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

Comparing libeio/eio.c (file contents):
Revision 1.81 by root, Tue Jul 5 20:34:42 2011 UTC vs.
Revision 1.144 by root, Tue Dec 27 09:58:44 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
156#endif 317#endif
157 318
158/* used for readlink etc. */ 319/* used for readlink etc. */
159#ifndef PATH_MAX 320#ifndef PATH_MAX
160# define PATH_MAX 4096 321# define PATH_MAX 0
161#endif 322#endif
323
324#ifndef EIO_PATH_MIN
325# define EIO_PATH_MIN 8160
326#endif
327
328#define EIO_PATH_MAX (PATH_MAX <= EIO_PATH_MIN ? EIO_PATH_MIN : PATH_MAX)
162 329
163/* buffer size for various temporary buffers */ 330/* buffer size for various temporary buffers */
164#define EIO_BUFSIZE 65536 331#define EIO_BUFSIZE 65536
165 332
166#define dBUF \ 333#define dBUF \
167 char *eio_buf; \
168 ETP_WORKER_LOCK (self); \
169 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 334 char *eio_buf = malloc (EIO_BUFSIZE); \
170 ETP_WORKER_UNLOCK (self); \
171 errno = ENOMEM; \ 335 errno = ENOMEM; \
172 if (!eio_buf) \ 336 if (!eio_buf) \
173 return -1; 337 return -1
174 338
175#define EIO_TICKS ((1000000 + 1023) >> 10) 339#define FUBd \
340 free (eio_buf)
341
342/*****************************************************************************/
343
344struct etp_tmpbuf;
345
346#if _POSIX_VERSION >= 200809L
347 #define HAVE_AT 1
348 #define WD2FD(wd) ((wd) ? (wd)->fd : AT_FDCWD)
349 #ifndef O_SEARCH
350 #define O_SEARCH O_RDONLY
351 #endif
352#else
353 #define HAVE_AT 0
354 static const char *wd_expand (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path);
355#endif
356
357struct eio_pwd
358{
359#if HAVE_AT
360 int fd;
361#endif
362 int len;
363 char str[1]; /* actually, a 0-terminated canonical path */
364};
365
366/*****************************************************************************/
176 367
177#define ETP_PRI_MIN EIO_PRI_MIN 368#define ETP_PRI_MIN EIO_PRI_MIN
178#define ETP_PRI_MAX EIO_PRI_MAX 369#define ETP_PRI_MAX EIO_PRI_MAX
179 370
371#define ETP_TYPE_QUIT -1
372#define ETP_TYPE_GROUP EIO_GROUP
373
374static void eio_nop_callback (void) { }
375static void (*eio_want_poll_cb)(void) = eio_nop_callback;
376static void (*eio_done_poll_cb)(void) = eio_nop_callback;
377
378#define ETP_WANT_POLL(pool) eio_want_poll_cb ()
379#define ETP_DONE_POLL(pool) eio_done_poll_cb ()
380
180struct etp_worker; 381struct etp_worker;
181
182#define ETP_REQ eio_req 382#define ETP_REQ eio_req
183#define ETP_DESTROY(req) eio_destroy (req) 383#define ETP_DESTROY(req) eio_destroy (req)
184static int eio_finish (eio_req *req); 384static int eio_finish (eio_req *req);
185#define ETP_FINISH(req) eio_finish (req) 385#define ETP_FINISH(req) eio_finish (req)
186static void eio_execute (struct etp_worker *self, eio_req *req); 386static void eio_execute (struct etp_worker *self, eio_req *req);
187#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) 387#define ETP_EXECUTE(wrk,req) eio_execute (wrk, req)
188 388
189#define ETP_WORKER_CLEAR(req) \ 389#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 }
201 390
202#define ETP_WORKER_COMMON \ 391static struct etp_pool eio_pool;
203 void *dbuf; \ 392#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}
684 393
685/*****************************************************************************/ 394/*****************************************************************************/
686 395
687static void 396static void
688grp_try_feed (eio_req *grp) 397grp_try_feed (eio_req *grp)
689{ 398{
690 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 399 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
691 { 400 {
692 grp->flags &= ~EIO_FLAG_GROUPADD; 401 grp->flags &= ~ETP_FLAG_GROUPADD;
693 402
694 EIO_FEED (grp); 403 EIO_FEED (grp);
695 404
696 /* stop if no progress has been made */ 405 /* stop if no progress has been made */
697 if (!(grp->flags & EIO_FLAG_GROUPADD)) 406 if (!(grp->flags & ETP_FLAG_GROUPADD))
698 { 407 {
699 grp->feed = 0; 408 grp->feed = 0;
700 break; 409 break;
701 } 410 }
702 } 411 }
709 418
710 /* call feeder, if applicable */ 419 /* call feeder, if applicable */
711 grp_try_feed (grp); 420 grp_try_feed (grp);
712 421
713 /* finish, if done */ 422 /* finish, if done */
714 if (!grp->size && grp->int1) 423 if (!grp->size && grp->flags & ETP_FLAG_DELAYED)
715 return eio_finish (grp); 424 return eio_finish (grp);
716 else 425 else
717 return 0; 426 return 0;
718} 427}
719 428
755} 464}
756 465
757void 466void
758eio_grp_cancel (eio_req *grp) 467eio_grp_cancel (eio_req *grp)
759{ 468{
760 for (grp = grp->grp_first; grp; grp = grp->grp_next) 469 etp_grp_cancel (EIO_POOL, grp);
761 eio_cancel (grp);
762} 470}
763 471
764void 472void
765eio_cancel (eio_req *req) 473eio_cancel (eio_req *req)
766{ 474{
767 etp_cancel (req); 475 etp_cancel (EIO_POOL, req);
768} 476}
769 477
770void 478void
771eio_submit (eio_req *req) 479eio_submit (eio_req *req)
772{ 480{
773 etp_submit (req); 481 etp_submit (EIO_POOL, req);
774} 482}
775 483
776unsigned int 484unsigned int
777eio_nreqs (void) 485eio_nreqs (void)
778{ 486{
779 return etp_nreqs (); 487 return etp_nreqs (EIO_POOL);
780} 488}
781 489
782unsigned int 490unsigned int
783eio_nready (void) 491eio_nready (void)
784{ 492{
785 return etp_nready (); 493 return etp_nready (EIO_POOL);
786} 494}
787 495
788unsigned int 496unsigned int
789eio_npending (void) 497eio_npending (void)
790{ 498{
791 return etp_npending (); 499 return etp_npending (EIO_POOL);
