have the same score) */
1465 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1466 }
1467
1468 break;
1469 }
1470
1471 /* now add the entry to our list(s) */
1472 name = D_NAME (entp);
1473
1474 /* skip . and .. entries */
1475 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1476 {
1477 int len = D_NAMLEN (entp) + 1;
1478
1479 while (ecb_expect_false (namesoffs + len > namesalloc))
1480 {
1481 namesalloc *= 2;
1482 req->ptr2 = names = realloc (names, namesalloc);
1483
1484 if (!names)
1485 break;
1486 }
1487
1488 memcpy (names + namesoffs, name, len);
1489
1490 if (dents)
1491 {
1492 struct eio_dirent *ent;
1493
1494 if (ecb_expect_false (dentoffs == dentalloc))
1242 { 1495 {
1243 /* in this case, all is known, and we just put dirs first and sort them */ 1496 dentalloc *= 2;
1497 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1498
1499 if (!dents)
1500 break;
1501 }
1502
1244 eio_dirent *oth = dents + dentoffs; 1503 ent = dents + dentoffs;
1245 eio_dirent *dir = dents;
1246 1504
1247 /* now partition dirs to the front, and non-dirs to the back */ 1505 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1248 /* by walking from both sides and swapping if necessary */ 1506 ent->namelen = len - 1;
1249 /* also clear score, so it doesn't influence sorting */ 1507 ent->inode = D_INO (entp);
1250 while (oth > dir) 1508
1509 inode_bits |= ent->inode;
1510
1511 switch (D_TYPE (entp))
1512 {
1513 default:
1514 ent->type = EIO_DT_UNKNOWN;
1515 flags |= EIO_READDIR_FOUND_UNKNOWN;
1516 break;
1517
1518 #ifdef DT_FIFO
1519 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1520 #endif
1521 #ifdef DT_CHR
1522 case DT_CHR: ent->type = EIO_DT_CHR; break;
1523 #endif
1524 #ifdef DT_MPC
1525 case DT_MPC: ent->type = EIO_DT_MPC; break;
1526 #endif
1527 #ifdef DT_DIR
1528 case DT_DIR: ent->type = EIO_DT_DIR; break;
1529 #endif
1530 #ifdef DT_NAM
1531 case DT_NAM: ent->type = EIO_DT_NAM; break;
1532 #endif
1533 #ifdef DT_BLK
1534 case DT_BLK: ent->type = EIO_DT_BLK; break;
1535 #endif
1536 #ifdef DT_MPB
1537 case DT_MPB: ent->type = EIO_DT_MPB; break;
1538 #endif
1539 #ifdef DT_REG
1540 case DT_REG: ent->type = EIO_DT_REG; break;
1541 #endif
1542 #ifdef DT_NWK
1543 case DT_NWK: ent->type = EIO_DT_NWK; break;
1544 #endif
1545 #ifdef DT_CMP
1546 case DT_CMP: ent->type = EIO_DT_CMP; break;
1547 #endif
1548 #ifdef DT_LNK
1549 case DT_LNK: ent->type = EIO_DT_LNK; break;
1550 #endif
1551 #ifdef DT_SOCK
1552 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1553 #endif
1554 #ifdef DT_DOOR
1555 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1556 #endif
1557 #ifdef DT_WHT
1558 case DT_WHT: ent->type = EIO_DT_WHT; break;
1559 #endif
1560 }
1561
1562 ent->score = 7;
1563
1564 if (flags & EIO_READDIR_DIRS_FIRST)
1565 {
1566 if (ent->type == EIO_DT_UNKNOWN)
1251 { 1567 {
1252 if (dir->type == EIO_DT_DIR) 1568 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1253 ++dir; 1569 ent->score = 1;
1254 else if ((--oth)->type == EIO_DT_DIR) 1570 else if (!strchr (name, '.')) /* absence of dots indicate likely dirs */
1255 { 1571 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1256 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1257
1258 ++dir;
1259 }
1260 } 1572 }
1261 1573 else if (ent->type == EIO_DT_DIR)
1262 /* now sort the dirs only */ 1574 ent->score = 0;
1263 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1264 } 1575 }
1265
1266 break;
1267 }
1268
1269 /* now add the entry to our list(s) */
1270 name = entp->d_name;
1271
1272 /* skip . and .. entries */
1273 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1274 {
1275 int len = D_NAMLEN (entp) + 1;
1276
1277 while (expect_false (namesoffs + len > namesalloc))
1278 {
1279 namesalloc *= 2;
1280 X_LOCK (wrklock);
1281 req->ptr2 = names = realloc (names, namesalloc);
1282 X_UNLOCK (wrklock);
1283
1284 if (!names)
1285 break;
1286 } 1576 }
1287 1577
