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

Comparing libeio/eio.c (file contents):
Revision 1.27 by root, Wed Oct 22 18:15:36 2008 UTC vs.
Revision 1.143 by root, Sat Dec 3 16:33:46 2016 UTC

1/* 1/*
2 * libeio implementation 2 * libeio implementation
3 * 3 *
4 * Copyright (c) 2007,2008 Marc Alexander Lehmann <libeio@schmorp.de> 4 * Copyright (c) 2007,2008,2009,2010,2011,2012,2013,2016 Marc Alexander Lehmann <libeio@schmorp.de>
5 * All rights reserved. 5 * All rights reserved.
6 * 6 *
7 * Redistribution and use in source and binary forms, with or without modifica- 7 * Redistribution and use in source and binary forms, with or without modifica-
8 * tion, are permitted provided that the following conditions are met: 8 * tion, are permitted provided that the following conditions are met:
9 * 9 *
10 * 1. Redistributions of source code must retain the above copyright notice, 10 * 1. Redistributions of source code must retain the above copyright notice,
11 * this list of conditions and the following disclaimer. 11 * this list of conditions and the following disclaimer.
12 * 12 *
13 * 2. Redistributions in binary form must reproduce the above copyright 13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the 14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution. 15 * documentation and/or other materials provided with the distribution.
16 * 16 *
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR IMPLIED 17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR IMPLIED
18 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MER- 18 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MER-
19 * CHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO 19 * CHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO
20 * EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPE- 20 * EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPE-
21 * CIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, 21 * CIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
… …
35 * and other provisions required by the GPL. If you do not delete the 35 * and other provisions required by the GPL. If you do not delete the
36 * provisions above, a recipient may use your version of this file under 36 * provisions above, a recipient may use your version of this file under
37 * either the BSD or the GPL. 37 * either the BSD or the GPL.
38 */ 38 */
39 39
40#ifndef _WIN32
41# include "config.h"
42#endif
43
40#include "eio.h" 44#include "eio.h"
41#include "xthread.h" 45#include "ecb.h"
42 46
43#include <errno.h> 47#include <errno.h>
44#include <stddef.h> 48#include <stddef.h>
45#include <stdlib.h> 49#include <stdlib.h>
46#include <string.h> 50#include <string.h>
… …
49#include <sys/stat.h> 53#include <sys/stat.h>
50#include <limits.h> 54#include <limits.h>
51#include <fcntl.h> 55#include <fcntl.h>
52#include <assert.h> 56#include <assert.h>
53 57
58/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
59/* intptr_t only comes from stdint.h, says idiot openbsd coder */
60#if HAVE_STDINT_H
61# include <stdint.h>
62#endif
63
64#ifndef ECANCELED
65# define ECANCELED EDOM
66#endif
67#ifndef ELOOP
68# define ELOOP EDOM
69#endif
70
71#if !defined(ENOTSOCK) && defined(WSAENOTSOCK)
72# define ENOTSOCK WSAENOTSOCK
73#endif
74
75static void eio_destroy (eio_req *req);
76
54#ifndef EIO_FINISH 77#ifndef EIO_FINISH
55# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0 78# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
56#endif 79#endif
57 80
58#ifndef EIO_DESTROY 81#ifndef EIO_DESTROY
… …
61 84
62#ifndef EIO_FEED 85#ifndef EIO_FEED
63# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0) 86# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)
64#endif 87#endif
65 88
89#ifndef EIO_FD_TO_WIN32_HANDLE
90# define EIO_FD_TO_WIN32_HANDLE(fd) _get_osfhandle (fd)
91#endif
92#ifndef EIO_WIN32_HANDLE_TO_FD
93# define EIO_WIN32_HANDLE_TO_FD(handle) _open_osfhandle (handle, 0)
94#endif
95
96#define EIO_ERRNO(errval,retval) ((errno = errval), retval)
97
98#define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1)
99
66#ifdef _WIN32 100#ifdef _WIN32
67 101
68 /*doh*/ 102 #undef PAGESIZE
103 #define PAGESIZE 4096 /* GetSystemInfo? */
104
105 /* TODO: look at how perl does stat (non-sloppy), unlink (ro-files), utime, link */
106
107 #ifdef EIO_STRUCT_STATI64
108 /* look at perl's non-sloppy stat */
109 #define stat(path,buf) _stati64 (path,buf)
110 #define fstat(fd,buf) _fstati64 (fd,buf)
111 #endif
112 #define lstat(path,buf) stat (path,buf)
113 #define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
114 #define mkdir(path,mode) _mkdir (path)
115 #define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
116
117 #define chmod(path,mode) _chmod (path, mode)
118 #define dup(fd) _dup (fd)
119 #define dup2(fd1,fd2) _dup2 (fd1, fd2)
120 #define pipe(fds) _pipe (fds, 4096, O_BINARY)
121
122 #define fcntl(fd,cmd,arg) EIO_ENOSYS ()
123 #define ioctl(fd,cmd,arg) EIO_ENOSYS ()
124 #define fchmod(fd,mode) EIO_ENOSYS ()
125 #define chown(path,uid,gid) EIO_ENOSYS ()
126 #define fchown(fd,uid,gid) EIO_ENOSYS ()
127 #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
128 #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
129 #define mknod(path,mode,dev) EIO_ENOSYS ()
130 #define sync() EIO_ENOSYS ()
131 #define readlink(path,buf,s) EIO_ENOSYS ()
132 #define statvfs(path,buf) EIO_ENOSYS ()
133 #define fstatvfs(fd,buf) EIO_ENOSYS ()
134
135 #define pread(fd,buf,count,offset) eio__pread (fd, buf, count, offset)
136 #define pwrite(fd,buf,count,offset) eio__pwrite (fd, buf, count, offset)
137
138 #if __GNUC__
139 typedef long long eio_off_t; /* signed for compatibility to msvc */
140 #else
141 typedef __int64 eio_off_t; /* unsigned not supported by msvc */
142 #endif
143
144 static eio_ssize_t
145 eio__pread (int fd, void *buf, eio_ssize_t count, eio_off_t offset)
146 {
147 OVERLAPPED o = { 0 };
148 DWORD got;
149
150 o.Offset = offset;
151 o.OffsetHigh = offset >> 32;
152
153 return ReadFile ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd), buf, count, &got, &o)
154 ? got : -1;
155 }
156
157 static eio_ssize_t
158 eio__pwrite (int fd, void *buf, eio_ssize_t count, eio_off_t offset)
159 {
160 OVERLAPPED o = { 0 };
161 DWORD got;
162
163 o.Offset = offset;
164 o.OffsetHigh = offset >> 32;
165
166 return WriteFile ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd), buf, count, &got, &o)
167 ? got : -1;
168 }
169
170 /* rename() uses MoveFile, which fails to overwrite */
171 #define rename(old,neu) eio__rename (old, neu)
172
173 static int
174 eio__rename (const char *old, const char *neu)
175 {
176 if (MoveFileEx (old, neu, MOVEFILE_REPLACE_EXISTING))
177 return 0;
178
179 /* should steal _dosmaperr */
180 switch (GetLastError ())
181 {
182 case ERROR_FILE_NOT_FOUND:
183 case ERROR_PATH_NOT_FOUND:
184 case ERROR_INVALID_DRIVE:
185 case ERROR_NO_MORE_FILES:
186 case ERROR_BAD_NETPATH:
187 case ERROR_BAD_NET_NAME:
188 case ERROR_BAD_PATHNAME:
189 case ERROR_FILENAME_EXCED_RANGE:
190 errno = ENOENT;
191 break;
192
193 default:
194 errno = EACCES;
195 break;
196 }
197
198 return -1;
199 }
200
201 /* we could even stat and see if it exists */
202 static int
203 symlink (const char *old, const char *neu)
204 {
205 #if WINVER >= 0x0600
206 int flags;
207
208 /* This tries out all combinations of SYMBOLIC_LINK_FLAG_DIRECTORY
209 * and SYMBOLIC_LINK_FLAG_ALLOW_UNPRIVILEGED_CREATE,
210 * with directory first.