792} 500}
793 501
794unsigned int ecb_cold 502unsigned int ecb_cold
795eio_nthreads (void) 503eio_nthreads (void)
796{ 504{
797 return etp_nthreads (); 505 return etp_nthreads (EIO_POOL);
798} 506}
799 507
800void ecb_cold 508void ecb_cold
801eio_set_max_poll_time (double nseconds) 509eio_set_max_poll_time (double nseconds)
802{ 510{
803 etp_set_max_poll_time (nseconds); 511 etp_set_max_poll_time (EIO_POOL, nseconds);
804} 512}
805 513
806void ecb_cold 514void ecb_cold
807eio_set_max_poll_reqs (unsigned int maxreqs) 515eio_set_max_poll_reqs (unsigned int maxreqs)
808{ 516{
809 etp_set_max_poll_reqs (maxreqs); 517 etp_set_max_poll_reqs (EIO_POOL, maxreqs);
810} 518}
811 519
812void ecb_cold 520void ecb_cold
813eio_set_max_idle (unsigned int nthreads) 521eio_set_max_idle (unsigned int nthreads)
814{ 522{
815 etp_set_max_idle (nthreads); 523 etp_set_max_idle (EIO_POOL, nthreads);
816} 524}
817 525
818void ecb_cold 526void ecb_cold
819eio_set_idle_timeout (unsigned int seconds) 527eio_set_idle_timeout (unsigned int seconds)
820{ 528{
821 etp_set_idle_timeout (seconds); 529 etp_set_idle_timeout (EIO_POOL, seconds);
822} 530}
823 531
824void ecb_cold 532void ecb_cold
825eio_set_min_parallel (unsigned int nthreads) 533eio_set_min_parallel (unsigned int nthreads)
826{ 534{
827 etp_set_min_parallel (nthreads); 535 etp_set_min_parallel (EIO_POOL, nthreads);
828} 536}
829 537
830void ecb_cold 538void ecb_cold
831eio_set_max_parallel (unsigned int nthreads) 539eio_set_max_parallel (unsigned int nthreads)
832{ 540{
833 etp_set_max_parallel (nthreads); 541 etp_set_max_parallel (EIO_POOL, nthreads);
834} 542}
835 543
836int eio_poll (void) 544int eio_poll (void)
837{ 545{
838 return etp_poll (); 546 return etp_poll (EIO_POOL);
839} 547}
840 548
841/*****************************************************************************/ 549/*****************************************************************************/
842/* work around various missing functions */ 550/* 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
882 551
883#ifndef HAVE_UTIMES 552#ifndef HAVE_UTIMES
884 553
885# undef utimes 554# undef utimes
886# define utimes(path,times) eio__utimes (path, times) 555# define utimes(path,times) eio__utimes (path, times)
919 588
920#if !HAVE_FDATASYNC 589#if !HAVE_FDATASYNC
921# undef fdatasync 590# undef fdatasync
922# define fdatasync(fd) fsync (fd) 591# define fdatasync(fd) fsync (fd)
923#endif 592#endif
593
594static int
595eio__syncfs (int fd)
596{
597 int res;
598
599#if HAVE_SYS_SYNCFS
600 res = (int)syscall (__NR_syncfs, (int)(fd));
601#else
602 res = EIO_ENOSYS ();
603#endif
604
605 if (res < 0 && errno == ENOSYS && fd >= 0)
606 sync ();
607
608 return res;
609}
924 610
925/* sync_file_range always needs emulation */ 611/* sync_file_range always needs emulation */
926static int 612static int
927eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags) 613eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
928{ 614{
948 /* even though we could play tricks with the flags, it's better to always 634 /* even though we could play tricks with the flags, it's better to always
949 * call fdatasync, as that matches the expectation of its users best */ 635 * call fdatasync, as that matches the expectation of its users best */
950 return fdatasync (fd); 636 return fdatasync (fd);
951} 637}
952 638
639static int
640eio__fallocate (int fd, int mode, off_t offset, size_t len)
641{
642#if HAVE_LINUX_FALLOCATE
643 return fallocate (fd, mode, offset, len);
644#else
645 return EIO_ENOSYS ();
646#endif
647}
648
953#if !HAVE_READAHEAD 649#if !HAVE_READAHEAD
954# undef readahead 650# undef readahead
955# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 651# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
956 652
957static ssize_t 653static eio_ssize_t
958eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) 654eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
959{ 655{
960 size_t todo = count; 656 size_t todo = count;
961 dBUF; 657 dBUF;
962 658
967 pread (fd, eio_buf, len, offset); 663 pread (fd, eio_buf, len, offset);
968 offset += len; 664 offset += len;
969 todo -= len; 665 todo -= len;
970 } 666 }
971 667
972 errno = 0; 668 FUBd;
669
670 /* linux's readahead basically only fails for EBADF or EINVAL (not mmappable) */
671 /* but not for e.g. EIO or eof, so we also never fail */
973 return count; 672 return 0;
974} 673}
975 674
976#endif 675#endif
977 676
978/* sendfile always needs emulation */ 677/* sendfile always needs emulation */
979static ssize_t 678static eio_ssize_t
980eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) 679eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
981{ 680{
982 ssize_t written = 0; 681 eio_ssize_t written = 0;
983 ssize_t res; 682 eio_ssize_t res;
984 683
985 if (!count) 684 if (!count)
986 return 0; 685 return 0;
987 686
988 for (;;) 687 for (;;)
989 { 688 {
689#ifdef __APPLE__
690# undef HAVE_SENDFILE /* broken, as everything on os x */
691#endif
990#if HAVE_SENDFILE 692#if HAVE_SENDFILE
991# if __linux 693# if __linux
992 off_t soffset = offset; 694 off_t soffset = offset;
993 res = sendfile (ofd, ifd, &soffset, count); 695 res = sendfile (ofd, ifd, &soffset, count);
994 696
1010 712
1011 /* according to source inspection, this is correct, and useful behaviour */ 713 /* according to source inspection, this is correct, and useful behaviour */
1012 if (sbytes) 714 if (sbytes)
