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

Comparing libeio/eio.c (file contents):
Revision 1.35 by root, Sat Jun 6 20:13:46 2009 UTC vs.
Revision 1.139 by root, Thu Jun 25 18:14:19 2015 UTC

1/* 1/*
2 * libeio implementation 2 * libeio implementation
3 * 3 *
4 * Copyright (c) 2007,2008,2009 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>
… …
53#include <sys/stat.h> 58#include <sys/stat.h>
54#include <limits.h> 59#include <limits.h>
55#include <fcntl.h> 60#include <fcntl.h>
56#include <assert.h> 61#include <assert.h>
57 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
58#ifndef EIO_FINISH 82#ifndef EIO_FINISH
59# 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
60#endif 84#endif
61 85
62#ifndef EIO_DESTROY 86#ifndef EIO_DESTROY
… …
65 89
66#ifndef EIO_FEED 90#ifndef EIO_FEED
67# 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)
68#endif 92#endif
69 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
70#ifdef _WIN32 105#ifdef _WIN32
71 106
72 /*doh*/ 107 #undef PAGESIZE
108 #define PAGESIZE 4096 /* GetSystemInfo? */
109
110 /* TODO: look at how perl does stat (non-sloppy), unlink (ro-files), utime, link */
111
112 #ifdef EIO_STRUCT_STATI64
113 /* look at perl's non-sloppy stat */
114 #define stat(path,buf) _stati64 (path,buf)
115 #define fstat(fd,buf) _fstati64 (fd,buf)
116 #endif
117 #define lstat(path,buf) stat (path,buf)
118 #define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
119 #define mkdir(path,mode) _mkdir (path)
120 #define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
121
122 #define chmod(path,mode) _chmod (path, mode)
123 #define dup(fd) _dup (fd)
124 #define dup2(fd1,fd2) _dup2 (fd1, fd2)
125 #define pipe(fds) _pipe (fds, 4096, O_BINARY)
126
127 #define fchmod(fd,mode) EIO_ENOSYS ()
128 #define chown(path,uid,gid) EIO_ENOSYS ()
129 #define fchown(fd,uid,gid) EIO_ENOSYS ()
130 #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
131 #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
132 #define mknod(path,mode,dev) EIO_ENOSYS ()
133 #define sync() EIO_ENOSYS ()
134 #define readlink(path,buf,s) EIO_ENOSYS ()
135 #define statvfs(path,buf) EIO_ENOSYS ()
136 #define fstatvfs(fd,buf) EIO_ENOSYS ()
137
138 #define pread(fd,buf,count,offset) eio__pread (fd, buf, count, offset)
139 #define pwrite(fd,buf,count,offset) eio__pwrite (fd, buf, count, offset)
140
141 #if __GNUC__
142 typedef long long eio_off_t; /* signed for compatibility to msvc */
143 #else
144 typedef __int64 eio_off_t; /* unsigned not supported by msvc */
145 #endif
146
147 static eio_ssize_t
148 eio__pread (int fd, void *buf, eio_ssize_t count, eio_off_t offset)
149 {
150 OVERLAPPED o = { 0 };
151 DWORD got;
152
153 o.Offset = offset;
154 o.OffsetHigh = offset >> 32;
155
156 return ReadFile ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd), buf, count, &got, &o)
157 ? got : -1;
158 }
159
160 static eio_ssize_t
161 eio__pwrite (int fd, void *buf, eio_ssize_t count, eio_off_t offset)
162 {
163 OVERLAPPED o = { 0 };
164 DWORD got;
165
166 o.Offset = offset;
167 o.OffsetHigh = offset >> 32;
168
169 return WriteFile ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd), buf, count, &got, &o)
170 ? got : -1;
171 }
172
173 /* rename() uses MoveFile, which fails to overwrite */
174 #define rename(old,neu) eio__rename (old, neu)
175
176 static int
177 eio__rename (const char *old, const char *neu)
178 {
179 if (MoveFileEx (old, neu, MOVEFILE_REPLACE_EXISTING))
180 return 0;
181
182 /* should steal _dosmaperr */
183 switch (GetLastError ())
184 {
185 case ERROR_FILE_NOT_FOUND:
186 case ERROR_PATH_NOT_FOUND:
187 case ERROR_INVALID_DRIVE:
188 case ERROR_NO_MORE_FILES:
189 case ERROR_BAD_NETPATH:
190 case ERROR_BAD_NET_NAME:
191 case ERROR_BAD_PATHNAME:
192 case ERROR_FILENAME_EXCED_RANGE:
193 errno = ENOENT;
194 break;
195
196 default:
197 errno = EACCES;
198 break;
199 }
200
201 return -1;
202 }
203
204 /* we could even stat and see if it exists */
205 static int
206 symlink (const char *old, const char *neu)
207 {
208 #if WINVER >= 0x0600
209 if (CreateSymbolicLink (neu, old, 1))
210 return 0;
211
212 if (CreateSymbolicLink (neu, old, 0))
213 return 0;
214 #endif
215
216 return EIO_ERRNO (ENOENT, -1);
217 }
218
219 /* POSIX API only, causing trouble for win32 apps */
220 #define CreateHardLink(neu,old,flags) 0 /* not really creating hardlink, still using relative paths? */
221 #define CreateSymbolicLink(neu,old,flags) 0 /* vista+ only */
222
223 struct statvfs
224 {
225 int dummy;
226 };
227
228 #define DT_DIR EIO_DT_DIR
229 #define DT_REG EIO_DT_REG
230 #define D_NAME(entp) entp.cFileName
231 #define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
232
73#else 233#else
74 234
75# include "config.h"
76# include <sys/time.h> 235 #include <sys/time.h>
77# include <sys/select.h> 236 #include <sys/select.h>
78# include <unistd.h> 237 #include <unistd.h>
238 #include <signal.h>
239 #include <dirent.h>
240
241 #ifdef ANDROID
242 #include <sys/vfs.h>
243 #define statvfs statfs
244 #define fstatvfs fstatfs
245 #include <asm/page.h> /* supposedly limits.h does #define PAGESIZE PAGESIZE */
246 #else
247 #include <sys/statvfs.h>
248 #endif
249
250 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
251 #include <sys/mman.h>
252 #endif
253
254 #define D_NAME(entp) entp->d_name
255
256 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
257 #if __FreeBSD__ || __NetBSD__ || __OpenBSD__
258 #define _DIRENT_HAVE_D_TYPE /* sigh */
259 #define D_INO(de) (de)->d_fileno
260 #define D_NAMLEN(de) (de)->d_namlen
261 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
262 #define D_INO(de) (de)->d_ino
263 #endif
264
265 #ifdef _D_EXACT_NAMLEN
266 #undef D_NAMLEN
267 #define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
268 #endif
269
270 #ifdef _DIRENT_HAVE_D_TYPE
271 #define D_TYPE(de) (de)->d_type
272 #endif
273
274 #ifndef EIO_STRUCT_DIRENT
275 #define EIO_STRUCT_DIRENT struct dirent
276 #endif
277
278#endif
279
280#if HAVE_UTIMES
79# include <utime.h> 281# include <utime.h>
80# include <signal.h>
81# include <dirent.h>
82
83/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
84# if defined(__FreeBSD__) || defined(__NetBSD__) || defined(__OpenBSD__)
85# define _DIRENT_HAVE_D_TYPE /* sigh */
86# define D_INO(de) (de)->d_fileno
87# define D_NAMLEN(de) (de)->d_namlen
88# elif defined(__linux) || defined(d_ino) || _XOPEN_SOURCE >= 600
89# define D_INO(de) (de)->d_ino
90# endif 282#endif
91 283
92#ifdef _D_EXACT_NAMLEN 284#if HAVE_SYS_SYSCALL_H
93# undef D_NAMLEN 285# include <sys/syscall.h>
94# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
95#endif
96
97# ifdef _DIRENT_HAVE_D_TYPE
98# define D_TYPE(de) (de)->d_type
99# endif 286#endif
100 287
101# ifndef EIO_STRUCT_DIRENT 288#if HAVE_SYS_PRCTL_H
102# define EIO_STRUCT_DIRENT struct dirent 289# include <sys/prctl.h>
103# endif
104
105#endif 290#endif
106 291
107#if HAVE_SENDFILE 292#if HAVE_SENDFILE
108# if __linux 293# if __linux
109# include <sys/sendfile.h> 294# include <sys/sendfile.h>
110# elif __freebsd 295# elif __FreeBSD__ || defined __APPLE__
111# include <sys/socket.h> 296# include <sys/socket.h>
112# include <sys/uio.h> 297# include <sys/uio.h>
113# elif __hpux 298# elif __hpux
114# include <sys/socket.h> 299# include <sys/socket.h>
115# elif __solaris /* not yet */ 300# elif __solaris
116# include <sys/sendfile.h> 301# include <sys/sendfile.h>
117# else 302# else
118# error sendfile support requested but not available 303# error sendfile support requested but not available
119# endif 304# endif
120#endif 305#endif
… …
124#endif 309#endif
125#ifndef D_INO 310#ifndef D_INO
126# define D_INO(de) 0 311# define D_INO(de) 0
127#endif 312#endif
128#ifndef D_NAMLEN 313#ifndef D_NAMLEN
129# define D_NAMLEN(de) strlen ((de)->d_name) 314# define D_NAMLEN(entp) strlen (D_NAME (entp))
130#endif 315#endif
131
132/* number of seconds after which an idle threads exit */
133#define IDLE_TIMEOUT 10
134 316
135/* used for struct dirent, AIX doesn't provide it */ 317/* used for struct dirent, AIX doesn't provide it */
136#ifndef NAME_MAX 318#ifndef NAME_MAX
137# define NAME_MAX 4096 319# define NAME_MAX 4096
138#endif 320#endif
139 321
322/* used for readlink etc. */
323#ifndef PATH_MAX
324# define PATH_MAX 4096
325#endif
326
140/* buffer size for various temporary buffers */ 327/* buffer size for various temporary buffers */
141#define EIO_BUFSIZE 65536 328#define EIO_BUFSIZE 65536
142 329
143#define dBUF \ 330#define dBUF \
144 char *eio_buf; \
145 ETP_WORKER_LOCK (self); \
146 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 331 char *eio_buf = malloc (EIO_BUFSIZE); \
147 ETP_WORKER_UNLOCK (self); \
148 errno = ENOMEM; \ 332 errno = ENOMEM; \
149 if (!eio_buf) \ 333 if (!eio_buf) \
150 return -1; 334 return -1
151 335
152#define EIO_TICKS ((1000000 + 1023) >> 10) 336#define FUBd \
337 free (eio_buf)
153 338
154/*****************************************************************************/ 339/*****************************************************************************/
155 340
156#if __GNUC__ >= 3 341struct etp_tmpbuf;
157# define expect(expr,value) __builtin_expect ((expr),(value)) 342
343#if _POSIX_VERSION >= 200809L
344 #define HAVE_AT 1
345 #define WD2FD(wd) ((wd) ? (wd)->fd : AT_FDCWD)
346 #ifndef O_SEARCH
347 #define O_SEARCH O_RDONLY
348 #endif
158#else 349#else
159# define expect(expr,value) (expr) 350 #define HAVE_AT 0
351 static const char *wd_expand (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path);
160#endif 352#endif
161 353
162#define expect_false(expr) expect ((expr) != 0, 0) 354struct eio_pwd
163#define expect_true(expr) expect ((expr) != 0, 1) 355{
356#if HAVE_AT
357 int fd;
358#endif
359 int len;
360 char str[1]; /* actually, a 0-terminated canonical path */
361};
164 362
165/*****************************************************************************/ 363/*****************************************************************************/
166 364
167#define ETP_PRI_MIN EIO_PRI_MIN 365#define ETP_PRI_MIN EIO_PRI_MIN
168#define ETP_PRI_MAX EIO_PRI_MAX 366#define ETP_PRI_MAX EIO_PRI_MAX
169 367
368#define ETP_TYPE_QUIT -1
369#define ETP_TYPE_GROUP EIO_GROUP
370
371static void eio_nop_callback (void) { }
372static void (*eio_want_poll_cb)(void) = eio_nop_callback;
373static void (*eio_done_poll_cb)(void) = eio_nop_callback;
374
375#define ETP_WANT_POLL(pool) eio_want_poll_cb ()
376#define ETP_DONE_POLL(pool) eio_done_poll_cb ()