211 */
212 for (flags = 3; flags >= 0; --flags)
213 if (CreateSymbolicLink (neu, old, flags))
214 return 0;
215 #endif
216
217 return EIO_ERRNO (ENOENT, -1);
218 }
219
220 /* POSIX API only, causing trouble for win32 apps */
221 #define CreateHardLink(neu,old,flags) 0 /* not really creating hardlink, still using relative paths? */
222 #define CreateSymbolicLink(neu,old,flags) 0 /* vista+ only */
223
224 struct statvfs
225 {
226 int dummy;
227 };
228
229 #define DT_DIR EIO_DT_DIR
230 #define DT_REG EIO_DT_REG
231 #define D_NAME(entp) entp.cFileName
232 #define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
69 233
70#else 234#else
71 235
72# include "config.h"
73# include <sys/time.h> 236 #include <sys/time.h>
74# include <sys/select.h> 237 #include <sys/select.h>
75# include <unistd.h> 238 #include <unistd.h>
239 #include <signal.h>
240 #include <dirent.h>
241
242 #ifdef ANDROID
243 #include <sys/vfs.h>
244 #define statvfs statfs
245 #define fstatvfs fstatfs
246 #include <asm/page.h> /* supposedly limits.h does #define PAGESIZE PAGESIZE */
247 #else
248 #include <sys/statvfs.h>
249 #endif
250
251 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
252 #include <sys/mman.h>
253 #endif
254
255 #define D_NAME(entp) entp->d_name
256
257 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
258 #if __FreeBSD__ || __NetBSD__ || __OpenBSD__
259 #define _DIRENT_HAVE_D_TYPE /* sigh */
260 #define D_INO(de) (de)->d_fileno
261 #define D_NAMLEN(de) (de)->d_namlen
262 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
263 #define D_INO(de) (de)->d_ino
264 #endif
265
266 #ifdef _D_EXACT_NAMLEN
267 #undef D_NAMLEN
268 #define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
269 #endif
270
271 #ifdef _DIRENT_HAVE_D_TYPE
272 #define D_TYPE(de) (de)->d_type
273 #endif
274
275 #ifndef EIO_STRUCT_DIRENT
276 #define EIO_STRUCT_DIRENT struct dirent
277 #endif
278
279#endif
280
281#if HAVE_UTIMES
76# include <utime.h> 282# include <utime.h>
77# include <signal.h>
78# include <dirent.h>
79
80# ifndef EIO_STRUCT_DIRENT
81# define EIO_STRUCT_DIRENT struct dirent
82# endif 283#endif
83 284
285#if HAVE_SYS_SYSCALL_H
286# include <sys/syscall.h>
84#endif 287#endif
85 288
86#if HAVE_SENDFILE 289#if HAVE_SENDFILE
87# if __linux 290# if __linux
88# include <sys/sendfile.h> 291# include <sys/sendfile.h>
89# elif __freebsd 292# elif __FreeBSD__ || defined __APPLE__
90# include <sys/socket.h> 293# include <sys/socket.h>
91# include <sys/uio.h> 294# include <sys/uio.h>
92# elif __hpux 295# elif __hpux
93# include <sys/socket.h> 296# include <sys/socket.h>
94# elif __solaris /* not yet */ 297# elif __solaris
95# include <sys/sendfile.h> 298# include <sys/sendfile.h>
96# else 299# else
97# error sendfile support requested but not available 300# error sendfile support requested but not available
98# endif 301# endif
99#endif 302#endif
100 303
101/* number of seconds after which an idle threads exit */ 304#ifndef D_TYPE
102#define IDLE_TIMEOUT 10 305# define D_TYPE(de) 0
306#endif
307#ifndef D_INO
308# define D_INO(de) 0
309#endif
310#ifndef D_NAMLEN
311# define D_NAMLEN(entp) strlen (D_NAME (entp))
312#endif
103 313
104/* used for struct dirent, AIX doesn't provide it */ 314/* used for struct dirent, AIX doesn't provide it */
105#ifndef NAME_MAX 315#ifndef NAME_MAX
106# define NAME_MAX 4096 316# define NAME_MAX 4096
107#endif 317#endif
108 318
319/* used for readlink etc. */
320#ifndef PATH_MAX
321# define PATH_MAX 4096
322#endif
323
109/* buffer size for various temporary buffers */ 324/* buffer size for various temporary buffers */
110#define EIO_BUFSIZE 65536 325#define EIO_BUFSIZE 65536
111 326
112#define dBUF \ 327#define dBUF \
113 char *eio_buf; \
114 ETP_WORKER_LOCK (self); \
115 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 328 char *eio_buf = malloc (EIO_BUFSIZE); \
116 ETP_WORKER_UNLOCK (self); \
117 errno = ENOMEM; \ 329 errno = ENOMEM; \
118 if (!eio_buf) \ 330 if (!eio_buf) \
119 return -1; 331 return -1
120 332
121#define EIO_TICKS ((1000000 + 1023) >> 10) 333#define FUBd \
334 free (eio_buf)
122 335
123/*****************************************************************************/ 336/*****************************************************************************/
124 337
125#if __GNUC__ >= 3 338struct etp_tmpbuf;
126# define expect(expr,value) __builtin_expect ((expr),(value)) 339
340#if _POSIX_VERSION >= 200809L
341 #define HAVE_AT 1
342 #define WD2FD(wd) ((wd) ? (wd)->fd : AT_FDCWD)
343 #ifndef O_SEARCH
344 #define O_SEARCH O_RDONLY
345 #endif
127#else 346#else
128# define expect(expr,value) (expr) 347 #define HAVE_AT 0
348 static const char *wd_expand (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path);
129#endif 349#endif
130 350
131#define expect_false(expr) expect ((expr) != 0, 0) 351struct eio_pwd
132#define expect_true(expr) expect ((expr) != 0, 1) 352{
353#if HAVE_AT
354 int fd;
355#endif
356 int len;
357 char str[1]; /* actually, a 0-terminated canonical path */
358};
133 359
134/*****************************************************************************/ 360/*****************************************************************************/
135 361
136#define ETP_PRI_MIN EIO_PRI_MIN 362#define ETP_PRI_MIN EIO_PRI_MIN
137#define ETP_PRI_MAX EIO_PRI_MAX 363#define ETP_PRI_MAX EIO_PRI_MAX
138 364
365#define ETP_TYPE_QUIT -1
366#define ETP_TYPE_GROUP EIO_GROUP
367
368static void eio_nop_callback (void) { }
369static void (*eio_want_poll_cb)(void) = eio_nop_callback;
370static void (*eio_done_poll_cb)(void) = eio_nop_callback;
371
372#define ETP_WANT_POLL(pool) eio_want_poll_cb ()
373#define ETP_DONE_POLL(pool) eio_done_poll_cb ()
374
139struct etp_worker; 375struct etp_worker;
140
141#define ETP_REQ eio_req 376#define ETP_REQ eio_req
142#define ETP_DESTROY(req) eio_destroy (req) 377#define ETP_DESTROY(req) eio_destroy (req)
143static int eio_finish (eio_req *req); 378static int eio_finish (eio_req *req);
144#define ETP_FINISH(req) eio_finish (req) 379#define ETP_FINISH(req) eio_finish (req)
145static void eio_execute (struct etp_worker *self, eio_req *req); 380static void eio_execute (struct etp_worker *self, eio_req *req);
146#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) 381#define ETP_EXECUTE(wrk,req) eio_execute (wrk, req)
147 382
148#define ETP_WORKER_CLEAR(req) \ 383#include "etp.c"
149 if (wrk->dbuf) \ 384
150 { \ 385static struct etp_pool eio_pool;
151 free (wrk->dbuf); \ 386#define EIO_POOL (&eio_pool)
152 wrk->dbuf = 0; \
153 } \
154 \
155 if (wrk->dirp) \
156 { \
157 closedir (wrk->dirp); \
158 wrk->dirp = 0; \
159 }
160#define ETP_WORKER_COMMON \
161 void *dbuf; \
162 DIR *dirp;
163 387
164/*****************************************************************************/ 388/*****************************************************************************/
165 389
166#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
167
168/* calculcate time difference in ~1/EIO_TICKS of a second */
169static int tvdiff (struct timeval *tv1, struct timeval *tv2)
170{
171 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
172 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
173}
174
175static unsigned int started, idle, wanted = 4;
176
177static void (*want_poll_cb) (void);
178static void (*done_poll_cb) (void);
179
180static unsigned int max_poll_time; /* reslock */
181static unsigned int max_poll_reqs; /* reslock */
182
183static volatile unsigned int nreqs; /* reqlock */
184static volatile unsigned int nready; /* reqlock */
185static volatile unsigned int npending; /* reqlock */
186static volatile unsigned int max_idle = 4;
187
188static mutex_t wrklock = X_MUTEX_INIT;
189static mutex_t reslock = X_MUTEX_INIT;
190static mutex_t reqlock = X_MUTEX_INIT;
191static cond_t reqwait = X_COND_INIT;
192
193#if !HAVE_PREADWRITE
194/*
195 * make our pread/pwrite emulation safe against themselves, but not against
196 * normal read/write by using a mutex. slows down execution a lot,
197 * but that's your problem, not mine.
198 */
199static mutex_t preadwritelock = X_MUTEX_INIT;
200#endif
201
202typedef struct etp_worker
203{
204 /* locked by wrklock */
205 struct etp_worker *prev, *next;
206
207 thread_t tid;
208
209 /* locked by reslock, reqlock or wrklock */
210 ETP_REQ *req; /* currently processed request */
211
212 ETP_WORKER_COMMON
213} etp_worker;
214
215static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */
216
217#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
218#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
219
220/* worker threads management */
221
222static void etp_worker_clear (etp_worker *wrk)
223{
224 ETP_WORKER_CLEAR (wrk);
225}
226
227static void etp_worker_free (etp_worker *wrk)
228{
229 wrk->next->prev = wrk->prev;
230 wrk->prev->next = wrk->next;
231
232 free (wrk);
233}
234
235static unsigned int etp_nreqs (void)
236{
237 int retval;
238 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
239 retval = nreqs;
240 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
241 return retval;
242}
243
244static unsigned int etp_nready (void)
245{
246 unsigned int retval;
247
248 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
249 retval = nready;
250 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
251
252 return retval;
253}
254
255static unsigned int etp_npending (void)
256{
257 unsigned int retval;
258
259 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
260 retval = npending;
261 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
262
263 return retval;
264}
265
266static unsigned int etp_nthreads (void)
267{
268 unsigned int retval;
269
270 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
271 retval = started;
272 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
273
274 return retval;
275}
276
277/*
278 * a somewhat faster data structure might be nice, but
279 * with 8 priorities this actually needs <20 insns
280 * per shift, the most expensive operation.