1288 memcpy (names + namesoffs, name, len);
1289
1290 if (dents)
1291 {
1292 struct eio_dirent *ent;
1293
1294 if (expect_false (dentoffs == dentalloc))
1295 {
1296 dentalloc *= 2;
1297 X_LOCK (wrklock);
1298 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1299 X_UNLOCK (wrklock);
1300
1301 if (!dents)
1302 break;
1303 }
1304
1305 ent = dents + dentoffs;
1306
1307 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1308 ent->namelen = len - 1;
1309 ent->inode = D_INO (entp);
1310
1311 inode_bits |= ent->inode;
1312
1313 switch (D_TYPE (entp))
1314 {
1315 default:
1316 ent->type = EIO_DT_UNKNOWN;
1317 flags |= EIO_READDIR_FOUND_UNKNOWN;
1318 break;
1319
1320 #ifdef DT_FIFO
1321 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1322 #endif
1323 #ifdef DT_CHR
1324 case DT_CHR: ent->type = EIO_DT_CHR; break;
1325 #endif
1326 #ifdef DT_MPC
1327 case DT_MPC: ent->type = EIO_DT_MPC; break;
1328 #endif
1329 #ifdef DT_DIR
1330 case DT_DIR: ent->type = EIO_DT_DIR; break;
1331 #endif
1332 #ifdef DT_NAM
1333 case DT_NAM: ent->type = EIO_DT_NAM; break;
1334 #endif
1335 #ifdef DT_BLK
1336 case DT_BLK: ent->type = EIO_DT_BLK; break;
1337 #endif
1338 #ifdef DT_MPB
1339 case DT_MPB: ent->type = EIO_DT_MPB; break;
1340 #endif
1341 #ifdef DT_REG
1342 case DT_REG: ent->type = EIO_DT_REG; break;
1343 #endif
1344 #ifdef DT_NWK
1345 case DT_NWK: ent->type = EIO_DT_NWK; break;
1346 #endif
1347 #ifdef DT_CMP
1348 case DT_CMP: ent->type = EIO_DT_CMP; break;
1349 #endif
1350 #ifdef DT_LNK
1351 case DT_LNK: ent->type = EIO_DT_LNK; break;
1352 #endif
1353 #ifdef DT_SOCK
1354 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1355 #endif
1356 #ifdef DT_DOOR
1357 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1358 #endif
1359 #ifdef DT_WHT
1360 case DT_WHT: ent->type = EIO_DT_WHT; break;
1361 #endif
1362 }
1363
1364 ent->score = 7;
1365
1366 if (flags & EIO_READDIR_DIRS_FIRST)
1367 {
1368 if (ent->type == EIO_DT_UNKNOWN)
1369 {
1370 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1371 ent->score = 1;
1372 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1373 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1374 }
1375 else if (ent->type == EIO_DT_DIR)
1376 ent->score = 0;
1377 }
1378 }
1379
1380 namesoffs += len; 1578 namesoffs += len;
1381 ++dentoffs; 1579 ++dentoffs;
1382 } 1580 }
1383 1581
1384 if (EIO_CANCELLED (req)) 1582 if (EIO_CANCELLED (req))
1385 { 1583 {
1386 errno = ECANCELED; 1584 errno = ECANCELED;
1387 break; 1585 break;
1388 } 1586 }
1587
1588#ifdef _WIN32
1589 if (!FindNextFile (dirp, &entp))
1590 {
1591 FindClose (dirp);
1592 dirp = 0;
1389 } 1593 }
1594#endif
1595 }
1390} 1596}
1391 1597
1392#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) 1598/*****************************************************************************/
1393# define eio__msync(a,b,c) ((errno = ENOSYS), -1) 1599/* working directory stuff */
1394#else 1600/* various deficiencies in the posix 2008 api force us to */
1601/* keep the absolute path in string form at all times */
1602/* fuck yeah. */
1395 1603
1396int 1604#if !HAVE_AT
1397eio__msync (void *mem, size_t len, int flags) 1605
1606/* a bit like realpath, but usually faster because it doesn'T have to return */
1607/* an absolute or canonical path */
1608static const char *
1609wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
1398{ 1610{
1399 if (EIO_MS_ASYNC != MS_SYNC 1611 if (!wd || *path == '/')
1400 || EIO_MS_INVALIDATE != MS_INVALIDATE 1612 return path;
1401 || EIO_MS_SYNC != MS_SYNC) 1613
1614 if (path [0] == '.' && !path [1])
1615 return wd->str;
1616
1617 {
1618 int l1 = wd->len;
1619 int l2 = strlen (path);
1620
1621 char *res = tmpbuf_get (tmpbuf, l1 + l2 + 2);
1622
1623 memcpy (res, wd->str, l1);
1624 res [l1] = '/';
1625 memcpy (res + l1 + 1, path, l2 + 1);
1626
1627 return res;
1628 }
1629}
1630
1631#endif
1632
1633static eio_wd
