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

Comparing libeio/eio.c (file contents):
Revision 1.83 by root, Thu Jul 7 22:36:18 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 */
… …
68#endif 62#endif
69 63
70#ifndef ECANCELED 64#ifndef ECANCELED
71# define ECANCELED EDOM 65# define ECANCELED EDOM
72#endif 66#endif
67#ifndef ELOOP
68# define ELOOP EDOM
69#endif
70
71#if !defined(ENOTSOCK) && defined(WSAENOTSOCK)
72# define ENOTSOCK WSAENOTSOCK
73#endif
73 74
74static void eio_destroy (eio_req *req); 75static void eio_destroy (eio_req *req);
75 76
76#ifndef EIO_FINISH 77#ifndef EIO_FINISH
77# 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
… …
83 84
84#ifndef EIO_FEED 85#ifndef EIO_FEED
85# 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)
86#endif 87#endif
87 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
88#ifdef _WIN32 100#ifdef _WIN32
89 101
90 /*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
91#else 234#else
92 235
93# include <sys/time.h> 236 #include <sys/time.h>
94# include <sys/select.h> 237 #include <sys/select.h>
95# 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
96# include <utime.h> 282# include <utime.h>
97# include <signal.h>
98# include <dirent.h>
99
100#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
101# include <sys/mman.h>
102#endif
103
104/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
105# if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
106# define _DIRENT_HAVE_D_TYPE /* sigh */
107# define D_INO(de) (de)->d_fileno
108# define D_NAMLEN(de) (de)->d_namlen
109# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
110# define D_INO(de) (de)->d_ino
111# endif 283#endif
112 284
113#ifdef _D_EXACT_NAMLEN 285#if HAVE_SYS_SYSCALL_H
114# undef D_NAMLEN 286# include <sys/syscall.h>
115# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
116#endif
117
118# ifdef _DIRENT_HAVE_D_TYPE
119# define D_TYPE(de) (de)->d_type
120# endif
121
122# ifndef EIO_STRUCT_DIRENT
123# define EIO_STRUCT_DIRENT struct dirent
124# endif
125
126#endif 287#endif
127 288
128#if HAVE_SENDFILE 289#if HAVE_SENDFILE
129# if __linux 290# if __linux
130# include <sys/sendfile.h> 291# include <sys/sendfile.h>
… …
145#endif 306#endif
146#ifndef D_INO 307#ifndef D_INO
147# define D_INO(de) 0 308# define D_INO(de) 0
148#endif 309#endif
149#ifndef D_NAMLEN 310#ifndef D_NAMLEN
150# define D_NAMLEN(de) strlen ((de)->d_name) 311# define D_NAMLEN(entp) strlen (D_NAME (entp))
151#endif 312#endif
152 313
153/* used for struct dirent, AIX doesn't provide it */ 314/* used for struct dirent, AIX doesn't provide it */
154#ifndef NAME_MAX 315#ifndef NAME_MAX
155# define NAME_MAX 4096 316# define NAME_MAX 4096
… …
161#endif 322#endif
162 323
163/* buffer size for various temporary buffers */ 324/* buffer size for various temporary buffers */
164#define EIO_BUFSIZE 65536 325#define EIO_BUFSIZE 65536
165 326
166#define dBUF \ 327#define dBUF \
167 char *eio_buf; \
168 ETP_WORKER_LOCK (self); \
169 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 328 char *eio_buf = malloc (EIO_BUFSIZE); \
170 ETP_WORKER_UNLOCK (self); \
171 errno = ENOMEM; \ 329 errno = ENOMEM; \
172 if (!eio_buf) \ 330 if (!eio_buf) \
173 return -1; 331 return -1
174 332
175#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/*****************************************************************************/
176 361
177#define ETP_PRI_MIN EIO_PRI_MIN 362#define ETP_PRI_MIN EIO_PRI_MIN
178#define ETP_PRI_MAX EIO_PRI_MAX 363#define ETP_PRI_MAX EIO_PRI_MAX
179 364
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
180struct etp_worker; 375struct etp_worker;
181
182#define ETP_REQ eio_req 376#define ETP_REQ eio_req
183#define ETP_DESTROY(req) eio_destroy (req) 377#define ETP_DESTROY(req) eio_destroy (req)
184static int eio_finish (eio_req *req); 378static int eio_finish (eio_req *req);
185#define ETP_FINISH(req) eio_finish (req) 379#define ETP_FINISH(req) eio_finish (req)
186static void eio_execute (struct etp_worker *self, eio_req *req); 380static void eio_execute (struct etp_worker *self, eio_req *req);
187#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) 381#define ETP_EXECUTE(wrk,req) eio_execute (wrk, req)
188 382
189#define ETP_WORKER_CLEAR(req) \ 383#include "etp.c"
190 if (wrk->dbuf) \
191 { \
192 free (wrk->dbuf); \
193 wrk->dbuf = 0; \
194 } \
195 \
196 if (wrk->dirp) \
197 { \
198 closedir (wrk->dirp); \
199 wrk->dirp = 0; \
200 }
201 384
202#define ETP_WORKER_COMMON \ 385static struct etp_pool eio_pool;
203 void *dbuf; \ 386#define EIO_POOL (&eio_pool)
204 DIR *dirp;
205
206/*****************************************************************************/
207
208#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
209
210/* calculate time difference in ~1/EIO_TICKS of a second */
211ecb_inline int
212tvdiff (struct timeval *tv1, struct timeval *tv2)
213{
214 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
215 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
216}
217
218static unsigned int started, idle, wanted = 4;
219
220static void (*want_poll_cb) (void);
221static void (*done_poll_cb) (void);
222
223static unsigned int max_poll_time; /* reslock */
224static unsigned int max_poll_reqs; /* reslock */
225
226static volatile unsigned int nreqs; /* reqlock */
227static volatile unsigned int nready; /* reqlock */
228static volatile unsigned int npending; /* reqlock */
229static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
230static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
231
232static xmutex_t wrklock;
233static xmutex_t reslock;
234static xmutex_t reqlock;
235static xcond_t reqwait;
236
237#if !HAVE_PREADWRITE
238/*
239 * make our pread/pwrite emulation safe against themselves, but not against
240 * normal read/write by using a mutex. slows down execution a lot,
241 * but that's your problem, not mine.
242 */
243static xmutex_t preadwritelock = X_MUTEX_INIT;
244#endif
245
246typedef struct etp_worker
247{
248 /* locked by wrklock */
249 struct etp_worker *prev, *next;
250
251 xthread_t tid;
252
253 /* locked by reslock, reqlock or wrklock */
254 ETP_REQ *req; /* currently processed request */
255
256 ETP_WORKER_COMMON
257} etp_worker;
258
259static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */
260
261#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
262#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
263
264/* worker threads management */
265
266static void ecb_cold
267etp_worker_clear (etp_worker *wrk)
268{
269 ETP_WORKER_CLEAR (wrk);
270}
271
272static void ecb_cold
273etp_worker_free (etp_worker *wrk)
274{
275 wrk->next->prev = wrk->prev;
276 wrk->prev->next = wrk->next;
277
278 free (wrk);
279}
280
281static unsigned int
282etp_nreqs (void)
283{
284 int retval;
285 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
286 retval = nreqs;
287 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
288 return retval;
289}
290
291static unsigned int
292etp_nready (void)
293{
294 unsigned int retval;
295
296 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
297 retval = nready;
298 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
299
300 return retval;
301}
302
303static unsigned int
304etp_npending (void)
305{
306 unsigned int retval;
307
308 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
309 retval = npending;
310 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
311
312 return retval;
313}
314
315static unsigned int
316etp_nthreads (void)
317{
318 unsigned int retval;
319
320 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
321 retval = started;
322 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
323
324 return retval;
325}
326
327/*
328 * a somewhat faster data structure might be nice, but
329 * with 8 priorities this actually needs <20 insns
330 * per shift, the most expensive operation.
