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

Comparing libeio/eio.c (file contents):
Revision 1.74 by root, Mon Jun 20 07:28:15 2011 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) 2007,2008,2009,2010,2011 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,
… …
42#endif 42#endif
43 43
44#include "eio.h" 44#include "eio.h"
45#include "ecb.h" 45#include "ecb.h"
46 46
47#ifdef EIO_STACKSIZE
48# define XTHREAD_STACKSIZE EIO_STACKSIZE
49#endif
50#include "xthread.h"
51
52#include <errno.h> 47#include <errno.h>
53#include <stddef.h> 48#include <stddef.h>
54#include <stdlib.h> 49#include <stdlib.h>
55#include <string.h> 50#include <string.h>
56#include <errno.h> 51#include <errno.h>
57#include <sys/types.h> 52#include <sys/types.h>
58#include <sys/stat.h> 53#include <sys/stat.h>
59#include <sys/statvfs.h>
60#include <limits.h> 54#include <limits.h>
61#include <fcntl.h> 55#include <fcntl.h>
62#include <assert.h> 56#include <assert.h>
63 57
64/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */ 58/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
65/* intptr_t only comes form stdint.h, says idiot openbsd coder */ 59/* intptr_t only comes from stdint.h, says idiot openbsd coder */
66#if HAVE_STDINT_H 60#if HAVE_STDINT_H
67# include <stdint.h> 61# include <stdint.h>
68#endif 62#endif
69 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
70#ifndef EIO_FINISH 77#ifndef EIO_FINISH
71# 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
72#endif 79#endif
73 80
74#ifndef EIO_DESTROY 81#ifndef EIO_DESTROY
… …
77 84
78#ifndef EIO_FEED 85#ifndef EIO_FEED
79# 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)
80#endif 87#endif
81 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
82#ifdef _WIN32 100#ifdef _WIN32
83 101
84 /*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)
233
85#else 234#else
86 235
87# include <sys/time.h> 236 #include <sys/time.h>
88# include <sys/select.h> 237 #include <sys/select.h>
89# 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
90# include <utime.h> 282# include <utime.h>
91# include <signal.h>
92# include <dirent.h>
93
94#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
95# include <sys/mman.h>
96#endif
97
98/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
99# if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
100# define _DIRENT_HAVE_D_TYPE /* sigh */
101# define D_INO(de) (de)->d_fileno
102# define D_NAMLEN(de) (de)->d_namlen
103# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
104# define D_INO(de) (de)->d_ino
105# endif 283#endif
106 284
107#ifdef _D_EXACT_NAMLEN 285#if HAVE_SYS_SYSCALL_H
108# undef D_NAMLEN 286# include <sys/syscall.h>
109# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
110#endif
111
112# ifdef _DIRENT_HAVE_D_TYPE
113# define D_TYPE(de) (de)->d_type
114# endif
115
116# ifndef EIO_STRUCT_DIRENT
117# define EIO_STRUCT_DIRENT struct dirent
118# endif
119
120#endif 287#endif
121 288
122#if HAVE_SENDFILE 289#if HAVE_SENDFILE
123# if __linux 290# if __linux
124# include <sys/sendfile.h> 291# include <sys/sendfile.h>
… …
139#endif 306#endif
140#ifndef D_INO 307#ifndef D_INO
141# define D_INO(de) 0 308# define D_INO(de) 0
142#endif 309#endif
143#ifndef D_NAMLEN 310#ifndef D_NAMLEN
144# define D_NAMLEN(de) strlen ((de)->d_name) 311# define D_NAMLEN(entp) strlen (D_NAME (entp))
145#endif 312#endif
146 313
147/* used for struct dirent, AIX doesn't provide it */ 314/* used for struct dirent, AIX doesn't provide it */
148#ifndef NAME_MAX 315#ifndef NAME_MAX
149# define NAME_MAX 4096 316# define NAME_MAX 4096
… …
155#endif 322#endif
156 323
157/* buffer size for various temporary buffers */ 324/* buffer size for various temporary buffers */
158#define EIO_BUFSIZE 65536 325#define EIO_BUFSIZE 65536
159 326
160#define dBUF \ 327#define dBUF \
161 char *eio_buf; \
162 ETP_WORKER_LOCK (self); \
163 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 328 char *eio_buf = malloc (EIO_BUFSIZE); \
164 ETP_WORKER_UNLOCK (self); \
165 errno = ENOMEM; \ 329 errno = ENOMEM; \
166 if (!eio_buf) \ 330 if (!eio_buf) \
167 return -1; 331 return -1
168 332
169#define EIO_TICKS ((1000000 + 1023) >> 10) 333#define FUBd \
334 free (eio_buf)
335
336/*****************************************************************************/
337
338struct etp_tmpbuf;
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
346#else
347 #define HAVE_AT 0
348 static const char *wd_expand (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path);
349#endif
350
351struct eio_pwd
352{
353#if HAVE_AT
354 int fd;
355#endif
356 int len;
357 char str[1]; /* actually, a 0-terminated canonical path */
358};
359
360/*****************************************************************************/
170 361
171#define ETP_PRI_MIN EIO_PRI_MIN 362#define ETP_PRI_MIN EIO_PRI_MIN
172#define ETP_PRI_MAX EIO_PRI_MAX 363#define ETP_PRI_MAX EIO_PRI_MAX
173 364
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
174struct etp_worker; 375struct etp_worker;
175
176#define ETP_REQ eio_req 376#define ETP_REQ eio_req
177#define ETP_DESTROY(req) eio_destroy (req) 377#define ETP_DESTROY(req) eio_destroy (req)
178static int eio_finish (eio_req *req); 378static int eio_finish (eio_req *req);
179#define ETP_FINISH(req) eio_finish (req) 379#define ETP_FINISH(req) eio_finish (req)
180static void eio_execute (struct etp_worker *self, eio_req *req); 380static void eio_execute (struct etp_worker *self, eio_req *req);
181#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) 381#define ETP_EXECUTE(wrk,req) eio_execute (wrk, req)
182 382
183#define ETP_WORKER_CLEAR(req) \ 383#include "etp.c"
184 if (wrk->dbuf) \
185 { \
186 free (wrk->dbuf); \
187 wrk->dbuf = 0; \
188 } \
189 \
190 if (wrk->dirp) \
191 { \
192 closedir (wrk->dirp); \
193 wrk->dirp = 0; \
194 }
195 384
196#define ETP_WORKER_COMMON \ 385static struct etp_pool eio_pool;
197 void *dbuf; \ 386#define EIO_POOL (&eio_pool)
198 DIR *dirp;
199
200/*****************************************************************************/
201
202#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
203
204/* calculate time difference in ~1/EIO_TICKS of a second */
205ecb_inline int
206tvdiff (struct timeval *tv1, struct timeval *tv2)
207{
208 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
209 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
210}
211
212static unsigned int started, idle, wanted = 4;
213
214static void (*want_poll_cb) (void);
215static void (*done_poll_cb) (void);
216
217static unsigned int max_poll_time; /* reslock */
218static unsigned int max_poll_reqs; /* reslock */
219
220static volatile unsigned int nreqs; /* reqlock */
221static volatile unsigned int nready; /* reqlock */
222static volatile unsigned int npending; /* reqlock */
223static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
224static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
225
226static xmutex_t wrklock;
227static xmutex_t reslock;
228static xmutex_t reqlock;
229static xcond_t reqwait;
230
231#if !HAVE_PREADWRITE
232/*
233 * make our pread/pwrite emulation safe against themselves, but not against
234 * normal read/write by using a mutex. slows down execution a lot,
235 * but that's your problem, not mine.