1634eio__wd_open_sync (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
1635{
1636 int fd;
1637 eio_wd res;
1638 int len = eio__realpath (tmpbuf, wd, path);
1639
1640 if (len < 0)
1641 return EIO_INVALID_WD;
1642
1643#if HAVE_AT
1644 fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1645
1646 if (fd < 0)
1647 return EIO_INVALID_WD;
1648#endif
1649
1650 res = malloc (sizeof (*res) + len); /* one extra 0-byte */
1651
1652#if HAVE_AT
1653 res->fd = fd;
1654#endif
1655
1656 res->len = len;
1657 memcpy (res->str, tmpbuf->ptr, len);
1658 res->str [len] = 0;
1659
1660 return res;
1661}
1662
1663eio_wd
1664eio_wd_open_sync (eio_wd wd, const char *path)
1665{
1666 struct tmpbuf tmpbuf = { 0 };
1667 wd = eio__wd_open_sync (&tmpbuf, wd, path);
1668 free (tmpbuf.ptr);
1669
1670 return wd;
1671}
1672
1673void
1674eio_wd_close_sync (eio_wd wd)
1675{
1676 if (wd != EIO_INVALID_WD && wd != EIO_CWD)
1402 { 1677 {
1403 flags = 0 1678 #if HAVE_AT
1404 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0) 1679 close (wd->fd);
1405 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0) 1680 #endif
1406 | (flags & EIO_MS_SYNC ? MS_SYNC : 0); 1681 free (wd);
1407 } 1682 }
1408
1409 return msync (mem, len, flags);
1410} 1683}
1411 1684
1412#endif 1685#if HAVE_AT
1413 1686
1414int 1687/* they forgot these */
1415eio__mtouch (void *mem, size_t len, int flags)
1416{
1417 intptr_t addr = (intptr_t)mem;
1418 intptr_t end = addr + len;
1419#ifdef PAGESIZE
1420 const intptr_t page = PAGESIZE;
1421#else
1422 static intptr_t page;
1423 1688
1424 if (!page) 1689static int
1425 page = sysconf (_SC_PAGESIZE); 1690eio__truncateat (int dirfd, const char *path, off_t length)
1426#endif 1691{
1692 int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
1693 int res;
1427 1694
1428 /* round down to start of page, although this is probably useless */ 1695 if (fd < 0)
1429 addr &= ~(page - 1); /* assume page size is always a power of two */ 1696 return fd;
1430 1697
1431 if (addr < end) 1698 res = ftruncate (fd, length);
1432 if (flags & EIO_MT_MODIFY) /* modify */ 1699 close (fd);
1433 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
1434 else
1435 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
1436
1437 return 0; 1700 return res;
1438} 1701}
1702
1703static int
1704eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
1705{
1706 int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
1707 int res;
1708
1709 if (fd < 0)
1710 return fd;
1711
1712 res = fstatvfs (fd, buf);
1713 close (fd);
1714 return res;
1715
1716}
1717
1718#endif
1439 1719
1440/*****************************************************************************/ 1720/*****************************************************************************/
1441 1721
1442#define ALLOC(len) \ 1722#define ALLOC(len) \
1443 if (!req->ptr2) \ 1723 if (!req->ptr2) \
… …
1452 req->result = -1; \ 1732 req->result = -1; \
1453 break; \ 1733 break; \
1454 } \ 1734 } \
1455 } 1735 }
1456 1736
1737static void ecb_noinline ecb_cold
1738etp_proc_init (void)
1739{
1740#if HAVE_PRCTL_SET_NAME
1741 /* provide a more sensible "thread name" */
1742 char name[16 + 1];
1743 const int namelen = sizeof (name) - 1;
1744 int len;
1745
1746 prctl (PR_GET_NAME, (unsigned long)name, 0, 0, 0);
1747 name [namelen] = 0;
1748 len = strlen (name);
1749 strcpy (name + (len <= namelen - 4 ? len : namelen - 4), "/eio");
1750 prctl (PR_SET_NAME, (unsigned long)name, 0, 0, 0);
1751#endif
1752}
1753
1754/* TODO: move somehow to etp.c */
1457X_THREAD_PROC (etp_proc) 1755X_THREAD_PROC (etp_proc)
1458{ 1756{
1459 ETP_REQ *req; 1757 ETP_REQ *req;
1460 struct timespec ts; 1758 struct timespec ts;
1461 etp_worker *self = (etp_worker *)thr_arg; 1759 etp_worker *self = (etp_worker *)thr_arg;
1462 1760
1761 etp_proc_init ();
1762
1463 /* try to distribute timeouts somewhat randomly */ 1763 /* try to distribute timeouts somewhat evenly */