331 */
332typedef struct {
333 ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */
334 int size;
335} etp_reqq;
336
337static etp_reqq req_queue;
338static etp_reqq res_queue;
339
340static int ecb_noinline
341reqq_push (etp_reqq *q, ETP_REQ *req)
342{
343 int pri = req->pri;
344 req->next = 0;
345
346 if (q->qe[pri])
347 {
348 q->qe[pri]->next = req;
349 q->qe[pri] = req;
350 }
351 else
352 q->qe[pri] = q->qs[pri] = req;
353
354 return q->size++;
355}
356
357static ETP_REQ * ecb_noinline
358reqq_shift (etp_reqq *q)
359{
360 int pri;
361
362 if (!q->size)
363 return 0;
364
365 --q->size;
366
367 for (pri = ETP_NUM_PRI; pri--; )
368 {
369 eio_req *req = q->qs[pri];
370
371 if (req)
372 {
373 if (!(q->qs[pri] = (eio_req *)req->next))
374 q->qe[pri] = 0;
375
376 return req;
377 }
378 }
379
380 abort ();
381}
382
383static void ecb_cold
384etp_thread_init (void)
385{
386 X_MUTEX_CREATE (wrklock);
387 X_MUTEX_CREATE (reslock);
388 X_MUTEX_CREATE (reqlock);
389 X_COND_CREATE (reqwait);
390}
391
392static void ecb_cold
393etp_atfork_prepare (void)
394{
395 X_LOCK (wrklock);
396 X_LOCK (reqlock);
397 X_LOCK (reslock);
398#if !HAVE_PREADWRITE
399 X_LOCK (preadwritelock);
400#endif
401}
402
403static void ecb_cold
404etp_atfork_parent (void)
405{
406#if !HAVE_PREADWRITE
407 X_UNLOCK (preadwritelock);
408#endif
409 X_UNLOCK (reslock);
410 X_UNLOCK (reqlock);
411 X_UNLOCK (wrklock);
412}
413
414static void ecb_cold
415etp_atfork_child (void)
416{
417 ETP_REQ *prv;
418
419 while ((prv = reqq_shift (&req_queue)))
420 ETP_DESTROY (prv);
421
422 while ((prv = reqq_shift (&res_queue)))
423 ETP_DESTROY (prv);
424
425 while (wrk_first.next != &wrk_first)
426 {
427 etp_worker *wrk = wrk_first.next;
428
429 if (wrk->req)
430 ETP_DESTROY (wrk->req);
431
432 etp_worker_clear (wrk);
433 etp_worker_free (wrk);
434 }
435
436 started = 0;
437 idle = 0;
438 nreqs = 0;
439 nready = 0;
440 npending = 0;
441
442 etp_thread_init ();
443}
444
445static void ecb_cold
446etp_once_init (void)
447{
448 etp_thread_init ();
449 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
450}
451
452static int ecb_cold
453etp_init (void (*want_poll)(void), void (*done_poll)(void))
454{
455 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
456
457 pthread_once (&doinit, etp_once_init);
458
459 want_poll_cb = want_poll;
460 done_poll_cb = done_poll;
461
462 return 0;
463}
464
465X_THREAD_PROC (etp_proc);
466
467static void ecb_cold
468etp_start_thread (void)
469{
470 etp_worker *wrk = calloc (1, sizeof (etp_worker));
471
472 /*TODO*/
473 assert (("unable to allocate worker thread data", wrk));
474
475 X_LOCK (wrklock);
476
477 if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
478 {
479 wrk->prev = &wrk_first;
480 wrk->next = wrk_first.next;
481 wrk_first.next->prev = wrk;
482 wrk_first.next = wrk;
483 ++started;
484 }
485 else
486 free (wrk);
487
488 X_UNLOCK (wrklock);
489}
490
491static void
492etp_maybe_start_thread (void)
493{
494 if (ecb_expect_true (etp_nthreads () >= wanted))
495 return;
496
497 /* todo: maybe use idle here, but might be less exact */
498 if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
499 return;
500
501 etp_start_thread ();
502}
503
504static void ecb_cold
505etp_end_thread (void)
506{
507 eio_req *req = calloc (1, sizeof (eio_req));
508
509 req->type = -1;
510 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
511
512 X_LOCK (reqlock);
513 reqq_push (&req_queue, req);
514 X_COND_SIGNAL (reqwait);
515 X_UNLOCK (reqlock);
516
517 X_LOCK (wrklock);
518 --started;
519 X_UNLOCK (wrklock);
520}
521
522static int
523etp_poll (void)
524{
525 unsigned int maxreqs;
526 unsigned int maxtime;
527 struct timeval tv_start, tv_now;
528
529 X_LOCK (reslock);
530 maxreqs = max_poll_reqs;
531 maxtime = max_poll_time;
532 X_UNLOCK (reslock);
533
534 if (maxtime)
535 gettimeofday (&tv_start, 0);
536
537 for (;;)
538 {
539 ETP_REQ *req;
540
541 etp_maybe_start_thread ();
542
543 X_LOCK (reslock);
544 req = reqq_shift (&res_queue);
545
546 if (req)
547 {
548 --npending;
549
550 if (!res_queue.size && done_poll_cb)
551 done_poll_cb ();
552 }
553
554 X_UNLOCK (reslock);
555
556 if (!req)
557 return 0;
558
559 X_LOCK (reqlock);
560 --nreqs;
561 X_UNLOCK (reqlock);
562
563 if (ecb_expect_false (req->type == EIO_GROUP && req->size))
564 {
565 req->int1 = 1; /* mark request as delayed */
566 continue;
567 }
568 else
569 {
570 int res = ETP_FINISH (req);
571 if (ecb_expect_false (res))
572 return res;
573 }
574
575 if (ecb_expect_false (maxreqs && !--maxreqs))
576 break;
577
578 if (maxtime)
579 {
580 gettimeofday (&tv_now, 0);
581
582 if (tvdiff (&tv_start, &tv_now) >= maxtime)
583 break;
584 }
585 }
586
587 errno = EAGAIN;
588 return -1;
589}
590
591static void
592etp_cancel (ETP_REQ *req)
593{
594 req->cancelled = 1;
595
596 eio_grp_cancel (req);
597}
598
599static void
600etp_submit (ETP_REQ *req)
601{
602 req->pri -= ETP_PRI_MIN;
603
604 if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
605 if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
606
607 if (ecb_expect_false (req->type == EIO_GROUP))
608 {
609 /* I hope this is worth it :/ */
610 X_LOCK (reqlock);
611 ++nreqs;
612 X_UNLOCK (reqlock);
613
614 X_LOCK (reslock);
615
616 ++npending;
617
618 if (!reqq_push (&res_queue, req) && want_poll_cb)
619 want_poll_cb ();
620
621 X_UNLOCK (reslock);
622 }
623 else
624 {
625 X_LOCK (reqlock);
626 ++nreqs;
627 ++nready;
628 reqq_push (&req_queue, req);
629 X_COND_SIGNAL (reqwait);
630 X_UNLOCK (reqlock);
631
632 etp_maybe_start_thread ();
633 }
634}
635
636static void ecb_cold
637etp_set_max_poll_time (double nseconds)
638{
639 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
640 max_poll_time = nseconds * EIO_TICKS;
641 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
642}
643
644static void ecb_cold
645etp_set_max_poll_reqs (unsigned int maxreqs)
646{
647 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
648 max_poll_reqs = maxreqs;
649 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
650}
651
652static void ecb_cold
653etp_set_max_idle (unsigned int nthreads)
654{
655 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
656 max_idle = nthreads;
657 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
658}
659
660static void ecb_cold
661etp_set_idle_timeout (unsigned int seconds)
662{
663 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
664 idle_timeout = seconds;
665 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
666}
667
668static void ecb_cold
669etp_set_min_parallel (unsigned int nthreads)
670{
671 if (wanted < nthreads)
672 wanted = nthreads;
673}
674
675static void ecb_cold
676etp_set_max_parallel (unsigned int nthreads)
677{
678 if (wanted > nthreads)
679 wanted = nthreads;
680
681 while (started > wanted)
682 etp_end_thread ();
683}
684 387
685/*****************************************************************************/ 388/*****************************************************************************/
686 389
687static void 390static void
688grp_try_feed (eio_req *grp) 391grp_try_feed (eio_req *grp)
689{ 392{