1013 res = sbytes; 715 res = sbytes;
1014 716
1015# elif defined (__APPLE__) && 0 /* broken, as everything on os x */ 717# elif defined __APPLE__
1016 off_t sbytes = count; 718 off_t sbytes = count;
1017 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0); 719 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
1018 720
1019 /* according to the manpage, sbytes is always valid */ 721 /* according to the manpage, sbytes is always valid */
1020 if (sbytes) 722 if (sbytes)
1037 if (res < 0 && sbytes) 739 if (res < 0 && sbytes)
1038 res = sbytes; 740 res = sbytes;
1039 741
1040# endif 742# endif
1041 743
1042#elif defined (_WIN32) 744#elif defined (_WIN32) && 0
1043 /* does not work, just for documentation of what would need to be done */ 745 /* does not work, just for documentation of what would need to be done */
1044 /* actually, cannot be done like this, as TransmitFile changes the file offset, */ 746 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
1045 /* libeio guarantees that the file offset does not change, and windows */ 747 /* libeio guarantees that the file offset does not change, and windows */
1046 /* has no way to get an independent handle to the same file description */ 748 /* has no way to get an independent handle to the same file description */
1047 HANDLE h = TO_SOCKET (ifd); 749 HANDLE h = TO_SOCKET (ifd);
1048 SetFilePointer (h, offset, 0, FILE_BEGIN); 750 SetFilePointer (h, offset, 0, FILE_BEGIN);
1049 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0); 751 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1050 752
1051#else 753#else
1052 res = -1; 754 res = EIO_ENOSYS ();
1053 errno = ENOSYS;
1054#endif 755#endif
1055 756
1056 /* we assume sendfile can copy at least 128mb in one go */ 757 /* we assume sendfile can copy at least 128mb in one go */
1057 if (res <= 128 * 1024 * 1024) 758 if (res <= 128 * 1024 * 1024)
1058 { 759 {
1080 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 781 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1081 /* BSDs */ 782 /* BSDs */
1082#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */ 783#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1083 || errno == ENOTSUP 784 || errno == ENOTSUP
1084#endif 785#endif
786#ifdef EOPNOTSUPP /* windows */
1085 || errno == EOPNOTSUPP /* BSDs */ 787 || errno == EOPNOTSUPP /* BSDs */
788#endif
1086#if __solaris 789#if __solaris
1087 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 790 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
1088#endif 791#endif
1089 ) 792 )
1090 ) 793 )
1094 797
1095 res = 0; 798 res = 0;
1096 799
1097 while (count) 800 while (count)
1098 { 801 {
1099 ssize_t cnt; 802 eio_ssize_t cnt;
1100 803
1101 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 804 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
1102 805
1103 if (cnt <= 0) 806 if (cnt <= 0)
1104 { 807 {
1116 819
1117 offset += cnt; 820 offset += cnt;
1118 res += cnt; 821 res += cnt;
1119 count -= cnt; 822 count -= cnt;
1120 } 823 }
824
825 FUBd;
1121 } 826 }
1122 827
1123 return res; 828 return res;
1124} 829}
1125 830
1152 /* round up length */ 857 /* round up length */
1153 *length = (*length + mask) & ~mask; 858 *length = (*length + mask) & ~mask;
1154} 859}
1155 860
1156#if !_POSIX_MEMLOCK 861#if !_POSIX_MEMLOCK
1157# define eio__mlockall(a) ((errno = ENOSYS), -1) 862# define eio__mlockall(a) EIO_ENOSYS ()
1158#else 863#else
1159 864
1160static int 865static int
1161eio__mlockall (int flags) 866eio__mlockall (int flags)
1162{ 867{
1176 return mlockall (flags); 881 return mlockall (flags);
1177} 882}
1178#endif 883#endif
1179 884
1180#if !_POSIX_MEMLOCK_RANGE 885#if !_POSIX_MEMLOCK_RANGE
1181# define eio__mlock(a,b) ((errno = ENOSYS), -1) 886# define eio__mlock(a,b) EIO_ENOSYS ()
1182#else 887#else
1183 888
1184static int 889static int
1185eio__mlock (void *addr, size_t length) 890eio__mlock (void *addr, size_t length)
1186{ 891{
1190} 895}
1191 896
1192#endif 897#endif
1193 898
1194#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) 899#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1195# define eio__msync(a,b,c) ((errno = ENOSYS), -1) 900# define eio__msync(a,b,c) EIO_ENOSYS ()
1196#else 901#else
1197 902
1198static int 903static int
1199eio__msync (void *mem, size_t len, int flags) 904eio__msync (void *mem, size_t len, int flags)
1200{ 905{
1241 946
1242/*****************************************************************************/ 947/*****************************************************************************/
1243/* requests implemented outside eio_execute, because they are so large */ 948/* requests implemented outside eio_execute, because they are so large */
1244 949
1245static void 950static void
1246eio__realpath (eio_req *req, etp_worker *self) 951eio__lseek (eio_req *req)
1247{ 952{
1248 char *rel = req->ptr1; 953 /* this usually gets optimised away completely, or your compiler sucks, */
954 /* or the whence constants really are not 0, 1, 2 */
955 int whence = req->int2 == EIO_SEEK_SET ? SEEK_SET
956 : req->int2 == EIO_SEEK_CUR ? SEEK_CUR
957 : req->int2 == EIO_SEEK_END ? SEEK_END
958 : req->int2;
959
960 req->offs = lseek (req->int1, req->offs, whence);
961 req->result = req->offs == (off_t)-1 ? -1 : 0;
962}
963
964/* result will always end up in tmpbuf, there is always space for adding a 0-byte */
965static int
966eio__realpath (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