1464 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 1764 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1465 1765
1466 for (;;) 1766 for (;;)
1467 { 1767 {
1768 ts.tv_sec = 0;
1769
1468 X_LOCK (reqlock); 1770 X_LOCK (reqlock);
1469 1771
1470 for (;;) 1772 for (;;)
1471 { 1773 {
1472 self->req = req = reqq_shift (&req_queue); 1774 req = reqq_shift (&req_queue);
1473 1775
1474 if (req) 1776 if (req)
1475 break; 1777 break;
1476 1778
1779 if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
1780 {
1781 X_UNLOCK (reqlock);
1782 X_LOCK (wrklock);
1783 --started;
1784 X_UNLOCK (wrklock);
1785 goto quit;
1786 }
1787
1477 ++idle; 1788 ++idle;
1478 1789
1479 ts.tv_sec = time (0) + IDLE_TIMEOUT; 1790 if (idle <= max_idle)
1480 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 1791 /* we are allowed to idle, so do so without any timeout */
1792 X_COND_WAIT (reqwait, reqlock);
1793 else
1481 { 1794 {
1482 if (idle > max_idle) 1795 /* initialise timeout once */
1483 { 1796 if (!ts.tv_sec)
1484 --idle; 1797 ts.tv_sec = time (0) + idle_timeout;
1485 X_UNLOCK (reqlock);
1486 X_LOCK (wrklock);
1487 --started;
1488 X_UNLOCK (wrklock);
1489 goto quit;
1490 }
1491 1798
1492 /* we are allowed to idle, so do so without any timeout */
1493 X_COND_WAIT (reqwait, reqlock); 1799 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
1800 ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1494 } 1801 }
1495 1802
1496 --idle; 1803 --idle;
1497 } 1804 }
1498 1805
1499 --nready; 1806 --nready;
1500 1807
1501 X_UNLOCK (reqlock); 1808 X_UNLOCK (reqlock);
1502 1809
1503 if (req->type < 0) 1810 if (req->type == ETP_TYPE_QUIT)
1504 goto quit; 1811 goto quit;
1505 1812
1506 if (!EIO_CANCELLED (req))
1507 ETP_EXECUTE (self, req); 1813 ETP_EXECUTE (self, req);
1508 1814
1509 X_LOCK (reslock); 1815 X_LOCK (reslock);
1510 1816
1511 ++npending; 1817 ++npending;
1512 1818
1513 if (!reqq_push (&res_queue, req) && want_poll_cb) 1819 if (!reqq_push (&res_queue, req) && want_poll_cb)
1514 want_poll_cb (); 1820 want_poll_cb ();
1515 1821
1516 self->req = 0;
1517 etp_worker_clear (self); 1822 etp_worker_clear (self);
1518 1823
1519 X_UNLOCK (reslock); 1824 X_UNLOCK (reslock);
1520 } 1825 }
1521 1826
1522quit: 1827quit:
1828 free (req);
1829
1523 X_LOCK (wrklock); 1830 X_LOCK (wrklock);
1524 etp_worker_free (self); 1831 etp_worker_free (self);
1525 X_UNLOCK (wrklock); 1832 X_UNLOCK (wrklock);
1526 1833
1527 return 0; 1834 return 0;
1528} 1835}
1529 1836
1530/*****************************************************************************/ 1837/*****************************************************************************/
1531 1838
1839int ecb_cold
1532int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1840eio_init (void (*want_poll)(void), void (*done_poll)(void))
1533{ 1841{
1534 return etp_init (want_poll, done_poll); 1842 return etp_init (want_poll, done_poll);
1535} 1843}
1536 1844
1845ecb_inline void
1537static void eio_api_destroy (eio_req *req) 1846eio_api_destroy (eio_req *req)
1538{ 1847{
1539 free (req); 1848 free (req);
1540} 1849}
1541 1850
1542#define REQ(rtype) \ 1851#define REQ(rtype) \
1543 eio_req *req; \ 1852 eio_req *req; \
1544 \ 1853 \
1545 req = (eio_req *)calloc (1, sizeof *req); \ 1854 req = (eio_req *)calloc (1, sizeof *req); \
1546 if (!req) \ 1855 if (!req) \
1547 return 0; \ 1856 return 0; \
… …
1561 { \ 1870 { \
1562 eio_api_destroy (req); \ 1871 eio_api_destroy (req); \
1563 return 0; \ 1872 return 0; \
1564 } 1873 }
1565 1874
1875#define SINGLEDOT(ptr) (0[(char *)(ptr)] == '.' && !1[(char *)(ptr)])
1876
1877static void
1566static void eio_execute (etp_worker *self, eio_req *req) 1878eio_execute (etp_worker *self, eio_req *req)
1567{ 1879{
1568 errno = 0; 1880#if HAVE_AT
1881 int dirfd;
1882#else
1883 const char *path;
1884#endif
1885
1886 if (ecb_expect_false (EIO_CANCELLED (req)))
1887 {
1888 req->result = -1;