236 */
237static xmutex_t preadwritelock = X_MUTEX_INIT;
238#endif
239
240typedef struct etp_worker
241{
242 /* locked by wrklock */
243 struct etp_worker *prev, *next;
244
245 xthread_t tid;
246
247 /* locked by reslock, reqlock or wrklock */
248 ETP_REQ *req; /* currently processed request */
249
250 ETP_WORKER_COMMON
251} etp_worker;
252
253static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */
254
255#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
256#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
257
258/* worker threads management */
259
260static void ecb_cold
261etp_worker_clear (etp_worker *wrk)
262{
263 ETP_WORKER_CLEAR (wrk);
264}
265
266static void ecb_cold
267etp_worker_free (etp_worker *wrk)
268{
269 wrk->next->prev = wrk->prev;
270 wrk->prev->next = wrk->next;
271
272 free (wrk);
273}
274
275static unsigned int
276etp_nreqs (void)
277{
278 int retval;
279 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
280 retval = nreqs;
281 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
282 return retval;
283}
284
285static unsigned int
286etp_nready (void)
287{
288 unsigned int retval;
289
290 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
291 retval = nready;
292 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
293
294 return retval;
295}
296
297static unsigned int
298etp_npending (void)
299{
300 unsigned int retval;
301
302 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
303 retval = npending;
304 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
305
306 return retval;
307}
308
309static unsigned int
310etp_nthreads (void)
311{
312 unsigned int retval;
313
314 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
315 retval = started;
316 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
317
318 return retval;
319}
320
321/*
322 * a somewhat faster data structure might be nice, but
323 * with 8 priorities this actually needs <20 insns
324 * per shift, the most expensive operation.
325 */
326typedef struct {
327 ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */
328 int size;
329} etp_reqq;
330
331static etp_reqq req_queue;
332static etp_reqq res_queue;
333
334static int ecb_noinline
335reqq_push (etp_reqq *q, ETP_REQ *req)
336{
337 int pri = req->pri;
338 req->next = 0;
339
340 if (q->qe[pri])
341 {
342 q->qe[pri]->next = req;
343 q->qe[pri] = req;
344 }
345 else
346 q->qe[pri] = q->qs[pri] = req;
347
348 return q->size++;
349}
350
351static ETP_REQ * ecb_noinline
352reqq_shift (etp_reqq *q)
353{
354 int pri;
355
356 if (!q->size)
357 return 0;
358
359 --q->size;
360
361 for (pri = ETP_NUM_PRI; pri--; )
362 {
363 eio_req *req = q->qs[pri];
364
365 if (req)
366 {
367 if (!(q->qs[pri] = (eio_req *)req->next))
368 q->qe[pri] = 0;
369
370 return req;
371 }
372 }
373
374 abort ();
375}
376
377static void ecb_cold
378etp_thread_init (void)
379{
380 X_MUTEX_CREATE (wrklock);
381 X_MUTEX_CREATE (reslock);
382 X_MUTEX_CREATE (reqlock);
383 X_COND_CREATE (reqwait);
384}
385
386static void ecb_cold
387etp_atfork_prepare (void)
388{
389 X_LOCK (wrklock);
390 X_LOCK (reqlock);
391 X_LOCK (reslock);
392#if !HAVE_PREADWRITE
393 X_LOCK (preadwritelock);
394#endif
395}
396
397static void ecb_cold
398etp_atfork_parent (void)
399{
400#if !HAVE_PREADWRITE
401 X_UNLOCK (preadwritelock);
402#endif
403 X_UNLOCK (reslock);
404 X_UNLOCK (reqlock);
405 X_UNLOCK (wrklock);
406}
407
408static void ecb_cold
409etp_atfork_child (void)
410{
411 ETP_REQ *prv;
412
413 while ((prv = reqq_shift (&req_queue)))
414 ETP_DESTROY (prv);
415
416 while ((prv = reqq_shift (&res_queue)))
417 ETP_DESTROY (prv);
418
419 while (wrk_first.next != &wrk_first)
420 {
421 etp_worker *wrk = wrk_first.next;
422
423 if (wrk->req)
424 ETP_DESTROY (wrk->req);
425
426 etp_worker_clear (wrk);
427 etp_worker_free (wrk);
428 }
429
430 started = 0;
431 idle = 0;
432 nreqs = 0;
433 nready = 0;
434 npending = 0;
435
436 etp_thread_init ();
437}
438
439static void ecb_cold
440etp_once_init (void)
441{
442 etp_thread_init ();
443 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
444}
445
446static int ecb_cold
447etp_init (void (*want_poll)(void), void (*done_poll)(void))
448{
449 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
450
451 pthread_once (&doinit, etp_once_init);
452
453 want_poll_cb = want_poll;
454 done_poll_cb = done_poll;
455
456 return 0;
457}
458
459X_THREAD_PROC (etp_proc);
460
461static void ecb_cold
462etp_start_thread (void)
463{
464 etp_worker *wrk = calloc (1, sizeof (etp_worker));
465
466 /*TODO*/
467 assert (("unable to allocate worker thread data", wrk));
468
469 X_LOCK (wrklock);
470
471 if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
472 {
473 wrk->prev = &wrk_first;
474 wrk->next = wrk_first.next;
475 wrk_first.next->prev = wrk;
476 wrk_first.next = wrk;
477 ++started;
478 }
479 else
480 free (wrk);
481
482 X_UNLOCK (wrklock);
483}
484
485static void
486etp_maybe_start_thread (void)
487{
488 if (ecb_expect_true (etp_nthreads () >= wanted))
489 return;
490
491 /* todo: maybe use idle here, but might be less exact */
492 if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
493 return;
494
495 etp_start_thread ();
496}
497
498static void ecb_cold
499etp_end_thread (void)
500{
501 eio_req *req = calloc (1, sizeof (eio_req));
502
503 req->type = -1;
504 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
505
506 X_LOCK (reqlock);
507 reqq_push (&req_queue, req);
508 X_COND_SIGNAL (reqwait);
509 X_UNLOCK (reqlock);
510
511 X_LOCK (wrklock);
512 --started;
513 X_UNLOCK (wrklock);
514}
515
516static int
517etp_poll (void)
518{
519 unsigned int maxreqs;
520 unsigned int maxtime;
521 struct timeval tv_start, tv_now;
522
523 X_LOCK (reslock);
524 maxreqs = max_poll_reqs;
525 maxtime = max_poll_time;
526 X_UNLOCK (reslock);
527
528 if (maxtime)
529 gettimeofday (&tv_start, 0);
530
531 for (;;)
532 {
533 ETP_REQ *req;
534
535 etp_maybe_start_thread ();
536
537 X_LOCK (reslock);
538 req = reqq_shift (&res_queue);
539
540 if (req)
541 {
542 --npending;
543
544 if (!res_queue.size && done_poll_cb)
545 done_poll_cb ();
546 }
547
548 X_UNLOCK (reslock);
549
550 if (!req)
551 return 0;
552
553 X_LOCK (reqlock);
554 --nreqs;
555 X_UNLOCK (reqlock);
556
557 if (ecb_expect_false (req->type == EIO_GROUP && req->size))
558 {
559 req->int1 = 1; /* mark request as delayed */
560 continue;
561 }
562 else
563 {
564 int res = ETP_FINISH (req);
565 if (ecb_expect_false (res))
566 return res;
567 }
568
569 if (ecb_expect_false (maxreqs && !--maxreqs))
570 break;
571
572 if (maxtime)
573 {
574 gettimeofday (&tv_now, 0);
575
576 if (tvdiff (&tv_start, &tv_now) >= maxtime)
577 break;
578 }
579 }
580
581 errno = EAGAIN;
582 return -1;
583}
584
585static void
586etp_cancel (ETP_REQ *req)
587{
588 X_LOCK (wrklock);
589 req->flags |= EIO_FLAG_CANCELLED;
590 X_UNLOCK (wrklock);
591
592 eio_grp_cancel (req);
593}
594
595static void
596etp_submit (ETP_REQ *req)
597{
598 req->pri -= ETP_PRI_MIN;
599
600 if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
601 if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
602
603 if (ecb_expect_false (req->type == EIO_GROUP))
604 {
605 /* I hope this is worth it :/ */
606 X_LOCK (reqlock);
607 ++nreqs;
608 X_UNLOCK (reqlock);
609
610 X_LOCK (reslock);
611
612 ++npending;