281 */
282typedef struct {
283 ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */
284 int size;
285} etp_reqq;
286
287static etp_reqq req_queue;
288static etp_reqq res_queue;
289
290static int reqq_push (etp_reqq *q, ETP_REQ *req)
291{
292 int pri = req->pri;
293 req->next = 0;
294
295 if (q->qe[pri])
296 {
297 q->qe[pri]->next = req;
298 q->qe[pri] = req;
299 }
300 else
301 q->qe[pri] = q->qs[pri] = req;
302
303 return q->size++;
304}
305
306static ETP_REQ *reqq_shift (etp_reqq *q)
307{
308 int pri;
309
310 if (!q->size)
311 return 0;
312
313 --q->size;
314
315 for (pri = ETP_NUM_PRI; pri--; )
316 {
317 eio_req *req = q->qs[pri];
318
319 if (req)
320 {
321 if (!(q->qs[pri] = (eio_req *)req->next))
322 q->qe[pri] = 0;
323
324 return req;
325 }
326 }
327
328 abort ();
329}
330
331static void etp_atfork_prepare (void)
332{
333 X_LOCK (wrklock);
334 X_LOCK (reqlock);
335 X_LOCK (reslock);
336#if !HAVE_PREADWRITE
337 X_LOCK (preadwritelock);
338#endif
339}
340
341static void etp_atfork_parent (void)
342{
343#if !HAVE_PREADWRITE
344 X_UNLOCK (preadwritelock);
345#endif
346 X_UNLOCK (reslock);
347 X_UNLOCK (reqlock);
348 X_UNLOCK (wrklock);
349}
350
351static void etp_atfork_child (void)
352{
353 ETP_REQ *prv;
354
355 while ((prv = reqq_shift (&req_queue)))
356 ETP_DESTROY (prv);
357
358 while ((prv = reqq_shift (&res_queue)))
359 ETP_DESTROY (prv);
360
361 while (wrk_first.next != &wrk_first)
362 {
363 etp_worker *wrk = wrk_first.next;
364
365 if (wrk->req)
366 ETP_DESTROY (wrk->req);
367
368 etp_worker_clear (wrk);
369 etp_worker_free (wrk);
370 }
371
372 started = 0;
373 idle = 0;
374 nreqs = 0;
375 nready = 0;
376 npending = 0;
377
378 etp_atfork_parent ();
379}
380
381static void 390static void
382etp_once_init (void)
383{
384 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
385}
386
387static int
388etp_init (void (*want_poll)(void), void (*done_poll)(void))
389{
390 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
391
392 pthread_once (&doinit, etp_once_init);
393
394 want_poll_cb = want_poll;
395 done_poll_cb = done_poll;
396
397 return 0;
398}
399
400X_THREAD_PROC (etp_proc);
401
402static void etp_start_thread (void)
403{
404 etp_worker *wrk = calloc (1, sizeof (etp_worker));
405
406 /*TODO*/
407 assert (("unable to allocate worker thread data", wrk));
408
409 X_LOCK (wrklock);
410
411 if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
412 {
413 wrk->prev = &wrk_first;
414 wrk->next = wrk_first.next;
415 wrk_first.next->prev = wrk;
416 wrk_first.next = wrk;
417 ++started;
418 }
419 else
420 free (wrk);
421
422 X_UNLOCK (wrklock);
423}
424
425static void etp_maybe_start_thread (void)
426{
427 if (expect_true (etp_nthreads () >= wanted))
428 return;
429
430 /* todo: maybe use idle here, but might be less exact */
431 if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
432 return;
433
434 etp_start_thread ();
435}
436
437static void etp_end_thread (void)
438{
439 eio_req *req = calloc (1, sizeof (eio_req));
440
441 req->type = -1;
442 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
443
444 X_LOCK (reqlock);
445 reqq_push (&req_queue, req);
446 X_COND_SIGNAL (reqwait);
447 X_UNLOCK (reqlock);
448
449 X_LOCK (wrklock);
450 --started;
451 X_UNLOCK (wrklock);
452}
453
454static int etp_poll (void)
455{
456 unsigned int maxreqs;
457 unsigned int maxtime;
458 struct timeval tv_start, tv_now;
459
460 X_LOCK (reslock);
461 maxreqs = max_poll_reqs;
462 maxtime = max_poll_time;
463 X_UNLOCK (reslock);
464
465 if (maxtime)
466 gettimeofday (&tv_start, 0);
467
468 for (;;)
469 {
470 ETP_REQ *req;
471
472 etp_maybe_start_thread ();
473
474 X_LOCK (reslock);
475 req = reqq_shift (&res_queue);
476
477 if (req)
478 {
479 --npending;
480
481 if (!res_queue.size && done_poll_cb)
482 done_poll_cb ();
483 }
484
485 X_UNLOCK (reslock);
486
487 if (!req)
488 return 0;
489
490 X_LOCK (reqlock);
491 --nreqs;
492 X_UNLOCK (reqlock);
493
494 if (expect_false (req->type == EIO_GROUP && req->size))
495 {
496 req->int1 = 1; /* mark request as delayed */
497 continue;
498 }
499 else
500 {
501 int res = ETP_FINISH (req);
502 if (expect_false (res))
503 return res;
504 }
505
506 if (expect_false (maxreqs && !--maxreqs))
507 break;
508
509 if (maxtime)
510 {
511 gettimeofday (&tv_now, 0);
512
513 if (tvdiff (&tv_start, &tv_now) >= maxtime)
514 break;
515 }
516 }
517
518 errno = EAGAIN;
519 return -1;
520}
521
522static void etp_cancel (ETP_REQ *req)
523{
524 X_LOCK (wrklock);
525 req->flags |= EIO_FLAG_CANCELLED;
526 X_UNLOCK (wrklock);
527
528 eio_grp_cancel (req);
529}
530
531static void etp_submit (ETP_REQ *req)
532{
533 req->pri -= ETP_PRI_MIN;
534
535 if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
536 if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
537
538 if (expect_false (req->type == EIO_GROUP))
539 {
540 /* I hope this is worth it :/ */
541 X_LOCK (reqlock);
542 ++nreqs;
543 X_UNLOCK (reqlock);
544
545 X_LOCK (reslock);
546
547 ++npending;
548
549 if (!reqq_push (&res_queue, req) && want_poll_cb)
550 want_poll_cb ();
551
552 X_UNLOCK (reslock);
553 }
554 else
555 {
556 X_LOCK (reqlock);
557 ++nreqs;
558 ++nready;
559 reqq_push (&req_queue, req);
560 X_COND_SIGNAL (reqwait);
561 X_UNLOCK (reqlock);
562
563 etp_maybe_start_thread ();
564 }
565}
566
567static void etp_set_max_poll_time (double nseconds)
568{
569 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
570 max_poll_time = nseconds;
571 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
572}
573
574static void etp_set_max_poll_reqs (unsigned int maxreqs)
575{
576 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
577 max_poll_reqs = maxreqs;
578 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
579}
580
581static void etp_set_max_idle (unsigned int nthreads)
582{
583 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
584 max_idle = nthreads <= 0 ? 1 : nthreads;
585 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
586}
587
588static void etp_set_min_parallel (unsigned int nthreads)
589{
590 if (wanted < nthreads)
591 wanted = nthreads;
592}
593
594static void etp_set_max_parallel (unsigned int nthreads)
595{
596 if (wanted > nthreads)
597 wanted = nthreads;
598
599 while (started > wanted)
600 etp_end_thread ();
601}
602
603/*****************************************************************************/
604
605static void grp_try_feed (eio_req *grp) 391grp_try_feed (eio_req *grp)
606{ 392{
607 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 393 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
608 { 394 {
609 grp->flags &= ~EIO_FLAG_GROUPADD; 395 grp->flags &= ~ETP_FLAG_GROUPADD;
610 396
611 EIO_FEED (grp); 397 EIO_FEED (grp);
612 398
613 /* stop if no progress has been made */ 399 /* stop if no progress has been made */
614 if (!(grp->flags & EIO_FLAG_GROUPADD)) 400 if (!(grp->flags & ETP_FLAG_GROUPADD))
615 { 401 {
616 grp->feed = 0; 402 grp->feed = 0;
617 break; 403 break;
618 } 404 }
619 } 405 }
620} 406}
621 407
408static int
622static int grp_dec (eio_req *grp) 409grp_dec (eio_req *grp)
623{ 410{
624 --grp->size; 411 --grp->size;
625 412
626 /* call feeder, if applicable */ 413 /* call feeder, if applicable */
627 grp_try_feed (grp); 414 grp_try_feed (grp);
628 415
629 /* finish, if done */ 416 /* finish, if done */
630 if (!grp->size && grp->int1) 417 if (!grp->size && grp->flags & ETP_FLAG_DELAYED)
631 return eio_finish (grp); 418 return eio_finish (grp);
632 else 419 else
633 return 0; 420 return 0;
634} 421}
635 422
423static void
636void eio_destroy (eio_req *req) 424eio_destroy (eio_req *req)
637{ 425{
638 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); 426 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
639 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); 427 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
640 428
641 EIO_DESTROY (req); 429 EIO_DESTROY (req);
642} 430}
643 431
432static int
644static int eio_finish (eio_req *req) 433eio_finish (eio_req *req)
645{ 434{
646 int res = EIO_FINISH (req); 435 int res = EIO_FINISH (req);
647 436
648 if (req->grp) 437 if (req->grp)
649 { 438 {
… …
657 if (grp->grp_first == req) 446 if (grp->grp_first == req)
658 grp->grp_first = req->grp_next; 447 grp->grp_first = req->grp_next;
659 448
660 res2 = grp_dec (grp); 449 res2 = grp_dec (grp);
661 450
662 if (!res && res2) 451 if (!res)
663 res = res2; 452 res = res2;
664 } 453 }
665 454
666 eio_destroy (req); 455 eio_destroy (req);
667 456
668 return res; 457 return res;
669} 458}
670 459
460void
671void eio_grp_cancel (eio_req *grp) 461eio_grp_cancel (eio_req *grp)
672{ 462{
673 for (grp = grp->grp_first; grp; grp = grp->grp_next) 463 etp_grp_cancel (EIO_POOL, grp);
674 eio_cancel (grp);
675} 464}
676 465
466void
677void eio_cancel (eio_req *req) 467eio_cancel (eio_req *req)
678{ 468{
679 etp_cancel (req); 469 etp_cancel (EIO_POOL, req);
680} 470}
681 471
472void
682void eio_submit (eio_req *req) 473eio_submit (eio_req *req)
683{ 474{
684 etp_submit (req); 475 etp_submit (EIO_POOL, req);
685} 476}
686 477
687unsigned int eio_nreqs (void) 478unsigned int
479eio_nreqs (void)
688{ 480{
689 return etp_nreqs (); 481 return etp_nreqs (EIO_POOL);
690} 482}
691 483
692unsigned int eio_nready (void) 484unsigned int
485eio_nready (void)
693{ 486{
694 return etp_nready (); 487 return etp_nready (EIO_POOL);
695} 488}
696 489
697unsigned int eio_npending (void) 490unsigned int
491eio_npending (void)
698{ 492{
699 return etp_npending (); 493 return etp_npending (EIO_POOL);
700} 494}
701 495
702unsigned int eio_nthreads (void) 496unsigned int ecb_cold
497eio_nthreads (void)
703{ 498{
704 return etp_nthreads (); 499 return etp_nthreads (EIO_POOL);
705} 500}
706 501
502void ecb_cold
707void eio_set_max_poll_time (double nseconds) 503eio_set_max_poll_time (double nseconds)
708{ 504{
709 etp_set_max_poll_time (nseconds); 505 etp_set_max_poll_time (EIO_POOL, nseconds);
710} 506}
711 507
508void ecb_cold
712void eio_set_max_poll_reqs (unsigned int maxreqs) 509eio_set_max_poll_reqs (unsigned int maxreqs)
713{ 510{
714 etp_set_max_poll_reqs (maxreqs); 511 etp_set_max_poll_reqs (EIO_POOL, maxreqs);
715} 512}
716 513
514void ecb_cold
717void eio_set_max_idle (unsigned int nthreads) 515eio_set_max_idle (unsigned int nthreads)
718{ 516{