377
170struct etp_worker; 378struct etp_worker;
171
172#define ETP_REQ eio_req 379#define ETP_REQ eio_req
173#define ETP_DESTROY(req) eio_destroy (req) 380#define ETP_DESTROY(req) eio_destroy (req)
174static int eio_finish (eio_req *req); 381static int eio_finish (eio_req *req);
175#define ETP_FINISH(req) eio_finish (req) 382#define ETP_FINISH(req) eio_finish (req)
176static void eio_execute (struct etp_worker *self, eio_req *req); 383static void eio_execute (struct etp_worker *self, eio_req *req);
177#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) 384#define ETP_EXECUTE(wrk,req) eio_execute (wrk, req)
178 385
179#define ETP_WORKER_CLEAR(req) \ 386#include "etp.c"
180 if (wrk->dbuf) \
181 { \
182 free (wrk->dbuf); \
183 wrk->dbuf = 0; \
184 } \
185 \
186 if (wrk->dirp) \
187 { \
188 closedir (wrk->dirp); \
189 wrk->dirp = 0; \
190 }
191 387
192#define ETP_WORKER_COMMON \ 388static struct etp_pool eio_pool;
193 void *dbuf; \ 389#define EIO_POOL (&eio_pool)
194 DIR *dirp;
195 390
196/*****************************************************************************/ 391/*****************************************************************************/
197 392
198#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
199
200/* calculcate time difference in ~1/EIO_TICKS of a second */
201static int tvdiff (struct timeval *tv1, struct timeval *tv2)
202{
203 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
204 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
205}
206
207static unsigned int started, idle, wanted = 4;
208
209static void (*want_poll_cb) (void);
210static void (*done_poll_cb) (void);
211
212static unsigned int max_poll_time; /* reslock */
213static unsigned int max_poll_reqs; /* reslock */
214
215static volatile unsigned int nreqs; /* reqlock */
216static volatile unsigned int nready; /* reqlock */
217static volatile unsigned int npending; /* reqlock */
218static volatile unsigned int max_idle = 4;
219
220static mutex_t wrklock = X_MUTEX_INIT;
221static mutex_t reslock = X_MUTEX_INIT;
222static mutex_t reqlock = X_MUTEX_INIT;
223static cond_t reqwait = X_COND_INIT;
224
225#if !HAVE_PREADWRITE
226/*
227 * make our pread/pwrite emulation safe against themselves, but not against
228 * normal read/write by using a mutex. slows down execution a lot,
229 * but that's your problem, not mine.
230 */
231static mutex_t preadwritelock = X_MUTEX_INIT;
232#endif
233
234typedef struct etp_worker
235{
236 /* locked by wrklock */
237 struct etp_worker *prev, *next;
238
239 thread_t tid;
240
241 /* locked by reslock, reqlock or wrklock */
242 ETP_REQ *req; /* currently processed request */
243
244 ETP_WORKER_COMMON
245} etp_worker;
246
247static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */
248
249#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
250#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
251
252/* worker threads management */
253
254static void etp_worker_clear (etp_worker *wrk)
255{
256 ETP_WORKER_CLEAR (wrk);
257}
258
259static void etp_worker_free (etp_worker *wrk)
260{
261 wrk->next->prev = wrk->prev;
262 wrk->prev->next = wrk->next;
263
264 free (wrk);
265}
266
267static unsigned int etp_nreqs (void)
268{
269 int retval;
270 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
271 retval = nreqs;
272 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
273 return retval;
274}
275
276static unsigned int etp_nready (void)
277{
278 unsigned int retval;
279
280 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
281 retval = nready;
282 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
283
284 return retval;
285}
286
287static unsigned int etp_npending (void)
288{
289 unsigned int retval;
290
291 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
292 retval = npending;
293 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
294
295 return retval;
296}
297
298static unsigned int etp_nthreads (void)
299{
300 unsigned int retval;
301
302 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
303 retval = started;
304 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
305
306 return retval;
307}
308
309/*
310 * a somewhat faster data structure might be nice, but
311 * with 8 priorities this actually needs <20 insns
312 * per shift, the most expensive operation.
313 */
314typedef struct {
315 ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */
316 int size;
317} etp_reqq;
318
319static etp_reqq req_queue;
320static etp_reqq res_queue;
321
322static int reqq_push (etp_reqq *q, ETP_REQ *req)
323{
324 int pri = req->pri;
325 req->next = 0;
326
327 if (q->qe[pri])
328 {
329 q->qe[pri]->next = req;
330 q->qe[pri] = req;
331 }
332 else
333 q->qe[pri] = q->qs[pri] = req;
334
335 return q->size++;
336}
337
338static ETP_REQ *reqq_shift (etp_reqq *q)
339{
340 int pri;
341
342 if (!q->size)
343 return 0;
344
345 --q->size;
346
347 for (pri = ETP_NUM_PRI; pri--; )
348 {
349 eio_req *req = q->qs[pri];
350
351 if (req)
352 {
353 if (!(q->qs[pri] = (eio_req *)req->next))
354 q->qe[pri] = 0;
355
356 return req;
357 }
358 }
359
360 abort ();
361}
362
363static void etp_atfork_prepare (void)
364{
365 X_LOCK (wrklock);
366 X_LOCK (reqlock);
367 X_LOCK (reslock);
368#if !HAVE_PREADWRITE
369 X_LOCK (preadwritelock);
370#endif
371}
372
373static void etp_atfork_parent (void)
374{
375#if !HAVE_PREADWRITE
376 X_UNLOCK (preadwritelock);
377#endif
378 X_UNLOCK (reslock);
379 X_UNLOCK (reqlock);
380 X_UNLOCK (wrklock);
381}
382
383static void etp_atfork_child (void)
384{
385 ETP_REQ *prv;
386
387 while ((prv = reqq_shift (&req_queue)))
388 ETP_DESTROY (prv);
389
390 while ((prv = reqq_shift (&res_queue)))
391 ETP_DESTROY (prv);
392
393 while (wrk_first.next != &wrk_first)
394 {
395 etp_worker *wrk = wrk_first.next;
396
397 if (wrk->req)
398 ETP_DESTROY (wrk->req);
399
400 etp_worker_clear (wrk);
401 etp_worker_free (wrk);
402 }
403
404 started = 0;
405 idle = 0;
406 nreqs = 0;
407 nready = 0;
408 npending = 0;
409
410 etp_atfork_parent ();
411}
412
413static void 393static void
414etp_once_init (void)
415{
416 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
417}
418
419static int
420etp_init (void (*want_poll)(void), void (*done_poll)(void))
421{
422 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
423
424 pthread_once (&doinit, etp_once_init);
425
426 want_poll_cb = want_poll;
427 done_poll_cb = done_poll;
428
429 return 0;
430}
431
432X_THREAD_PROC (etp_proc);
433
434static void etp_start_thread (void)
435{
436 etp_worker *wrk = calloc (1, sizeof (etp_worker));
437
438 /*TODO*/
439 assert (("unable to allocate worker thread data", wrk));
440
441 X_LOCK (wrklock);
442
443 if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
444 {
445 wrk->prev = &wrk_first;
446 wrk->next = wrk_first.next;
447 wrk_first.next->prev = wrk;
448 wrk_first.next = wrk;
449 ++started;
450 }
451 else
452 free (wrk);
453
454 X_UNLOCK (wrklock);
455}
456
457static void etp_maybe_start_thread (void)
458{
459 if (expect_true (etp_nthreads () >= wanted))
460 return;
461
462 /* todo: maybe use idle here, but might be less exact */
463 if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
464 return;
465
466 etp_start_thread ();
467}
468
469static void etp_end_thread (void)
470{
471 eio_req *req = calloc (1, sizeof (eio_req));
472
473 req->type = -1;
474 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
475
476 X_LOCK (reqlock);
477 reqq_push (&req_queue, req);
478 X_COND_SIGNAL (reqwait);
479 X_UNLOCK (reqlock);
480
481 X_LOCK (wrklock);
482 --started;
483 X_UNLOCK (wrklock);
484}
485
486static int etp_poll (void)
487{
488 unsigned int maxreqs;
489 unsigned int maxtime;
490 struct timeval tv_start, tv_now;
491
492 X_LOCK (reslock);
493 maxreqs = max_poll_reqs;
494 maxtime = max_poll_time;
495 X_UNLOCK (reslock);
496
497 if (maxtime)
498 gettimeofday (&tv_start, 0);
499
500 for (;;)
501 {
502 ETP_REQ *req;
503
504 etp_maybe_start_thread ();
505
506 X_LOCK (reslock);
507 req = reqq_shift (&res_queue);
508
509 if (req)
510 {
511 --npending;
512
513 if (!res_queue.size && done_poll_cb)
514 done_poll_cb ();
515 }
516
517 X_UNLOCK (reslock);
518
519 if (!req)
520 return 0;
521
522 X_LOCK (reqlock);
523 --nreqs;
524 X_UNLOCK (reqlock);
525
526 if (expect_false (req->type == EIO_GROUP && req->size))
527 {
528 req->int1 = 1; /* mark request as delayed */
529 continue;
530 }
531 else
532 {
533 int res = ETP_FINISH (req);
534 if (expect_false (res))
535 return res;
536 }
537
538 if (expect_false (maxreqs && !--maxreqs))
539 break;
540
541 if (maxtime)
542 {
543 gettimeofday (&tv_now, 0);
544
545 if (tvdiff (&tv_start, &tv_now) >= maxtime)
546 break;
547 }
548 }
549
550 errno = EAGAIN;
551 return -1;
552}
553
554static void etp_cancel (ETP_REQ *req)
555{
556 X_LOCK (wrklock);
557 req->flags |= EIO_FLAG_CANCELLED;
558 X_UNLOCK (wrklock);
559
560 eio_grp_cancel (req);
561}
562
563static void etp_submit (ETP_REQ *req)
564{
565 req->pri -= ETP_PRI_MIN;
566
567 if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
568 if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
569
570 if (expect_false (req->type == EIO_GROUP))
571 {
572 /* I hope this is worth it :/ */
573 X_LOCK (reqlock);
574 ++nreqs;
575 X_UNLOCK (reqlock);
576
577 X_LOCK (reslock);
578
579 ++npending;
580
581 if (!reqq_push (&res_queue, req) && want_poll_cb)
582 want_poll_cb ();
583
584 X_UNLOCK (reslock);
585 }
586 else
587 {
588 X_LOCK (reqlock);
589 ++nreqs;
590 ++nready;
591 reqq_push (&req_queue, req);
592 X_COND_SIGNAL (reqwait);
593 X_UNLOCK (reqlock);
594
595 etp_maybe_start_thread ();
596 }
597}
598
599static void etp_set_max_poll_time (double nseconds)
600{
601 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
602 max_poll_time = nseconds;
603 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
604}
605
606static void etp_set_max_poll_reqs (unsigned int maxreqs)
607{
608 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
609 max_poll_reqs = maxreqs;
610 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
611}
612
613static void etp_set_max_idle (unsigned int nthreads)
614{
615 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
616 max_idle = nthreads <= 0 ? 1 : nthreads;
617 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
618}
619
620static void etp_set_min_parallel (unsigned int nthreads)
621{
622 if (wanted < nthreads)
623 wanted = nthreads;
624}
625
626static void etp_set_max_parallel (unsigned int nthreads)
627{