1889 req->errorno = ECANCELED;
1890 return;
1891 }
1892
1893 if (ecb_expect_false (req->wd == EIO_INVALID_WD))
1894 {
1895 req->result = -1;
1896 req->errorno = ENOENT;
1897 return;
1898 }
1899
1900 if (req->type >= EIO_OPEN)
1901 {
1902 #if HAVE_AT
1903 dirfd = WD2FD (req->wd);
1904 #else
1905 path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1906 #endif
1907 }
1569 1908
1570 switch (req->type) 1909 switch (req->type)
1571 { 1910 {
1911 case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
1912 req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
1913 break;
1914 case EIO_WD_CLOSE: req->result = 0;
1915 eio_wd_close_sync (req->wd); break;
1916
1917 case EIO_SEEK: eio__lseek (req); break;
1572 case EIO_READ: ALLOC (req->size); 1918 case EIO_READ: ALLOC (req->size);
1573 req->result = req->offs >= 0 1919 req->result = req->offs >= 0
1574 ? pread (req->int1, req->ptr2, req->size, req->offs) 1920 ? pread (req->int1, req->ptr2, req->size, req->offs)
1575 : read (req->int1, req->ptr2, req->size); break; 1921 : read (req->int1, req->ptr2, req->size); break;
1576 case EIO_WRITE: req->result = req->offs >= 0 1922 case EIO_WRITE: req->result = req->offs >= 0
1577 ? pwrite (req->int1, req->ptr2, req->size, req->offs) 1923 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1578 : write (req->int1, req->ptr2, req->size); break; 1924 : write (req->int1, req->ptr2, req->size); break;
1579 1925
1580 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break; 1926 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1581 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break; 1927 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
1928
1929#if HAVE_AT
1582 1930
1583 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1931 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1584 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 1932 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break;
1585 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1933 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1586 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 1934 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, AT_SYMLINK_NOFOLLOW); break;
1587 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1935 case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break;
1588 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; 1936 case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break;
1937 case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break;
1938 case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break;
1589 1939
1940 case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
1941 case EIO_RMDIR: /* complications arise because "." cannot be removed, so we might have to expand */
1942 req->result = req->wd && SINGLEDOT (req->ptr1)
1943 ? rmdir (req->wd->str)
1944 : unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break;
1945 case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); break;
1946 case EIO_RENAME: /* complications arise because "." cannot be renamed, so we might have to expand */
1947 req->result = req->wd && SINGLEDOT (req->ptr1)
1948 ? rename (req->wd->str, req->ptr2)
1949 : renameat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2); break;
1950 case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2, 0); break;
1951 case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
1952 case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1953 case EIO_READLINK: ALLOC (PATH_MAX);
1954 req->result = readlinkat (dirfd, req->ptr1, req->ptr2, PATH_MAX); break;
1590 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS)); 1955 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1591 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break; 1956 req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1592 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1593 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1594