719 etp_set_max_idle (nthreads); 517 etp_set_max_idle (EIO_POOL, nthreads);
720} 518}
721 519
520void ecb_cold
521eio_set_idle_timeout (unsigned int seconds)
522{
523 etp_set_idle_timeout (EIO_POOL, seconds);
524}
525
526void ecb_cold
722void eio_set_min_parallel (unsigned int nthreads) 527eio_set_min_parallel (unsigned int nthreads)
723{ 528{
724 etp_set_min_parallel (nthreads); 529 etp_set_min_parallel (EIO_POOL, nthreads);
725} 530}
726 531
532void ecb_cold
727void eio_set_max_parallel (unsigned int nthreads) 533eio_set_max_parallel (unsigned int nthreads)
728{ 534{
729 etp_set_max_parallel (nthreads); 535 etp_set_max_parallel (EIO_POOL, nthreads);
730} 536}
731 537
732int eio_poll (void) 538int eio_poll (void)
733{ 539{
734 return etp_poll (); 540 return etp_poll (EIO_POOL);
735} 541}
736 542
737/*****************************************************************************/ 543/*****************************************************************************/
738/* work around various missing functions */ 544/* work around various missing functions */
739 545
740#if !HAVE_PREADWRITE
741# undef pread
742# undef pwrite
743# define pread eio__pread
744# define pwrite eio__pwrite
745
746static ssize_t
747eio__pread (int fd, void *buf, size_t count, off_t offset)
748{
749 ssize_t res;
750 off_t ooffset;
751
752 X_LOCK (preadwritelock);
753 ooffset = lseek (fd, 0, SEEK_CUR);
754 lseek (fd, offset, SEEK_SET);
755 res = read (fd, buf, count);
756 lseek (fd, ooffset, SEEK_SET);
757 X_UNLOCK (preadwritelock);
758
759 return res;
760}
761
762static ssize_t
763eio__pwrite (int fd, void *buf, size_t count, off_t offset)
764{
765 ssize_t res;
766 off_t ooffset;
767
768 X_LOCK (preadwritelock);
769 ooffset = lseek (fd, 0, SEEK_CUR);
770 lseek (fd, offset, SEEK_SET);
771 res = write (fd, buf, count);
772 lseek (fd, offset, SEEK_SET);
773 X_UNLOCK (preadwritelock);
774
775 return res;
776}
777#endif
778
779#ifndef HAVE_FUTIMES 546#ifndef HAVE_UTIMES
780 547
781# undef utimes 548# undef utimes
782# undef futimes
783# define utimes(path,times) eio__utimes (path, times) 549# define utimes(path,times) eio__utimes (path, times)
784# define futimes(fd,times) eio__futimes (fd, times)
785 550
786static int 551static int
787eio__utimes (const char *filename, const struct timeval times[2]) 552eio__utimes (const char *filename, const struct timeval times[2])
788{ 553{
789 if (times) 554 if (times)
… …
797 } 562 }
798 else 563 else
799 return utime (filename, 0); 564 return utime (filename, 0);
800} 565}
801 566
567#endif
568
569#ifndef HAVE_FUTIMES
570
571# undef futimes
572# define futimes(fd,times) eio__futimes (fd, times)
573
574static int
802static int eio__futimes (int fd, const struct timeval tv[2]) 575eio__futimes (int fd, const struct timeval tv[2])
803{ 576{
804 errno = ENOSYS; 577 errno = ENOSYS;
805 return -1; 578 return -1;
806} 579}
807 580
… …
810#if !HAVE_FDATASYNC 583#if !HAVE_FDATASYNC
811# undef fdatasync 584# undef fdatasync
812# define fdatasync(fd) fsync (fd) 585# define fdatasync(fd) fsync (fd)
813#endif 586#endif
814 587
588static int
589eio__syncfs (int fd)
590{
591 int res;
592
593#if HAVE_SYS_SYNCFS
594 res = (int)syscall (__NR_syncfs, (int)(fd));
595#else
596 res = EIO_ENOSYS ();
597#endif
598
599 if (res < 0 && errno == ENOSYS && fd >= 0)
600 sync ();
601
602 return res;
603}
604
815/* sync_file_range always needs emulation */ 605/* sync_file_range always needs emulation */
816int 606static int
817eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags) 607eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
818{ 608{
819#if HAVE_SYNC_FILE_RANGE 609#if HAVE_SYNC_FILE_RANGE
820 int res; 610 int res;
821 611
… …
829 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0); 619 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
830 } 620 }
831 621
832 res = sync_file_range (fd, offset, nbytes, flags); 622 res = sync_file_range (fd, offset, nbytes, flags);
833 623
834 if (res != ENOSYS) 624 if (!res || errno != ENOSYS)
835 return res; 625 return res;
836#endif 626#endif
837 627
838 /* even though we could play tricks with the flags, it's better to always 628 /* even though we could play tricks with the flags, it's better to always
839 * call fdatasync, as thta matches the expectation of it's users best */ 629 * call fdatasync, as that matches the expectation of its users best */
840 return fdatasync (fd); 630 return fdatasync (fd);
631}
632
633static int
634eio__fallocate (int fd, int mode, off_t offset, size_t len)
635{
636#if HAVE_LINUX_FALLOCATE
637 return fallocate (fd, mode, offset, len);
638#else
639 return EIO_ENOSYS ();
640#endif
841} 641}
842 642
843#if !HAVE_READAHEAD 643#if !HAVE_READAHEAD
844# undef readahead 644# undef readahead
845# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 645# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
846 646
847static ssize_t 647static eio_ssize_t
848eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) 648eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
849{ 649{
850 size_t todo = count; 650 size_t todo = count;
851 dBUF; 651 dBUF;
852 652
… …
857 pread (fd, eio_buf, len, offset); 657 pread (fd, eio_buf, len, offset);
858 offset += len; 658 offset += len;
859 todo -= len; 659 todo -= len;
860 } 660 }
861 661
862 errno = 0; 662 FUBd;
663
664 /* linux's readahead basically only fails for EBADF or EINVAL (not mmappable) */
665 /* but not for e.g. EIO or eof, so we also never fail */
863 return count; 666 return 0;
864} 667}
865 668
866#endif 669#endif
867 670
868/* sendfile always needs emulation */ 671/* sendfile always needs emulation */
869static ssize_t 672static eio_ssize_t
870eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) 673eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
871{ 674{
675 eio_ssize_t written = 0;
872 ssize_t res; 676 eio_ssize_t res;
873 677
874 if (!count) 678 if (!count)
875 return 0; 679 return 0;
876 680
681 for (;;)
682 {
683#ifdef __APPLE__
684# undef HAVE_SENDFILE /* broken, as everything on os x */
685#endif
877#if HAVE_SENDFILE 686#if HAVE_SENDFILE
878# if __linux 687# if __linux
688 off_t soffset = offset;
879 res = sendfile (ofd, ifd, &offset, count); 689 res = sendfile (ofd, ifd, &soffset, count);
880 690
881# elif __freebsd 691# elif __FreeBSD__
882 /* 692 /*
883 * Of course, the freebsd sendfile is a dire hack with no thoughts 693 * Of course, the freebsd sendfile is a dire hack with no thoughts
884 * wasted on making it similar to other I/O functions. 694 * wasted on making it similar to other I/O functions.
885 */ 695 */
886 {
887 off_t sbytes; 696 off_t sbytes;
888 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 697 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
889 698
890 if (res < 0 && sbytes) 699 #if 0 /* according to the manpage, this is correct, but broken behaviour */
891 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 700 /* freebsd' sendfile will return 0 on success */
701 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
702 /* not on e.g. EIO or EPIPE - sounds broken */
703 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
892 res = sbytes; 704 res = sbytes;
893 } 705 #endif
706
707 /* according to source inspection, this is correct, and useful behaviour */
708 if (sbytes)
709 res = sbytes;
710
711# elif defined __APPLE__
712 off_t sbytes = count;
713 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
714
715 /* according to the manpage, sbytes is always valid */
716 if (sbytes)
717 res = sbytes;
894 718
895# elif __hpux 719# elif __hpux
896 res = sendfile (ofd, ifd, offset, count, 0, 0); 720 res = sendfile (ofd, ifd, offset, count, 0, 0);
897 721
898# elif __solaris 722# elif __solaris
899 {
900 struct sendfilevec vec; 723 struct sendfilevec vec;
901 size_t sbytes; 724 size_t sbytes;
902 725
903 vec.sfv_fd = ifd; 726 vec.sfv_fd = ifd;
904 vec.sfv_flag = 0; 727 vec.sfv_flag = 0;
905 vec.sfv_off = offset; 728 vec.sfv_off = offset;
906 vec.sfv_len = count; 729 vec.sfv_len = count;
907 730
908 res = sendfilev (ofd, &vec, 1, &sbytes); 731 res = sendfilev (ofd, &vec, 1, &sbytes);
909 732
910 if (res < 0 && sbytes) 733 if (res < 0 && sbytes)
911 res = sbytes; 734 res = sbytes;
912 }
913 735
914# endif 736# endif
737
738#elif defined (_WIN32) && 0
739 /* does not work, just for documentation of what would need to be done */
740 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
741 /* libeio guarantees that the file offset does not change, and windows */
742 /* has no way to get an independent handle to the same file description */
743 HANDLE h = TO_SOCKET (ifd);
744 SetFilePointer (h, offset, 0, FILE_BEGIN);
745 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
746
915#else 747#else
916 res = -1; 748 res = EIO_ENOSYS ();
917 errno = ENOSYS;
918#endif 749#endif
919 750
751 /* we assume sendfile can copy at least 128mb in one go */
752 if (res <= 128 * 1024 * 1024)
753 {
754 if (res > 0)
755 written += res;
756
757 if (written)
758 return written;
759
760 break;
761 }
762 else
763 {
764 /* if we requested more, then probably the kernel was lazy */
765 written += res;
766 offset += res;
767 count -= res;
768
769 if (!count)
770 return written;
771 }
772 }
773
920 if (res < 0 774 if (res < 0
921 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 775 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
776 /* BSDs */
777#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
778 || errno == ENOTSUP
779#endif
780#ifdef EOPNOTSUPP /* windows */
781 || errno == EOPNOTSUPP /* BSDs */
782#endif
922#if __solaris 783#if __solaris
923 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 784 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
924#endif 785#endif
925 ) 786 )
926 ) 787 )
… …
930 791
931 res = 0; 792 res = 0;
932 793
933 while (count) 794 while (count)
934 { 795 {
935 ssize_t cnt; 796 eio_ssize_t cnt;
936 797
937 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 798 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
938 799
939 if (cnt <= 0) 800 if (cnt <= 0)
940 { 801 {
… …
952 813
953 offset += cnt; 814 offset += cnt;
954 res += cnt; 815 res += cnt;
955 count -= cnt; 816 count -= cnt;
956 } 817 }
818
819 FUBd;
957 } 820 }
958 821
959 return res; 822 return res;
823}
824
825#ifdef PAGESIZE
826# define eio_pagesize() PAGESIZE
827#else