690 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 393 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
691 { 394 {
692 grp->flags &= ~EIO_FLAG_GROUPADD; 395 grp->flags &= ~ETP_FLAG_GROUPADD;
693 396
694 EIO_FEED (grp); 397 EIO_FEED (grp);
695 398
696 /* stop if no progress has been made */ 399 /* stop if no progress has been made */
697 if (!(grp->flags & EIO_FLAG_GROUPADD)) 400 if (!(grp->flags & ETP_FLAG_GROUPADD))
698 { 401 {
699 grp->feed = 0; 402 grp->feed = 0;
700 break; 403 break;
701 } 404 }
702 } 405 }
… …
709 412
710 /* call feeder, if applicable */ 413 /* call feeder, if applicable */
711 grp_try_feed (grp); 414 grp_try_feed (grp);
712 415
713 /* finish, if done */ 416 /* finish, if done */
714 if (!grp->size && grp->int1) 417 if (!grp->size && grp->flags & ETP_FLAG_DELAYED)
715 return eio_finish (grp); 418 return eio_finish (grp);
716 else 419 else
717 return 0; 420 return 0;
718} 421}
719 422
… …
755} 458}
756 459
757void 460void
758eio_grp_cancel (eio_req *grp) 461eio_grp_cancel (eio_req *grp)
759{ 462{
760 for (grp = grp->grp_first; grp; grp = grp->grp_next) 463 etp_grp_cancel (EIO_POOL, grp);
761 eio_cancel (grp);
762} 464}
763 465
764void 466void
765eio_cancel (eio_req *req) 467eio_cancel (eio_req *req)
766{ 468{
767 etp_cancel (req); 469 etp_cancel (EIO_POOL, req);
768} 470}
769 471
770void 472void
771eio_submit (eio_req *req) 473eio_submit (eio_req *req)
772{ 474{
773 etp_submit (req); 475 etp_submit (EIO_POOL, req);
774} 476}
775 477
776unsigned int 478unsigned int
777eio_nreqs (void) 479eio_nreqs (void)
778{ 480{
779 return etp_nreqs (); 481 return etp_nreqs (EIO_POOL);
780} 482}
781 483
782unsigned int 484unsigned int
783eio_nready (void) 485eio_nready (void)
784{ 486{
785 return etp_nready (); 487 return etp_nready (EIO_POOL);
786} 488}
787 489
788unsigned int 490unsigned int
789eio_npending (void) 491eio_npending (void)
790{ 492{
791 return etp_npending (); 493 return etp_npending (EIO_POOL);
792} 494}
793 495
794unsigned int ecb_cold 496unsigned int ecb_cold
795eio_nthreads (void) 497eio_nthreads (void)
796{ 498{
797 return etp_nthreads (); 499 return etp_nthreads (EIO_POOL);
798} 500}
799 501
800void ecb_cold 502void ecb_cold
801eio_set_max_poll_time (double nseconds) 503eio_set_max_poll_time (double nseconds)
802{ 504{
803 etp_set_max_poll_time (nseconds); 505 etp_set_max_poll_time (EIO_POOL, nseconds);
804} 506}
805 507
806void ecb_cold 508void ecb_cold
807eio_set_max_poll_reqs (unsigned int maxreqs) 509eio_set_max_poll_reqs (unsigned int maxreqs)
808{ 510{
809 etp_set_max_poll_reqs (maxreqs); 511 etp_set_max_poll_reqs (EIO_POOL, maxreqs);
810} 512}
811 513
812void ecb_cold 514void ecb_cold
813eio_set_max_idle (unsigned int nthreads) 515eio_set_max_idle (unsigned int nthreads)
814{ 516{
815 etp_set_max_idle (nthreads); 517 etp_set_max_idle (EIO_POOL, nthreads);
816} 518}
817 519
818void ecb_cold 520void ecb_cold
819eio_set_idle_timeout (unsigned int seconds) 521eio_set_idle_timeout (unsigned int seconds)
820{ 522{
821 etp_set_idle_timeout (seconds); 523 etp_set_idle_timeout (EIO_POOL, seconds);
822} 524}
823 525
824void ecb_cold 526void ecb_cold
825eio_set_min_parallel (unsigned int nthreads) 527eio_set_min_parallel (unsigned int nthreads)
826{ 528{
827 etp_set_min_parallel (nthreads); 529 etp_set_min_parallel (EIO_POOL, nthreads);
828} 530}
829 531
830void ecb_cold 532void ecb_cold
831eio_set_max_parallel (unsigned int nthreads) 533eio_set_max_parallel (unsigned int nthreads)
832{ 534{
833 etp_set_max_parallel (nthreads); 535 etp_set_max_parallel (EIO_POOL, nthreads);
834} 536}
835 537
836int eio_poll (void) 538int eio_poll (void)
837{ 539{
838 return etp_poll (); 540 return etp_poll (EIO_POOL);
839} 541}
840 542
841/*****************************************************************************/ 543/*****************************************************************************/
842/* work around various missing functions */ 544/* work around various missing functions */
843
844#if !HAVE_PREADWRITE
845# undef pread
846# undef pwrite
847# define pread eio__pread
848# define pwrite eio__pwrite
849
850static ssize_t
851eio__pread (int fd, void *buf, size_t count, off_t offset)
852{
853 ssize_t res;
854 off_t ooffset;
855
856 X_LOCK (preadwritelock);
857 ooffset = lseek (fd, 0, SEEK_CUR);
858 lseek (fd, offset, SEEK_SET);
859 res = read (fd, buf, count);
860 lseek (fd, ooffset, SEEK_SET);
861 X_UNLOCK (preadwritelock);
862
863 return res;
864}
865
866static ssize_t
867eio__pwrite (int fd, void *buf, size_t count, off_t offset)
868{
869 ssize_t res;
870 off_t ooffset;
871
872 X_LOCK (preadwritelock);
873 ooffset = lseek (fd, 0, SEEK_CUR);
874 lseek (fd, offset, SEEK_SET);
875 res = write (fd, buf, count);
876 lseek (fd, ooffset, SEEK_SET);
877 X_UNLOCK (preadwritelock);
878
879 return res;
880}
881#endif
882 545
883#ifndef HAVE_UTIMES 546#ifndef HAVE_UTIMES
884 547
885# undef utimes 548# undef utimes
886# define utimes(path,times) eio__utimes (path, times) 549# define utimes(path,times) eio__utimes (path, times)
… …
919 582
920#if !HAVE_FDATASYNC 583#if !HAVE_FDATASYNC
921# undef fdatasync 584# undef fdatasync
922# define fdatasync(fd) fsync (fd) 585# define fdatasync(fd) fsync (fd)
923#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}
924 604
925/* sync_file_range always needs emulation */ 605/* sync_file_range always needs emulation */
926static int 606static int
927eio__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)
928{ 608{
… …
949 * call fdatasync, as that matches the expectation of its users best */ 629 * call fdatasync, as that matches the expectation of its users best */
950 return fdatasync (fd); 630 return fdatasync (fd);
951} 631}
952 632
953static int 633static int
954eio__fallocate (int fd, int mode, off_t offset, size_t nbytes) 634eio__fallocate (int fd, int mode, off_t offset, size_t len)
955{ 635{
956#if HAVE_FALLOCATE 636#if HAVE_LINUX_FALLOCATE
957 return fallocate (fd, offset, nbytes, flags); 637 return fallocate (fd, mode, offset, len);
958#else 638#else
959 errno = ENOSYS; 639 return EIO_ENOSYS ();
960 return -1;
961#endif 640#endif
962} 641}
963 642
964#if !HAVE_READAHEAD 643#if !HAVE_READAHEAD
965# undef readahead 644# undef readahead
966# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 645# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
967 646
968static ssize_t 647static eio_ssize_t
969eio__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)
970{ 649{
971 size_t todo = count; 650 size_t todo = count;
972 dBUF; 651 dBUF;
973 652
… …
978 pread (fd, eio_buf, len, offset); 657 pread (fd, eio_buf, len, offset);
979 offset += len; 658 offset += len;
980 todo -= len; 659 todo -= len;
981 } 660 }
982 661
983 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 */
984 return count; 666 return 0;
985} 667}
986 668
987#endif 669#endif
988 670
989/* sendfile always needs emulation */ 671/* sendfile always needs emulation */
990static ssize_t 672static eio_ssize_t
991eio__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)
992{ 674{
993 ssize_t written = 0; 675 eio_ssize_t written = 0;