1595 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1596 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1597 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1598 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1599 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1600 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1601
1602 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
1603 case EIO_CLOSE: req->result = close (req->int1); break;
1604 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1605 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
1606 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1607 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1608 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1609 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1610 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1611 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break;
1612
1613 case EIO_READLINK: ALLOC (NAME_MAX);
1614 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
1615
1616 case EIO_SYNC: req->result = 0; sync (); break;
1617 case EIO_FSYNC: req->result = fsync (req->int1); break;
1618 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1619 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1620 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;
1621 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1622
1623 case EIO_READDIR: eio__scandir (req, self); break;
1624
1625 case EIO_BUSY: 1957 case EIO_UTIME:
1626#ifdef _WIN32 1958 case EIO_FUTIME:
1627 Sleep (req->nv1 * 1000.); 1959 {
1960 struct timespec ts[2];
1961 struct timespec *times;
1962
1963 if (req->nv1 != -1. || req->nv2 != -1.)
1964 {
1965 ts[0].tv_sec = req->nv1;
1966 ts[0].tv_nsec = (req->nv1 - ts[0].tv_sec) * 1e9;
1967 ts[1].tv_sec = req->nv2;
1968 ts[1].tv_nsec = (req->nv2 - ts[1].tv_sec) * 1e9;
1969
1970 times = ts;
1971 }
1972 else
1973 times = 0;
1974
1975 req->result = req->type == EIO_FUTIME
1976 ? futimens (req->int1, times)
1977 : utimensat (dirfd, req->ptr1, times, 0);
1978 }
1979 break;
1980
1628#else 1981#else
1629 {
1630 struct timeval tv;
1631 1982
1632 tv.tv_sec = req->nv1; 1983 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1633 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; 1984 req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
1985 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1986 req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
1987 case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
1988 case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
1989 case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
1990 case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
1634 1991
1635 req->result = select (0, 0, 0, 0, &tv); 1992 case EIO_UNLINK: req->result = unlink (path ); break;
1636 } 1993 case EIO_RMDIR: req->result = rmdir (path ); break;
1637#endif 1994 case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
1638 break; 1995 case EIO_RENAME: req->result = rename (path , req->ptr2); break;
1996 case EIO_LINK: req->result = link (path , req->ptr2); break;
1997 case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
1998 case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
1999 case EIO_READLINK: ALLOC (PATH_MAX);
2000 req->result = readlink (path, req->ptr2, PATH_MAX); break;
2001 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2002 req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
1639 2003
1640 case EIO_UTIME: 2004 case EIO_UTIME:
1641 case EIO_FUTIME: 2005 case EIO_FUTIME:
1642 { 2006 {
1643 struct timeval tv[2]; 2007 struct timeval tv[2];
1644 struct timeval *times; 2008 struct timeval *times;
1645 2009
1646 if (req->nv1 != -1. || req->nv2 != -1.) 2010 if (req->nv1 != -1. || req->nv2 != -1.)