628 if (wanted > nthreads)
629 wanted = nthreads;
630
631 while (started > wanted)
632 etp_end_thread ();
633}
634
635/*****************************************************************************/
636
637static void grp_try_feed (eio_req *grp) 394grp_try_feed (eio_req *grp)
638{ 395{
639 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 396 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
640 { 397 {
641 grp->flags &= ~EIO_FLAG_GROUPADD; 398 grp->flags &= ~ETP_FLAG_GROUPADD;
642 399
643 EIO_FEED (grp); 400 EIO_FEED (grp);
644 401
645 /* stop if no progress has been made */ 402 /* stop if no progress has been made */
646 if (!(grp->flags & EIO_FLAG_GROUPADD)) 403 if (!(grp->flags & ETP_FLAG_GROUPADD))
647 { 404 {
648 grp->feed = 0; 405 grp->feed = 0;
649 break; 406 break;
650 } 407 }
651 } 408 }
652} 409}
653 410
411static int
654static int grp_dec (eio_req *grp) 412grp_dec (eio_req *grp)
655{ 413{
656 --grp->size; 414 --grp->size;
657 415
658 /* call feeder, if applicable */ 416 /* call feeder, if applicable */
659 grp_try_feed (grp); 417 grp_try_feed (grp);
660 418
661 /* finish, if done */ 419 /* finish, if done */
662 if (!grp->size && grp->int1) 420 if (!grp->size && grp->flags & ETP_FLAG_DELAYED)
663 return eio_finish (grp); 421 return eio_finish (grp);
664 else 422 else
665 return 0; 423 return 0;
666} 424}
667 425
426static void
668void eio_destroy (eio_req *req) 427eio_destroy (eio_req *req)
669{ 428{
670 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); 429 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
671 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); 430 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
672 431
673 EIO_DESTROY (req); 432 EIO_DESTROY (req);
674} 433}
675 434
435static int
676static int eio_finish (eio_req *req) 436eio_finish (eio_req *req)
677{ 437{
678 int res = EIO_FINISH (req); 438 int res = EIO_FINISH (req);
679 439
680 if (req->grp) 440 if (req->grp)
681 { 441 {
… …
689 if (grp->grp_first == req) 449 if (grp->grp_first == req)
690 grp->grp_first = req->grp_next; 450 grp->grp_first = req->grp_next;
691 451
692 res2 = grp_dec (grp); 452 res2 = grp_dec (grp);
693 453
694 if (!res && res2) 454 if (!res)
695 res = res2; 455 res = res2;
696 } 456 }
697 457
698 eio_destroy (req); 458 eio_destroy (req);
699 459
700 return res; 460 return res;
701} 461}
702 462
463void
703void eio_grp_cancel (eio_req *grp) 464eio_grp_cancel (eio_req *grp)
704{ 465{
705 for (grp = grp->grp_first; grp; grp = grp->grp_next) 466 etp_grp_cancel (EIO_POOL, grp);
706 eio_cancel (grp);
707} 467}
708 468
469void
709void eio_cancel (eio_req *req) 470eio_cancel (eio_req *req)
710{ 471{
711 etp_cancel (req); 472 etp_cancel (EIO_POOL, req);
712} 473}
713 474
475void
714void eio_submit (eio_req *req) 476eio_submit (eio_req *req)
715{ 477{
716 etp_submit (req); 478 etp_submit (EIO_POOL, req);
717} 479}
718 480
719unsigned int eio_nreqs (void) 481unsigned int
482eio_nreqs (void)
720{ 483{
721 return etp_nreqs (); 484 return etp_nreqs (EIO_POOL);
722} 485}
723 486
724unsigned int eio_nready (void) 487unsigned int
488eio_nready (void)
725{ 489{
726 return etp_nready (); 490 return etp_nready (EIO_POOL);
727} 491}
728 492
729unsigned int eio_npending (void) 493unsigned int
494eio_npending (void)
730{ 495{
731 return etp_npending (); 496 return etp_npending (EIO_POOL);
732} 497}
733 498
734unsigned int eio_nthreads (void) 499unsigned int ecb_cold
500eio_nthreads (void)
735{ 501{
736 return etp_nthreads (); 502 return etp_nthreads (EIO_POOL);
737} 503}
738 504
505void ecb_cold
739void eio_set_max_poll_time (double nseconds) 506eio_set_max_poll_time (double nseconds)
740{ 507{
741 etp_set_max_poll_time (nseconds); 508 etp_set_max_poll_time (EIO_POOL, nseconds);
742} 509}
743 510
511void ecb_cold
744void eio_set_max_poll_reqs (unsigned int maxreqs) 512eio_set_max_poll_reqs (unsigned int maxreqs)
745{ 513{
746 etp_set_max_poll_reqs (maxreqs); 514 etp_set_max_poll_reqs (EIO_POOL, maxreqs);
747} 515}
748 516
517void ecb_cold
749void eio_set_max_idle (unsigned int nthreads) 518eio_set_max_idle (unsigned int nthreads)
750{ 519{
751 etp_set_max_idle (nthreads); 520 etp_set_max_idle (EIO_POOL, nthreads);
752} 521}
753 522
523void ecb_cold
524eio_set_idle_timeout (unsigned int seconds)
525{
526 etp_set_idle_timeout (EIO_POOL, seconds);
527}
528
529void ecb_cold
754void eio_set_min_parallel (unsigned int nthreads) 530eio_set_min_parallel (unsigned int nthreads)
755{ 531{
756 etp_set_min_parallel (nthreads); 532 etp_set_min_parallel (EIO_POOL, nthreads);
757} 533}
758 534
535void ecb_cold
759void eio_set_max_parallel (unsigned int nthreads) 536eio_set_max_parallel (unsigned int nthreads)
760{ 537{
761 etp_set_max_parallel (nthreads); 538 etp_set_max_parallel (EIO_POOL, nthreads);
762} 539}
763 540
764int eio_poll (void) 541int eio_poll (void)
765{ 542{
766 return etp_poll (); 543 return etp_poll (EIO_POOL);
767} 544}
768 545
769/*****************************************************************************/ 546/*****************************************************************************/
770/* work around various missing functions */ 547/* work around various missing functions */
771 548
772#if !HAVE_PREADWRITE
773# undef pread
774# undef pwrite
775# define pread eio__pread
776# define pwrite eio__pwrite
777
778static ssize_t
779eio__pread (int fd, void *buf, size_t count, off_t offset)
780{
781 ssize_t res;
782 off_t ooffset;
783
784 X_LOCK (preadwritelock);
785 ooffset = lseek (fd, 0, SEEK_CUR);
786 lseek (fd, offset, SEEK_SET);
787 res = read (fd, buf, count);
788 lseek (fd, ooffset, SEEK_SET);
789 X_UNLOCK (preadwritelock);
790
791 return res;
792}
793
794static ssize_t
795eio__pwrite (int fd, void *buf, size_t count, off_t offset)
796{
797 ssize_t res;
798 off_t ooffset;
799
800 X_LOCK (preadwritelock);
801 ooffset = lseek (fd, 0, SEEK_CUR);
802 lseek (fd, offset, SEEK_SET);
803 res = write (fd, buf, count);
804 lseek (fd, ooffset, SEEK_SET);
805 X_UNLOCK (preadwritelock);
806
807 return res;
808}
809#endif
810
811#ifndef HAVE_FUTIMES 549#ifndef HAVE_UTIMES
812 550
813# undef utimes 551# undef utimes
814# undef futimes
815# define utimes(path,times) eio__utimes (path, times) 552# define utimes(path,times) eio__utimes (path, times)
816# define futimes(fd,times) eio__futimes (fd, times)
817 553
818static int 554static int
819eio__utimes (const char *filename, const struct timeval times[2]) 555eio__utimes (const char *filename, const struct timeval times[2])
820{ 556{
821 if (times) 557 if (times)
… …
829 } 565 }
830 else 566 else
831 return utime (filename, 0); 567 return utime (filename, 0);
832} 568}
833 569
570#endif
571
572#ifndef HAVE_FUTIMES
573
574# undef futimes
575# define futimes(fd,times) eio__futimes (fd, times)
576
577static int
834static int eio__futimes (int fd, const struct timeval tv[2]) 578eio__futimes (int fd, const struct timeval tv[2])
835{ 579{
836 errno = ENOSYS; 580 errno = ENOSYS;
837 return -1; 581 return -1;
838} 582}
839 583
… …
842#if !HAVE_FDATASYNC 586#if !HAVE_FDATASYNC
843# undef fdatasync 587# undef fdatasync
844# define fdatasync(fd) fsync (fd) 588# define fdatasync(fd) fsync (fd)
845#endif 589#endif
846 590
591static int
592eio__syncfs (int fd)
593{
594 int res;
595
596#if HAVE_SYS_SYNCFS
597 res = (int)syscall (__NR_syncfs, (int)(fd));
598#else
599 res = EIO_ENOSYS ();
600#endif
601
602 if (res < 0 && errno == ENOSYS && fd >= 0)
603 sync ();
604
605 return res;
606}
607
847/* sync_file_range always needs emulation */ 608/* sync_file_range always needs emulation */
848int 609static int
849eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags) 610eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
850{ 611{
851#if HAVE_SYNC_FILE_RANGE 612#if HAVE_SYNC_FILE_RANGE
852 int res; 613 int res;
853 614
… …
866 if (!res || errno != ENOSYS) 627 if (!res || errno != ENOSYS)
867 return res; 628 return res;
868#endif 629#endif
869 630
870 /* even though we could play tricks with the flags, it's better to always 631 /* even though we could play tricks with the flags, it's better to always
871 * call fdatasync, as thta matches the expectation of it's users best */ 632 * call fdatasync, as that matches the expectation of its users best */
872 return fdatasync (fd); 633 return fdatasync (fd);
634}
635
636static int
637eio__fallocate (int fd, int mode, off_t offset, size_t len)
638{
639#if HAVE_LINUX_FALLOCATE
640 return fallocate (fd, mode, offset, len);
641#else
642 return EIO_ENOSYS ();
643#endif
873} 644}
874 645
875#if !HAVE_READAHEAD 646#if !HAVE_READAHEAD
876# undef readahead 647# undef readahead
877# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 648# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
878 649
879static ssize_t 650static eio_ssize_t
880eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) 651eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
881{ 652{
882 size_t todo = count; 653 size_t todo = count;
883 dBUF; 654 dBUF;
884 655
… …
889 pread (fd, eio_buf, len, offset); 660 pread (fd, eio_buf, len, offset);
890 offset += len; 661 offset += len;
891 todo -= len; 662 todo -= len;
892 } 663 }
893 664
894 errno = 0; 665 FUBd;
666
667 /* linux's readahead basically only fails for EBADF or EINVAL (not mmappable) */
668 /* but not for e.g. EIO or eof, so we also never fail */
895 return count; 669 return 0;
896} 670}
897 671
898#endif 672#endif
899 673
900/* sendfile always needs emulation */ 674/* sendfile always needs emulation */
901static ssize_t 675static eio_ssize_t
902eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) 676eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
903{ 677{
678 eio_ssize_t written = 0;
904 ssize_t res; 679 eio_ssize_t res;
905 680
906 if (!count) 681 if (!count)
907 return 0; 682 return 0;
908 683
684 for (;;)
685 {
686#ifdef __APPLE__
687# undef HAVE_SENDFILE /* broken, as everything on os x */
688#endif
909#if HAVE_SENDFILE 689#if HAVE_SENDFILE
910# if __linux 690# if __linux
691 off_t soffset = offset;
911 res = sendfile (ofd, ifd, &offset, count); 692 res = sendfile (ofd, ifd, &soffset, count);
912 693
913# elif __freebsd 694# elif __FreeBSD__
914 /* 695 /*
915 * Of course, the freebsd sendfile is a dire hack with no thoughts 696 * Of course, the freebsd sendfile is a dire hack with no thoughts
916 * wasted on making it similar to other I/O functions. 697 * wasted on making it similar to other I/O functions.