613
614 if (!reqq_push (&res_queue, req) && want_poll_cb)
615 want_poll_cb ();
616
617 X_UNLOCK (reslock);
618 }
619 else
620 {
621 X_LOCK (reqlock);
622 ++nreqs;
623 ++nready;
624 reqq_push (&req_queue, req);
625 X_COND_SIGNAL (reqwait);
626 X_UNLOCK (reqlock);
627
628 etp_maybe_start_thread ();
629 }
630}
631
632static void ecb_cold
633etp_set_max_poll_time (double nseconds)
634{
635 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
636 max_poll_time = nseconds * EIO_TICKS;
637 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
638}
639
640static void ecb_cold
641etp_set_max_poll_reqs (unsigned int maxreqs)
642{
643 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
644 max_poll_reqs = maxreqs;
645 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
646}
647
648static void ecb_cold
649etp_set_max_idle (unsigned int nthreads)
650{
651 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
652 max_idle = nthreads;
653 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
654}
655
656static void ecb_cold
657etp_set_idle_timeout (unsigned int seconds)
658{
659 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
660 idle_timeout = seconds;
661 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
662}
663
664static void ecb_cold
665etp_set_min_parallel (unsigned int nthreads)
666{
667 if (wanted < nthreads)
668 wanted = nthreads;
669}
670
671static void ecb_cold
672etp_set_max_parallel (unsigned int nthreads)
673{
674 if (wanted > nthreads)
675 wanted = nthreads;
676
677 while (started > wanted)
678 etp_end_thread ();
679}
680 387
681/*****************************************************************************/ 388/*****************************************************************************/
682 389
683static void 390static void
684grp_try_feed (eio_req *grp) 391grp_try_feed (eio_req *grp)
685{ 392{
686 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 393 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
687 { 394 {
688 grp->flags &= ~EIO_FLAG_GROUPADD; 395 grp->flags &= ~ETP_FLAG_GROUPADD;
689 396
690 EIO_FEED (grp); 397 EIO_FEED (grp);
691 398
692 /* stop if no progress has been made */ 399 /* stop if no progress has been made */
693 if (!(grp->flags & EIO_FLAG_GROUPADD)) 400 if (!(grp->flags & ETP_FLAG_GROUPADD))
694 { 401 {
695 grp->feed = 0; 402 grp->feed = 0;
696 break; 403 break;
697 } 404 }
698 } 405 }
… …
705 412
706 /* call feeder, if applicable */ 413 /* call feeder, if applicable */
707 grp_try_feed (grp); 414 grp_try_feed (grp);
708 415
709 /* finish, if done */ 416 /* finish, if done */
710 if (!grp->size && grp->int1) 417 if (!grp->size && grp->flags & ETP_FLAG_DELAYED)
711 return eio_finish (grp); 418 return eio_finish (grp);
712 else 419 else
713 return 0; 420 return 0;
714} 421}
715 422
716void 423static void
717eio_destroy (eio_req *req) 424eio_destroy (eio_req *req)
718{ 425{
719 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); 426 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
720 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); 427 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
721 428
… …
739 if (grp->grp_first == req) 446 if (grp->grp_first == req)
740 grp->grp_first = req->grp_next; 447 grp->grp_first = req->grp_next;
741 448
742 res2 = grp_dec (grp); 449 res2 = grp_dec (grp);
743 450
744 if (!res && res2) 451 if (!res)
745 res = res2; 452 res = res2;
746 } 453 }
747 454
748 eio_destroy (req); 455 eio_destroy (req);
749 456
… …
751} 458}
752 459
753void 460void
754eio_grp_cancel (eio_req *grp) 461eio_grp_cancel (eio_req *grp)
755{ 462{
756 for (grp = grp->grp_first; grp; grp = grp->grp_next) 463 etp_grp_cancel (EIO_POOL, grp);
757 eio_cancel (grp);
758} 464}
759 465
760void 466void
761eio_cancel (eio_req *req) 467eio_cancel (eio_req *req)
762{ 468{
763 etp_cancel (req); 469 etp_cancel (EIO_POOL, req);
764} 470}
765 471
766void 472void
767eio_submit (eio_req *req) 473eio_submit (eio_req *req)
768{ 474{
769 etp_submit (req); 475 etp_submit (EIO_POOL, req);
770} 476}
771 477
772unsigned int 478unsigned int
773eio_nreqs (void) 479eio_nreqs (void)
774{ 480{
775 return etp_nreqs (); 481 return etp_nreqs (EIO_POOL);
776} 482}
777 483
778unsigned int 484unsigned int
779eio_nready (void) 485eio_nready (void)
780{ 486{
781 return etp_nready (); 487 return etp_nready (EIO_POOL);
782} 488}
783 489
784unsigned int 490unsigned int
785eio_npending (void) 491eio_npending (void)
786{ 492{
787 return etp_npending (); 493 return etp_npending (EIO_POOL);
788} 494}
789 495
790unsigned int ecb_cold 496unsigned int ecb_cold
791eio_nthreads (void) 497eio_nthreads (void)
792{ 498{
793 return etp_nthreads (); 499 return etp_nthreads (EIO_POOL);
794} 500}
795 501
796void ecb_cold 502void ecb_cold
797eio_set_max_poll_time (double nseconds) 503eio_set_max_poll_time (double nseconds)
798{ 504{
799 etp_set_max_poll_time (nseconds); 505 etp_set_max_poll_time (EIO_POOL, nseconds);
800} 506}
801 507
802void ecb_cold 508void ecb_cold
803eio_set_max_poll_reqs (unsigned int maxreqs) 509eio_set_max_poll_reqs (unsigned int maxreqs)
804{ 510{
805 etp_set_max_poll_reqs (maxreqs); 511 etp_set_max_poll_reqs (EIO_POOL, maxreqs);
806} 512}
807 513
808void ecb_cold 514void ecb_cold
809eio_set_max_idle (unsigned int nthreads) 515eio_set_max_idle (unsigned int nthreads)
810{ 516{
811 etp_set_max_idle (nthreads); 517 etp_set_max_idle (EIO_POOL, nthreads);
812} 518}
813 519
814void ecb_cold 520void ecb_cold
815eio_set_idle_timeout (unsigned int seconds) 521eio_set_idle_timeout (unsigned int seconds)
816{ 522{
817 etp_set_idle_timeout (seconds); 523 etp_set_idle_timeout (EIO_POOL, seconds);
818} 524}
819 525
820void ecb_cold 526void ecb_cold
821eio_set_min_parallel (unsigned int nthreads) 527eio_set_min_parallel (unsigned int nthreads)
822{ 528{
823 etp_set_min_parallel (nthreads); 529 etp_set_min_parallel (EIO_POOL, nthreads);
824} 530}
825 531
826void ecb_cold 532void ecb_cold
827eio_set_max_parallel (unsigned int nthreads) 533eio_set_max_parallel (unsigned int nthreads)
828{ 534{
829 etp_set_max_parallel (nthreads); 535 etp_set_max_parallel (EIO_POOL, nthreads);
830} 536}
831 537
832int eio_poll (void) 538int eio_poll (void)
833{ 539{
834 return etp_poll (); 540 return etp_poll (EIO_POOL);
835} 541}
836 542
837/*****************************************************************************/ 543/*****************************************************************************/
838/* work around various missing functions */ 544/* work around various missing functions */
839
840#if !HAVE_PREADWRITE
841# undef pread
842# undef pwrite
843# define pread eio__pread
844# define pwrite eio__pwrite
845
846static ssize_t
847eio__pread (int fd, void *buf, size_t count, off_t offset)
848{
849 ssize_t res;
850 off_t ooffset;
851
852 X_LOCK (preadwritelock);
853 ooffset = lseek (fd, 0, SEEK_CUR);
854 lseek (fd, offset, SEEK_SET);
855 res = read (fd, buf, count);
856 lseek (fd, ooffset, SEEK_SET);
857 X_UNLOCK (preadwritelock);
858
859 return res;
860}
861
862static ssize_t
863eio__pwrite (int fd, void *buf, size_t count, off_t offset)
864{
865 ssize_t res;
866 off_t ooffset;
867
868 X_LOCK (preadwritelock);
869 ooffset = lseek (fd, 0, SEEK_CUR);
870 lseek (fd, offset, SEEK_SET);
871 res = write (fd, buf, count);