967{
1249 char *res; 968 char *res;
969 const char *rel = path;
1250 char *tmp1, *tmp2; 970 char *tmp1, *tmp2;
1251#if SYMLOOP_MAX > 32 971#if SYMLOOP_MAX > 32
1252 int symlinks = SYMLOOP_MAX; 972 int symlinks = SYMLOOP_MAX;
1253#else 973#else
1254 int symlinks = 32; 974 int symlinks = 32;
1255#endif 975#endif
1256 976
1257 req->result = -1;
1258
1259 errno = EINVAL; 977 errno = EINVAL;
1260 if (!rel) 978 if (!rel)
1261 return; 979 return -1;
1262 980
1263 errno = ENOENT; 981 errno = ENOENT;
1264 if (!*rel) 982 if (!*rel)
1265 return; 983 return -1;
1266 984
1267 if (!req->ptr2) 985 res = etp_tmpbuf_get (tmpbuf, EIO_PATH_MAX * 3);
1268 { 986#ifdef _WIN32
1269 X_LOCK (wrklock); 987 if (_access (rel, 4) != 0)
1270 req->flags |= EIO_FLAG_PTR2_FREE;
1271 X_UNLOCK (wrklock);
1272 req->ptr2 = malloc (PATH_MAX * 3);
1273
1274 errno = ENOMEM;
1275 if (!req->ptr2)
1276 return; 988 return -1;
1277 }
1278 989
1279 res = req->ptr2; 990 symlinks = GetFullPathName (rel, EIO_PATH_MAX * 3, res, 0);
991
992 errno = ENAMETOOLONG;
993 if (symlinks >= EIO_PATH_MAX * 3)
994 return -1;
995
996 errno = EIO;
997 if (symlinks <= 0)
998 return -1;
999
1000 return symlinks;
1001
1002#else
1280 tmp1 = res + PATH_MAX; 1003 tmp1 = res + EIO_PATH_MAX;
1281 tmp2 = tmp1 + PATH_MAX; 1004 tmp2 = tmp1 + EIO_PATH_MAX;
1282 1005
1283#if 0 /* disabled, the musl way to do things is just too racy */ 1006#if 0 /* disabled, the musl way to do things is just too racy */
1284#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME) 1007#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1285 /* on linux we may be able to ask the kernel */ 1008 /* on linux we may be able to ask the kernel */
1286 { 1009 {
1287 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME); 1010 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1288 1011
1289 if (fd >= 0) 1012 if (fd >= 0)
1290 { 1013 {
1291 sprintf (tmp1, "/proc/self/fd/%d", fd); 1014 sprintf (tmp1, "/proc/self/fd/%d", fd);
1292 req->result = readlink (tmp1, res, PATH_MAX); 1015 req->result = readlink (tmp1, res, EIO_PATH_MAX);
1016 /* here we should probably stat the open file and the disk file, to make sure they still match */
1293 close (fd); 1017 close (fd);
1294
1295 /* here we should probably stat the open file and the disk file, to make sure they still match */
1296 1018
1297 if (req->result > 0) 1019 if (req->result > 0)
1298 goto done; 1020 goto done;
1299 } 1021 }
1300 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO) 1022 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1301 return; 1023 return -1;
1302 } 1024 }
1303#endif 1025#endif
1304#endif 1026#endif
1305 1027
1306 if (*rel != '/') 1028 if (*rel != '/')
1307 { 1029 {
1308 if (!getcwd (res, PATH_MAX)) 1030 int len;
1031
1032 errno = ENOENT;
1033 if (wd == EIO_INVALID_WD)
1309 return; 1034 return -1;
1035
1036 if (wd == EIO_CWD)
1037 {
1038 if (!getcwd (res, EIO_PATH_MAX))
1039 return -1;
1040
1041 len = strlen (res);
1042 }
1043 else
1044 memcpy (res, wd->str, len = wd->len);
1310 1045
1311 if (res [1]) /* only use if not / */ 1046 if (res [1]) /* only use if not / */
1312 res += strlen (res); 1047 res += len;
1313 } 1048 }
1314 1049
1315 while (*rel) 1050 while (*rel)
1316 { 1051 {
1317 ssize_t len, linklen; 1052 eio_ssize_t len, linklen;
1318 char *beg = rel; 1053 const char *beg = rel;
1319 1054
1320 while (*rel && *rel != '/') 1055 while (*rel && *rel != '/')
1321 ++rel; 1056 ++rel;
1322 1057
1323 len = rel - beg; 1058 len = rel - beg;
1335 1070
1336 if (beg [1] == '.' && len == 2) 1071 if (beg [1] == '.' && len == 2)
1337 { 1072 {
1338 /* .. - back up one component, if possible */ 1073 /* .. - back up one component, if possible */
1339 1074
1340 while (res != req->ptr2) 1075 while (res != tmpbuf->ptr)
1341 if (*--res == '/') 1076 if (*--res == '/')
1342 break; 1077 break;
1343 1078
1344 continue; 1079 continue;
1345 } 1080 }
1346 } 1081 }
1347 1082
1348 errno = ENAMETOOLONG; 1083 errno = ENAMETOOLONG;
1349 if (res + 1 + len + 1 >= tmp1) 1084 if (res + 1 + len + 1 >= tmp1)
1350 return; 1085 return -1;
1351 1086
1352 /* copy one component */ 1087 /* copy one component */
1353 *res = '/'; 1088 *res = '/';
1354 memcpy (res + 1, beg, len); 1089 memcpy (res + 1, beg, len);
1355 1090
1356 /* zero-terminate, for readlink */ 1091 /* zero-terminate, for readlink */
1357 res [len + 1] = 0; 1092 res [len + 1] = 0;
1358 1093
1359 /* now check if it's a symlink */ 1094 /* now check if it's a symlink */
1360 linklen = readlink (req->ptr2, tmp1, PATH_MAX); 1095 linklen = readlink (tmpbuf->ptr, tmp1, EIO_PATH_MAX);
1361 1096
1362 if (linklen < 0) 1097 if (linklen < 0)
1363 { 1098 {
1364 if (errno != EINVAL) 1099 if (errno != EINVAL)
1365 return; 1100 return -1;
1366 1101
1367 /* it's a normal directory. hopefully */ 1102 /* it's a normal directory. hopefully */
1368 res += len + 1; 1103 res += len + 1;
1369 } 1104 }
1370 else 1105 else
1371 { 1106 {
1372 /* yay, it was a symlink - build new path in tmp2 */ 1107 /* yay, it was a symlink - build new path in tmp2 */
1373 int rellen = strlen (rel); 1108 int rellen = strlen (rel);
1374 1109
1375 errno = ENAMETOOLONG; 1110 errno = ENAMETOOLONG;
1376 if (linklen + 1 + rellen >= PATH_MAX) 1111 if (linklen + 1 + rellen >= EIO_PATH_MAX) /* also catch linklen >= EIO_PATH_MAX */
1377 return; 1112 return -1;
1378 1113
1379 errno = ELOOP; 1114 errno = ELOOP;
1380 if (!--symlinks) 1115 if (!--symlinks)
1381 return; 1116 return -1;