917 */ 698 */
918 {
919 off_t sbytes; 699 off_t sbytes;
920 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 700 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
921 701
922 if (res < 0 && sbytes) 702 #if 0 /* according to the manpage, this is correct, but broken behaviour */
923 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 703 /* freebsd' sendfile will return 0 on success */
704 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
705 /* not on e.g. EIO or EPIPE - sounds broken */
706 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
924 res = sbytes; 707 res = sbytes;
925 } 708 #endif
709
710 /* according to source inspection, this is correct, and useful behaviour */
711 if (sbytes)
712 res = sbytes;
713
714# elif defined __APPLE__
715 off_t sbytes = count;
716 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
717
718 /* according to the manpage, sbytes is always valid */
719 if (sbytes)
720 res = sbytes;
926 721
927# elif __hpux 722# elif __hpux
928 res = sendfile (ofd, ifd, offset, count, 0, 0); 723 res = sendfile (ofd, ifd, offset, count, 0, 0);
929 724
930# elif __solaris 725# elif __solaris
931 {
932 struct sendfilevec vec; 726 struct sendfilevec vec;
933 size_t sbytes; 727 size_t sbytes;
934 728
935 vec.sfv_fd = ifd; 729 vec.sfv_fd = ifd;
936 vec.sfv_flag = 0; 730 vec.sfv_flag = 0;
937 vec.sfv_off = offset; 731 vec.sfv_off = offset;
938 vec.sfv_len = count; 732 vec.sfv_len = count;
939 733
940 res = sendfilev (ofd, &vec, 1, &sbytes); 734 res = sendfilev (ofd, &vec, 1, &sbytes);
941 735
942 if (res < 0 && sbytes) 736 if (res < 0 && sbytes)
943 res = sbytes; 737 res = sbytes;
944 }
945 738
946# endif 739# endif
740
741#elif defined (_WIN32) && 0
742 /* does not work, just for documentation of what would need to be done */
743 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
744 /* libeio guarantees that the file offset does not change, and windows */
745 /* has no way to get an independent handle to the same file description */
746 HANDLE h = TO_SOCKET (ifd);
747 SetFilePointer (h, offset, 0, FILE_BEGIN);
748 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
749
947#else 750#else
948 res = -1; 751 res = EIO_ENOSYS ();
949 errno = ENOSYS;
950#endif 752#endif
951 753
754 /* we assume sendfile can copy at least 128mb in one go */
755 if (res <= 128 * 1024 * 1024)
756 {
757 if (res > 0)
758 written += res;
759
760 if (written)
761 return written;
762
763 break;
764 }
765 else
766 {
767 /* if we requested more, then probably the kernel was lazy */
768 written += res;
769 offset += res;
770 count -= res;
771
772 if (!count)
773 return written;
774 }
775 }
776
952 if (res < 0 777 if (res < 0
953 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 778 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
779 /* BSDs */
780#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
781 || errno == ENOTSUP
782#endif
783#ifdef EOPNOTSUPP /* windows */
784 || errno == EOPNOTSUPP /* BSDs */
785#endif
954#if __solaris 786#if __solaris
955 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 787 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
956#endif 788#endif
957 ) 789 )
958 ) 790 )
… …
962 794
963 res = 0; 795 res = 0;
964 796
965 while (count) 797 while (count)
966 { 798 {
967 ssize_t cnt; 799 eio_ssize_t cnt;
968 800
969 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 801 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
970 802
971 if (cnt <= 0) 803 if (cnt <= 0)
972 { 804 {
… …
984 816
985 offset += cnt; 817 offset += cnt;
986 res += cnt; 818 res += cnt;
987 count -= cnt; 819 count -= cnt;
988 } 820 }
821
822 FUBd;
989 } 823 }
990 824
991 return res; 825 return res;
992} 826}
993 827
828#ifdef PAGESIZE
829# define eio_pagesize() PAGESIZE
830#else
831static intptr_t
832eio_pagesize (void)
833{
834 static intptr_t page;
835
836 if (!page)
837 page = sysconf (_SC_PAGESIZE);
838
839 return page;
840}
841#endif
842
843static void
844eio_page_align (void **addr, size_t *length)
845{
846 intptr_t mask = eio_pagesize () - 1;
847
848 /* round down addr */
849 intptr_t adj = mask & (intptr_t)*addr;
850
851 *addr = (void *)((intptr_t)*addr - adj);
852 *length += adj;
853
854 /* round up length */
855 *length = (*length + mask) & ~mask;
856}
857
858#if !_POSIX_MEMLOCK
859# define eio__mlockall(a) EIO_ENOSYS ()
860#else
861
994static int 862static int
863eio__mlockall (int flags)
864{
865 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
866 extern int mallopt (int, int);
867 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
868 #endif
869
870 if (EIO_MCL_CURRENT != MCL_CURRENT
871 || EIO_MCL_FUTURE != MCL_FUTURE)
872 {
873 flags = 0
874 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
875 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
876 }
877
878 return mlockall (flags);
879}
880#endif
881
882#if !_POSIX_MEMLOCK_RANGE
883# define eio__mlock(a,b) EIO_ENOSYS ()
884#else
885
886static int
887eio__mlock (void *addr, size_t length)
888{
889 eio_page_align (&addr, &length);
890
891 return mlock (addr, length);
892}
893
894#endif
895
896#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
897# define eio__msync(a,b,c) EIO_ENOSYS ()
898#else
899
900static int
901eio__msync (void *mem, size_t len, int flags)
902{
903 eio_page_align (&mem, &len);
904
905 if (EIO_MS_ASYNC != MS_SYNC
906 || EIO_MS_INVALIDATE != MS_INVALIDATE
907 || EIO_MS_SYNC != MS_SYNC)
908 {
909 flags = 0
910 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
911 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
912 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
913 }
914
915 return msync (mem, len, flags);
916}
917
918#endif
919
920static int
921eio__mtouch (eio_req *req)
922{
923 void *mem = req->ptr2;
924 size_t len = req->size;
925 int flags = req->int1;
926
927 eio_page_align (&mem, &len);
928
929 {
930 intptr_t addr = (intptr_t)mem;
931 intptr_t end = addr + len;
932 intptr_t page = eio_pagesize ();
933
934 if (addr < end)
935 if (flags & EIO_MT_MODIFY) /* modify */
936 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
937 else
938 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
939 }
940
941 return 0;
942}
943
944/*****************************************************************************/
945/* requests implemented outside eio_execute, because they are so large */
946
947static void
948eio__lseek (eio_req *req)
949{
950 /* this usually gets optimised away completely, or your compiler sucks, */
951 /* or the whence constants really are not 0, 1, 2 */
952 int whence = req->int2 == EIO_SEEK_SET ? SEEK_SET
953 : req->int2 == EIO_SEEK_CUR ? SEEK_CUR
954 : req->int2 == EIO_SEEK_END ? SEEK_END
955 : req->int2;
956
957 req->offs = lseek (req->int1, req->offs, whence);
958 req->result = req->offs == (off_t)-1 ? -1 : 0;
959}
960
961/* result will always end up in tmpbuf, there is always space for adding a 0-byte */
962static int
963eio__realpath (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
964{
965 char *res;
966 const char *rel = path;
967 char *tmp1, *tmp2;
968#if SYMLOOP_MAX > 32
969 int symlinks = SYMLOOP_MAX;
970#else
971 int symlinks = 32;
972#endif
973
974 errno = EINVAL;
975 if (!rel)
976 return -1;
977
978 errno = ENOENT;
979 if (!*rel)
980 return -1;
981
982 res = etp_tmpbuf_get (tmpbuf, PATH_MAX * 3);
983#ifdef _WIN32
984 if (_access (rel, 4) != 0)
985 return -1;
986
987 symlinks = GetFullPathName (rel, PATH_MAX * 3, res, 0);
988
989 errno = ENAMETOOLONG;
990 if (symlinks >= PATH_MAX * 3)
991 return -1;
992
993 errno = EIO;
994 if (symlinks <= 0)
995 return -1;
996
997 return symlinks;
998
999#else
1000 tmp1 = res + PATH_MAX;
1001 tmp2 = tmp1 + PATH_MAX;
1002
1003#if 0 /* disabled, the musl way to do things is just too racy */
1004#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1005 /* on linux we may be able to ask the kernel */
1006 {
1007 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1008
1009 if (fd >= 0)
1010 {
1011 sprintf (tmp1, "/proc/self/fd/%d", fd);
1012 req->result = readlink (tmp1, res, PATH_MAX);
1013 /* here we should probably stat the open file and the disk file, to make sure they still match */
1014 close (fd);
1015
1016 if (req->result > 0)
1017 goto done;
1018 }
1019 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1020 return -1;
1021 }
1022#endif
1023#endif
1024
1025 if (*rel != '/')
1026 {
1027 int len;
1028
1029 errno = ENOENT;
1030 if (wd == EIO_INVALID_WD)
1031 return -1;
1032
1033 if (wd == EIO_CWD)
1034 {
1035 if (!getcwd (res, PATH_MAX))
1036 return -1;
1037
1038 len = strlen (res);
1039 }
1040 else
1041 memcpy (res, wd->str, len = wd->len);
1042
1043 if (res [1]) /* only use if not / */
1044 res += len;
1045 }
1046
1047 while (*rel)
1048 {
1049 eio_ssize_t len, linklen;
1050 const char *beg = rel;
1051
1052 while (*rel && *rel != '/')
1053 ++rel;
1054
1055 len = rel - beg;
1056
1057 if (!len) /* skip slashes */
1058 {
1059 ++rel;
1060 continue;
1061 }
1062
1063 if (beg [0] == '.')