872 lseek (fd, ooffset, SEEK_SET);
873 X_UNLOCK (preadwritelock);
874
875 return res;
876}
877#endif
878 545
879#ifndef HAVE_UTIMES 546#ifndef HAVE_UTIMES
880 547
881# undef utimes 548# undef utimes
882# define utimes(path,times) eio__utimes (path, times) 549# define utimes(path,times) eio__utimes (path, times)
… …
915 582
916#if !HAVE_FDATASYNC 583#if !HAVE_FDATASYNC
917# undef fdatasync 584# undef fdatasync
918# define fdatasync(fd) fsync (fd) 585# define fdatasync(fd) fsync (fd)
919#endif 586#endif
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}
920 604
921/* sync_file_range always needs emulation */ 605/* sync_file_range always needs emulation */
922static int 606static int
923eio__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)
924{ 608{
… …
944 /* 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
945 * call fdatasync, as that matches the expectation of its users best */ 629 * call fdatasync, as that matches the expectation of its users best */
946 return fdatasync (fd); 630 return fdatasync (fd);
947} 631}
948 632
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
641}
642
949#if !HAVE_READAHEAD 643#if !HAVE_READAHEAD
950# undef readahead 644# undef readahead
951# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 645# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
952 646
953static ssize_t 647static eio_ssize_t
954eio__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)
955{ 649{
956 size_t todo = count; 650 size_t todo = count;
957 dBUF; 651 dBUF;
958 652
… …
963 pread (fd, eio_buf, len, offset); 657 pread (fd, eio_buf, len, offset);
964 offset += len; 658 offset += len;
965 todo -= len; 659 todo -= len;
966 } 660 }
967 661
968 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 */
969 return count; 666 return 0;
970} 667}
971 668
972#endif 669#endif
973 670
974/* sendfile always needs emulation */ 671/* sendfile always needs emulation */
975static ssize_t 672static eio_ssize_t
976eio__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)
977{ 674{
978 ssize_t written = 0; 675 eio_ssize_t written = 0;
979 ssize_t res; 676 eio_ssize_t res;
980 677
981 if (!count) 678 if (!count)
982 return 0; 679 return 0;
983 680
984 for (;;) 681 for (;;)
985 { 682 {
683#ifdef __APPLE__
684# undef HAVE_SENDFILE /* broken, as everything on os x */
685#endif
986#if HAVE_SENDFILE 686#if HAVE_SENDFILE
987# if __linux 687# if __linux
988 off_t soffset = offset; 688 off_t soffset = offset;
989 res = sendfile (ofd, ifd, &soffset, count); 689 res = sendfile (ofd, ifd, &soffset, count);
990 690
… …
1006 706
1007 /* according to source inspection, this is correct, and useful behaviour */ 707 /* according to source inspection, this is correct, and useful behaviour */
1008 if (sbytes) 708 if (sbytes)
1009 res = sbytes; 709 res = sbytes;
1010 710
1011# elif defined (__APPLE__) 711# elif defined __APPLE__
1012 off_t sbytes = count; 712 off_t sbytes = count;
1013 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0); 713 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
1014 714
1015 /* according to the manpage, sbytes is always valid */ 715 /* according to the manpage, sbytes is always valid */
1016 if (sbytes) 716 if (sbytes)
… …
1033 if (res < 0 && sbytes) 733 if (res < 0 && sbytes)
1034 res = sbytes; 734 res = sbytes;
1035 735
1036# endif 736# endif
1037 737
1038#elif defined (_WIN32) 738#elif defined (_WIN32) && 0
1039 /* does not work, just for documentation of what would need to be done */ 739 /* does not work, just for documentation of what would need to be done */
1040 /* actually, cannot be done like this, as TransmitFile changes the file offset, */ 740 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
1041 /* libeio guarantees that the file offset does not change, and windows */ 741 /* libeio guarantees that the file offset does not change, and windows */
1042 /* has no way to get an independent handle to the same file description */ 742 /* has no way to get an independent handle to the same file description */
1043 HANDLE h = TO_SOCKET (ifd); 743 HANDLE h = TO_SOCKET (ifd);
1044 SetFilePointer (h, offset, 0, FILE_BEGIN); 744 SetFilePointer (h, offset, 0, FILE_BEGIN);
1045 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0); 745 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1046 746
1047#else 747#else
1048 res = -1; 748 res = EIO_ENOSYS ();
1049 errno = ENOSYS;
1050#endif 749#endif
1051 750
1052 /* we assume sendfile can copy at least 128mb in one go */ 751 /* we assume sendfile can copy at least 128mb in one go */
1053 if (res <= 128 * 1024 * 1024) 752 if (res <= 128 * 1024 * 1024)
1054 { 753 {
… …
1076 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 775 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1077 /* BSDs */ 776 /* BSDs */
1078#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */ 777#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1079 || errno == ENOTSUP 778 || errno == ENOTSUP
1080#endif 779#endif
780#ifdef EOPNOTSUPP /* windows */
1081 || errno == EOPNOTSUPP /* BSDs */ 781 || errno == EOPNOTSUPP /* BSDs */
782#endif
1082#if __solaris 783#if __solaris
1083 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 784 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
1084#endif 785#endif
1085 ) 786 )
1086 ) 787 )
… …
1090 791
1091 res = 0; 792 res = 0;
1092 793
1093 while (count) 794 while (count)
1094 { 795 {
1095 ssize_t cnt; 796 eio_ssize_t cnt;
1096 797
1097 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);
1098 799
1099 if (cnt <= 0) 800 if (cnt <= 0)
1100 { 801 {
… …
1112 813
1113 offset += cnt; 814 offset += cnt;
1114 res += cnt; 815 res += cnt;
1115 count -= cnt; 816 count -= cnt;
1116 } 817 }
818
819 FUBd;
1117 } 820 }
1118 821
1119 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
1120} 1130}
1121 1131
1122static signed char 1132static signed char
1123eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) 1133eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1124{ 1134{
… …
1132 1142
1133#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */ 1143#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1134#define EIO_SORT_FAST 60 /* when to only use insertion sort */ 1144#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1135 1145
1136static void 1146static void
1137eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1147eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1138{ 1148{
1139 unsigned char bits [9 + sizeof (ino_t) * 8]; 1149 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1140 unsigned char *bit = bits; 1150 unsigned char *bit = bits;
1141 1151
1142 assert (CHAR_BIT == 8); 1152 assert (CHAR_BIT == 8);
1143 assert (sizeof (eio_dirent) * 8 < 256); 1153 assert (sizeof (eio_dirent) * 8 < 256);
1144 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */ 1154 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
… …
1146 1156
1147 if (size <= EIO_SORT_FAST) 1157 if (size <= EIO_SORT_FAST)
1148 return; 1158 return;
1149 1159
1150 /* first prepare an array of bits to test in our radix sort */ 1160 /* first prepare an array of bits to test in our radix sort */
1151 /* try to take endianness into account, as well as differences in ino_t sizes */ 1161 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1152 /* inode_bits must contain all inodes ORed together */ 1162 /* inode_bits must contain all inodes ORed together */