828static intptr_t
829eio_pagesize (void)
830{
831 static intptr_t page;
832
833 if (!page)
834 page = sysconf (_SC_PAGESIZE);
835
836 return page;
837}
838#endif
839
840static void
841eio_page_align (void **addr, size_t *length)
842{
843 intptr_t mask = eio_pagesize () - 1;
844
845 /* round down addr */
846 intptr_t adj = mask & (intptr_t)*addr;
847
848 *addr = (void *)((intptr_t)*addr - adj);
849 *length += adj;
850
851 /* round up length */
852 *length = (*length + mask) & ~mask;
853}
854
855#if !_POSIX_MEMLOCK
856# define eio__mlockall(a) EIO_ENOSYS ()
857#else
858
859static int
860eio__mlockall (int flags)
861{
862 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
863 extern int mallopt (int, int);
864 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
865 #endif
866
867 if (EIO_MCL_CURRENT != MCL_CURRENT
868 || EIO_MCL_FUTURE != MCL_FUTURE)
869 {
870 flags = 0
871 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
872 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
873 }
874
875 return mlockall (flags);
876}
877#endif
878
879#if !_POSIX_MEMLOCK_RANGE
880# define eio__mlock(a,b) EIO_ENOSYS ()
881#else
882
883static int
884eio__mlock (void *addr, size_t length)
885{
886 eio_page_align (&addr, &length);
887
888 return mlock (addr, length);
889}
890
891#endif
892
893#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
894# define eio__msync(a,b,c) EIO_ENOSYS ()
895#else
896
897static int
898eio__msync (void *mem, size_t len, int flags)
899{
900 eio_page_align (&mem, &len);
901
902 if (EIO_MS_ASYNC != MS_SYNC
903 || EIO_MS_INVALIDATE != MS_INVALIDATE
904 || EIO_MS_SYNC != MS_SYNC)
905 {
906 flags = 0
907 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
908 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
909 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
910 }
911
912 return msync (mem, len, flags);
913}
914
915#endif
916
917static int
918eio__mtouch (eio_req *req)
919{
920 void *mem = req->ptr2;
921 size_t len = req->size;
922 int flags = req->int1;
923
924 eio_page_align (&mem, &len);
925
926 {
927 intptr_t addr = (intptr_t)mem;
928 intptr_t end = addr + len;
929 intptr_t page = eio_pagesize ();
930
931 if (addr < end)
932 if (flags & EIO_MT_MODIFY) /* modify */
933 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
934 else
935 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
936 }
937
938 return 0;
939}
940
941/*****************************************************************************/
942/* requests implemented outside eio_execute, because they are so large */
943
944static void
945eio__lseek (eio_req *req)
946{
947 /* this usually gets optimised away completely, or your compiler sucks, */
948 /* or the whence constants really are not 0, 1, 2 */
949 int whence = req->int2 == EIO_SEEK_SET ? SEEK_SET
950 : req->int2 == EIO_SEEK_CUR ? SEEK_CUR
951 : req->int2 == EIO_SEEK_END ? SEEK_END
952 : req->int2;
953
954 req->offs = lseek (req->int1, req->offs, whence);
955 req->result = req->offs == (off_t)-1 ? -1 : 0;
956}
957
958/* result will always end up in tmpbuf, there is always space for adding a 0-byte */
959static int
960eio__realpath (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
961{
962 char *res;
963 const char *rel = path;
964 char *tmp1, *tmp2;
965#if SYMLOOP_MAX > 32
966 int symlinks = SYMLOOP_MAX;
967#else
968 int symlinks = 32;
969#endif
970
971 errno = EINVAL;
972 if (!rel)
973 return -1;
974
975 errno = ENOENT;
976 if (!*rel)
977 return -1;
978
979 res = etp_tmpbuf_get (tmpbuf, PATH_MAX * 3);
980#ifdef _WIN32
981 if (_access (rel, 4) != 0)
982 return -1;
983
984 symlinks = GetFullPathName (rel, PATH_MAX * 3, res, 0);
985
986 errno = ENAMETOOLONG;
987 if (symlinks >= PATH_MAX * 3)
988 return -1;
989
990 errno = EIO;
991 if (symlinks <= 0)
992 return -1;
993
994 return symlinks;
995
996#else
997 tmp1 = res + PATH_MAX;
998 tmp2 = tmp1 + PATH_MAX;
999
1000#if 0 /* disabled, the musl way to do things is just too racy */
1001#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1002 /* on linux we may be able to ask the kernel */
1003 {
1004 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1005
1006 if (fd >= 0)
1007 {
1008 sprintf (tmp1, "/proc/self/fd/%d", fd);
1009 req->result = readlink (tmp1, res, PATH_MAX);
1010 /* here we should probably stat the open file and the disk file, to make sure they still match */
1011 close (fd);
1012
1013 if (req->result > 0)
1014 goto done;
1015 }
1016 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1017 return -1;
1018 }
1019#endif
1020#endif
1021
1022 if (*rel != '/')
1023 {
1024 int len;
1025
1026 errno = ENOENT;
1027 if (wd == EIO_INVALID_WD)
1028 return -1;
1029
1030 if (wd == EIO_CWD)
1031 {
1032 if (!getcwd (res, PATH_MAX))
1033 return -1;
1034
1035 len = strlen (res);
1036 }
1037 else
1038 memcpy (res, wd->str, len = wd->len);
1039
1040 if (res [1]) /* only use if not / */
1041 res += len;
1042 }
1043
1044 while (*rel)
1045 {
1046 eio_ssize_t len, linklen;
1047 const char *beg = rel;
1048
1049 while (*rel && *rel != '/')
1050 ++rel;
1051
1052 len = rel - beg;
1053
1054 if (!len) /* skip slashes */
1055 {
1056 ++rel;
1057 continue;
1058 }
1059
1060 if (beg [0] == '.')
1061 {
1062 if (len == 1)
1063 continue; /* . - nop */
1064
1065 if (beg [1] == '.' && len == 2)
1066 {
1067 /* .. - back up one component, if possible */
1068
1069 while (res != tmpbuf->ptr)
1070 if (*--res == '/')
1071 break;
1072
1073 continue;
1074 }
1075 }
1076
1077 errno = ENAMETOOLONG;
1078 if (res + 1 + len + 1 >= tmp1)
1079 return -1;
1080
1081 /* copy one component */
1082 *res = '/';
1083 memcpy (res + 1, beg, len);
1084
1085 /* zero-terminate, for readlink */
1086 res [len + 1] = 0;
1087
1088 /* now check if it's a symlink */
1089 linklen = readlink (tmpbuf->ptr, tmp1, PATH_MAX);
1090
1091 if (linklen < 0)
1092 {
1093 if (errno != EINVAL)
1094 return -1;
1095
1096 /* it's a normal directory. hopefully */
1097 res += len + 1;
1098 }
1099 else
1100 {
1101 /* yay, it was a symlink - build new path in tmp2 */
1102 int rellen = strlen (rel);
1103
1104 errno = ENAMETOOLONG;
1105 if (linklen + 1 + rellen >= PATH_MAX)
1106 return -1;
1107
1108 errno = ELOOP;
1109 if (!--symlinks)
1110 return -1;
1111
1112 if (*tmp1 == '/')
1113 res = tmpbuf->ptr; /* symlink resolves to an absolute path */
1114
1115 /* we need to be careful, as rel might point into tmp2 already */
1116 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1117 tmp2 [linklen] = '/';
1118 memcpy (tmp2, tmp1, linklen);
1119
1120 rel = tmp2;
1121 }
1122 }
1123
1124 /* special case for the lone root path */
1125 if (res == tmpbuf->ptr)
1126 *res++ = '/';
1127
1128 return res - (char *)tmpbuf->ptr;
1129#endif
1130}
1131
1132static signed char
1133eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1134{
1135 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1136 : a->inode < b->inode ? -1
1137 : a->inode > b->inode ? 1
1138 : 0;
1139}
1140
1141#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1142
1143#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1144#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1145
1146static void
1147eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1148{
1149 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1150 unsigned char *bit = bits;
1151
1152 assert (CHAR_BIT == 8);
1153 assert (sizeof (eio_dirent) * 8 < 256);
1154 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1155 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1156
1157 if (size <= EIO_SORT_FAST)
1158 return;
1159
1160 /* first prepare an array of bits to test in our radix sort */
1161 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1162 /* inode_bits must contain all inodes ORed together */
1163 /* which is used to skip bits that are 0 everywhere, which is very common */
1164 {
1165 eio_ino_t endianness;
1166 int i, j;
1167
1168 /* we store the byte offset of byte n into byte n of "endianness" */
1169 for (i = 0; i < sizeof (eio_ino_t); ++i)
1170 ((unsigned char *)&endianness)[i] = i;
1171
1172 *bit++ = 0;
1173
1174 for (i = 0; i < sizeof (eio_ino_t); ++i)
1175 {
1176 /* shifting off the byte offsets out of "endianness" */
1177 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1178 endianness >>= 8;
1179
1180 for (j = 0; j < 8; ++j)
1181 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1182 *bit++ = offs + j;
1183 }
1184
1185 for (j = 0; j < 8; ++j)
1186 if (score_bits & (1 << j))
1187 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1188 }
1189
1190 /* now actually do the sorting (a variant of MSD radix sort) */
1191 {
1192 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1193 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1194 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1195 int stk_idx = 0;
1196
1197 base_stk [stk_idx] = dents;
1198 end_stk [stk_idx] = dents + size;
1199 bit_stk [stk_idx] = bit - 1;
1200
1201 do
1202 {
1203 base = base_stk [stk_idx];
1204 end = end_stk [stk_idx];
1205 bit = bit_stk [stk_idx];
1206
1207 for (;;)
1208 {
1209 unsigned char O = *bit >> 3;
1210 unsigned char M = 1 << (*bit & 7);
1211
1212 eio_dirent *a = base;
1213 eio_dirent *b = end;
1214
1215 if (b - a < EIO_SORT_CUTOFF)
1216 break;
1217
1218 /* now bit-partition the array on the bit */
1219 /* this ugly asymmetric loop seems to perform much better than typical */
1220 /* partition algos found in the literature */
1221 do
1222 if (!(((unsigned char *)a)[O] & M))
1223 ++a;
1224 else if (!(((unsigned char *)--b)[O] & M))
1225 {
1226 eio_dirent tmp = *a; *a = *b; *b = tmp;
1227 ++a;
1228 }
1229 while (b > a);
1230
1231 /* next bit, or stop, if no bits left in this path */
1232 if (!*--bit)
1233 break;
1234
1235 base_stk [stk_idx] = a;
1236 end_stk [stk_idx] = end;
1237 bit_stk [stk_idx] = bit;
1238 ++stk_idx;
1239
1240 end = a;
1241 }
1242 }
1243 while (stk_idx--);
1244 }
1245}
1246
1247static void
1248eio_dent_insertion_sort (eio_dirent *dents, int size)
1249{
1250 /* first move the smallest element to the front, to act as a sentinel */
1251 {
1252 int i;
1253 eio_dirent *min = dents;
1254
1255 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1256 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1257 if (EIO_DENT_CMP (dents [i], <, *min))
1258 min = &dents [i];
1259
1260 /* swap elements 0 and j (minimum) */