994 ssize_t res; 676 eio_ssize_t res;
995 677
996 if (!count) 678 if (!count)
997 return 0; 679 return 0;
998 680
999 for (;;) 681 for (;;)
… …
1024 706
1025 /* according to source inspection, this is correct, and useful behaviour */ 707 /* according to source inspection, this is correct, and useful behaviour */
1026 if (sbytes) 708 if (sbytes)
1027 res = sbytes; 709 res = sbytes;
1028 710
1029# elif defined (__APPLE__) 711# elif defined __APPLE__
1030 off_t sbytes = count; 712 off_t sbytes = count;
1031 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0); 713 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
1032 714
1033 /* according to the manpage, sbytes is always valid */ 715 /* according to the manpage, sbytes is always valid */
1034 if (sbytes) 716 if (sbytes)
… …
1051 if (res < 0 && sbytes) 733 if (res < 0 && sbytes)
1052 res = sbytes; 734 res = sbytes;
1053 735
1054# endif 736# endif
1055 737
1056#elif defined (_WIN32) 738#elif defined (_WIN32) && 0
1057 /* 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 */
1058 /* 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, */
1059 /* libeio guarantees that the file offset does not change, and windows */ 741 /* libeio guarantees that the file offset does not change, and windows */
1060 /* 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 */
1061 HANDLE h = TO_SOCKET (ifd); 743 HANDLE h = TO_SOCKET (ifd);
1062 SetFilePointer (h, offset, 0, FILE_BEGIN); 744 SetFilePointer (h, offset, 0, FILE_BEGIN);
1063 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0); 745 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1064 746
1065#else 747#else
1066 res = -1; 748 res = EIO_ENOSYS ();
1067 errno = ENOSYS;
1068#endif 749#endif
1069 750
1070 /* we assume sendfile can copy at least 128mb in one go */ 751 /* we assume sendfile can copy at least 128mb in one go */
1071 if (res <= 128 * 1024 * 1024) 752 if (res <= 128 * 1024 * 1024)
1072 { 753 {
… …
1094 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 775 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1095 /* BSDs */ 776 /* BSDs */
1096#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... */
1097 || errno == ENOTSUP 778 || errno == ENOTSUP
1098#endif 779#endif
780#ifdef EOPNOTSUPP /* windows */
1099 || errno == EOPNOTSUPP /* BSDs */ 781 || errno == EOPNOTSUPP /* BSDs */
782#endif
1100#if __solaris 783#if __solaris
1101 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 784 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
1102#endif 785#endif
1103 ) 786 )
1104 ) 787 )
… …
1108 791
1109 res = 0; 792 res = 0;
1110 793
1111 while (count) 794 while (count)
1112 { 795 {
1113 ssize_t cnt; 796 eio_ssize_t cnt;
1114 797
1115 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);
1116 799
1117 if (cnt <= 0) 800 if (cnt <= 0)
1118 { 801 {
… …
1130 813
1131 offset += cnt; 814 offset += cnt;
1132 res += cnt; 815 res += cnt;
1133 count -= cnt; 816 count -= cnt;
1134 } 817 }
818
819 FUBd;
1135 } 820 }
1136 821
1137 return res; 822 return res;
1138} 823}
1139 824
… …
1166 /* round up length */ 851 /* round up length */
1167 *length = (*length + mask) & ~mask; 852 *length = (*length + mask) & ~mask;
1168} 853}
1169 854
1170#if !_POSIX_MEMLOCK 855#if !_POSIX_MEMLOCK
1171# define eio__mlockall(a) ((errno = ENOSYS), -1) 856# define eio__mlockall(a) EIO_ENOSYS ()
1172#else 857#else
1173 858
1174static int 859static int
1175eio__mlockall (int flags) 860eio__mlockall (int flags)
1176{ 861{
… …
1190 return mlockall (flags); 875 return mlockall (flags);
1191} 876}
1192#endif 877#endif
1193 878
1194#if !_POSIX_MEMLOCK_RANGE 879#if !_POSIX_MEMLOCK_RANGE
1195# define eio__mlock(a,b) ((errno = ENOSYS), -1) 880# define eio__mlock(a,b) EIO_ENOSYS ()
1196#else 881#else
1197 882
1198static int 883static int
1199eio__mlock (void *addr, size_t length) 884eio__mlock (void *addr, size_t length)
1200{ 885{
… …
1204} 889}
1205 890
1206#endif 891#endif
1207 892
1208#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) 893#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1209# define eio__msync(a,b,c) ((errno = ENOSYS), -1) 894# define eio__msync(a,b,c) EIO_ENOSYS ()
1210#else 895#else
1211 896
1212static int 897static int
1213eio__msync (void *mem, size_t len, int flags) 898eio__msync (void *mem, size_t len, int flags)
1214{ 899{
… …
1255 940
1256/*****************************************************************************/ 941/*****************************************************************************/
1257/* requests implemented outside eio_execute, because they are so large */ 942/* requests implemented outside eio_execute, because they are so large */
1258 943
1259static void 944static void
1260eio__realpath (eio_req *req, etp_worker *self) 945eio__lseek (eio_req *req)
1261{ 946{
1262 char *rel = req->ptr1; 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{
1263 char *res; 962 char *res;
963 const char *rel = path;
1264 char *tmp1, *tmp2; 964 char *tmp1, *tmp2;
1265#if SYMLOOP_MAX > 32 965#if SYMLOOP_MAX > 32
1266 int symlinks = SYMLOOP_MAX; 966 int symlinks = SYMLOOP_MAX;
1267#else 967#else
1268 int symlinks = 32; 968 int symlinks = 32;
1269#endif 969#endif
1270 970
1271 req->result = -1;
1272
1273 errno = EINVAL; 971 errno = EINVAL;
1274 if (!rel) 972 if (!rel)
1275 return; 973 return -1;
1276 974
1277 errno = ENOENT; 975 errno = ENOENT;
1278 if (!*rel) 976 if (!*rel)
1279 return; 977 return -1;
1280 978
1281 if (!req->ptr2) 979 res = etp_tmpbuf_get (tmpbuf, PATH_MAX * 3);
1282 { 980#ifdef _WIN32
1283 X_LOCK (wrklock); 981 if (_access (rel, 4) != 0)
1284 req->flags |= EIO_FLAG_PTR2_FREE;
1285 X_UNLOCK (wrklock);
1286 req->ptr2 = malloc (PATH_MAX * 3);
1287
1288 errno = ENOMEM;
1289 if (!req->ptr2)
1290 return; 982 return -1;
1291 }
1292 983
1293 res = req->ptr2; 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
1294 tmp1 = res + PATH_MAX; 997 tmp1 = res + PATH_MAX;
1295 tmp2 = tmp1 + PATH_MAX; 998 tmp2 = tmp1 + PATH_MAX;
1296 999
1297#if 0 /* disabled, the musl way to do things is just too racy */ 1000#if 0 /* disabled, the musl way to do things is just too racy */
1298#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME) 1001#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
… …
1302 1005
1303 if (fd >= 0) 1006 if (fd >= 0)
1304 { 1007 {
1305 sprintf (tmp1, "/proc/self/fd/%d", fd); 1008 sprintf (tmp1, "/proc/self/fd/%d", fd);
1306 req->result = readlink (tmp1, res, PATH_MAX); 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 */
1307 close (fd); 1011 close (fd);
1308
1309 /* here we should probably stat the open file and the disk file, to make sure they still match */
1310 1012
1311 if (req->result > 0) 1013 if (req->result > 0)
1312 goto done; 1014 goto done;
1313 } 1015 }
1314 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO) 1016 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1315 return; 1017 return -1;
1316 } 1018 }
1317#endif 1019#endif
1318#endif 1020#endif
1319 1021
1320 if (*rel != '/') 1022 if (*rel != '/')
1321 { 1023 {
1024 int len;
1025
1026 errno = ENOENT;
1027 if (wd == EIO_INVALID_WD)
1028 return -1;