1153 /* which is used to skip bits that are 0 everywhere, which is very common */ 1163 /* which is used to skip bits that are 0 everywhere, which is very common */
1154 { 1164 {
1155 ino_t endianness; 1165 eio_ino_t endianness;
1156 int i, j; 1166 int i, j;
1157 1167
1158 /* we store the byte offset of byte n into byte n of "endianness" */ 1168 /* we store the byte offset of byte n into byte n of "endianness" */
1159 for (i = 0; i < sizeof (ino_t); ++i) 1169 for (i = 0; i < sizeof (eio_ino_t); ++i)
1160 ((unsigned char *)&endianness)[i] = i; 1170 ((unsigned char *)&endianness)[i] = i;
1161 1171
1162 *bit++ = 0; 1172 *bit++ = 0;
1163 1173
1164 for (i = 0; i < sizeof (ino_t); ++i) 1174 for (i = 0; i < sizeof (eio_ino_t); ++i)
1165 { 1175 {
1166 /* shifting off the byte offsets out of "endianness" */ 1176 /* shifting off the byte offsets out of "endianness" */
1167 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8; 1177 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1168 endianness >>= 8; 1178 endianness >>= 8;
1169 1179
1170 for (j = 0; j < 8; ++j) 1180 for (j = 0; j < 8; ++j)
1171 if (inode_bits & (((ino_t)1) << (i * 8 + j))) 1181 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1172 *bit++ = offs + j; 1182 *bit++ = offs + j;
1173 } 1183 }
1174 1184
1175 for (j = 0; j < 8; ++j) 1185 for (j = 0; j < 8; ++j)
1176 if (score_bits & (1 << j)) 1186 if (score_bits & (1 << j))
1177 *bit++ = offsetof (eio_dirent, score) * 8 + j; 1187 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1178 } 1188 }
1179 1189
1180 /* now actually do the sorting (a variant of MSD radix sort) */ 1190 /* now actually do the sorting (a variant of MSD radix sort) */
1181 { 1191 {
1182 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base; 1192 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1183 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end; 1193 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1184 unsigned char *bit_stk [9 + sizeof (ino_t) * 8]; 1194 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1185 int stk_idx = 0; 1195 int stk_idx = 0;
1186 1196
1187 base_stk [stk_idx] = dents; 1197 base_stk [stk_idx] = dents;
1188 end_stk [stk_idx] = dents + size; 1198 end_stk [stk_idx] = dents + size;
1189 bit_stk [stk_idx] = bit - 1; 1199 bit_stk [stk_idx] = bit - 1;
… …
1268 } 1278 }
1269 } 1279 }
1270} 1280}
1271 1281
1272static void 1282static void
1273eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1283eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1274{ 1284{
1275 if (size <= 1) 1285 if (size <= 1)
1276 return; /* our insertion sort relies on size > 0 */ 1286 return; /* our insertion sort relies on size > 0 */
1277 1287
1278 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */ 1288 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
… …
1286 1296
1287/* read a full directory */ 1297/* read a full directory */
1288static void 1298static void
1289eio__scandir (eio_req *req, etp_worker *self) 1299eio__scandir (eio_req *req, etp_worker *self)
1290{ 1300{
1291 DIR *dirp;
1292 EIO_STRUCT_DIRENT *entp;
1293 char *name, *names; 1301 char *name, *names;
1294 int namesalloc = 4096; 1302 int namesalloc = 4096 - sizeof (void *) * 4;
1295 int namesoffs = 0; 1303 int namesoffs = 0;
1296 int flags = req->int1; 1304 int flags = req->int1;
1297 eio_dirent *dents = 0; 1305 eio_dirent *dents = 0;
1298 int dentalloc = 128; 1306 int dentalloc = 128;
1299 int dentoffs = 0; 1307 int dentoffs = 0;
1300 ino_t inode_bits = 0; 1308 eio_ino_t inode_bits = 0;
1309#ifdef _WIN32
1310 HANDLE dirp;
1311 WIN32_FIND_DATA entp;
1312#else
1313 DIR *dirp;
1314 EIO_STRUCT_DIRENT *entp;
1315#endif
1301 1316
1302 req->result = -1; 1317 req->result = -1;
1303 1318
1304 if (!(flags & EIO_READDIR_DENTS)) 1319 if (!(flags & EIO_READDIR_DENTS))
1305 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); 1320 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1306 1321
1307 X_LOCK (wrklock); 1322#ifdef _WIN32
1308 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 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)
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
1309 self->dirp = dirp = opendir (req->ptr1); 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
1310 1390
1311 if (req->flags & EIO_FLAG_PTR1_FREE) 1391 if (req->flags & EIO_FLAG_PTR1_FREE)
1312 free (req->ptr1); 1392 free (req->ptr1);
1313 1393
1314 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1394 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1315 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; 1395 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1316 req->ptr2 = names = malloc (namesalloc); 1396 req->ptr2 = names = malloc (namesalloc);
1317 X_UNLOCK (wrklock);
1318 1397
1319 if (dirp && names && (!flags || dents)) 1398 if (!names || (flags && !dents))
1399 return;
1400
1320 for (;;) 1401 for (;;)
1321 { 1402 {
1403 int done;
1404
1405#ifdef _WIN32
1406 done = !dirp;
1407#else
1322 errno = 0; 1408 errno = 0;
1323 entp = readdir (dirp); 1409 entp = readdir (dirp);
1410 done = !entp;
1411#endif
1324 1412
1325 if (!entp) 1413 if (done)
1326 { 1414 {
1415#ifndef _WIN32
1416 int old_errno = errno;
1417 closedir (dirp);
1418 errno = old_errno;
1419
1327 if (errno) 1420 if (errno)
1328 break; 1421 break;
1422#endif
1329 1423
1330 /* sort etc. */ 1424 /* sort etc. */
1331 req->int1 = flags; 1425 req->int1 = flags;
1332 req->result = dentoffs; 1426 req->result = dentoffs;
1333 1427
1334 if (flags & EIO_READDIR_STAT_ORDER) 1428 if (flags & EIO_READDIR_STAT_ORDER)
1335 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits); 1429 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1336 else if (flags & EIO_READDIR_DIRS_FIRST) 1430 else if (flags & EIO_READDIR_DIRS_FIRST)
1337 if (flags & EIO_READDIR_FOUND_UNKNOWN) 1431 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1338 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */ 1432 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1339 else 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
1457 break;
1458 }
1459
1460 /* now add the entry to our list(s) */
1461 name = D_NAME (entp);
1462
1463 /* skip . and .. entries */
1464 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1465 {
1466 int len = D_NAMLEN (entp) + 1;
1467
1468 while (ecb_expect_false (namesoffs + len > namesalloc))
1469 {
1470 namesalloc *= 2;
1471 req->ptr2 = names = realloc (names, namesalloc);
1472
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))
1340 { 1484 {
1341 /* in this case, all is known, and we just put dirs first and sort them */ 1485 dentalloc *= 2;
1486 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1487
1488 if (!dents)
1489 break;
1490 }
1491
1342 eio_dirent *oth = dents + dentoffs; 1492 ent = dents + dentoffs;
1343 eio_dirent *dir = dents;
1344 1493
1345 /* now partition dirs to the front, and non-dirs to the back */ 1494 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1346 /* by walking from both sides and swapping if necessary */ 1495 ent->namelen = len - 1;
1347 while (oth > dir) 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)
1348 { 1556 {