1382 1117
1383 if (*tmp1 == '/') 1118 if (*tmp1 == '/')
1384 res = req->ptr2; /* symlink resolves to an absolute path */ 1119 res = tmpbuf->ptr; /* symlink resolves to an absolute path */
1385 1120
1386 /* we need to be careful, as rel might point into tmp2 already */ 1121 /* we need to be careful, as rel might point into tmp2 already */
1387 memmove (tmp2 + linklen + 1, rel, rellen + 1); 1122 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1388 tmp2 [linklen] = '/'; 1123 tmp2 [linklen] = '/';
1389 memcpy (tmp2, tmp1, linklen); 1124 memcpy (tmp2, tmp1, linklen);
1391 rel = tmp2; 1126 rel = tmp2;
1392 } 1127 }
1393 } 1128 }
1394 1129
1395 /* special case for the lone root path */ 1130 /* special case for the lone root path */
1396 if (res == req->ptr2) 1131 if (res == tmpbuf->ptr)
1397 *res++ = '/'; 1132 *res++ = '/';
1398 1133
1399 req->result = res - (char *)req->ptr2; 1134 return res - (char *)tmpbuf->ptr;
1400 1135#endif
1401done:
1402 req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
1403} 1136}
1404 1137
1405static signed char 1138static signed char
1406eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) 1139eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1407{ 1140{
1415 1148
1416#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */ 1149#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1417#define EIO_SORT_FAST 60 /* when to only use insertion sort */ 1150#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1418 1151
1419static void 1152static void
1420eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1153eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1421{ 1154{
1422 unsigned char bits [9 + sizeof (ino_t) * 8]; 1155 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1423 unsigned char *bit = bits; 1156 unsigned char *bit = bits;
1424 1157
1425 assert (CHAR_BIT == 8); 1158 assert (CHAR_BIT == 8);
1426 assert (sizeof (eio_dirent) * 8 < 256); 1159 assert (sizeof (eio_dirent) * 8 < 256);
1427 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */ 1160 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1429 1162
1430 if (size <= EIO_SORT_FAST) 1163 if (size <= EIO_SORT_FAST)
1431 return; 1164 return;
1432 1165
1433 /* first prepare an array of bits to test in our radix sort */ 1166 /* first prepare an array of bits to test in our radix sort */
1434 /* try to take endianness into account, as well as differences in ino_t sizes */ 1167 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1435 /* inode_bits must contain all inodes ORed together */ 1168 /* inode_bits must contain all inodes ORed together */
1436 /* which is used to skip bits that are 0 everywhere, which is very common */ 1169 /* which is used to skip bits that are 0 everywhere, which is very common */
1437 { 1170 {
1438 ino_t endianness; 1171 eio_ino_t endianness;
1439 int i, j; 1172 int i, j;
1440 1173
1441 /* we store the byte offset of byte n into byte n of "endianness" */ 1174 /* we store the byte offset of byte n into byte n of "endianness" */
1442 for (i = 0; i < sizeof (ino_t); ++i) 1175 for (i = 0; i < sizeof (eio_ino_t); ++i)
1443 ((unsigned char *)&endianness)[i] = i; 1176 ((unsigned char *)&endianness)[i] = i;
1444 1177
1445 *bit++ = 0; 1178 *bit++ = 0;
1446 1179
1447 for (i = 0; i < sizeof (ino_t); ++i) 1180 for (i = 0; i < sizeof (eio_ino_t); ++i)
1448 { 1181 {
1449 /* shifting off the byte offsets out of "endianness" */ 1182 /* shifting off the byte offsets out of "endianness" */
1450 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8; 1183 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1451 endianness >>= 8; 1184 endianness >>= 8;
1452 1185
1453 for (j = 0; j < 8; ++j) 1186 for (j = 0; j < 8; ++j)
1454 if (inode_bits & (((ino_t)1) << (i * 8 + j))) 1187 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1455 *bit++ = offs + j; 1188 *bit++ = offs + j;
1456 } 1189 }
1457 1190
1458 for (j = 0; j < 8; ++j) 1191 for (j = 0; j < 8; ++j)
1459 if (score_bits & (1 << j)) 1192 if (score_bits & (1 << j))
1460 *bit++ = offsetof (eio_dirent, score) * 8 + j; 1193 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1461 } 1194 }
1462 1195
1463 /* now actually do the sorting (a variant of MSD radix sort) */ 1196 /* now actually do the sorting (a variant of MSD radix sort) */
1464 { 1197 {
1465 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base; 1198 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1466 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end; 1199 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1467 unsigned char *bit_stk [9 + sizeof (ino_t) * 8]; 1200 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1468 int stk_idx = 0; 1201 int stk_idx = 0;
1469 1202
1470 base_stk [stk_idx] = dents; 1203 base_stk [stk_idx] = dents;
1471 end_stk [stk_idx] = dents + size; 1204 end_stk [stk_idx] = dents + size;
1472 bit_stk [stk_idx] = bit - 1; 1205 bit_stk [stk_idx] = bit - 1;
1551 } 1284 }
1552 } 1285 }
1553} 1286}
1554 1287
1555static void 1288static void
1556eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1289eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1557{ 1290{
1558 if (size <= 1) 1291 if (size <= 1)
1559 return; /* our insertion sort relies on size > 0 */ 1292 return; /* our insertion sort relies on size > 0 */
1560 1293
1561 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */ 1294 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1569 1302