1647 { 2011 {
1648 tv[0].tv_sec = req->nv1; 2012 tv[0].tv_sec = req->nv1;
1649 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.; 2013 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
1650 tv[1].tv_sec = req->nv2; 2014 tv[1].tv_sec = req->nv2;
1651 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.; 2015 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
1652 2016
1653 times = tv; 2017 times = tv;
1654 } 2018 }
1655 else 2019 else
1656 times = 0; 2020 times = 0;
1657
1658 2021
1659 req->result = req->type == EIO_FUTIME 2022 req->result = req->type == EIO_FUTIME
1660 ? futimes (req->int1, times) 2023 ? futimes (req->int1, times)
1661 : utimes (req->ptr1, times); 2024 : utimes (req->ptr1, times);
1662 } 2025 }
1663 break; 2026 break;
1664 2027
2028#endif
2029
2030 case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
2031 {
2032 ALLOC (req->result);
2033 memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
2034 }
2035 break;
2036
2037 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2038 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
2039
2040 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2041 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2042
2043 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
2044 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
2045 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
2046
2047 case EIO_CLOSE: req->result = close (req->int1); break;
2048 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
2049 case EIO_SYNC: req->result = 0; sync (); break;
2050 case EIO_FSYNC: req->result = fsync (req->int1); break;
2051 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
2052 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
2053 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
2054 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
2055 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
2056 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
2057 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
2058 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
2059
2060 case EIO_READDIR: eio__scandir (req, self); break;
2061
2062 case EIO_BUSY:
2063#ifdef _WIN32
2064 Sleep (req->nv1 * 1e3);
2065#else
2066 {
2067 struct timeval tv;
2068
2069 tv.tv_sec = req->nv1;
2070 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
2071
2072 req->result = select (0, 0, 0, 0, &tv);
2073 }
2074#endif
2075 break;
2076
1665 case EIO_GROUP: 2077 case EIO_GROUP:
1666 abort (); /* handled in eio_request */ 2078 abort (); /* handled in eio_request */
1667 2079
1668 case EIO_NOP: 2080 case EIO_NOP:
1669 req->result = 0; 2081 req->result = 0;
1670 break; 2082 break;
1671 2083
1672 case EIO_CUSTOM: 2084 case EIO_CUSTOM:
1673 ((void (*)(eio_req *))req->feed) (req); 2085 req->feed (req);
1674 break; 2086 break;
1675 2087
1676 default: 2088 default:
1677 req->result = -1; 2089 req->result = EIO_ENOSYS ();
1678 break; 2090 break;
1679 } 2091 }
1680 2092
1681 req->errorno = errno; 2093 req->errorno = errno;
1682} 2094}
1683 2095
1684#ifndef EIO_NO_WRAPPERS 2096#ifndef EIO_NO_WRAPPERS
1685 2097
2098eio_req *eio_wd_open (const char *path, int pri, eio_cb cb, void *data)
2099{
2100 REQ (EIO_WD_OPEN); PATH; SEND;
2101}
2102
2103eio_req *eio_wd_close (eio_wd wd, int pri, eio_cb cb, void *data)
2104{
2105 REQ (EIO_WD_CLOSE); req->wd = wd; SEND;
2106}
2107
1686eio_req *eio_nop (int pri, eio_cb cb, void *data) 2108eio_req *eio_nop (int pri, eio_cb cb, void *data)
1687{ 2109{
1688 REQ (EIO_NOP); SEND; 2110 REQ (EIO_NOP); SEND;
1689} 2111}
1690 2112
… …
1706eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2128eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1707{ 2129{
1708 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2130 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1709} 2131}
1710 2132
2133eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2134{
2135 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2136}
2137
2138eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
2139{
2140 REQ (EIO_SYNCFS); req->int1 = fd; SEND;
2141}
2142
2143eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2144{
2145 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2146}
2147
1711eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2148eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1712{ 2149{
1713 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2150 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1714} 2151}
1715 2152
1716eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) 2153eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
1717{ 2154{
1718 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; 2155 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
1719} 2156}
1720 2157
1721eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) 2158eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
1722{ 2159{
1723 REQ (EIO_FDATASYNC); req->int1 = fd; SEND; 2160 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2161}
2162
2163eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2164{
2165 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
1724} 2166}
1725 2167
1726eio_req *eio_close (int fd, int pri, eio_cb cb, void *data) 2168eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
1727{ 2169{
1728 REQ (EIO_CLOSE); req->int1 = fd; SEND; 2170 REQ (EIO_CLOSE); req->int1 = fd; SEND;
… …
1731eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data) 2173eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
1732{ 2174{
1733 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND; 2175 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