1349 if (dir->type == EIO_DT_DIR) 1557 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1350 ++dir; 1558 ent->score = 1;
1351 else if ((--oth)->type == EIO_DT_DIR) 1559 else if (!strchr (name, '.')) /* absence of dots indicate likely dirs */
1352 { 1560 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1353 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1354
1355 ++dir;
1356 }
1357 } 1561 }
1358 1562 else if (ent->type == EIO_DT_DIR)
1359 /* now sort the dirs only (dirs all have the same score) */ 1563 ent->score = 0;
1360 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1361 } 1564 }
1362
1363 break;
1364 }
1365
1366 /* now add the entry to our list(s) */
1367 name = entp->d_name;
1368
1369 /* skip . and .. entries */
1370 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1371 {
1372 int len = D_NAMLEN (entp) + 1;
1373
1374 while (ecb_expect_false (namesoffs + len > namesalloc))
1375 {
1376 namesalloc *= 2;
1377 X_LOCK (wrklock);
1378 req->ptr2 = names = realloc (names, namesalloc);
1379 X_UNLOCK (wrklock);
1380
1381 if (!names)
1382 break;
1383 } 1565 }
1384 1566
1385 memcpy (names + namesoffs, name, len);
1386
1387 if (dents)
1388 {
1389 struct eio_dirent *ent;
1390
1391 if (ecb_expect_false (dentoffs == dentalloc))
1392 {
1393 dentalloc *= 2;
1394 X_LOCK (wrklock);
1395 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1396 X_UNLOCK (wrklock);
1397
1398 if (!dents)
1399 break;
1400 }
1401
1402 ent = dents + dentoffs;
1403
1404 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1405 ent->namelen = len - 1;
1406 ent->inode = D_INO (entp);
1407
1408 inode_bits |= ent->inode;
1409
1410 switch (D_TYPE (entp))
1411 {
1412 default:
1413 ent->type = EIO_DT_UNKNOWN;
1414 flags |= EIO_READDIR_FOUND_UNKNOWN;
1415 break;
1416
1417 #ifdef DT_FIFO
1418 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1419 #endif
1420 #ifdef DT_CHR
1421 case DT_CHR: ent->type = EIO_DT_CHR; break;
1422 #endif
1423 #ifdef DT_MPC
1424 case DT_MPC: ent->type = EIO_DT_MPC; break;
1425 #endif
1426 #ifdef DT_DIR
1427 case DT_DIR: ent->type = EIO_DT_DIR; break;
1428 #endif
1429 #ifdef DT_NAM
1430 case DT_NAM: ent->type = EIO_DT_NAM; break;
1431 #endif
1432 #ifdef DT_BLK
1433 case DT_BLK: ent->type = EIO_DT_BLK; break;
1434 #endif
1435 #ifdef DT_MPB
1436 case DT_MPB: ent->type = EIO_DT_MPB; break;
1437 #endif
1438 #ifdef DT_REG
1439 case DT_REG: ent->type = EIO_DT_REG; break;
1440 #endif
1441 #ifdef DT_NWK
1442 case DT_NWK: ent->type = EIO_DT_NWK; break;
1443 #endif
1444 #ifdef DT_CMP
1445 case DT_CMP: ent->type = EIO_DT_CMP; break;
1446 #endif
1447 #ifdef DT_LNK
1448 case DT_LNK: ent->type = EIO_DT_LNK; break;
1449 #endif
1450 #ifdef DT_SOCK
1451 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1452 #endif
1453 #ifdef DT_DOOR
1454 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1455 #endif
1456 #ifdef DT_WHT
1457 case DT_WHT: ent->type = EIO_DT_WHT; break;
1458 #endif
1459 }
1460
1461 ent->score = 7;
1462
1463 if (flags & EIO_READDIR_DIRS_FIRST)
1464 {
1465 if (ent->type == EIO_DT_UNKNOWN)
1466 {
1467 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1468 ent->score = 1;
1469 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1470 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1471 }
1472 else if (ent->type == EIO_DT_DIR)
1473 ent->score = 0;
1474 }
1475 }
1476
1477 namesoffs += len; 1567 namesoffs += len;
1478 ++dentoffs; 1568 ++dentoffs;
1479 } 1569 }
1480 1570
1481 if (EIO_CANCELLED (req)) 1571 if (EIO_CANCELLED (req))
1482 { 1572 {
1483 errno = ECANCELED; 1573 errno = ECANCELED;
1484 break; 1574 break;
1485 } 1575 }
1576
1577#ifdef _WIN32
1578 if (!FindNextFile (dirp, &entp))
1579 {
1580 FindClose (dirp);
1581 dirp = 0;
1486 } 1582 }
1583#endif
1584 }
1487} 1585}
1488 1586
1489#ifdef PAGESIZE 1587/*****************************************************************************/
1490# define eio_pagesize() PAGESIZE 1588/* working directory stuff */
1491#else 1589/* various deficiencies in the posix 2008 api force us to */
1492static intptr_t 1590/* keep the absolute path in string form at all times */
1493eio_pagesize (void) 1591/* fuck yeah. */
1494{
1495 static intptr_t page;
1496 1592
1497 if (!page) 1593#if !HAVE_AT
1498 page = sysconf (_SC_PAGESIZE);
1499 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
1606 {
1607 int l1 = wd->len;
1608 int l2 = strlen (path);
1609
1610 char *res = etp_tmpbuf_get (tmpbuf, l1 + l2 + 2);
1611
1612 memcpy (res, wd->str, l1);
1613 res [l1] = '/';
1614 memcpy (res + l1 + 1, path, l2 + 1);
1615
1616 return res;
1617 }
1618}
1619
1620#endif
1621
1622static eio_wd
1623eio__wd_open_sync (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
1624{
1625 int fd;
1626 eio_wd res;
1627 int len = eio__realpath (tmpbuf, wd, path);
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
1500 return page; 1649 return res;
1501} 1650}
1502#endif
1503 1651
1504static void 1652eio_wd
1505eio_page_align (void **addr, size_t *length) 1653eio_wd_open_sync (eio_wd wd, const char *path)
1506{ 1654{
1507 intptr_t mask = eio_pagesize () - 1; 1655 struct etp_tmpbuf tmpbuf = { };
1656 wd = eio__wd_open_sync (&tmpbuf, wd, path);
1657 free (tmpbuf.ptr);
1508 1658
1509 /* round down addr */ 1659 return wd;
1510 intptr_t adj = mask & (intptr_t)*addr;
1511
1512 *addr = (void *)((intptr_t)*addr - adj);
1513 *length += adj;
1514
1515 /* round up length */
1516 *length = (*length + mask) & ~mask;
1517} 1660}
1518 1661
1519#if !_POSIX_MEMLOCK 1662void
1520# define eio__mlockall(a) ((errno = ENOSYS), -1) 1663eio_wd_close_sync (eio_wd wd)
1521#else 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 */
1522 1677
1523static int 1678static int
1524eio__mlockall (int flags) 1679eio__truncateat (int dirfd, const char *path, off_t length)
1525{ 1680{
1526 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7 1681 int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
1527 extern int mallopt (int, int); 1682 int res;
1528 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1529 #endif
1530 1683
1531 if (EIO_MCL_CURRENT != MCL_CURRENT 1684 if (fd < 0)
1532 || EIO_MCL_FUTURE != MCL_FUTURE) 1685 return fd;
1533 {
1534 flags = 0
1535 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1536 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1537 }
1538 1686
1539 return mlockall (flags); 1687 res = ftruncate (fd, length);
1688 close (fd);
1689 return res;
1540} 1690}
1541#endif
1542
1543#if !_POSIX_MEMLOCK_RANGE
1544# define eio__mlock(a,b) ((errno = ENOSYS), -1)
1545#else
1546 1691
1547static int 1692static int
1548eio__mlock (void *addr, size_t length) 1693eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
1549{ 1694{
1550 eio_page_align (&addr, &length); 1695 int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
1696 int res;
1551 1697
1552 return mlock (addr, length); 1698 if (fd < 0)
1553} 1699 return fd;
1554 1700
1555#endif 1701 res = fstatvfs (fd, buf);
1556 1702 close (fd);
1557#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1558# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
1559#else
1560
1561static int
1562eio__msync (void *mem, size_t len, int flags)
1563{
1564 eio_page_align (&mem, &len);
1565
1566 if (EIO_MS_ASYNC != MS_SYNC
1567 || EIO_MS_INVALIDATE != MS_INVALIDATE
1568 || EIO_MS_SYNC != MS_SYNC)
1569 {
1570 flags = 0