1029
1030 if (wd == EIO_CWD)
1031 {
1322 if (!getcwd (res, PATH_MAX)) 1032 if (!getcwd (res, PATH_MAX))
1323 return; 1033 return -1;
1034
1035 len = strlen (res);
1036 }
1037 else
1038 memcpy (res, wd->str, len = wd->len);
1324 1039
1325 if (res [1]) /* only use if not / */ 1040 if (res [1]) /* only use if not / */
1326 res += strlen (res); 1041 res += len;
1327 } 1042 }
1328 1043
1329 while (*rel) 1044 while (*rel)
1330 { 1045 {
1331 ssize_t len, linklen; 1046 eio_ssize_t len, linklen;
1332 char *beg = rel; 1047 const char *beg = rel;
1333 1048
1334 while (*rel && *rel != '/') 1049 while (*rel && *rel != '/')
1335 ++rel; 1050 ++rel;
1336 1051
1337 len = rel - beg; 1052 len = rel - beg;
… …
1349 1064
1350 if (beg [1] == '.' && len == 2) 1065 if (beg [1] == '.' && len == 2)
1351 { 1066 {
1352 /* .. - back up one component, if possible */ 1067 /* .. - back up one component, if possible */
1353 1068
1354 while (res != req->ptr2) 1069 while (res != tmpbuf->ptr)
1355 if (*--res == '/') 1070 if (*--res == '/')
1356 break; 1071 break;
1357 1072
1358 continue; 1073 continue;
1359 } 1074 }
1360 } 1075 }
1361 1076
1362 errno = ENAMETOOLONG; 1077 errno = ENAMETOOLONG;
1363 if (res + 1 + len + 1 >= tmp1) 1078 if (res + 1 + len + 1 >= tmp1)
1364 return; 1079 return -1;
1365 1080
1366 /* copy one component */ 1081 /* copy one component */
1367 *res = '/'; 1082 *res = '/';
1368 memcpy (res + 1, beg, len); 1083 memcpy (res + 1, beg, len);
1369 1084
1370 /* zero-terminate, for readlink */ 1085 /* zero-terminate, for readlink */
1371 res [len + 1] = 0; 1086 res [len + 1] = 0;
1372 1087
1373 /* now check if it's a symlink */ 1088 /* now check if it's a symlink */
1374 linklen = readlink (req->ptr2, tmp1, PATH_MAX); 1089 linklen = readlink (tmpbuf->ptr, tmp1, PATH_MAX);
1375 1090
1376 if (linklen < 0) 1091 if (linklen < 0)
1377 { 1092 {
1378 if (errno != EINVAL) 1093 if (errno != EINVAL)
1379 return; 1094 return -1;
1380 1095
1381 /* it's a normal directory. hopefully */ 1096 /* it's a normal directory. hopefully */
1382 res += len + 1; 1097 res += len + 1;
1383 } 1098 }
1384 else 1099 else
… …
1386 /* yay, it was a symlink - build new path in tmp2 */ 1101 /* yay, it was a symlink - build new path in tmp2 */
1387 int rellen = strlen (rel); 1102 int rellen = strlen (rel);
1388 1103
1389 errno = ENAMETOOLONG; 1104 errno = ENAMETOOLONG;
1390 if (linklen + 1 + rellen >= PATH_MAX) 1105 if (linklen + 1 + rellen >= PATH_MAX)
1391 return; 1106 return -1;
1392 1107
1393 errno = ELOOP; 1108 errno = ELOOP;
1394 if (!--symlinks) 1109 if (!--symlinks)
1395 return; 1110 return -1;
1396 1111
1397 if (*tmp1 == '/') 1112 if (*tmp1 == '/')
1398 res = req->ptr2; /* symlink resolves to an absolute path */ 1113 res = tmpbuf->ptr; /* symlink resolves to an absolute path */
1399 1114
1400 /* we need to be careful, as rel might point into tmp2 already */ 1115 /* we need to be careful, as rel might point into tmp2 already */
1401 memmove (tmp2 + linklen + 1, rel, rellen + 1); 1116 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1402 tmp2 [linklen] = '/'; 1117 tmp2 [linklen] = '/';
1403 memcpy (tmp2, tmp1, linklen); 1118 memcpy (tmp2, tmp1, linklen);
… …
1405 rel = tmp2; 1120 rel = tmp2;
1406 } 1121 }
1407 } 1122 }
1408 1123
1409 /* special case for the lone root path */ 1124 /* special case for the lone root path */
1410 if (res == req->ptr2) 1125 if (res == tmpbuf->ptr)
1411 *res++ = '/'; 1126 *res++ = '/';
1412 1127
1413 req->result = res - (char *)req->ptr2; 1128 return res - (char *)tmpbuf->ptr;
1414 1129#endif
1415done:
1416 req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
1417} 1130}
1418 1131
1419static signed char 1132static signed char
1420eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) 1133eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1421{ 1134{
… …
1429 1142
1430#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 */
1431#define EIO_SORT_FAST 60 /* when to only use insertion sort */ 1144#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1432 1145
1433static void 1146static void
1434eio_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)
1435{ 1148{
1436 unsigned char bits [9 + sizeof (ino_t) * 8]; 1149 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1437 unsigned char *bit = bits; 1150 unsigned char *bit = bits;
1438 1151
1439 assert (CHAR_BIT == 8); 1152 assert (CHAR_BIT == 8);
1440 assert (sizeof (eio_dirent) * 8 < 256); 1153 assert (sizeof (eio_dirent) * 8 < 256);
1441 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */ 1154 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
… …
1443 1156
1444 if (size <= EIO_SORT_FAST) 1157 if (size <= EIO_SORT_FAST)
1445 return; 1158 return;
1446 1159
1447 /* 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 */
1448 /* 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 */
1449 /* inode_bits must contain all inodes ORed together */ 1162 /* inode_bits must contain all inodes ORed together */
1450 /* 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 */
1451 { 1164 {
1452 ino_t endianness; 1165 eio_ino_t endianness;
1453 int i, j; 1166 int i, j;
1454 1167
1455 /* 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" */
1456 for (i = 0; i < sizeof (ino_t); ++i) 1169 for (i = 0; i < sizeof (eio_ino_t); ++i)
1457 ((unsigned char *)&endianness)[i] = i; 1170 ((unsigned char *)&endianness)[i] = i;
1458 1171
1459 *bit++ = 0; 1172 *bit++ = 0;
1460 1173
1461 for (i = 0; i < sizeof (ino_t); ++i) 1174 for (i = 0; i < sizeof (eio_ino_t); ++i)
1462 { 1175 {
1463 /* shifting off the byte offsets out of "endianness" */ 1176 /* shifting off the byte offsets out of "endianness" */
1464 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8; 1177 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1465 endianness >>= 8; 1178 endianness >>= 8;
1466 1179
1467 for (j = 0; j < 8; ++j) 1180 for (j = 0; j < 8; ++j)
1468 if (inode_bits & (((ino_t)1) << (i * 8 + j))) 1181 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1469 *bit++ = offs + j; 1182 *bit++ = offs + j;
1470 } 1183 }
1471 1184
1472 for (j = 0; j < 8; ++j) 1185 for (j = 0; j < 8; ++j)
1473 if (score_bits & (1 << j)) 1186 if (score_bits & (1 << j))
1474 *bit++ = offsetof (eio_dirent, score) * 8 + j; 1187 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1475 } 1188 }
1476 1189
1477 /* now actually do the sorting (a variant of MSD radix sort) */ 1190 /* now actually do the sorting (a variant of MSD radix sort) */
1478 { 1191 {
1479 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base; 1192 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1480 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end; 1193 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1481 unsigned char *bit_stk [9 + sizeof (ino_t) * 8]; 1194 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1482 int stk_idx = 0; 1195 int stk_idx = 0;
1483 1196
1484 base_stk [stk_idx] = dents; 1197 base_stk [stk_idx] = dents;
1485 end_stk [stk_idx] = dents + size; 1198 end_stk [stk_idx] = dents + size;
1486 bit_stk [stk_idx] = bit - 1; 1199 bit_stk [stk_idx] = bit - 1;
… …