1064 {
1065 if (len == 1)
1066 continue; /* . - nop */
1067
1068 if (beg [1] == '.' && len == 2)
1069 {
1070 /* .. - back up one component, if possible */
1071
1072 while (res != tmpbuf->ptr)
1073 if (*--res == '/')
1074 break;
1075
1076 continue;
1077 }
1078 }
1079
1080 errno = ENAMETOOLONG;
1081 if (res + 1 + len + 1 >= tmp1)
1082 return -1;
1083
1084 /* copy one component */
1085 *res = '/';
1086 memcpy (res + 1, beg, len);
1087
1088 /* zero-terminate, for readlink */
1089 res [len + 1] = 0;
1090
1091 /* now check if it's a symlink */
1092 linklen = readlink (tmpbuf->ptr, tmp1, PATH_MAX);
1093
1094 if (linklen < 0)
1095 {
1096 if (errno != EINVAL)
1097 return -1;
1098
1099 /* it's a normal directory. hopefully */
1100 res += len + 1;
1101 }
1102 else
1103 {
1104 /* yay, it was a symlink - build new path in tmp2 */
1105 int rellen = strlen (rel);
1106
1107 errno = ENAMETOOLONG;
1108 if (linklen + 1 + rellen >= PATH_MAX)
1109 return -1;
1110
1111 errno = ELOOP;
1112 if (!--symlinks)
1113 return -1;
1114
1115 if (*tmp1 == '/')
1116 res = tmpbuf->ptr; /* symlink resolves to an absolute path */
1117
1118 /* we need to be careful, as rel might point into tmp2 already */
1119 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1120 tmp2 [linklen] = '/';
1121 memcpy (tmp2, tmp1, linklen);
1122
1123 rel = tmp2;
1124 }
1125 }
1126
1127 /* special case for the lone root path */
1128 if (res == tmpbuf->ptr)
1129 *res++ = '/';
1130
1131 return res - (char *)tmpbuf->ptr;
1132#endif
1133}
1134
1135static signed char
995eio_dent_cmp (const void *a_, const void *b_) 1136eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
996{ 1137{
997 const eio_dirent *a = (const eio_dirent *)a_; 1138 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
998 const eio_dirent *b = (const eio_dirent *)b_; 1139 : a->inode < b->inode ? -1
1140 : a->inode > b->inode ? 1
1141 : 0;
1142}
999 1143
1000 return (int)b->score - (int)a->score ? (int)b->score - (int)a->score 1144#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1001 : a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0; /* int might be < ino_t */ 1145
1146#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1147#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1148
1149static void
1150eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1151{
1152 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1153 unsigned char *bit = bits;
1154
1155 assert (CHAR_BIT == 8);
1156 assert (sizeof (eio_dirent) * 8 < 256);
1157 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1158 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1159
1160 if (size <= EIO_SORT_FAST)
1161 return;
1162
1163 /* first prepare an array of bits to test in our radix sort */
1164 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1165 /* inode_bits must contain all inodes ORed together */
1166 /* which is used to skip bits that are 0 everywhere, which is very common */
1167 {
1168 eio_ino_t endianness;
1169 int i, j;
1170
1171 /* we store the byte offset of byte n into byte n of "endianness" */
1172 for (i = 0; i < sizeof (eio_ino_t); ++i)
1173 ((unsigned char *)&endianness)[i] = i;
1174
1175 *bit++ = 0;
1176
1177 for (i = 0; i < sizeof (eio_ino_t); ++i)
1178 {
1179 /* shifting off the byte offsets out of "endianness" */
1180 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1181 endianness >>= 8;
1182
1183 for (j = 0; j < 8; ++j)
1184 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1185 *bit++ = offs + j;
1186 }
1187
1188 for (j = 0; j < 8; ++j)
1189 if (score_bits & (1 << j))
1190 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1191 }
1192
1193 /* now actually do the sorting (a variant of MSD radix sort) */
1194 {
1195 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1196 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1197 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1198 int stk_idx = 0;
1199
1200 base_stk [stk_idx] = dents;
1201 end_stk [stk_idx] = dents + size;
1202 bit_stk [stk_idx] = bit - 1;
1203
1204 do
1205 {
1206 base = base_stk [stk_idx];
1207 end = end_stk [stk_idx];
1208 bit = bit_stk [stk_idx];
1209
1210 for (;;)
1211 {
1212 unsigned char O = *bit >> 3;
1213 unsigned char M = 1 << (*bit & 7);
1214
1215 eio_dirent *a = base;
1216 eio_dirent *b = end;
1217
1218 if (b - a < EIO_SORT_CUTOFF)
1219 break;
1220
1221 /* now bit-partition the array on the bit */
1222 /* this ugly asymmetric loop seems to perform much better than typical */
1223 /* partition algos found in the literature */
1224 do
1225 if (!(((unsigned char *)a)[O] & M))
1226 ++a;
1227 else if (!(((unsigned char *)--b)[O] & M))
1228 {
1229 eio_dirent tmp = *a; *a = *b; *b = tmp;
1230 ++a;
1231 }
1232 while (b > a);
1233
1234 /* next bit, or stop, if no bits left in this path */
1235 if (!*--bit)
1236 break;
1237
1238 base_stk [stk_idx] = a;
1239 end_stk [stk_idx] = end;
1240 bit_stk [stk_idx] = bit;
1241 ++stk_idx;
1242
1243 end = a;
1244 }
1245 }
1246 while (stk_idx--);
1247 }
1248}
1249
1250static void
1251eio_dent_insertion_sort (eio_dirent *dents, int size)
1252{
1253 /* first move the smallest element to the front, to act as a sentinel */
1254 {
1255 int i;
1256 eio_dirent *min = dents;
1257
1258 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1259 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1260 if (EIO_DENT_CMP (dents [i], <, *min))
1261 min = &dents [i];
1262
1263 /* swap elements 0 and j (minimum) */
1264 {
1265 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1266 }
1267 }
1268
1269 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1270 {
1271 eio_dirent *i, *j;
1272
1273 for (i = dents + 1; i < dents + size; ++i)
1274 {
1275 eio_dirent value = *i;
1276
1277 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1278 j [1] = j [0];
1279
1280 j [1] = value;
1281 }
1282 }
1283}
1284
1285static void
1286eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1287{
1288 if (size <= 1)
1289 return; /* our insertion sort relies on size > 0 */
1290
1291 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1292 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1293 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1294
1295 /* use an insertion sort at the end, or for small arrays, */
1296 /* as insertion sort is more efficient for small partitions */
1297 eio_dent_insertion_sort (dents, size);
1002} 1298}
1003 1299
1004/* read a full directory */ 1300/* read a full directory */
1005static void 1301static void
1006eio__scandir (eio_req *req, etp_worker *self) 1302eio__scandir (eio_req *req, etp_worker *self)
1007{ 1303{
1008 DIR *dirp;
1009 EIO_STRUCT_DIRENT *entp;
1010 unsigned char *name, *names; 1304 char *name, *names;
1011 int namesalloc = 4096; 1305 int namesalloc = 4096 - sizeof (void *) * 4;
1012 int namesoffs = 0; 1306 int namesoffs = 0;
1013 int flags = req->int1; 1307 int flags = req->int1;
1014 eio_dirent *dents = 0; 1308 eio_dirent *dents = 0;
1015 int dentalloc = 128; 1309 int dentalloc = 128;
1016 int dentoffs = 0; 1310 int dentoffs = 0;
1311 eio_ino_t inode_bits = 0;
1312#ifdef _WIN32
1313 HANDLE dirp;
1314 WIN32_FIND_DATA entp;
1315#else
1316 DIR *dirp;
1317 EIO_STRUCT_DIRENT *entp;
1318#endif
1017 1319
1018 req->result = -1; 1320 req->result = -1;
1019 1321
1020 if (!(flags & EIO_READDIR_DENTS)) 1322 if (!(flags & EIO_READDIR_DENTS))
1021 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); 1323 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1022 1324
1023 X_LOCK (wrklock); 1325#ifdef _WIN32
1024 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 1326 {
1327 int len = strlen ((const char *)req->ptr1);
1328 char *path = malloc (MAX_PATH);
1329 const char *fmt;
1330 const char *reqpath = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1331
1332 if (!len)
1333 fmt = "./*";
1334 else if (reqpath[len - 1] == '/' || reqpath[len - 1] == '\\')
1335 fmt = "%s*";
1336 else
1337 fmt = "%s/*";
1338
1339 _snprintf (path, MAX_PATH, fmt, reqpath);
1340 dirp = FindFirstFile (path, &entp);
1341 free (path);
1342
1343 if (dirp == INVALID_HANDLE_VALUE)
1344 {
1345 /* should steal _dosmaperr */
1346 switch (GetLastError ())
1347 {
1348 case ERROR_FILE_NOT_FOUND:
1349 req->result = 0;
1350 break;
1351
1352 case ERROR_INVALID_NAME:
1353 case ERROR_PATH_NOT_FOUND:
1354 case ERROR_NO_MORE_FILES:
1355 errno = ENOENT;
1356 break;
1357
1358 case ERROR_NOT_ENOUGH_MEMORY:
1359 errno = ENOMEM;
1360 break;
1361
1362 default:
1363 errno = EINVAL;
1364 break;
1365 }
1366
1367 return;
1368 }
1369 }
1370#else
1371 #if HAVE_AT
1372 if (req->wd)
1373 {
1374 int fd = openat (WD2FD (req->wd), req->ptr1, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1375
1376 if (fd < 0)
1377 return;
1378
1379 dirp = fdopendir (fd);
1380
1381 if (!dirp)
1382 close (fd);
1383 }
1384 else
1025 self->dirp = dirp = opendir (req->ptr1); 1385 dirp = opendir (req->ptr1);
1386 #else
1387 dirp = opendir (wd_expand (&self->tmpbuf, req->wd, req->ptr1));
1388 #endif
1389
1390 if (!dirp)
1391 return;
1392#endif
1393
1394 if (req->flags & EIO_FLAG_PTR1_FREE)
1395 free (req->ptr1);
1396
1026 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1397 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1027 req->ptr1 = names = malloc (namesalloc);
1028 req->ptr2 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; 1398 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1029 X_UNLOCK (wrklock); 1399 req->ptr2 = names = malloc (namesalloc);
1030 1400
1031 if (dirp && names && (!flags || dents)) 1401 if (!names || (flags && !dents))
1402 return;
1403
1032 for (;;) 1404 for (;;)
1033 { 1405 {
1406 int done;
1407
1408#ifdef _WIN32
1409 done = !dirp;
1410#else
1034 errno = 0; 1411 errno = 0;
1035 entp = readdir (dirp); 1412 entp = readdir (dirp);
1413 done = !entp;
1414#endif
1036 1415
1037 if (!entp) 1416 if (done)
1038 { 1417 {
1418#ifndef _WIN32
1419 int old_errno = errno;
1420 closedir (dirp);
1421 errno = old_errno;
1422
1039 if (errno) 1423 if (errno)
1040 break; 1424 break;
1425#endif
1041 1426
1042 /* sort etc. */ 1427 /* sort etc. */
1043 req->int1 = flags; 1428 req->int1 = flags;
1044 req->result = dentoffs; 1429 req->result = dentoffs;
1045 1430
1046 if (dents)
1047 {