1571 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1572 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1573 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1574 }
1575
1576 return msync (mem, len, flags);
1577}
1578
1579#endif
1580
1581static int
1582eio__mtouch (eio_req *req)
1583{
1584 void *mem = req->ptr2;
1585 size_t len = req->size;
1586 int flags = req->int1;
1587
1588 eio_page_align (&mem, &len);
1589
1590 {
1591 intptr_t addr = (intptr_t)mem;
1592 intptr_t end = addr + len;
1593 intptr_t page = eio_pagesize ();
1594
1595 if (addr < end)
1596 if (flags & EIO_MT_MODIFY) /* modify */
1597 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1598 else
1599 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1600 }
1601
1602 return 0; 1703 return res;
1704
1603} 1705}
1706
1707#endif
1604 1708
1605/*****************************************************************************/ 1709/*****************************************************************************/
1606 1710
1607#define ALLOC(len) \ 1711#define ALLOC(len) \
1608 if (!req->ptr2) \ 1712 if (!req->ptr2) \
1609 { \ 1713 { \
1610 X_LOCK (wrklock); \ 1714 X_LOCK (EIO_POOL->wrklock); \
1611 req->flags |= EIO_FLAG_PTR2_FREE; \ 1715 req->flags |= EIO_FLAG_PTR2_FREE; \
1612 X_UNLOCK (wrklock); \ 1716 X_UNLOCK (EIO_POOL->wrklock); \
1613 req->ptr2 = malloc (len); \ 1717 req->ptr2 = malloc (len); \
1614 if (!req->ptr2) \ 1718 if (!req->ptr2) \
1615 { \ 1719 { \
1616 errno = ENOMEM; \ 1720 errno = ENOMEM; \
1617 req->result = -1; \ 1721 req->result = -1; \
1618 break; \ 1722 break; \
1619 } \ 1723 } \
1620 } 1724 }
1621 1725
1622X_THREAD_PROC (etp_proc)
1623{
1624 ETP_REQ *req;
1625 struct timespec ts;
1626 etp_worker *self = (etp_worker *)thr_arg;
1627
1628 /* try to distribute timeouts somewhat randomly */
1629 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1630
1631 for (;;)
1632 {
1633 X_LOCK (reqlock);
1634
1635 for (;;)
1636 {
1637 self->req = req = reqq_shift (&req_queue);
1638
1639 if (req)
1640 break;
1641
1642 ++idle;
1643
1644 ts.tv_sec = time (0) + idle_timeout;
1645 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
1646 {
1647 if (idle > max_idle)
1648 {
1649 --idle;
1650 X_UNLOCK (reqlock);
1651 X_LOCK (wrklock);
1652 --started;
1653 X_UNLOCK (wrklock);
1654 goto quit;
1655 }
1656
1657 /* we are allowed to idle, so do so without any timeout */
1658 X_COND_WAIT (reqwait, reqlock);
1659 }
1660
1661 --idle;
1662 }
1663
1664 --nready;
1665
1666 X_UNLOCK (reqlock);
1667
1668 if (req->type < 0)
1669 goto quit;
1670
1671 if (!EIO_CANCELLED (req))
1672 ETP_EXECUTE (self, req);
1673
1674 X_LOCK (reslock);
1675
1676 ++npending;
1677
1678 if (!reqq_push (&res_queue, req) && want_poll_cb)
1679 want_poll_cb ();
1680
1681 self->req = 0;
1682 etp_worker_clear (self);
1683
1684 X_UNLOCK (reslock);
1685 }
1686
1687quit:
1688 X_LOCK (wrklock);
1689 etp_worker_free (self);
1690 X_UNLOCK (wrklock);
1691
1692 return 0;
1693}
1694
1695/*****************************************************************************/ 1726/*****************************************************************************/
1696 1727
1697int ecb_cold 1728int ecb_cold
1698eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1729eio_init (void (*want_poll)(void), void (*done_poll)(void))
1699{ 1730{
1700 return etp_init (want_poll, done_poll); 1731 eio_want_poll_cb = want_poll;
1732 eio_done_poll_cb = done_poll;
1733
1734 return etp_init (EIO_POOL, 0, 0, 0);
1701} 1735}
1702 1736
1703ecb_inline void 1737ecb_inline void
1704eio_api_destroy (eio_req *req) 1738eio_api_destroy (eio_req *req)
1705{ 1739{
1706 free (req); 1740 free (req);
1707} 1741}
1708 1742
1709#define REQ(rtype) \ 1743#define REQ(rtype) \
1710 eio_req *req; \ 1744 eio_req *req; \
1711 \ 1745 \
1712 req = (eio_req *)calloc (1, sizeof *req); \ 1746 req = (eio_req *)calloc (1, sizeof *req); \
1713 if (!req) \ 1747 if (!req) \
1714 return 0; \ 1748 return 0; \
… …
1728 { \ 1762 { \
1729 eio_api_destroy (req); \ 1763 eio_api_destroy (req); \
1730 return 0; \ 1764 return 0; \
1731 } 1765 }
1732 1766
1767#define SINGLEDOT(ptr) (0[(char *)(ptr)] == '.' && !1[(char *)(ptr)])
1768
1733static void 1769static void
1734eio_execute (etp_worker *self, eio_req *req) 1770eio_execute (etp_worker *self, eio_req *req)
1735{ 1771{
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 }
1800
1736 switch (req->type) 1801 switch (req->type)
1737 { 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;
1738 case EIO_READ: ALLOC (req->size); 1810 case EIO_READ: ALLOC (req->size);
1739 req->result = req->offs >= 0 1811 req->result = req->offs >= 0
1740 ? pread (req->int1, req->ptr2, req->size, req->offs) 1812 ? pread (req->int1, req->ptr2, req->size, req->offs)
1741 : read (req->int1, req->ptr2, req->size); break; 1813 : read (req->int1, req->ptr2, req->size); break;
1742 case EIO_WRITE: req->result = req->offs >= 0 1814 case EIO_WRITE: req->result = req->offs >= 0
1743 ? pwrite (req->int1, req->ptr2, req->size, req->offs) 1815 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1744 : write (req->int1, req->ptr2, req->size); break; 1816 : write (req->int1, req->ptr2, req->size); break;
1745 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
1746 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;
1747 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
1748 1825
1749 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1826 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1750 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;
1751 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1828 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1829 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, AT_SYMLINK_NOFOLLOW); break;
1830 case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break;
1831 case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break;
1832 case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break;
1833 case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break;
1834
1835 case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
1836 case EIO_RMDIR: /* complications arise because "." cannot be removed, so we might have to expand */
1837 req->result = req->wd && SINGLEDOT (req->ptr1)
1838 ? rmdir (req->wd->str)
1839 : unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break;
1840 case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); break;
1841 case EIO_RENAME: /* complications arise because "." cannot be renamed, so we might have to expand */
1842 req->result = req->wd && SINGLEDOT (req->ptr1)
1843 ? rename (req->wd->str, req->ptr2)
1844 : renameat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2); break;
1845 case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2, 0); break;
1846 case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
1847 case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1848 case EIO_READLINK: ALLOC (PATH_MAX);
1849 req->result = readlinkat (dirfd, req->ptr1, req->ptr2, PATH_MAX); break;
1850 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1851 req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1852 case EIO_UTIME:
1853 case EIO_FUTIME:
1854 {
1855 struct timespec ts[2];
1856 struct timespec *times;
1857
1858 if (req->nv1 != -1. || req->nv2 != -1.)