1565 } 1278 }
1566 } 1279 }
1567} 1280}
1568 1281
1569static void 1282static void
1570eio_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)
1571{ 1284{
1572 if (size <= 1) 1285 if (size <= 1)
1573 return; /* our insertion sort relies on size > 0 */ 1286 return; /* our insertion sort relies on size > 0 */
1574 1287
1575 /* 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 */
… …
1583 1296
1584/* read a full directory */ 1297/* read a full directory */
1585static void 1298static void
1586eio__scandir (eio_req *req, etp_worker *self) 1299eio__scandir (eio_req *req, etp_worker *self)
1587{ 1300{
1588 DIR *dirp;
1589 EIO_STRUCT_DIRENT *entp;
1590 char *name, *names; 1301 char *name, *names;
1591 int namesalloc = 4096; 1302 int namesalloc = 4096 - sizeof (void *) * 4;
1592 int namesoffs = 0; 1303 int namesoffs = 0;
1593 int flags = req->int1; 1304 int flags = req->int1;
1594 eio_dirent *dents = 0; 1305 eio_dirent *dents = 0;
1595 int dentalloc = 128; 1306 int dentalloc = 128;
1596 int dentoffs = 0; 1307 int dentoffs = 0;
1597 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
1598 1316
1599 req->result = -1; 1317 req->result = -1;
1600 1318
1601 if (!(flags & EIO_READDIR_DENTS)) 1319 if (!(flags & EIO_READDIR_DENTS))
1602 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); 1320 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1603 1321
1604 X_LOCK (wrklock); 1322#ifdef _WIN32
1605 /* 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
1606 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
1607 1390
1608 if (req->flags & EIO_FLAG_PTR1_FREE) 1391 if (req->flags & EIO_FLAG_PTR1_FREE)
1609 free (req->ptr1); 1392 free (req->ptr1);
1610 1393
1611 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1394 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1612 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; 1395 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1613 req->ptr2 = names = malloc (namesalloc); 1396 req->ptr2 = names = malloc (namesalloc);
1614 X_UNLOCK (wrklock);
1615 1397
1616 if (dirp && names && (!flags || dents)) 1398 if (!names || (flags && !dents))
1399 return;
1400
1617 for (;;) 1401 for (;;)
1618 { 1402 {
1403 int done;
1404
1405#ifdef _WIN32
1406 done = !dirp;
1407#else
1619 errno = 0; 1408 errno = 0;
1620 entp = readdir (dirp); 1409 entp = readdir (dirp);
1410 done = !entp;
1411#endif
1621 1412
1622 if (!entp) 1413 if (done)
1623 { 1414 {
1415#ifndef _WIN32
1416 int old_errno = errno;
1417 closedir (dirp);
1418 errno = old_errno;
1419
1624 if (errno) 1420 if (errno)
1625 break; 1421 break;
1422#endif
1626 1423
1627 /* sort etc. */ 1424 /* sort etc. */
1628 req->int1 = flags; 1425 req->int1 = flags;
1629 req->result = dentoffs; 1426 req->result = dentoffs;
1630 1427
1631 if (flags & EIO_READDIR_STAT_ORDER) 1428 if (flags & EIO_READDIR_STAT_ORDER)
1632 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);
1633 else if (flags & EIO_READDIR_DIRS_FIRST) 1430 else if (flags & EIO_READDIR_DIRS_FIRST)
1634 if (flags & EIO_READDIR_FOUND_UNKNOWN) 1431 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1635 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 */
1636 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))
1637 { 1484 {
1638 /* 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
1639 eio_dirent *oth = dents + dentoffs; 1492 ent = dents + dentoffs;
1640 eio_dirent *dir = dents;
1641 1493
1642 /* 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 */
1643 /* by walking from both sides and swapping if necessary */ 1495 ent->namelen = len - 1;
1644 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)
1645 { 1556 {
1646 if (dir->type == EIO_DT_DIR) 1557 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1647 ++dir; 1558 ent->score = 1;
1648 else if ((--oth)->type == EIO_DT_DIR) 1559 else if (!strchr (name, '.')) /* absence of dots indicate likely dirs */
1649 { 1560 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1650 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1651
1652 ++dir;
1653 }
1654 } 1561 }
1655 1562 else if (ent->type == EIO_DT_DIR)
1656 /* now sort the dirs only (dirs all have the same score) */ 1563 ent->score = 0;
1657 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1658 } 1564 }
1659
1660 break;
1661 }
1662
1663 /* now add the entry to our list(s) */
1664 name = entp->d_name;
1665
1666 /* skip . and .. entries */
1667 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1668 {
1669 int len = D_NAMLEN (entp) + 1;
1670
1671 while (ecb_expect_false (namesoffs + len > namesalloc))
1672 {
1673 namesalloc *= 2;
1674 X_LOCK (wrklock);
1675 req->ptr2 = names = realloc (names, namesalloc);
1676 X_UNLOCK (wrklock);
1677
1678 if (!names)
1679 break;
1680 } 1565 }
1681 1566
1682 memcpy (names + namesoffs, name, len);
1683
1684 if (dents)
1685 {
1686 struct eio_dirent *ent;
1687
1688 if (ecb_expect_false (dentoffs == dentalloc))
1689 {
1690 dentalloc *= 2;
1691 X_LOCK (wrklock);
1692 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1693 X_UNLOCK (wrklock);
1694
1695 if (!dents)
1696 break;
1697 }
1698
1699 ent = dents + dentoffs;
1700
1701 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1702 ent->namelen = len - 1;
1703 ent->inode = D_INO (entp);
1704
1705 inode_bits |= ent->inode;
1706
1707 switch (D_TYPE (entp))
1708 {
1709 default:
1710 ent->type = EIO_DT_UNKNOWN;
1711 flags |= EIO_READDIR_FOUND_UNKNOWN;
1712 break;
1713
1714 #ifdef DT_FIFO
1715 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1716 #endif
1717 #ifdef DT_CHR
1718 case DT_CHR: ent->type = EIO_DT_CHR; break;
1719 #endif
1720 #ifdef DT_MPC
1721 case DT_MPC: ent->type = EIO_DT_MPC; break;
1722 #endif
1723 #ifdef DT_DIR
1724 case DT_DIR: ent->type = EIO_DT_DIR; break;
1725 #endif
1726 #ifdef DT_NAM
1727 case DT_NAM: ent->type = EIO_DT_NAM; break;
1728 #endif
1729 #ifdef DT_BLK
1730 case DT_BLK: ent->type = EIO_DT_BLK; break;
1731 #endif
1732 #ifdef DT_MPB
1733 case DT_MPB: ent->type = EIO_DT_MPB; break;
1734 #endif
1735 #ifdef DT_REG
1736 case DT_REG: ent->type = EIO_DT_REG; break;
1737 #endif
1738 #ifdef DT_NWK
1739 case DT_NWK: ent->type = EIO_DT_NWK; break;
1740 #endif
1741 #ifdef DT_CMP
1742 case DT_CMP: ent->type = EIO_DT_CMP; break;
1743 #endif
1744 #ifdef DT_LNK
1745 case DT_LNK: ent->type = EIO_DT_LNK; break;
1746 #endif
1747 #ifdef DT_SOCK
1748 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1749 #endif
1750 #ifdef DT_DOOR
1751 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1752 #endif
1753 #ifdef DT_WHT
1754 case DT_WHT: ent->type = EIO_DT_WHT; break;
1755 #endif
1756 }
1757
1758 ent->score = 7;
1759
1760 if (flags & EIO_READDIR_DIRS_FIRST)
1761 {
1762 if (ent->type == EIO_DT_UNKNOWN)
1763 {
1764 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1765 ent->score = 1;
1766 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1767 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1768 }
1769 else if (ent->type == EIO_DT_DIR)
1770 ent->score = 0;
1771 }
1772 }
1773
1774 namesoffs += len; 1567 namesoffs += len;
1775 ++dentoffs; 1568 ++dentoffs;
1776 } 1569 }
1777 1570