1734} 2176}
1735 2177
2178eio_req *eio_seek (int fd, off_t offset, int whence, int pri, eio_cb cb, void *data)
2179{
2180 REQ (EIO_SEEK); req->int1 = fd; req->offs = offset; req->int2 = whence; SEND;
2181}
2182
1736eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data) 2183eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1737{ 2184{
1738 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND; 2185 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1739} 2186}
1740 2187
… …
1766eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) 2213eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1767{ 2214{
1768 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2215 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1769} 2216}
1770 2217
1771eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2218eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1772{ 2219{
1773 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2220 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1774} 2221}
1775 2222
1776eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) 2223eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
… …
1796eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data) 2243eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1797{ 2244{
1798 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2245 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1799} 2246}
1800 2247
1801eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2248eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1802{ 2249{
1803 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2250 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1804} 2251}
1805 2252
1806eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) 2253eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
… …
1822eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data) 2269eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1823{ 2270{
1824 return eio__1path (EIO_READLINK, path, pri, cb, data); 2271 return eio__1path (EIO_READLINK, path, pri, cb, data);
1825} 2272}
1826 2273
2274eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2275{
2276 return eio__1path (EIO_REALPATH, path, pri, cb, data);
2277}
2278
1827eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) 2279eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1828{ 2280{
1829 return eio__1path (EIO_STAT, path, pri, cb, data); 2281 return eio__1path (EIO_STAT, path, pri, cb, data);
1830} 2282}
1831 2283
… …
1854 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND; 2306 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1855} 2307}
1856 2308
1857eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) 2309eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1858{ 2310{
1859 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; 2311 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1860} 2312}
1861 2313
1862static eio_req * 2314static eio_req *
1863eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2315eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1864{ 2316{
… …
1888eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2340eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1889{ 2341{
1890 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2342 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1891} 2343}
1892 2344
1893eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) 2345eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
1894{ 2346{
1895 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND; 2347 REQ (EIO_CUSTOM); req->feed = execute; SEND;
1896} 2348}
1897 2349
1898#endif 2350#endif
1899 2351
1900eio_req *eio_grp (eio_cb cb, void *data) 2352eio_req *eio_grp (eio_cb cb, void *data)
… …
1909#undef SEND 2361#undef SEND
1910 2362
1911/*****************************************************************************/ 2363/*****************************************************************************/
1912/* grp functions */ 2364/* grp functions */
1913 2365
2366void
1914void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2367eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1915{ 2368{
1916 grp->int2 = limit; 2369 grp->int2 = limit;
1917 grp->feed = feed; 2370 grp->feed = feed;
1918 2371
1919 grp_try_feed (grp); 2372 grp_try_feed (grp);
1920} 2373}
1921 2374
2375void
1922void eio_grp_limit (eio_req *grp, int limit) 2376eio_grp_limit (eio_req *grp, int limit)
1923{ 2377{
1924 grp->int2 = limit; 2378 grp->int2 = limit;
1925 2379
1926 grp_try_feed (grp); 2380 grp_try_feed (grp);
1927} 2381}
1928 2382
2383void
1929void eio_grp_add (eio_req *grp, eio_req *req) 2384eio_grp_add (eio_req *grp, eio_req *req)
1930{ 2385{
1931 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2386 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1932 2387
1933 grp->flags |= EIO_FLAG_GROUPADD; 2388 grp->flags |= EIO_FLAG_GROUPADD;
1934 2389
… …
1945} 2400}
1946 2401
1947/*****************************************************************************/ 2402/*****************************************************************************/
1948/* misc garbage */ 2403/* misc garbage */
1949 2404
2405eio_ssize_t
1950ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2406eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1951{ 2407{
1952 etp_worker wrk;
1953 ssize_t ret;
1954
1955 wrk.dbuf = 0;
1956
1957 ret = eio__sendfile (ofd, ifd, offset, count, &wrk); 2408 return eio__sendfile (ofd, ifd, offset, count);
1958
1959 if (wrk.dbuf)
1960 free (wrk.dbuf);
1961
1962 return ret;
1963} 2409}
1964 2410

Diff Legend

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