1261 {
1262 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1263 }
1264 }
1265
1266 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1267 {
1268 eio_dirent *i, *j;
1269
1270 for (i = dents + 1; i < dents + size; ++i)
1271 {
1272 eio_dirent value = *i;
1273
1274 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1275 j [1] = j [0];
1276
1277 j [1] = value;
1278 }
1279 }
1280}
1281
1282static void
1283eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1284{
1285 if (size <= 1)
1286 return; /* our insertion sort relies on size > 0 */
1287
1288 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1289 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1290 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1291
1292 /* use an insertion sort at the end, or for small arrays, */
1293 /* as insertion sort is more efficient for small partitions */
1294 eio_dent_insertion_sort (dents, size);
960} 1295}
961 1296
962/* read a full directory */ 1297/* read a full directory */
963static void 1298static void
964eio__scandir (eio_req *req, etp_worker *self) 1299eio__scandir (eio_req *req, etp_worker *self)
965{ 1300{
1301 char *name, *names;
1302 int namesalloc = 4096 - sizeof (void *) * 4;
1303 int namesoffs = 0;
1304 int flags = req->int1;
1305 eio_dirent *dents = 0;
1306 int dentalloc = 128;
1307 int dentoffs = 0;
1308 eio_ino_t inode_bits = 0;
1309#ifdef _WIN32
1310 HANDLE dirp;
1311 WIN32_FIND_DATA entp;
1312#else
966 DIR *dirp; 1313 DIR *dirp;
967 EIO_STRUCT_DIRENT *entp; 1314 EIO_STRUCT_DIRENT *entp;
968 char *name, *names; 1315#endif
969 int memlen = 4096;
970 int memofs = 0;
971 int res = 0;
972 1316
973 X_LOCK (wrklock); 1317 req->result = -1;
974 /* the corresponding closedir is in ETP_WORKER_CLEAR */
975 self->dirp = dirp = opendir (req->ptr1);
976 req->flags |= EIO_FLAG_PTR2_FREE;
977 req->ptr2 = names = malloc (memlen);
978 X_UNLOCK (wrklock);
979 1318
980 if (dirp && names) 1319 if (!(flags & EIO_READDIR_DENTS))
981 for (;;) 1320 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1321
1322#ifdef _WIN32
1323 {
1324 int len = strlen ((const char *)req->ptr1);
1325 char *path = malloc (MAX_PATH);
1326 const char *fmt;
1327 const char *reqpath = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1328
1329 if (!len)
1330 fmt = "./*";
1331 else if (reqpath[len - 1] == '/' || reqpath[len - 1] == '\\')
1332 fmt = "%s*";
1333 else
1334 fmt = "%s/*";
1335
1336 _snprintf (path, MAX_PATH, fmt, reqpath);
1337 dirp = FindFirstFile (path, &entp);
1338 free (path);
1339
1340 if (dirp == INVALID_HANDLE_VALUE)
1341 {
1342 /* should steal _dosmaperr */
1343 switch (GetLastError ())
1344 {
1345 case ERROR_FILE_NOT_FOUND:
1346 req->result = 0;
1347 break;
1348
1349 case ERROR_INVALID_NAME:
1350 case ERROR_PATH_NOT_FOUND:
1351 case ERROR_NO_MORE_FILES:
1352 errno = ENOENT;
1353 break;
1354
1355 case ERROR_NOT_ENOUGH_MEMORY:
1356 errno = ENOMEM;
1357 break;
1358
1359 default:
1360 errno = EINVAL;
1361 break;
1362 }
1363
1364 return;
1365 }
1366 }
1367#else
1368 #if HAVE_AT
1369 if (req->wd)
982 { 1370 {
1371 int fd = openat (WD2FD (req->wd), req->ptr1, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1372
1373 if (fd < 0)
1374 return;
1375
1376 dirp = fdopendir (fd);
1377
1378 if (!dirp)
1379 close (fd);
1380 }
1381 else
1382 dirp = opendir (req->ptr1);
1383 #else
1384 dirp = opendir (wd_expand (&self->tmpbuf, req->wd, req->ptr1));
1385 #endif
1386
1387 if (!dirp)
1388 return;
1389#endif
1390
1391 if (req->flags & EIO_FLAG_PTR1_FREE)
1392 free (req->ptr1);
1393
1394 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1395 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1396 req->ptr2 = names = malloc (namesalloc);
1397
1398 if (!names || (flags && !dents))
1399 return;
1400
1401 for (;;)
1402 {
1403 int done;
1404
1405#ifdef _WIN32
1406 done = !dirp;
1407#else
983 errno = 0; 1408 errno = 0;
984 entp = readdir (dirp); 1409 entp = readdir (dirp);
1410 done = !entp;
1411#endif
985 1412
1413 if (done)
1414 {
1415#ifndef _WIN32
1416 int old_errno = errno;
1417 closedir (dirp);
1418 errno = old_errno;
1419
986 if (!entp) 1420 if (errno)
1421 break;
1422#endif
1423
1424 /* sort etc. */
1425 req->int1 = flags;
1426 req->result = dentoffs;
1427
1428 if (flags & EIO_READDIR_STAT_ORDER)
1429 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1430 else if (flags & EIO_READDIR_DIRS_FIRST)
1431 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1432 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1433 else
1434 {
1435 /* in this case, all is known, and we just put dirs first and sort them */
1436 eio_dirent *oth = dents + dentoffs;
1437 eio_dirent *dir = dents;
1438
1439 /* now partition dirs to the front, and non-dirs to the back */
1440 /* by walking from both sides and swapping if necessary */
1441 while (oth > dir)
1442 {
1443 if (dir->type == EIO_DT_DIR)
1444 ++dir;
1445 else if ((--oth)->type == EIO_DT_DIR)
1446 {
1447 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1448
1449 ++dir;
1450 }
1451 }
1452
1453 /* now sort the dirs only (dirs all have the same score) */
1454 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1455 }
1456
987 break; 1457 break;
1458 }
988 1459
989 name = entp->d_name; 1460 /* now add the entry to our list(s) */
1461 name = D_NAME (entp);
990 1462
1463 /* skip . and .. entries */
991 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1464 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
992 { 1465 {
993 int len = strlen (name) + 1; 1466 int len = D_NAMLEN (entp) + 1;
994 1467
995 res++; 1468 while (ecb_expect_false (namesoffs + len > namesalloc))
996
997 while (memofs + len > memlen)
998 { 1469 {
999 memlen *= 2; 1470 namesalloc *= 2;
1000 X_LOCK (wrklock);
1001 req->ptr2 = names = realloc (names, memlen); 1471 req->ptr2 = names = realloc (names, namesalloc);
1002 X_UNLOCK (wrklock);
1003 1472
1004 if (!names) 1473 if (!names)
1474 break;
1475 }
1476
1477 memcpy (names + namesoffs, name, len);
1478
1479 if (dents)
1480 {
1481 struct eio_dirent *ent;
1482
1483 if (ecb_expect_false (dentoffs == dentalloc))
1484 {
1485 dentalloc *= 2;
1486 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1487
1488 if (!dents)
1005 break; 1489 break;
1006 } 1490 }
1007 1491
1008 memcpy (names + memofs, name, len); 1492 ent = dents + dentoffs;
1009 memofs += len; 1493
1494 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1495 ent->namelen = len - 1;
1496 ent->inode = D_INO (entp);
1497
1498 inode_bits |= ent->inode;
1499
1500 switch (D_TYPE (entp))
1501 {
1502 default:
1503 ent->type = EIO_DT_UNKNOWN;
1504 flags |= EIO_READDIR_FOUND_UNKNOWN;
1505 break;
1506
1507 #ifdef DT_FIFO
1508 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1509 #endif
1510 #ifdef DT_CHR
1511 case DT_CHR: ent->type = EIO_DT_CHR; break;
1512 #endif
1513 #ifdef DT_MPC
1514 case DT_MPC: ent->type = EIO_DT_MPC; break;
1515 #endif
1516 #ifdef DT_DIR
1517 case DT_DIR: ent->type = EIO_DT_DIR; break;
1518 #endif
1519 #ifdef DT_NAM
1520 case DT_NAM: ent->type = EIO_DT_NAM; break;
1521 #endif
1522 #ifdef DT_BLK
1523 case DT_BLK: ent->type = EIO_DT_BLK; break;
1524 #endif
1525 #ifdef DT_MPB
1526 case DT_MPB: ent->type = EIO_DT_MPB; break;
1527 #endif
1528 #ifdef DT_REG
1529 case DT_REG: ent->type = EIO_DT_REG; break;
1530 #endif
1531 #ifdef DT_NWK
1532 case DT_NWK: ent->type = EIO_DT_NWK; break;
1533 #endif
1534 #ifdef DT_CMP
1535 case DT_CMP: ent->type = EIO_DT_CMP; break;
1536 #endif
1537 #ifdef DT_LNK
1538 case DT_LNK: ent->type = EIO_DT_LNK; break;
1539 #endif
1540 #ifdef DT_SOCK
1541 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1542 #endif
1543 #ifdef DT_DOOR
1544 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1545 #endif
1546 #ifdef DT_WHT
1547 case DT_WHT: ent->type = EIO_DT_WHT; break;
1548 #endif
1549 }
1550
1551 ent->score = 7;
1552
1553 if (flags & EIO_READDIR_DIRS_FIRST)
1554 {
1555 if (ent->type == EIO_DT_UNKNOWN)
1556 {
1557 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1558 ent->score = 1;
1559 else if (!strchr (name, '.')) /* absence of dots indicate likely dirs */
1560 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1561 }
1562 else if (ent->type == EIO_DT_DIR)
1563 ent->score = 0;
1564 }
1010 } 1565 }
1566
1567 namesoffs += len;
1568 ++dentoffs;
1011 } 1569 }
1012 1570
1013 if (errno) 1571 if (EIO_CANCELLED (req))
1014 res = -1; 1572 {
1573 errno = ECANCELED;
1574 break;
1575 }
1576
1577#ifdef _WIN32
1578 if (!FindNextFile (dirp, &entp))
1579 {
1580 FindClose (dirp);
1581 dirp = 0;
1582 }
1583#endif
1584 }
1585}
1586
1587/*****************************************************************************/
1588/* working directory stuff */
1589/* various deficiencies in the posix 2008 api force us to */
1590/* keep the absolute path in string form at all times */
1591/* fuck yeah. */
1592
1593#if !HAVE_AT
1594
1595/* a bit like realpath, but usually faster because it doesn'T have to return */
1596/* an absolute or canonical path */
1597static const char *
1598wd_expand (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
1599{
1600 if (!wd || *path == '/')
1601 return path;
1602
1603 if (path [0] == '.' && !path [1])
1604 return wd->str;
1605
1015 1606 {
1016 req->result = res; 1607 int l1 = wd->len;
1017} 1608 int l2 = strlen (path);
1018 1609
1019#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) 1610 char *res = etp_tmpbuf_get (tmpbuf, l1 + l2 + 2);
1020# undef msync
1021# define msync(a,b,c) ENOSYS
1022#endif
1023 1611
1024int 1612 memcpy (res, wd->str, l1);
1025eio__mtouch (void *mem, size_t len, int flags) 1613 res [l1] = '/';
1026{ 1614 memcpy (res + l1 + 1, path, l2 + 1);
1027 intptr_t addr = (intptr_t)mem;
1028 intptr_t end = addr + len;
1029#ifdef PAGESIZE
1030 const intptr_t page = PAGESIZE;
1031#else
1032 static intptr_t page;
1033 1615
1034 if (!page) 1616 return res;
1035 page = sysconf (_SC_PAGESIZE); 1617 }
1036#endif 1618}
1037 1619
1038 addr &= ~(page - 1); /* assume page size is always a power of two */ 1620#endif
1039 1621
1040 if (addr < end) 1622static eio_wd
1041 if (flags) /* modify */ 1623eio__wd_open_sync (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
1042 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len); 1624{
1043 else 1625 int fd;