1778 if (EIO_CANCELLED (req)) 1571 if (EIO_CANCELLED (req))
1779 { 1572 {
1780 errno = ECANCELED; 1573 errno = ECANCELED;
1781 break; 1574 break;
1782 } 1575 }
1576
1577#ifdef _WIN32
1578 if (!FindNextFile (dirp, &entp))
1579 {
1580 FindClose (dirp);
1581 dirp = 0;
1783 } 1582 }
1583#endif
1584 }
1784} 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
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
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
1659 return wd;
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
1785 1708
1786/*****************************************************************************/ 1709/*****************************************************************************/
1787 1710
1788#define ALLOC(len) \ 1711#define ALLOC(len) \
1789 if (!req->ptr2) \ 1712 if (!req->ptr2) \
1790 { \ 1713 { \
1791 X_LOCK (wrklock); \ 1714 X_LOCK (EIO_POOL->wrklock); \
1792 req->flags |= EIO_FLAG_PTR2_FREE; \ 1715 req->flags |= EIO_FLAG_PTR2_FREE; \
1793 X_UNLOCK (wrklock); \ 1716 X_UNLOCK (EIO_POOL->wrklock); \
1794 req->ptr2 = malloc (len); \ 1717 req->ptr2 = malloc (len); \
1795 if (!req->ptr2) \ 1718 if (!req->ptr2) \
1796 { \ 1719 { \
1797 errno = ENOMEM; \ 1720 errno = ENOMEM; \
1798 req->result = -1; \ 1721 req->result = -1; \
1799 break; \ 1722 break; \
1800 } \ 1723 } \
1801 } 1724 }
1802 1725
1803X_THREAD_PROC (etp_proc)
1804{
1805 ETP_REQ *req;
1806 struct timespec ts;
1807 etp_worker *self = (etp_worker *)thr_arg;
1808
1809 /* try to distribute timeouts somewhat randomly */
1810 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1811
1812 for (;;)
1813 {
1814 X_LOCK (reqlock);
1815
1816 for (;;)
1817 {
1818 self->req = req = reqq_shift (&req_queue);
1819
1820 if (req)
1821 break;
1822
1823 ++idle;
1824
1825 ts.tv_sec = time (0) + idle_timeout;
1826 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
1827 {
1828 if (idle > max_idle)
1829 {
1830 --idle;
1831 X_UNLOCK (reqlock);
1832 X_LOCK (wrklock);
1833 --started;
1834 X_UNLOCK (wrklock);
1835 goto quit;
1836 }
1837
1838 /* we are allowed to idle, so do so without any timeout */
1839 X_COND_WAIT (reqwait, reqlock);
1840 }
1841
1842 --idle;
1843 }
1844
1845 --nready;
1846
1847 X_UNLOCK (reqlock);
1848
1849 if (req->type < 0)
1850 goto quit;
1851
1852 ETP_EXECUTE (self, req);
1853
1854 X_LOCK (reslock);
1855
1856 ++npending;
1857
1858 if (!reqq_push (&res_queue, req) && want_poll_cb)
1859 want_poll_cb ();
1860
1861 self->req = 0;
1862 etp_worker_clear (self);
1863
1864 X_UNLOCK (reslock);
1865 }
1866
1867quit:
1868 X_LOCK (wrklock);
1869 etp_worker_free (self);
1870 X_UNLOCK (wrklock);
1871
1872 return 0;
1873}
1874
1875/*****************************************************************************/ 1726/*****************************************************************************/
1876 1727
1877int ecb_cold 1728int ecb_cold
1878eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1729eio_init (void (*want_poll)(void), void (*done_poll)(void))
1879{ 1730{
1880 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);
1881} 1735}
1882 1736
1883ecb_inline void 1737ecb_inline void
1884eio_api_destroy (eio_req *req) 1738eio_api_destroy (eio_req *req)
1885{ 1739{
1886 free (req); 1740 free (req);
1887} 1741}
1888 1742
1889#define REQ(rtype) \ 1743#define REQ(rtype) \
1890 eio_req *req; \ 1744 eio_req *req; \
1891 \ 1745 \
1892 req = (eio_req *)calloc (1, sizeof *req); \ 1746 req = (eio_req *)calloc (1, sizeof *req); \
1893 if (!req) \ 1747 if (!req) \
1894 return 0; \ 1748 return 0; \
… …
1908 { \ 1762 { \
1909 eio_api_destroy (req); \ 1763 eio_api_destroy (req); \
1910 return 0; \ 1764 return 0; \
1911 } 1765 }
1912 1766
1767#define SINGLEDOT(ptr) (0[(char *)(ptr)] == '.' && !1[(char *)(ptr)])
1768
1913static void 1769static void
1914eio_execute (etp_worker *self, eio_req *req) 1770eio_execute (etp_worker *self, eio_req *req)
1915{ 1771{
1772#if HAVE_AT
1773 int dirfd;
1774#else
1775 const char *path;
1776#endif
1777
1916 if (ecb_expect_false (EIO_CANCELLED (req))) 1778 if (ecb_expect_false (EIO_CANCELLED (req)))
1917 { 1779 {
1918 req->result = -1; 1780 req->result = -1;
1919 req->errorno = ECANCELED; 1781 req->errorno = ECANCELED;
1920 return; 1782 return;
1921 } 1783 }
1922 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
1923 switch (req->type) 1801 switch (req->type)
1924 { 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;
1925 case EIO_READ: ALLOC (req->size); 1810 case EIO_READ: ALLOC (req->size);
1926 req->result = req->offs >= 0 1811 req->result = req->offs >= 0
1927 ? pread (req->int1, req->ptr2, req->size, req->offs) 1812 ? pread (req->int1, req->ptr2, req->size, req->offs)
1928 : read (req->int1, req->ptr2, req->size); break; 1813 : read (req->int1, req->ptr2, req->size); break;
1929 case EIO_WRITE: req->result = req->offs >= 0 1814 case EIO_WRITE: req->result = req->offs >= 0
1930 ? pwrite (req->int1, req->ptr2, req->size, req->offs) 1815 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1931 : write (req->int1, req->ptr2, req->size); break; 1816 : write (req->int1, req->ptr2, req->size); break;
1932 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
1933 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;
1934 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
1935 1825
1936 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1826 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1937 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;
1938 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));
1939 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
1940 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1932 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1941 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; 1933 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1942 1934
1943 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1944 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1945 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS)); 1935 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1946 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break; 1936 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1947 1937
1948 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1949 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;
1950 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1951 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;
1952 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1953 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break; 1940 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1954 1941
1955 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
1956 case EIO_CLOSE: req->result = close (req->int1); break; 1942 case EIO_CLOSE: req->result = close (req->int1); break;
1957 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break; 1943 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1958 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