1048 eio_dirent *ent = dents + dentoffs;
1049
1050 while (ent > dents)
1051 (--ent)->name = names + (size_t)ent->name;
1052 }
1053
1054 if (flags & EIO_READDIR_STAT_ORDER 1431 if (flags & EIO_READDIR_STAT_ORDER)
1055 || !(~flags & (EIO_READDIR_DIRS_FIRST | EIO_READDIR_FOUND_UNKNOWN))) 1432 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1056 {
1057 /* pray your qsort doesn't use quicksort */
1058 qsort (dents, dentoffs, sizeof (*dents), eio_dent_cmp); /* score depends of DIRS_FIRST */
1059 }
1060 else if (flags & EIO_READDIR_DIRS_FIRST) 1433 else if (flags & EIO_READDIR_DIRS_FIRST)
1434 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1435 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1436 else
1061 { 1437 {
1062 /* in this case, all is known, and we just put dirs first and sort them */ 1438 /* in this case, all is known, and we just put dirs first and sort them */
1063 eio_dirent *ent = dents + dentoffs; 1439 eio_dirent *oth = dents + dentoffs;
1064 eio_dirent *dir = dents; 1440 eio_dirent *dir = dents;
1065 1441
1066 /* now move dirs to the front, and non-dirs to the back */ 1442 /* now partition dirs to the front, and non-dirs to the back */
1067 /* by walkign from both sides and swapping if necessary */ 1443 /* by walking from both sides and swapping if necessary */
1068 while (ent > dir) 1444 while (oth > dir)
1069 { 1445 {
1070 if (dir->type == DT_DIR) 1446 if (dir->type == EIO_DT_DIR)
1071 ++dir; 1447 ++dir;
1072 else 1448 else if ((--oth)->type == EIO_DT_DIR)
1073 { 1449 {
1074 --ent;
1075
1076 if (ent->type == DT_DIR)
1077 {
1078 eio_dirent tmp = *dir; 1450 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1079 *dir = *ent;
1080 *ent = tmp;
1081 1451
1082 ++dir; 1452 ++dir;
1083 }
1084 } 1453 }
1085 } 1454 }
1086 1455
1087 /* now sort the dirs only */ 1456 /* now sort the dirs only (dirs all have the same score) */
1088 qsort (dents, dir - dents, sizeof (*dents), eio_dent_cmp); 1457 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1089 } 1458 }
1090 1459
1091 /* only provide the names array unless DENTS is specified */
1092 if (!(flags & EIO_READDIR_DENTS))
1093 {
1094 X_LOCK (wrklock);
1095 assert (!dents);
1096 req->ptr1 = 0;
1097 req->ptr2 = names;
1098 X_UNLOCK (wrklock);
1099 }
1100
1101 break; 1460 break;
1102 } 1461 }
1103 1462
1104 /* now add the entry to our list(s) */ 1463 /* now add the entry to our list(s) */
1105 name = entp->d_name; 1464 name = D_NAME (entp);
1106 1465
1107 /* skip . and .. entries */ 1466 /* skip . and .. entries */
1108 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1467 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1109 { 1468 {
1110 int len = D_NAMLEN (entp) + 1; 1469 int len = D_NAMLEN (entp) + 1;
1111 1470
1112 while (expect_false (namesoffs + len > namesalloc)) 1471 while (ecb_expect_false (namesoffs + len > namesalloc))
1113 { 1472 {
1114 namesalloc *= 2; 1473 namesalloc *= 2;
1115 X_LOCK (wrklock);
1116 req->ptr1 = names = realloc (names, namesalloc); 1474 req->ptr2 = names = realloc (names, namesalloc);
1117 X_UNLOCK (wrklock);
1118 1475
1119 if (!names) 1476 if (!names)
1120 break; 1477 break;
1121 } 1478 }
1122 1479
1123 memcpy (names + namesoffs, name, len); 1480 memcpy (names + namesoffs, name, len);
1124 1481
1125 if (dents) 1482 if (dents)
1126 { 1483 {
1127 struct eio_dirent *ent; 1484 struct eio_dirent *ent;
1128 1485
1129 if (expect_false (dentoffs == dentalloc)) 1486 if (ecb_expect_false (dentoffs == dentalloc))
1130 { 1487 {
1131 dentalloc *= 2; 1488 dentalloc *= 2;
1132 X_LOCK (wrklock);
1133 req->ptr2 = dents = realloc (dents, dentalloc * sizeof (eio_dirent)); 1489 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1134 X_UNLOCK (wrklock);
1135 1490
1136 if (!dents) 1491 if (!dents)
1137 break; 1492 break;
1138 } 1493 }
1139 1494
1140 ent = dents + dentoffs; 1495 ent = dents + dentoffs;
1141 1496
1142 ent->name = (char *)(size_t)namesoffs; /* rather dirtily we store the offset in the pointer */ 1497 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1143 ent->namelen = len - 1; 1498 ent->namelen = len - 1;
1144 ent->inode = D_INO (entp); 1499 ent->inode = D_INO (entp);
1145 1500
1501 inode_bits |= ent->inode;
1502
1146 switch (D_TYPE (entp)) 1503 switch (D_TYPE (entp))
1147 { 1504 {
1148 default: 1505 default:
1149 ent->type = EIO_DT_UNKNOWN; 1506 ent->type = EIO_DT_UNKNOWN;
1150 flags |= EIO_READDIR_FOUND_UNKNOWN; 1507 flags |= EIO_READDIR_FOUND_UNKNOWN;
1151 break; 1508 break;
1152 1509
1153 #ifdef DT_FIFO 1510 #ifdef DT_FIFO
1154 case DT_FIFO: ent->type = EIO_DT_FIFO; break; 1511 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1155 #endif 1512 #endif
1156 #ifdef DT_CHR 1513 #ifdef DT_CHR
1157 case DT_CHR: ent->type = EIO_DT_CHR; break; 1514 case DT_CHR: ent->type = EIO_DT_CHR; break;
1158 #endif 1515 #endif
1159 #ifdef DT_MPC 1516 #ifdef DT_MPC
1160 case DT_MPC: ent->type = EIO_DT_MPC; break; 1517 case DT_MPC: ent->type = EIO_DT_MPC; break;
1161 #endif 1518 #endif
1162 #ifdef DT_DIR 1519 #ifdef DT_DIR
1163 case DT_DIR: ent->type = EIO_DT_DIR; break; 1520 case DT_DIR: ent->type = EIO_DT_DIR; break;
1164 #endif 1521 #endif
1165 #ifdef DT_NAM 1522 #ifdef DT_NAM
1166 case DT_NAM: ent->type = EIO_DT_NAM; break; 1523 case DT_NAM: ent->type = EIO_DT_NAM; break;
1167 #endif 1524 #endif
1168 #ifdef DT_BLK 1525 #ifdef DT_BLK
1169 case DT_BLK: ent->type = EIO_DT_BLK; break; 1526 case DT_BLK: ent->type = EIO_DT_BLK; break;
1170 #endif 1527 #endif
1171 #ifdef DT_MPB 1528 #ifdef DT_MPB
1172 case DT_MPB: ent->type = EIO_DT_MPB; break; 1529 case DT_MPB: ent->type = EIO_DT_MPB; break;
1173 #endif 1530 #endif
1174 #ifdef DT_REG 1531 #ifdef DT_REG
1175 case DT_REG: ent->type = EIO_DT_REG; break; 1532 case DT_REG: ent->type = EIO_DT_REG; break;
1176 #endif 1533 #endif
1177 #ifdef DT_NWK 1534 #ifdef DT_NWK
1178 case DT_NWK: ent->type = EIO_DT_NWK; break; 1535 case DT_NWK: ent->type = EIO_DT_NWK; break;
1179 #endif 1536 #endif
1180 #ifdef DT_CMP 1537 #ifdef DT_CMP
1181 case DT_CMP: ent->type = EIO_DT_CMP; break; 1538 case DT_CMP: ent->type = EIO_DT_CMP; break;
1182 #endif 1539 #endif
1183 #ifdef DT_LNK 1540 #ifdef DT_LNK
1184 case DT_LNK: ent->type = EIO_DT_LNK; break; 1541 case DT_LNK: ent->type = EIO_DT_LNK; break;
1185 #endif 1542 #endif
1186 #ifdef DT_SOCK 1543 #ifdef DT_SOCK
1187 case DT_SOCK: ent->type = EIO_DT_SOCK; break; 1544 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1188 #endif 1545 #endif
1189 #ifdef DT_DOOR 1546 #ifdef DT_DOOR
1190 case DT_DOOR: ent->type = EIO_DT_DOOR; break; 1547 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1191 #endif 1548 #endif
1192 #ifdef DT_WHT 1549 #ifdef DT_WHT
1193 case DT_WHT: ent->type = EIO_DT_WHT; break; 1550 case DT_WHT: ent->type = EIO_DT_WHT; break;
1194 #endif 1551 #endif
1195 } 1552 }
1196 1553
1197 ent->score = 0; 1554 ent->score = 7;
1198 1555
1199 if (flags & EIO_READDIR_DIRS_FIRST) 1556 if (flags & EIO_READDIR_DIRS_FIRST)
1200 { 1557 {
1201 if (ent->type == EIO_DT_UNKNOWN) 1558 if (ent->type == EIO_DT_UNKNOWN)
1202 { 1559 {
1203 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */ 1560 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1204 ent->score = 98; 1561 ent->score = 1;
1205 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */ 1562 else if (!strchr (name, '.')) /* absence of dots indicate likely dirs */
1206 ent->score = len <= 2 ? len + 6 : len <= 4 ? 5 : len <= 7 ? 4 : 1; /* shorter == more likely dir, but avoid too many classes */ 1563 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1207 } 1564 }
1208 else if (ent->type == EIO_DT_DIR) 1565 else if (ent->type == EIO_DT_DIR)
1209 ent->score = 100; 1566 ent->score = 0;
1210 } 1567 }
1211 } 1568 }
1212 1569
1213 namesoffs += len; 1570 namesoffs += len;
1214 ++dentoffs; 1571 ++dentoffs;
1215 } 1572 }
1573
1574 if (EIO_CANCELLED (req))
1575 {
1576 errno = ECANCELED;
1577 break;
1216 } 1578 }
1217 else
1218 req->result = -1;
1219}
1220 1579
1221#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) 1580#ifdef _WIN32
1222# undef msync 1581 if (!FindNextFile (dirp, &entp))
1223# define msync(a,b,c) ((errno = ENOSYS), -1) 1582 {
1583 FindClose (dirp);
1584 dirp = 0;
1585 }
1224#endif 1586#endif
1587 }
1588}
1225 1589
1226int 1590/*****************************************************************************/
1227eio__mtouch (void *mem, size_t len, int flags) 1591/* working directory stuff */
1228{ 1592/* various deficiencies in the posix 2008 api force us to */
1229 intptr_t addr = (intptr_t)mem; 1593/* keep the absolute path in string form at all times */
1230 intptr_t end = addr + len; 1594/* fuck yeah. */
1231#ifdef PAGESIZE
1232 const intptr_t page = PAGESIZE;
1233#else
1234 static intptr_t page;
1235 1595
1236 if (!page) 1596#if !HAVE_AT
1237 page = sysconf (_SC_PAGESIZE);
1238#endif
1239 1597
1240 addr &= ~(page - 1); /* assume page size is always a power of two */ 1598/* a bit like realpath, but usually faster because it doesn'T have to return */
1599/* an absolute or canonical path */
1600static const char *
1601wd_expand (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
1602{
1603 if (!wd || *path == '/')
1604 return path;
1241 1605
1242 if (addr < end) 1606 if (path [0] == '.' && !path [1])
1243 if (flags) /* modify */ 1607 return wd->str;
1244 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
1245 else
1246 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
1247 1608
1609 {
1610 int l1 = wd->len;
1611 int l2 = strlen (path);
1612
1613 char *res = etp_tmpbuf_get (tmpbuf, l1 + l2 + 2);
1614
1615 memcpy (res, wd->str, l1);
1616 res [l1] = '/';
1617 memcpy (res + l1 + 1, path, l2 + 1);
1618
1619 return res;
1620 }
1621}
1622
1623#endif
1624
1625static eio_wd
1626eio__wd_open_sync (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
1627{
1628 int fd;
1629 eio_wd res;
1630 int len = eio__realpath (tmpbuf, wd, path);
1631
1632 if (len < 0)
1633 return EIO_INVALID_WD;
1634
1635#if HAVE_AT
1636 fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1637
1638 if (fd < 0)
1639 return EIO_INVALID_WD;
1640#endif
1641
1642 res = malloc (sizeof (*res) + len); /* one extra 0-byte */
1643
1644#if HAVE_AT
1645 res->fd = fd;
1646#endif
1647
1648 res->len = len;
1649 memcpy (res->str, tmpbuf->ptr, len);