1044 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len); 1626 eio_wd res;
1627 int len = eio__realpath (tmpbuf, wd, path);
1045 1628
1629 if (len < 0)
1630 return EIO_INVALID_WD;
1631
1632#if HAVE_AT
1633 fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1634
1635 if (fd < 0)
1636 return EIO_INVALID_WD;
1637#endif
1638
1639 res = malloc (sizeof (*res) + len); /* one extra 0-byte */
1640
1641#if HAVE_AT
1642 res->fd = fd;
1643#endif
1644
1645 res->len = len;
1646 memcpy (res->str, tmpbuf->ptr, len);
1647 res->str [len] = 0;
1648
1649 return res;
1650}
1651
1652eio_wd
1653eio_wd_open_sync (eio_wd wd, const char *path)
1654{
1655 struct etp_tmpbuf tmpbuf = { };
1656 wd = eio__wd_open_sync (&tmpbuf, wd, path);
1657 free (tmpbuf.ptr);
1658
1046 return 0; 1659 return wd;
1047} 1660}
1661
1662void
1663eio_wd_close_sync (eio_wd wd)
1664{
1665 if (wd != EIO_INVALID_WD && wd != EIO_CWD)
1666 {
1667 #if HAVE_AT
1668 close (wd->fd);
1669 #endif
1670 free (wd);
1671 }
1672}
1673
1674#if HAVE_AT
1675
1676/* they forgot these */
1677
1678static int
1679eio__truncateat (int dirfd, const char *path, off_t length)
1680{
1681 int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
1682 int res;
1683
1684 if (fd < 0)
1685 return fd;
1686
1687 res = ftruncate (fd, length);
1688 close (fd);
1689 return res;
1690}
1691
1692static int
1693eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
1694{
1695 int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
1696 int res;
1697
1698 if (fd < 0)
1699 return fd;
1700
1701 res = fstatvfs (fd, buf);
1702 close (fd);
1703 return res;
1704
1705}
1706
1707#endif
1048 1708
1049/*****************************************************************************/ 1709/*****************************************************************************/
1050 1710
1051#define ALLOC(len) \ 1711#define ALLOC(len) \
1052 if (!req->ptr2) \ 1712 if (!req->ptr2) \
1053 { \ 1713 { \
1054 X_LOCK (wrklock); \ 1714 X_LOCK (EIO_POOL->wrklock); \
1055 req->flags |= EIO_FLAG_PTR2_FREE; \ 1715 req->flags |= EIO_FLAG_PTR2_FREE; \
1056 X_UNLOCK (wrklock); \ 1716 X_UNLOCK (EIO_POOL->wrklock); \
1057 req->ptr2 = malloc (len); \ 1717 req->ptr2 = malloc (len); \
1058 if (!req->ptr2) \ 1718 if (!req->ptr2) \
1059 { \ 1719 { \
1060 errno = ENOMEM; \ 1720 errno = ENOMEM; \
1061 req->result = -1; \ 1721 req->result = -1; \
1062 break; \ 1722 break; \
1063 } \ 1723 } \
1064 } 1724 }
1065 1725
1066X_THREAD_PROC (etp_proc)
1067{
1068 ETP_REQ *req;
1069 struct timespec ts;
1070 etp_worker *self = (etp_worker *)thr_arg;
1071
1072 /* try to distribute timeouts somewhat randomly */
1073 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1074
1075 for (;;)
1076 {
1077 X_LOCK (reqlock);
1078
1079 for (;;)
1080 {
1081 self->req = req = reqq_shift (&req_queue);
1082
1083 if (req)
1084 break;
1085
1086 ++idle;
1087
1088 ts.tv_sec = time (0) + IDLE_TIMEOUT;
1089 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
1090 {
1091 if (idle > max_idle)
1092 {
1093 --idle;
1094 X_UNLOCK (reqlock);
1095 X_LOCK (wrklock);
1096 --started;
1097 X_UNLOCK (wrklock);
1098 goto quit;
1099 }
1100
1101 /* we are allowed to idle, so do so without any timeout */
1102 X_COND_WAIT (reqwait, reqlock);
1103 }
1104
1105 --idle;
1106 }
1107
1108 --nready;
1109
1110 X_UNLOCK (reqlock);
1111
1112 if (req->type < 0)
1113 goto quit;
1114
1115 if (!EIO_CANCELLED (req))
1116 ETP_EXECUTE (self, req);
1117
1118 X_LOCK (reslock);
1119
1120 ++npending;
1121
1122 if (!reqq_push (&res_queue, req) && want_poll_cb)
1123 want_poll_cb ();
1124
1125 self->req = 0;
1126 etp_worker_clear (self);
1127
1128 X_UNLOCK (reslock);
1129 }
1130
1131quit:
1132 X_LOCK (wrklock);
1133 etp_worker_free (self);
1134 X_UNLOCK (wrklock);
1135
1136 return 0;
1137}
1138
1139/*****************************************************************************/ 1726/*****************************************************************************/
1140 1727
1728int ecb_cold
1141int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1729eio_init (void (*want_poll)(void), void (*done_poll)(void))
1142{ 1730{
1143 return etp_init (want_poll, done_poll); 1731 eio_want_poll_cb = want_poll;
1144} 1732 eio_done_poll_cb = done_poll;
1145 1733
1734 return etp_init (EIO_POOL, 0, 0, 0);
1735}
1736
1737ecb_inline void
1146static void eio_api_destroy (eio_req *req) 1738eio_api_destroy (eio_req *req)
1147{ 1739{
1148 free (req); 1740 free (req);
1149} 1741}
1150 1742
1151#define REQ(rtype) \ 1743#define REQ(rtype) \
1152 eio_req *req; \ 1744 eio_req *req; \
1153 \ 1745 \
1154 req = (eio_req *)calloc (1, sizeof *req); \ 1746 req = (eio_req *)calloc (1, sizeof *req); \
1155 if (!req) \ 1747 if (!req) \
1156 return 0; \ 1748 return 0; \
… …
1170 { \ 1762 { \
1171 eio_api_destroy (req); \ 1763 eio_api_destroy (req); \
1172 return 0; \ 1764 return 0; \
1173 } 1765 }
1174 1766
1767#define SINGLEDOT(ptr) (0[(char *)(ptr)] == '.' && !1[(char *)(ptr)])
1768
1769static void
1175static void eio_execute (etp_worker *self, eio_req *req) 1770eio_execute (etp_worker *self, eio_req *req)
1176{ 1771{
1177 errno = 0; 1772#if HAVE_AT
1773 int dirfd;
1774#else
1775 const char *path;
1776#endif
1777
1778 if (ecb_expect_false (EIO_CANCELLED (req)))
1779 {
1780 req->result = -1;
1781 req->errorno = ECANCELED;
1782 return;
1783 }
1784
1785 if (ecb_expect_false (req->wd == EIO_INVALID_WD))
1786 {
1787 req->result = -1;
1788 req->errorno = ENOENT;
1789 return;
1790 }
1791
1792 if (req->type >= EIO_OPEN)
1793 {
1794 #if HAVE_AT
1795 dirfd = WD2FD (req->wd);
1796 #else
1797 path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1798 #endif
1799 }
1178 1800
1179 switch (req->type) 1801 switch (req->type)
1180 { 1802 {
1803 case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
1804 req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
1805 break;
1806 case EIO_WD_CLOSE: req->result = 0;
1807 eio_wd_close_sync (req->wd); break;
1808
1809 case EIO_SEEK: eio__lseek (req); break;
1181 case EIO_READ: ALLOC (req->size); 1810 case EIO_READ: ALLOC (req->size);
1182 req->result = req->offs >= 0 1811 req->result = req->offs >= 0
1183 ? pread (req->int1, req->ptr2, req->size, req->offs) 1812 ? pread (req->int1, req->ptr2, req->size, req->offs)
1184 : read (req->int1, req->ptr2, req->size); break; 1813 : read (req->int1, req->ptr2, req->size); break;
1185 case EIO_WRITE: req->result = req->offs >= 0 1814 case EIO_WRITE: req->result = req->offs >= 0
1186 ? pwrite (req->int1, req->ptr2, req->size, req->offs) 1815 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1187 : write (req->int1, req->ptr2, req->size); break; 1816 : write (req->int1, req->ptr2, req->size); break;
1188 1817
1818 case EIO_FCNTL: req->result = fcntl (req->int1, (int) req->int2, req->ptr2); break;
1819 case EIO_IOCTL: req->result = ioctl (req->int1, (unsigned long)req->int2, req->ptr2); break;
1820
1189 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;
1190 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
1191 1825
1192 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1826 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1193 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;
1194 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1828 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1195 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;
1196 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1197 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1198
1199 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;
1200 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1201 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;
1202 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1203 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;
1204 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1205
1206 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;
1207 case EIO_CLOSE: req->result = close (req->int1); break; 1834
1208 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1209 case EIO_UNLINK: req->result = unlink (req->ptr1); break; 1835 case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
1210 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;
1211 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;
1212 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;
1213 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;
1214 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; 1846 case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
1215 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;
1216
1217 case EIO_READLINK: ALLOC (NAME_MAX); 1848 case EIO_READLINK: ALLOC (PATH_MAX);
1218 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; 1849 req->result = readlinkat (dirfd, req->ptr1, req->ptr2, PATH_MAX); break;
1219 1850 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1220 case EIO_SYNC: req->result = 0; sync (); break; 1851 req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1221 case EIO_FSYNC: req->result = fsync (req->int1); break;
1222 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1223 case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break;
1224 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;
1225 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1226
1227 case EIO_READDIR: eio__scandir (req, self); break;
1228
1229 case EIO_BUSY: 1852 case EIO_UTIME:
1230#ifdef _WIN32 1853 case EIO_FUTIME:
1231 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
1232#else 1876#else
1233 {
1234 struct timeval tv;
1235 1877
1236 tv.tv_sec = req->nv1; 1878 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1237 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;
1238 1886
1239 req->result = select (0, 0, 0, 0, &tv); 1887 case EIO_UNLINK: req->result = unlink (path ); break;
1240 } 1888 case EIO_RMDIR: req->result = rmdir (path ); break;
1241#endif 1889 case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
1242 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;
1243 1898
1244 case EIO_UTIME: 1899 case EIO_UTIME:
1245 case EIO_FUTIME: 1900 case EIO_FUTIME:
1246 { 1901 {
1247 struct timeval tv[2]; 1902 struct timeval tv[2];
1248 struct timeval *times; 1903 struct timeval *times;
1249 1904
1250 if (req->nv1 != -1. || req->nv2 != -1.) 1905 if (req->nv1 != -1. || req->nv2 != -1.)