1859 {
1860 ts[0].tv_sec = req->nv1;
1861 ts[0].tv_nsec = (req->nv1 - ts[0].tv_sec) * 1e9;
1862 ts[1].tv_sec = req->nv2;
1863 ts[1].tv_nsec = (req->nv2 - ts[1].tv_sec) * 1e9;
1864
1865 times = ts;
1866 }
1867 else
1868 times = 0;
1869
1870 req->result = req->type == EIO_FUTIME
1871 ? futimens (req->int1, times)
1872 : utimensat (dirfd, req->ptr1, times, 0);
1873 }
1874 break;
1875
1876#else
1877
1878 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1879 req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
1880 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1752 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 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;
1886
1887 case EIO_UNLINK: req->result = unlink (path ); break;
1888 case EIO_RMDIR: req->result = rmdir (path ); break;
1889 case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); 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;
1898
1899 case EIO_UTIME:
1900 case EIO_FUTIME:
1901 {
1902 struct timeval tv[2];
1903 struct timeval *times;
1904
1905 if (req->nv1 != -1. || req->nv2 != -1.)
1906 {
1907 tv[0].tv_sec = req->nv1;
1908 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
1909 tv[1].tv_sec = req->nv2;
1910 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
1911
1912 times = tv;
1913 }
1914 else
1915 times = 0;
1916
1917 req->result = req->type == EIO_FUTIME
1918 ? futimes (req->int1, times)
1919 : utimes (req->ptr1, times);
1920 }
1921 break;
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
1753 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1932 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1754 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; 1933 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1755 1934
1756 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1757 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1758 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS)); 1935 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1759 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break; 1936 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1760 1937
1761 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1762 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; 1938 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1763 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1764 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break; 1939 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1765 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1766 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break; 1940 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1767 1941
1768 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
1769 case EIO_CLOSE: req->result = close (req->int1); break; 1942 case EIO_CLOSE: req->result = close (req->int1); break;
1770 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break; 1943 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1771 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
1772 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1773 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1774 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1775 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1776 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1777 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1778
1779 case EIO_READLINK: ALLOC (PATH_MAX);
1780 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1781
1782 case EIO_SYNC: req->result = 0; sync (); break; 1944 case EIO_SYNC: req->result = 0; sync (); break;
1783 case EIO_FSYNC: req->result = fsync (req->int1); break; 1945 case EIO_FSYNC: req->result = fsync (req->int1); break;
1784 case EIO_FDATASYNC: req->result = fdatasync (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;
1785 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break; 1949 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1786 case EIO_MTOUCH: req->result = eio__mtouch (req); break; 1950 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1787 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break; 1951 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1788 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break; 1952 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1789 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break; 1953 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
1790 1954
1791 case EIO_READDIR: eio__scandir (req, self); break; 1955 case EIO_READDIR: eio__scandir (req, self); break;
1792 1956
1793 case EIO_BUSY: 1957 case EIO_BUSY:
1794#ifdef _WIN32 1958#ifdef _WIN32
… …
1803 req->result = select (0, 0, 0, 0, &tv); 1967 req->result = select (0, 0, 0, 0, &tv);
1804 } 1968 }
1805#endif 1969#endif
1806 break; 1970 break;
1807 1971
1808 case EIO_UTIME: 1972#if 0
1809 case EIO_FUTIME:
1810 {
1811 struct timeval tv[2];
1812 struct timeval *times;
1813
1814 if (req->nv1 != -1. || req->nv2 != -1.)
1815 {
1816 tv[0].tv_sec = req->nv1;
1817 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
1818 tv[1].tv_sec = req->nv2;
1819 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
1820
1821 times = tv;
1822 }
1823 else
1824 times = 0;
1825
1826 req->result = req->type == EIO_FUTIME
1827 ? futimes (req->int1, times)
1828 : utimes (req->ptr1, times);
1829 }
1830 break;
1831
1832 case EIO_GROUP: 1973 case EIO_GROUP:
1833 abort (); /* handled in eio_request */ 1974 abort (); /* handled in eio_request */
1975#endif
1834 1976
1835 case EIO_NOP: 1977 case EIO_NOP:
1836 req->result = 0; 1978 req->result = 0;
1837 break; 1979 break;
1838 1980
1839 case EIO_CUSTOM: 1981 case EIO_CUSTOM:
1840 req->feed (req); 1982 req->feed (req);
1841 break; 1983 break;
1842 1984
1843 default: 1985 default:
1844 errno = ENOSYS;
1845 req->result = -1; 1986 req->result = EIO_ENOSYS ();
1846 break; 1987 break;
1847 } 1988 }
1848 1989
1849 req->errorno = errno; 1990 req->errorno = errno;
1850} 1991}
1851 1992
1852#ifndef EIO_NO_WRAPPERS 1993#ifndef EIO_NO_WRAPPERS
1853 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
1854eio_req *eio_nop (int pri, eio_cb cb, void *data) 2005eio_req *eio_nop (int pri, eio_cb cb, void *data)
1855{ 2006{
1856 REQ (EIO_NOP); SEND; 2007 REQ (EIO_NOP); SEND;
1857} 2008}
1858 2009
… …
1874eio_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)
1875{ 2026{
1876 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;
1877} 2028}
1878 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
1879eio_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)
1880{ 2046{
1881 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;
1882} 2048}
1883 2049
… …
1889eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data) 2055eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
1890{ 2056{
1891 REQ (EIO_MLOCKALL); req->int1 = flags; SEND; 2057 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
1892} 2058}
1893 2059
1894eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) 2060eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
1895{ 2061{
1896 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; 2062 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
1897}
1898
1899eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1900{
1901 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1902} 2063}
1903 2064
1904eio_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)
1905{ 2066{
1906 REQ (EIO_CLOSE); req->int1 = fd; SEND; 2067 REQ (EIO_CLOSE); req->int1 = fd; SEND;
… …
1909eio_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)
1910{ 2071{
1911 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;
1912} 2073}
1913 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
1914eio_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)
1915{ 2081{
1916 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;
1917} 2083}
1918 2084
1919eio_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)
1920{ 2086{
1921 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;
1922} 2088}
1923 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
1924eio_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)
1925{ 2101{
1926 REQ (EIO_FSTAT); req->int1 = fd; SEND; 2102 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1927} 2103}
1928 2104
… …
1944eio_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)
1945{ 2121{
1946 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2122 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1947} 2123}
1948 2124
1949eio_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)
1950{ 2126{
1951 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;
1952} 2128}
1953 2129
1954eio_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)
… …
1974eio_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)
1975{ 2151{
1976 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2152 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1977} 2153}
1978 2154
1979eio_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)
1980{ 2156{
1981 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;
1982} 2158}
1983 2159
1984eio_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)
… …
1998} 2174}
1999 2175
2000eio_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)
2001{ 2177{
2002 return eio__1path (EIO_READLINK, path, pri, cb, data); 2178 return eio__1path (EIO_READLINK, path, pri, cb, data);
2179}
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);
2003} 2184}
2004 2185
2005eio_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)
2006{ 2187{
2007 return eio__1path (EIO_STAT, path, pri, cb, data); 2188 return eio__1path (EIO_STAT, path, pri, cb, data);
… …
2109void 2290void
2110eio_grp_add (eio_req *grp, eio_req *req) 2291eio_grp_add (eio_req *grp, eio_req *req)
2111{ 2292{
2112 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));
2113 2294
2114 grp->flags |= EIO_FLAG_GROUPADD; 2295 grp->flags |= ETP_FLAG_GROUPADD;
2115 2296
2116 ++grp->size; 2297 ++grp->size;
2117 req->grp = grp; 2298 req->grp = grp;
2118 2299
2119 req->grp_prev = 0; 2300 req->grp_prev = 0;
… …
2126} 2307}
2127 2308
2128/*****************************************************************************/ 2309/*****************************************************************************/
2129/* misc garbage */ 2310/* misc garbage */
2130 2311
2131ssize_t 2312eio_ssize_t
2132eio_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)
2133{ 2314{
2134 etp_worker wrk;
2135 ssize_t ret;
2136
2137 wrk.dbuf = 0;
2138
2139 ret = eio__sendfile (ofd, ifd, offset, count, &wrk); 2315 return eio__sendfile (ofd, ifd, offset, count);
2140
2141 if (wrk.dbuf)
2142 free (wrk.dbuf);
2143
2144 return ret;
2145} 2316}
2146 2317

Diff Legend

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