1570/* read a full directory */ 1303/* read a full directory */
1571static void 1304static void
1572eio__scandir (eio_req *req, etp_worker *self) 1305eio__scandir (eio_req *req, etp_worker *self)
1573{ 1306{
1574 DIR *dirp;
1575 EIO_STRUCT_DIRENT *entp;
1576 char *name, *names; 1307 char *name, *names;
1577 int namesalloc = 4096; 1308 int namesalloc = 4096 - sizeof (void *) * 4;
1578 int namesoffs = 0; 1309 int namesoffs = 0;
1579 int flags = req->int1; 1310 int flags = req->int1;
1580 eio_dirent *dents = 0; 1311 eio_dirent *dents = 0;
1581 int dentalloc = 128; 1312 int dentalloc = 128;
1582 int dentoffs = 0; 1313 int dentoffs = 0;
1583 ino_t inode_bits = 0; 1314 eio_ino_t inode_bits = 0;
1315#ifdef _WIN32
1316 HANDLE dirp;
1317 WIN32_FIND_DATA entp;
1318#else
1319 DIR *dirp;
1320 EIO_STRUCT_DIRENT *entp;
1321#endif
1584 1322
1585 req->result = -1; 1323 req->result = -1;
1586 1324
1587 if (!(flags & EIO_READDIR_DENTS)) 1325 if (!(flags & EIO_READDIR_DENTS))
1588 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); 1326 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1589 1327
1590 X_LOCK (wrklock); 1328#ifdef _WIN32
1591 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 1329 {
1330 int len = strlen ((const char *)req->ptr1);
1331 char *path = malloc (MAX_PATH);
1332 const char *fmt;
1333 const char *reqpath = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1334
1335 if (!len)
1336 fmt = "./*";
1337 else if (reqpath[len - 1] == '/' || reqpath[len - 1] == '\\')
1338 fmt = "%s*";
1339 else
1340 fmt = "%s/*";
1341
1342 _snprintf (path, MAX_PATH, fmt, reqpath);
1343 dirp = FindFirstFile (path, &entp);
1344 free (path);
1345
1346 if (dirp == INVALID_HANDLE_VALUE)
1347 {
1348 /* should steal _dosmaperr */
1349 switch (GetLastError ())
1350 {
1351 case ERROR_FILE_NOT_FOUND:
1352 req->result = 0;
1353 break;
1354
1355 case ERROR_INVALID_NAME:
1356 case ERROR_PATH_NOT_FOUND:
1357 case ERROR_NO_MORE_FILES:
1358 errno = ENOENT;
1359 break;
1360
1361 case ERROR_NOT_ENOUGH_MEMORY:
1362 errno = ENOMEM;
1363 break;
1364
1365 default:
1366 errno = EINVAL;
1367 break;
1368 }
1369
1370 return;
1371 }
1372 }
1373#else
1374 #if HAVE_AT
1375 if (req->wd)
1376 {
1377 int fd = openat (WD2FD (req->wd), req->ptr1, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1378
1379 if (fd < 0)
1380 return;
1381
1382 dirp = fdopendir (fd);
1383
1384 if (!dirp)
1385 close (fd);
1386 }
1387 else
1592 self->dirp = dirp = opendir (req->ptr1); 1388 dirp = opendir (req->ptr1);
1389 #else
1390 dirp = opendir (wd_expand (&self->tmpbuf, req->wd, req->ptr1));
1391 #endif
1392
1393 if (!dirp)
1394 return;
1395#endif
1593 1396
1594 if (req->flags & EIO_FLAG_PTR1_FREE) 1397 if (req->flags & EIO_FLAG_PTR1_FREE)
1595 free (req->ptr1); 1398 free (req->ptr1);
1596 1399
1597 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1400 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1598 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; 1401 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1599 req->ptr2 = names = malloc (namesalloc); 1402 req->ptr2 = names = malloc (namesalloc);
1600 X_UNLOCK (wrklock);
1601 1403
1602 if (dirp && names && (!flags || dents)) 1404 if (!names || (flags && !dents))
1405 return;
1406
1603 for (;;) 1407 for (;;)
1604 { 1408 {
1409 int done;
1410
1411#ifdef _WIN32
1412 done = !dirp;
1413#else
1605 errno = 0; 1414 errno = 0;
1606 entp = readdir (dirp); 1415 entp = readdir (dirp);
1416 done = !entp;
1417#endif
1607 1418
1608 if (!entp) 1419 if (done)
1609 { 1420 {
1421#ifndef _WIN32
1422 int old_errno = errno;
1423 closedir (dirp);
1424 errno = old_errno;
1425
1610 if (errno) 1426 if (errno)
1611 break; 1427 break;
1428#endif
1612 1429
1613 /* sort etc. */ 1430 /* sort etc. */
1614 req->int1 = flags; 1431 req->int1 = flags;
1615 req->result = dentoffs; 1432 req->result = dentoffs;
1616 1433
1617 if (flags & EIO_READDIR_STAT_ORDER) 1434 if (flags & EIO_READDIR_STAT_ORDER)
1618 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits); 1435 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1619 else if (flags & EIO_READDIR_DIRS_FIRST) 1436 else if (flags & EIO_READDIR_DIRS_FIRST)
1620 if (flags & EIO_READDIR_FOUND_UNKNOWN) 1437 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1621 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */ 1438 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1622 else 1439 else
1440 {
1441 /* in this case, all is known, and we just put dirs first and sort them */
1442 eio_dirent *oth = dents + dentoffs;
1443 eio_dirent *dir = dents;
1444
1445 /* now partition dirs to the front, and non-dirs to the back */
1446 /* by walking from both sides and swapping if necessary */
1447 while (oth > dir)
1448 {
1449 if (dir->type == EIO_DT_DIR)
1450 ++dir;
1451 else if ((--oth)->type == EIO_DT_DIR)
1452 {
1453 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1454
1455 ++dir;
1456 }
1457 }
1458
1459 /* now sort the dirs only (dirs all have the same score) */
1460 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1461 }
1462
1463 break;
1464 }
1465
1466 /* now add the entry to our list(s) */
1467 name = D_NAME (entp);
1468
1469 /* skip . and .. entries */
1470 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1471 {
1472 int len = D_NAMLEN (entp) + 1;
1473
1474 while (ecb_expect_false (namesoffs + len > namesalloc))
1475 {
1476 namesalloc *= 2;
1477 req->ptr2 = names = realloc (names, namesalloc);
1478