1251 { 1906 {
1252 tv[0].tv_sec = req->nv1; 1907 tv[0].tv_sec = req->nv1;
1253 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.; 1908 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
1254 tv[1].tv_sec = req->nv2; 1909 tv[1].tv_sec = req->nv2;
1255 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.; 1910 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
1256 1911
1257 times = tv; 1912 times = tv;
1258 } 1913 }
1259 else 1914 else
1260 times = 0; 1915 times = 0;
1261
1262 1916
1263 req->result = req->type == EIO_FUTIME 1917 req->result = req->type == EIO_FUTIME
1264 ? futimes (req->int1, times) 1918 ? futimes (req->int1, times)
1265 : utimes (req->ptr1, times); 1919 : utimes (req->ptr1, times);
1266 } 1920 }
1267 break; 1921 break;
1268 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
1269 case EIO_GROUP: 1973 case EIO_GROUP:
1270 abort (); /* handled in eio_request */ 1974 abort (); /* handled in eio_request */
1975#endif
1271 1976
1272 case EIO_NOP: 1977 case EIO_NOP:
1273 req->result = 0; 1978 req->result = 0;
1274 break; 1979 break;
1275 1980
1276 case EIO_CUSTOM: 1981 case EIO_CUSTOM:
1277 ((void (*)(eio_req *))req->feed) (req); 1982 req->feed (req);
1278 break; 1983 break;
1279 1984
1280 default: 1985 default:
1281 req->result = -1; 1986 req->result = EIO_ENOSYS ();
1282 break; 1987 break;
1283 } 1988 }
1284 1989
1285 req->errorno = errno; 1990 req->errorno = errno;
1286} 1991}
1287 1992
1288#ifndef EIO_NO_WRAPPERS 1993#ifndef EIO_NO_WRAPPERS
1289 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
1290eio_req *eio_nop (int pri, eio_cb cb, void *data) 2005eio_req *eio_nop (int pri, eio_cb cb, void *data)
1291{ 2006{
1292 REQ (EIO_NOP); SEND; 2007 REQ (EIO_NOP); SEND;
1293} 2008}
1294 2009
… …
1310eio_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)
1311{ 2026{
1312 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;
1313} 2028}
1314 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
1315eio_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)
1316{ 2046{
1317 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;
1318} 2048}
1319 2049
1320eio_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)
1321{ 2051{
1322 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;
1323} 2053}
1324 2054
1325eio_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)
1326{ 2056{
1327 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;
1328} 2063}
1329 2064
1330eio_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)
1331{ 2066{
1332 REQ (EIO_CLOSE); req->int1 = fd; SEND; 2067 REQ (EIO_CLOSE); req->int1 = fd; SEND;
… …
1335eio_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)
1336{ 2071{
1337 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;
1338} 2073}
1339 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
1340eio_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)
1341{ 2081{
1342 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;
1343} 2083}
1344 2084
1345eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data) 2085eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1346{ 2086{
1347 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND; 2087 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1348} 2088}
1349 2089
2090eio_req *eio_fcntl (int fd, int cmd, void *arg, int pri, eio_cb cb, void *data)
2091{
2092 REQ (EIO_IOCTL); req->int1 = fd; req->int2 = cmd; req->ptr2 = arg; SEND;
2093}
2094
2095eio_req *eio_ioctl (int fd, unsigned long request, void *buf, int pri, eio_cb cb, void *data)
2096{
2097 REQ (EIO_IOCTL); req->int1 = fd; req->int2 = request; req->ptr2 = buf; SEND;
2098}
2099
1350eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) 2100eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1351{ 2101{
1352 REQ (EIO_FSTAT); req->int1 = fd; SEND; 2102 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1353} 2103}
1354 2104
2105eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
2106{
2107 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
2108}
2109
1355eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) 2110eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1356{ 2111{
1357 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; 2112 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1358} 2113}
1359 2114
… …
1365eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) 2120eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1366{ 2121{
1367 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2122 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1368} 2123}
1369 2124
1370eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2125eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1371{ 2126{
1372 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2127 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1373} 2128}
1374 2129
1375eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) 2130eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
… …
1395eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data) 2150eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1396{ 2151{
1397 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2152 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1398} 2153}
1399 2154
1400eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2155eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1401{ 2156{
1402 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2157 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1403} 2158}
1404 2159
1405eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) 2160eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
… …
1421eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data) 2176eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1422{ 2177{
1423 return eio__1path (EIO_READLINK, path, pri, cb, data); 2178 return eio__1path (EIO_READLINK, path, pri, cb, data);
1424} 2179}
1425 2180
2181eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2182{
2183 return eio__1path (EIO_REALPATH, path, pri, cb, data);
2184}
2185
1426eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) 2186eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1427{ 2187{
1428 return eio__1path (EIO_STAT, path, pri, cb, data); 2188 return eio__1path (EIO_STAT, path, pri, cb, data);
1429} 2189}
1430 2190
1431eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) 2191eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1432{ 2192{
1433 return eio__1path (EIO_LSTAT, path, pri, cb, data); 2193 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1434} 2194}
1435 2195
2196eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
2197{
2198 return eio__1path (EIO_STATVFS, path, pri, cb, data);
2199}
2200
1436eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) 2201eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1437{ 2202{
1438 return eio__1path (EIO_UNLINK, path, pri, cb, data); 2203 return eio__1path (EIO_UNLINK, path, pri, cb, data);
1439} 2204}
1440 2205
1441eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data) 2206eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1442{ 2207{
1443 return eio__1path (EIO_RMDIR, path, pri, cb, data); 2208 return eio__1path (EIO_RMDIR, path, pri, cb, data);
1444} 2209}
1445 2210
1446eio_req *eio_readdir (const char *path, int pri, eio_cb cb, void *data) 2211eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1447{ 2212{
1448 return eio__1path (EIO_READDIR, path, pri, cb, data); 2213 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1449} 2214}
1450 2215
1451eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) 2216eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1452{ 2217{
1453 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; 2218 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1454} 2219}
1455 2220
1456static eio_req * 2221static eio_req *
1457eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2222eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1458{ 2223{
… …
1482eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2247eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1483{ 2248{
1484 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2249 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1485} 2250}
1486 2251
1487eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) 2252eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
1488{ 2253{
1489 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND; 2254 REQ (EIO_CUSTOM); req->feed = execute; SEND;
1490} 2255}
1491 2256
1492#endif 2257#endif
1493 2258
1494eio_req *eio_grp (eio_cb cb, void *data) 2259eio_req *eio_grp (eio_cb cb, void *data)
… …
1503#undef SEND 2268#undef SEND
1504 2269
1505/*****************************************************************************/ 2270/*****************************************************************************/
1506/* grp functions */ 2271/* grp functions */
1507 2272
2273void
1508void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2274eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1509{ 2275{
1510 grp->int2 = limit; 2276 grp->int2 = limit;
1511 grp->feed = feed; 2277 grp->feed = feed;
1512 2278
1513 grp_try_feed (grp); 2279 grp_try_feed (grp);
1514} 2280}
1515 2281
2282void
1516void eio_grp_limit (eio_req *grp, int limit) 2283eio_grp_limit (eio_req *grp, int limit)
1517{ 2284{
1518 grp->int2 = limit; 2285 grp->int2 = limit;
1519 2286
1520 grp_try_feed (grp); 2287 grp_try_feed (grp);
1521} 2288}
1522 2289
2290void
1523void eio_grp_add (eio_req *grp, eio_req *req) 2291eio_grp_add (eio_req *grp, eio_req *req)
1524{ 2292{
1525 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2293 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1526 2294
1527 grp->flags |= EIO_FLAG_GROUPADD; 2295 grp->flags |= ETP_FLAG_GROUPADD;
1528 2296
1529 ++grp->size; 2297 ++grp->size;
1530 req->grp = grp; 2298 req->grp = grp;
1531 2299
1532 req->grp_prev = 0; 2300 req->grp_prev = 0;
… …
1539} 2307}
1540 2308
1541/*****************************************************************************/ 2309/*****************************************************************************/
1542/* misc garbage */ 2310/* misc garbage */
1543 2311
2312eio_ssize_t
1544ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2313eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1545{ 2314{
1546 etp_worker wrk;
1547
1548 wrk.dbuf = 0;
1549
1550 eio__sendfile (ofd, ifd, offset, count, &wrk); 2315 return eio__sendfile (ofd, ifd, offset, count);
1551
1552 if (wrk.dbuf)
1553 free (wrk.dbuf);
1554} 2316}
1555 2317

Diff Legend

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