1650 res->str [len] = 0;
1651
1652 return res;
1653}
1654
1655eio_wd
1656eio_wd_open_sync (eio_wd wd, const char *path)
1657{
1658 struct etp_tmpbuf tmpbuf = { };
1659 wd = eio__wd_open_sync (&tmpbuf, wd, path);
1660 free (tmpbuf.ptr);
1661
1248 return 0; 1662 return wd;
1249} 1663}
1664
1665void
1666eio_wd_close_sync (eio_wd wd)
1667{
1668 if (wd != EIO_INVALID_WD && wd != EIO_CWD)
1669 {
1670 #if HAVE_AT
1671 close (wd->fd);
1672 #endif
1673 free (wd);
1674 }
1675}
1676
1677#if HAVE_AT
1678
1679/* they forgot these */
1680
1681static int
1682eio__truncateat (int dirfd, const char *path, off_t length)
1683{
1684 int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
1685 int res;
1686
1687 if (fd < 0)
1688 return fd;
1689
1690 res = ftruncate (fd, length);
1691 close (fd);
1692 return res;
1693}
1694
1695static int
1696eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
1697{
1698 int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
1699 int res;
1700
1701 if (fd < 0)
1702 return fd;
1703
1704 res = fstatvfs (fd, buf);
1705 close (fd);
1706 return res;
1707
1708}
1709
1710#endif
1250 1711
1251/*****************************************************************************/ 1712/*****************************************************************************/
1252 1713
1253#define ALLOC(len) \ 1714#define ALLOC(len) \
1254 if (!req->ptr2) \ 1715 if (!req->ptr2) \
1255 { \ 1716 { \
1256 X_LOCK (wrklock); \ 1717 X_LOCK (EIO_POOL->wrklock); \
1257 req->flags |= EIO_FLAG_PTR2_FREE; \ 1718 req->flags |= EIO_FLAG_PTR2_FREE; \
1258 X_UNLOCK (wrklock); \ 1719 X_UNLOCK (EIO_POOL->wrklock); \
1259 req->ptr2 = malloc (len); \ 1720 req->ptr2 = malloc (len); \
1260 if (!req->ptr2) \ 1721 if (!req->ptr2) \
1261 { \ 1722 { \
1262 errno = ENOMEM; \ 1723 errno = ENOMEM; \
1263 req->result = -1; \ 1724 req->result = -1; \
1264 break; \ 1725 break; \
1265 } \ 1726 } \
1266 } 1727 }
1267 1728
1268X_THREAD_PROC (etp_proc)
1269{
1270 ETP_REQ *req;
1271 struct timespec ts;
1272 etp_worker *self = (etp_worker *)thr_arg;
1273
1274 /* try to distribute timeouts somewhat randomly */
1275 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1276
1277 for (;;)
1278 {
1279 X_LOCK (reqlock);
1280
1281 for (;;)
1282 {
1283 self->req = req = reqq_shift (&req_queue);
1284
1285 if (req)
1286 break;
1287
1288 ++idle;
1289
1290 ts.tv_sec = time (0) + IDLE_TIMEOUT;
1291 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
1292 {
1293 if (idle > max_idle)
1294 {
1295 --idle;
1296 X_UNLOCK (reqlock);
1297 X_LOCK (wrklock);
1298 --started;
1299 X_UNLOCK (wrklock);
1300 goto quit;
1301 }
1302
1303 /* we are allowed to idle, so do so without any timeout */
1304 X_COND_WAIT (reqwait, reqlock);
1305 }
1306
1307 --idle;
1308 }
1309
1310 --nready;
1311
1312 X_UNLOCK (reqlock);
1313
1314 if (req->type < 0)
1315 goto quit;
1316
1317 if (!EIO_CANCELLED (req))
1318 ETP_EXECUTE (self, req);
1319
1320 X_LOCK (reslock);
1321
1322 ++npending;
1323
1324 if (!reqq_push (&res_queue, req) && want_poll_cb)
1325 want_poll_cb ();
1326
1327 self->req = 0;
1328 etp_worker_clear (self);
1329
1330 X_UNLOCK (reslock);
1331 }
1332
1333quit:
1334 X_LOCK (wrklock);
1335 etp_worker_free (self);
1336 X_UNLOCK (wrklock);
1337
1338 return 0;
1339}
1340
1341/*****************************************************************************/ 1729/*****************************************************************************/
1342 1730
1731int ecb_cold
1343int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1732eio_init (void (*want_poll)(void), void (*done_poll)(void))
1344{ 1733{
1345 return etp_init (want_poll, done_poll); 1734 eio_want_poll_cb = want_poll;
1346} 1735 eio_done_poll_cb = done_poll;
1347 1736
1737 return etp_init (EIO_POOL, 0, 0, 0);
1738}
1739
1740ecb_inline void
1348static void eio_api_destroy (eio_req *req) 1741eio_api_destroy (eio_req *req)
1349{ 1742{
1350 free (req); 1743 free (req);
1351} 1744}
1352 1745
1353#define REQ(rtype) \ 1746#define REQ(rtype) \
1354 eio_req *req; \ 1747 eio_req *req; \
1355 \ 1748 \
1356 req = (eio_req *)calloc (1, sizeof *req); \ 1749 req = (eio_req *)calloc (1, sizeof *req); \
1357 if (!req) \ 1750 if (!req) \
1358 return 0; \ 1751 return 0; \
… …
1372 { \ 1765 { \
1373 eio_api_destroy (req); \ 1766 eio_api_destroy (req); \
1374 return 0; \ 1767 return 0; \
1375 } 1768 }
1376 1769
1770#define SINGLEDOT(ptr) (0[(char *)(ptr)] == '.' && !1[(char *)(ptr)])
1771
1772static void
1377static void eio_execute (etp_worker *self, eio_req *req) 1773eio_execute (etp_worker *self, eio_req *req)
1378{ 1774{
1379 errno = 0; 1775#if HAVE_AT
1776 int dirfd;
1777#else
1778 const char *path;
1779#endif
1780
1781 if (ecb_expect_false (EIO_CANCELLED (req)))
1782 {
1783 req->result = -1;
1784 req->errorno = ECANCELED;
1785 return;
1786 }
1787
1788 if (ecb_expect_false (req->wd == EIO_INVALID_WD))
1789 {
1790 req->result = -1;
1791 req->errorno = ENOENT;
1792 return;
1793 }
1794
1795 if (req->type >= EIO_OPEN)
1796 {
1797 #if HAVE_AT
1798 dirfd = WD2FD (req->wd);
1799 #else
1800 path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1801 #endif
1802 }
1380 1803
1381 switch (req->type) 1804 switch (req->type)
1382 { 1805 {
1806 case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
1807 req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
1808 break;
1809 case EIO_WD_CLOSE: req->result = 0;
1810 eio_wd_close_sync (req->wd); break;
1811
1812 case EIO_SEEK: eio__lseek (req); break;
1383 case EIO_READ: ALLOC (req->size); 1813 case EIO_READ: ALLOC (req->size);
1384 req->result = req->offs >= 0 1814 req->result = req->offs >= 0
1385 ? pread (req->int1, req->ptr2, req->size, req->offs) 1815 ? pread (req->int1, req->ptr2, req->size, req->offs)
1386 : read (req->int1, req->ptr2, req->size); break; 1816 : read (req->int1, req->ptr2, req->size); break;
1387 case EIO_WRITE: req->result = req->offs >= 0 1817 case EIO_WRITE: req->result = req->offs >= 0
1388 ? pwrite (req->int1, req->ptr2, req->size, req->offs) 1818 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1389 : write (req->int1, req->ptr2, req->size); break; 1819 : write (req->int1, req->ptr2, req->size); break;
1390 1820
1391 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break; 1821 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1392 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break; 1822 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
1823
1824#if HAVE_AT
1393 1825
1394 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1826 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1395 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 1827 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break;
1396 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1828 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1397 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 1829 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, AT_SYMLINK_NOFOLLOW); break;
1398 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1399 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1400
1401 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; 1830 case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break;
1402 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1403 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break; 1831 case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break;
1404 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1405 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break; 1832 case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break;
1406 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1407
1408 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break; 1833 case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break;
1409 case EIO_CLOSE: req->result = close (req->int1); break; 1834
1410 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1411 case EIO_UNLINK: req->result = unlink (req->ptr1); break; 1835 case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
1412 case EIO_RMDIR: req->result = rmdir (req->ptr1); break; 1836 case EIO_RMDIR: /* complications arise because "." cannot be removed, so we might have to expand */
1837 req->result = req->wd && SINGLEDOT (req->ptr1)
1838 ? rmdir (req->wd->str)
1839 : unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break;
1413 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; 1840 case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); break;
1414 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; 1841 case EIO_RENAME: /* complications arise because "." cannot be renamed, so we might have to expand */
1842 req->result = req->wd && SINGLEDOT (req->ptr1)
1843 ? rename (req->wd->str, req->ptr2)
1844 : renameat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2); break;
1415 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; 1845 case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2, 0); break;
1416 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; 1846 case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
1417 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break; 1847 case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1418
1419 case EIO_READLINK: ALLOC (NAME_MAX); 1848 case EIO_READLINK: ALLOC (PATH_MAX);
1420 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; 1849 req->result = readlinkat (dirfd, req->ptr1, req->ptr2, PATH_MAX); break;
1421 1850 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1422 case EIO_SYNC: req->result = 0; sync (); break; 1851 req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1423 case EIO_FSYNC: req->result = fsync (req->int1); break;
1424 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1425 case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break;
1426 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;
1427 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1428
1429 case EIO_READDIR: eio__scandir (req, self); break;
1430
1431 case EIO_BUSY: 1852 case EIO_UTIME:
1432#ifdef _WIN32 1853 case EIO_FUTIME:
1433 Sleep (req->nv1 * 1000.); 1854 {
1855 struct timespec ts[2];
1856 struct timespec *times;
1857
1858 if (req->nv1 != -1. || req->nv2 != -1.)