1479 if (!names)
1480 break;
1481 }
1482
1483 memcpy (names + namesoffs, name, len);
1484
1485 if (dents)
1486 {
1487 struct eio_dirent *ent;
1488
1489 if (ecb_expect_false (dentoffs == dentalloc))
1623 { 1490 {
1624 /* in this case, all is known, and we just put dirs first and sort them */ 1491 dentalloc *= 2;
1492 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1493
1494 if (!dents)
1495 break;
1496 }
1497
1625 eio_dirent *oth = dents + dentoffs; 1498 ent = dents + dentoffs;
1626 eio_dirent *dir = dents;
1627 1499
1628 /* now partition dirs to the front, and non-dirs to the back */ 1500 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1629 /* by walking from both sides and swapping if necessary */ 1501 ent->namelen = len - 1;
1630 while (oth > dir) 1502 ent->inode = D_INO (entp);
1503
1504 inode_bits |= ent->inode;
1505
1506 switch (D_TYPE (entp))
1507 {
1508 default:
1509 ent->type = EIO_DT_UNKNOWN;
1510 flags |= EIO_READDIR_FOUND_UNKNOWN;
1511 break;
1512
1513 #ifdef DT_FIFO
1514 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1515 #endif
1516 #ifdef DT_CHR
1517 case DT_CHR: ent->type = EIO_DT_CHR; break;
1518 #endif
1519 #ifdef DT_MPC
1520 case DT_MPC: ent->type = EIO_DT_MPC; break;
1521 #endif
1522 #ifdef DT_DIR
1523 case DT_DIR: ent->type = EIO_DT_DIR; break;
1524 #endif
1525 #ifdef DT_NAM
1526 case DT_NAM: ent->type = EIO_DT_NAM; break;
1527 #endif
1528 #ifdef DT_BLK
1529 case DT_BLK: ent->type = EIO_DT_BLK; break;
1530 #endif
1531 #ifdef DT_MPB
1532 case DT_MPB: ent->type = EIO_DT_MPB; break;
1533 #endif
1534 #ifdef DT_REG
1535 case DT_REG: ent->type = EIO_DT_REG; break;
1536 #endif
1537 #ifdef DT_NWK
1538 case DT_NWK: ent->type = EIO_DT_NWK; break;
1539 #endif
1540 #ifdef DT_CMP
1541 case DT_CMP: ent->type = EIO_DT_CMP; break;
1542 #endif
1543 #ifdef DT_LNK
1544 case DT_LNK: ent->type = EIO_DT_LNK; break;
1545 #endif
1546 #ifdef DT_SOCK
1547 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1548 #endif
1549 #ifdef DT_DOOR
1550 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1551 #endif
1552 #ifdef DT_WHT
1553 case DT_WHT: ent->type = EIO_DT_WHT; break;
1554 #endif
1555 }
1556
1557 ent->score = 7;
1558
1559 if (flags & EIO_READDIR_DIRS_FIRST)
1560 {
1561 if (ent->type == EIO_DT_UNKNOWN)
1631 { 1562 {
1632 if (dir->type == EIO_DT_DIR) 1563 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1633 ++dir; 1564 ent->score = 1;
1634 else if ((--oth)->type == EIO_DT_DIR) 1565 else if (!strchr (name, '.')) /* absence of dots indicate likely dirs */
1635 { 1566 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1636 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1637
1638 ++dir;
1639 }
1640 } 1567 }
1641 1568 else if (ent->type == EIO_DT_DIR)
1642 /* now sort the dirs only (dirs all have the same score) */ 1569 ent->score = 0;
1643 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1644 } 1570 }
1645
1646 break;
1647 }
1648
1649 /* now add the entry to our list(s) */
1650 name = entp->d_name;
1651
1652 /* skip . and .. entries */
1653 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1654 {
1655 int len = D_NAMLEN (entp) + 1;
1656
1657 while (ecb_expect_false (namesoffs + len > namesalloc))
1658 {
1659 namesalloc *= 2;
1660 X_LOCK (wrklock);
1661 req->ptr2 = names = realloc (names, namesalloc);
1662 X_UNLOCK (wrklock);
1663
1664 if (!names)
1665 break;
1666 } 1571 }
1667 1572
1668 memcpy (names + namesoffs, name, len);
1669
1670 if (dents)
1671 {
1672 struct eio_dirent *ent;
1673
1674 if (ecb_expect_false (dentoffs == dentalloc))
1675 {
1676 dentalloc *= 2;
1677 X_LOCK (wrklock);
1678 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1679 X_UNLOCK (wrklock);
1680
1681 if (!dents)
1682 break;
1683 }
1684
1685 ent = dents + dentoffs;
1686
1687 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1688 ent->namelen = len - 1;
1689 ent->inode = D_INO (entp);
1690
1691 inode_bits |= ent->inode;
1692
1693 switch (D_TYPE (entp))
1694 {
1695 default:
1696 ent->type = EIO_DT_UNKNOWN;
1697 flags |= EIO_READDIR_FOUND_UNKNOWN;
1698 break;
1699
1700 #ifdef DT_FIFO
1701 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1702 #endif
1703 #ifdef DT_CHR
1704 case DT_CHR: ent->type = EIO_DT_CHR; break;
1705 #endif
1706 #ifdef DT_MPC
1707 case DT_MPC: ent->type = EIO_DT_MPC; break;
1708 #endif
1709 #ifdef DT_DIR
1710 case DT_DIR: ent->type = EIO_DT_DIR; break;
1711 #endif
1712 #ifdef DT_NAM
1713 case DT_NAM: ent->type = EIO_DT_NAM; break;
1714 #endif
1715 #ifdef DT_BLK
1716 case DT_BLK: ent->type = EIO_DT_BLK; break;
1717 #endif
1718 #ifdef DT_MPB
1719 case DT_MPB: ent->type = EIO_DT_MPB; break;
1720 #endif
1721 #ifdef DT_REG
1722 case DT_REG: ent->type = EIO_DT_REG; break;
1723 #endif
1724 #ifdef DT_NWK
1725 case DT_NWK: ent->type = EIO_DT_NWK; break;
1726 #endif
1727 #ifdef DT_CMP
1728 case DT_CMP: ent->type = EIO_DT_CMP; break;
1729 #endif
1730 #ifdef DT_LNK
1731 case DT_LNK: ent->type = EIO_DT_LNK; break;
1732 #endif
1733 #ifdef DT_SOCK
1734 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1735 #endif
1736 #ifdef DT_DOOR
1737 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1738 #endif
1739 #ifdef DT_WHT
1740 case DT_WHT: ent->type = EIO_DT_WHT; break;
1741 #endif
1742 }
1743
1744 ent->score = 7;
1745
1746 if (flags & EIO_READDIR_DIRS_FIRST)
1747 {
1748 if (ent->type == EIO_DT_UNKNOWN)
1749 {
1750 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1751 ent->score = 1;