1959 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1960 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1961 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1962 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1963 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1964 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1965
1966 case EIO_REALPATH: eio__realpath (req, self); break;
1967
1968 case EIO_READLINK: ALLOC (PATH_MAX);
1969 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1970
1971 case EIO_SYNC: req->result = 0; sync (); break; 1944 case EIO_SYNC: req->result = 0; sync (); break;
1972 case EIO_FSYNC: req->result = fsync (req->int1); break; 1945 case EIO_FSYNC: req->result = fsync (req->int1); break;
1973 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;
1974 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;
1975 case EIO_MTOUCH: req->result = eio__mtouch (req); break; 1950 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1976 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;
1977 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break; 1952 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1978 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1979 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break; 1953 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
1980 1954
1981 case EIO_READDIR: eio__scandir (req, self); break; 1955 case EIO_READDIR: eio__scandir (req, self); break;
1982 1956
1983 case EIO_BUSY: 1957 case EIO_BUSY:
… …
1993 req->result = select (0, 0, 0, 0, &tv); 1967 req->result = select (0, 0, 0, 0, &tv);
1994 } 1968 }
1995#endif 1969#endif
1996 break; 1970 break;
1997 1971
1998 case EIO_UTIME: 1972#if 0
1999 case EIO_FUTIME:
2000 {
2001 struct timeval tv[2];
2002 struct timeval *times;
2003
2004 if (req->nv1 != -1. || req->nv2 != -1.)
2005 {
2006 tv[0].tv_sec = req->nv1;
2007 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
2008 tv[1].tv_sec = req->nv2;
2009 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
2010
2011 times = tv;
2012 }
2013 else
2014 times = 0;
2015
2016 req->result = req->type == EIO_FUTIME
2017 ? futimes (req->int1, times)
2018 : utimes (req->ptr1, times);
2019 }
2020 break;
2021
2022 case EIO_GROUP: 1973 case EIO_GROUP:
2023 abort (); /* handled in eio_request */ 1974 abort (); /* handled in eio_request */
1975#endif
2024 1976
2025 case EIO_NOP: 1977 case EIO_NOP:
2026 req->result = 0; 1978 req->result = 0;
2027 break; 1979 break;
2028 1980
2029 case EIO_CUSTOM: 1981 case EIO_CUSTOM:
2030 req->feed (req); 1982 req->feed (req);
2031 break; 1983 break;
2032 1984
2033 default: 1985 default:
2034 errno = ENOSYS;
2035 req->result = -1; 1986 req->result = EIO_ENOSYS ();
2036 break; 1987 break;
2037 } 1988 }
2038 1989
2039 req->errorno = errno; 1990 req->errorno = errno;
2040} 1991}
2041 1992
2042#ifndef EIO_NO_WRAPPERS 1993#ifndef EIO_NO_WRAPPERS
2043 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
2044eio_req *eio_nop (int pri, eio_cb cb, void *data) 2005eio_req *eio_nop (int pri, eio_cb cb, void *data)
2045{ 2006{
2046 REQ (EIO_NOP); SEND; 2007 REQ (EIO_NOP); SEND;
2047} 2008}
2048 2009
… …
2064eio_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)
2065{ 2026{
2066 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;
2067} 2028}
2068 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
2069eio_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)
2070{ 2046{
2071 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;
2072} 2048}
2073 2049
… …
2079eio_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)
2080{ 2056{
2081 REQ (EIO_MLOCKALL); req->int1 = flags; SEND; 2057 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2082} 2058}
2083 2059
2084eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2085{
2086 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2087}
2088
2089eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, 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)
2090{ 2061{
2091 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND; 2062 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
2092} 2063}
2093 2064
2094eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2095{
2096 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2097}
2098
2099eio_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)
2100{ 2066{
2101 REQ (EIO_CLOSE); req->int1 = fd; SEND; 2067 REQ (EIO_CLOSE); req->int1 = fd; SEND;
2102} 2068}
2103 2069
2104eio_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)
2105{ 2071{
2106 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;
2107} 2073}
2108 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
2109eio_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)
2110{ 2081{
2111 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;
2112} 2083}
2113 2084
2114eio_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)
2115{ 2086{
2116 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;
2117} 2088}
2118 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
2119eio_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)
2120{ 2101{
2121 REQ (EIO_FSTAT); req->int1 = fd; SEND; 2102 REQ (EIO_FSTAT); req->int1 = fd; SEND;
2122} 2103}
2123 2104
… …
2139eio_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)
2140{ 2121{
2141 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2122 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
2142} 2123}
2143 2124
2144eio_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)
2145{ 2126{
2146 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;
2147} 2128}
2148 2129
2149eio_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)
… …
2169eio_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)
2170{ 2151{
2171 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2152 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
2172} 2153}
2173 2154
2174eio_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)
2175{ 2156{
2176 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;
2177} 2158}
2178 2159
2179eio_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)
… …
2309void 2290void
2310eio_grp_add (eio_req *grp, eio_req *req) 2291eio_grp_add (eio_req *grp, eio_req *req)
2311{ 2292{
2312 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));
2313 2294
2314 grp->flags |= EIO_FLAG_GROUPADD; 2295 grp->flags |= ETP_FLAG_GROUPADD;
2315 2296
2316 ++grp->size; 2297 ++grp->size;
2317 req->grp = grp; 2298 req->grp = grp;
2318 2299
2319 req->grp_prev = 0; 2300 req->grp_prev = 0;
… …
2326} 2307}
2327 2308
2328/*****************************************************************************/ 2309/*****************************************************************************/
2329/* misc garbage */ 2310/* misc garbage */
2330 2311
2331ssize_t 2312eio_ssize_t
2332eio_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)
2333{ 2314{
2334 etp_worker wrk;
2335 ssize_t ret;
2336
2337 wrk.dbuf = 0;
2338
2339 ret = eio__sendfile (ofd, ifd, offset, count, &wrk); 2315 return eio__sendfile (ofd, ifd, offset, count);
2340
2341 if (wrk.dbuf)
2342 free (wrk.dbuf);
2343
2344 return ret;
2345} 2316}
2346 2317

Diff Legend

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