1859 {
1860 ts[0].tv_sec = req->nv1;
1861 ts[0].tv_nsec = (req->nv1 - ts[0].tv_sec) * 1e9;
1862 ts[1].tv_sec = req->nv2;
1863 ts[1].tv_nsec = (req->nv2 - ts[1].tv_sec) * 1e9;
1864
1865 times = ts;
1866 }
1867 else
1868 times = 0;
1869
1870 req->result = req->type == EIO_FUTIME
1871 ? futimens (req->int1, times)
1872 : utimensat (dirfd, req->ptr1, times, 0);
1873 }
1874 break;
1875
1434#else 1876#else
1435 {
1436 struct timeval tv;
1437 1877
1438 tv.tv_sec = req->nv1; 1878 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1439 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; 1879 req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
1880 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1881 req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
1882 case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
1883 case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
1884 case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
1885 case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
1440 1886
1441 req->result = select (0, 0, 0, 0, &tv); 1887 case EIO_UNLINK: req->result = unlink (path ); break;
1442 } 1888 case EIO_RMDIR: req->result = rmdir (path ); break;
1443#endif 1889 case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
1444 break; 1890 case EIO_RENAME: req->result = rename (path , req->ptr2); break;
1891 case EIO_LINK: req->result = link (path , req->ptr2); break;
1892 case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
1893 case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
1894 case EIO_READLINK: ALLOC (PATH_MAX);
1895 req->result = readlink (path, req->ptr2, PATH_MAX); break;
1896 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1897 req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
1445 1898
1446 case EIO_UTIME: 1899 case EIO_UTIME:
1447 case EIO_FUTIME: 1900 case EIO_FUTIME:
1448 { 1901 {
1449 struct timeval tv[2]; 1902 struct timeval tv[2];
1450 struct timeval *times; 1903 struct timeval *times;
1451 1904
1452 if (req->nv1 != -1. || req->nv2 != -1.) 1905 if (req->nv1 != -1. || req->nv2 != -1.)
1453 { 1906 {
1454 tv[0].tv_sec = req->nv1; 1907 tv[0].tv_sec = req->nv1;
1455 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.; 1908 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
1456 tv[1].tv_sec = req->nv2; 1909 tv[1].tv_sec = req->nv2;
1457 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.; 1910 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
1458 1911
1459 times = tv; 1912 times = tv;
1460 } 1913 }
1461 else 1914 else
1462 times = 0; 1915 times = 0;
1463
1464 1916
1465 req->result = req->type == EIO_FUTIME 1917 req->result = req->type == EIO_FUTIME
1466 ? futimes (req->int1, times) 1918 ? futimes (req->int1, times)
1467 : utimes (req->ptr1, times); 1919 : utimes (req->ptr1, times);
1468 } 1920 }
1469 break; 1921 break;
1470 1922
1923#endif
1924
1925 case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
1926 {
1927 ALLOC (req->result);
1928 memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
1929 }
1930 break;
1931
1932 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1933 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1934
1935 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1936 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1937
1938 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1939 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1940 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1941
1942 case EIO_CLOSE: req->result = close (req->int1); break;
1943 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1944 case EIO_SYNC: req->result = 0; sync (); break;
1945 case EIO_FSYNC: req->result = fsync (req->int1); break;
1946 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1947 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
1948 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1949 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1950 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1951 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1952 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1953 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
1954
1955 case EIO_READDIR: eio__scandir (req, self); break;
1956
1957 case EIO_BUSY:
1958#ifdef _WIN32
1959 Sleep (req->nv1 * 1e3);
1960#else
1961 {
1962 struct timeval tv;
1963
1964 tv.tv_sec = req->nv1;
1965 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1966
1967 req->result = select (0, 0, 0, 0, &tv);
1968 }
1969#endif
1970 break;
1971
1972#if 0
1471 case EIO_GROUP: 1973 case EIO_GROUP:
1472 abort (); /* handled in eio_request */ 1974 abort (); /* handled in eio_request */
1975#endif
1473 1976
1474 case EIO_NOP: 1977 case EIO_NOP:
1475 req->result = 0; 1978 req->result = 0;
1476 break; 1979 break;
1477 1980
1478 case EIO_CUSTOM: 1981 case EIO_CUSTOM:
1479 ((void (*)(eio_req *))req->feed) (req); 1982 req->feed (req);
1480 break; 1983 break;
1481 1984
1482 default: 1985 default:
1483 req->result = -1; 1986 req->result = EIO_ENOSYS ();
1484 break; 1987 break;
1485 } 1988 }
1486 1989
1487 req->errorno = errno; 1990 req->errorno = errno;
1488} 1991}
1489 1992
1490#ifndef EIO_NO_WRAPPERS 1993#ifndef EIO_NO_WRAPPERS
1491 1994
1995eio_req *eio_wd_open (const char *path, int pri, eio_cb cb, void *data)
1996{
1997 REQ (EIO_WD_OPEN); PATH; SEND;
1998}
1999
2000eio_req *eio_wd_close (eio_wd wd, int pri, eio_cb cb, void *data)
2001{
2002 REQ (EIO_WD_CLOSE); req->wd = wd; SEND;
2003}
2004
1492eio_req *eio_nop (int pri, eio_cb cb, void *data) 2005eio_req *eio_nop (int pri, eio_cb cb, void *data)
1493{ 2006{
1494 REQ (EIO_NOP); SEND; 2007 REQ (EIO_NOP); SEND;
1495} 2008}
1496 2009
… …
1512eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2025eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1513{ 2026{
1514 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2027 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1515} 2028}
1516 2029
2030eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2031{
2032 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2033}
2034
2035eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
2036{
2037 REQ (EIO_SYNCFS); req->int1 = fd; SEND;
2038}
2039
2040eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2041{
2042 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2043}
2044
1517eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2045eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1518{ 2046{
1519 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2047 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1520} 2048}
1521 2049
1522eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) 2050eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
1523{ 2051{
1524 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; 2052 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
1525} 2053}
1526 2054
1527eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) 2055eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
1528{ 2056{
1529 REQ (EIO_FDATASYNC); req->int1 = fd; SEND; 2057 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2058}
2059
2060eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2061{
2062 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
1530} 2063}
1531 2064
1532eio_req *eio_close (int fd, int pri, eio_cb cb, void *data) 2065eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
1533{ 2066{
1534 REQ (EIO_CLOSE); req->int1 = fd; SEND; 2067 REQ (EIO_CLOSE); req->int1 = fd; SEND;
… …
1537eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data) 2070eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
1538{ 2071{
1539 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND; 2072 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
1540} 2073}
1541 2074
2075eio_req *eio_seek (int fd, off_t offset, int whence, int pri, eio_cb cb, void *data)
2076{
2077 REQ (EIO_SEEK); req->int1 = fd; req->offs = offset; req->int2 = whence; SEND;
2078}
2079
1542eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data) 2080eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1543{ 2081{
1544 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND; 2082 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1545} 2083}
1546 2084
… …
1552eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) 2090eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1553{ 2091{
1554 REQ (EIO_FSTAT); req->int1 = fd; SEND; 2092 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1555} 2093}
1556 2094
2095eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
2096{
2097 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
2098}
2099
1557eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) 2100eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1558{ 2101{
1559 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; 2102 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1560} 2103}
1561 2104
… …
1567eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) 2110eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1568{ 2111{
1569 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2112 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1570} 2113}
1571 2114
1572eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2115eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1573{ 2116{
1574 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2117 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1575} 2118}
1576 2119
1577eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) 2120eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
… …
1597eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data) 2140eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1598{ 2141{
1599 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2142 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1600} 2143}
1601 2144
1602eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2145eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1603{ 2146{
1604 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2147 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1605} 2148}
1606 2149
1607eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) 2150eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
… …
1623eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data) 2166eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1624{ 2167{
1625 return eio__1path (EIO_READLINK, path, pri, cb, data); 2168 return eio__1path (EIO_READLINK, path, pri, cb, data);
1626} 2169}
1627 2170
2171eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2172{
2173 return eio__1path (EIO_REALPATH, path, pri, cb, data);
2174}
2175
1628eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) 2176eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1629{ 2177{
1630 return eio__1path (EIO_STAT, path, pri, cb, data); 2178 return eio__1path (EIO_STAT, path, pri, cb, data);
1631} 2179}
1632 2180
1633eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) 2181eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1634{ 2182{
1635 return eio__1path (EIO_LSTAT, path, pri, cb, data); 2183 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1636} 2184}
1637 2185
2186eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
2187{
2188 return eio__1path (EIO_STATVFS, path, pri, cb, data);
2189}
2190
1638eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) 2191eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1639{ 2192{
1640 return eio__1path (EIO_UNLINK, path, pri, cb, data); 2193 return eio__1path (EIO_UNLINK, path, pri, cb, data);
1641} 2194}
1642 2195
… …
1650 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND; 2203 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1651} 2204}
1652 2205
1653eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) 2206eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1654{ 2207{
1655 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; 2208 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1656} 2209}
1657 2210
1658static eio_req * 2211static eio_req *
1659eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2212eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1660{ 2213{
… …
1684eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2237eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1685{ 2238{
1686 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2239 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1687} 2240}
1688 2241
1689eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) 2242eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
1690{ 2243{
1691 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND; 2244 REQ (EIO_CUSTOM); req->feed = execute; SEND;
1692} 2245}
1693 2246
1694#endif 2247#endif
1695 2248
1696eio_req *eio_grp (eio_cb cb, void *data) 2249eio_req *eio_grp (eio_cb cb, void *data)
… …
1705#undef SEND 2258#undef SEND
1706 2259
1707/*****************************************************************************/ 2260/*****************************************************************************/
1708/* grp functions */ 2261/* grp functions */
1709 2262
2263void
1710void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2264eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1711{ 2265{
1712 grp->int2 = limit; 2266 grp->int2 = limit;
1713 grp->feed = feed; 2267 grp->feed = feed;
1714 2268
1715 grp_try_feed (grp); 2269 grp_try_feed (grp);
1716} 2270}
1717 2271
2272void
1718void eio_grp_limit (eio_req *grp, int limit) 2273eio_grp_limit (eio_req *grp, int limit)
1719{ 2274{
1720 grp->int2 = limit; 2275 grp->int2 = limit;
1721 2276
1722 grp_try_feed (grp); 2277 grp_try_feed (grp);
1723} 2278}
1724 2279
2280void
1725void eio_grp_add (eio_req *grp, eio_req *req) 2281eio_grp_add (eio_req *grp, eio_req *req)
1726{ 2282{
1727 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2283 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1728 2284
1729 grp->flags |= EIO_FLAG_GROUPADD; 2285 grp->flags |= ETP_FLAG_GROUPADD;
1730 2286
1731 ++grp->size; 2287 ++grp->size;
1732 req->grp = grp; 2288 req->grp = grp;
1733 2289
1734 req->grp_prev = 0; 2290 req->grp_prev = 0;
… …
1741} 2297}
1742 2298
1743/*****************************************************************************/ 2299/*****************************************************************************/
1744/* misc garbage */ 2300/* misc garbage */
1745 2301
2302eio_ssize_t
1746ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2303eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1747{ 2304{
1748 etp_worker wrk;
1749
1750 wrk.dbuf = 0;
1751
1752 eio__sendfile (ofd, ifd, offset, count, &wrk); 2305 return eio__sendfile (ofd, ifd, offset, count);
1753
1754 if (wrk.dbuf)
1755 free (wrk.dbuf);
1756} 2306}
1757 2307

Diff Legend

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