1752 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1753 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1754 }
1755 else if (ent->type == EIO_DT_DIR)
1756 ent->score = 0;
1757 }
1758 }
1759
1760 namesoffs += len; 1573 namesoffs += len;
1761 ++dentoffs; 1574 ++dentoffs;
1762 } 1575 }
1763 1576
1764 if (EIO_CANCELLED (req)) 1577 if (EIO_CANCELLED (req))
1765 { 1578 {
1766 errno = ECANCELED; 1579 errno = ECANCELED;
1767 break; 1580 break;
1768 } 1581 }
1582
1583#ifdef _WIN32
1584 if (!FindNextFile (dirp, &entp))
1585 {
1586 FindClose (dirp);
1587 dirp = 0;
1769 } 1588 }
1589#endif
1590 }
1770} 1591}
1592
1593/*****************************************************************************/
1594/* working directory stuff */
1595/* various deficiencies in the posix 2008 api force us to */
1596/* keep the absolute path in string form at all times */
1597/* fuck yeah. */
1598
1599#if !HAVE_AT
1600
1601/* a bit like realpath, but usually faster because it doesn'T have to return */
1602/* an absolute or canonical path */
1603static const char *
1604wd_expand (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
1605{
1606 if (!wd || *path == '/')
1607 return path;
1608
1609 if (path [0] == '.' && !path [1])
1610 return wd->str;
1611
1612 {
1613 int l1 = wd->len;
1614 int l2 = strlen (path);
1615
1616 char *res = etp_tmpbuf_get (tmpbuf, l1 + l2 + 2);
1617
1618 memcpy (res, wd->str, l1);
1619 res [l1] = '/';
1620 memcpy (res + l1 + 1, path, l2 + 1);
1621
1622 return res;
1623 }
1624}
1625
1626#endif
1627
1628static eio_wd
1629eio__wd_open_sync (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
1630{
1631 int fd;
1632 eio_wd res;
1633 int len = eio__realpath (tmpbuf, wd, path);
1634
1635 if (len < 0)
1636 return EIO_INVALID_WD;
1637
1638#if HAVE_AT
1639 fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1640
1641 if (fd < 0)
1642 return EIO_INVALID_WD;
1643#endif
1644
1645 res = malloc (sizeof (*res) + len); /* one extra 0-byte */
1646
1647#if HAVE_AT
1648 res->fd = fd;
1649#endif
1650
1651 res->len = len;
1652 memcpy (res->str, tmpbuf->ptr, len);
1653 res->str [len] = 0;
1654
1655 return res;
1656}
1657
1658eio_wd
1659eio_wd_open_sync (eio_wd wd, const char *path)
1660{
1661 struct etp_tmpbuf tmpbuf = { };
1662 wd = eio__wd_open_sync (&tmpbuf, wd, path);
1663 free (tmpbuf.ptr);
1664
1665 return wd;
1666}
1667
1668void
1669eio_wd_close_sync (eio_wd wd)
1670{
1671 if (wd != EIO_INVALID_WD && wd != EIO_CWD)
1672 {
1673 #if HAVE_AT
1674 close (wd->fd);
1675 #endif
1676 free (wd);
1677 }
1678}
1679
1680#if HAVE_AT
1681
1682/* they forgot these */
1683
1684static int
1685eio__truncateat (int dirfd, const char *path, off_t length)
1686{
1687 int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
1688 int res;
1689
1690 if (fd < 0)
1691 return fd;
1692
1693 res = ftruncate (fd, length);
1694 close (fd);
1695 return res;
1696}
1697
1698static int
1699eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
1700{
1701 int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
1702 int res;
1703
1704 if (fd < 0)
1705 return fd;
1706
1707 res = fstatvfs (fd, buf);
1708 close (fd);
1709 return res;
1710
1711}
1712
1713#endif
1771 1714
1772/*****************************************************************************/ 1715/*****************************************************************************/
1773 1716
1774#define ALLOC(len) \ 1717#define ALLOC(len) \
1775 if (!req->ptr2) \ 1718 if (!req->ptr2) \
1776 { \ 1719 { \
1777 X_LOCK (wrklock); \ 1720 X_LOCK (EIO_POOL->wrklock); \
1778 req->flags |= EIO_FLAG_PTR2_FREE; \ 1721 req->flags |= EIO_FLAG_PTR2_FREE; \
1779 X_UNLOCK (wrklock); \ 1722 X_UNLOCK (EIO_POOL->wrklock); \
1780 req->ptr2 = malloc (len); \ 1723 req->ptr2 = malloc (len); \
1781 if (!req->ptr2) \ 1724 if (!req->ptr2) \
1782 { \ 1725 { \
1783 errno = ENOMEM; \ 1726 errno = ENOMEM; \
1784 req->result = -1; \ 1727 req->result = -1; \
1785 break; \ 1728 break; \
1786 } \ 1729 } \
1787 } 1730 }
1788 1731
1789X_THREAD_PROC (etp_proc)
1790{
1791 ETP_REQ *req;
1792 struct timespec ts;
1793 etp_worker *self = (etp_worker *)thr_arg;
1794
1795 /* try to distribute timeouts somewhat randomly */
1796 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1797
1798 for (;;)
1799 {
1800 X_LOCK (reqlock);
1801
1802 for (;;)
1803 {
1804 self->req = req = reqq_shift (&req_queue);
1805
1806 if (req)
1807 break;
1808
1809 ++idle;
1810
1811 ts.tv_sec = time (0) + idle_timeout;
1812 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
1813 {
1814 if (idle > max_idle)
1815 {
1816 --idle;
1817 X_UNLOCK (reqlock);
1818 X_LOCK (wrklock);
1819 --started;
1820 X_UNLOCK (wrklock);
1821 goto quit;
1822 }
1823
1824 /* we are allowed to idle, so do so without any timeout */
1825 X_COND_WAIT (reqwait, reqlock);
1826 }
1827
1828 --idle;
1829 }
1830
1831 --nready;
1832
1833 X_UNLOCK (reqlock);
1834
1835 if (req->type < 0)
1836 goto quit;
1837
1838 ETP_EXECUTE (self, req);
1839
1840 X_LOCK (reslock);
1841
1842 ++npending;
1843
1844 if (!reqq_push (&res_queue, req) && want_poll_cb)
1845 want_poll_cb ();
1846
1847 self->req = 0;
1848 etp_worker_clear (self);
1849
1850 X_UNLOCK (reslock);
1851 }
1852
1853quit:
1854 X_LOCK (wrklock);
1855 etp_worker_free (self);
1856 X_UNLOCK (wrklock);
1857
1858 return 0;
1859}
1860
1861/*****************************************************************************/ 1732/*****************************************************************************/
1862 1733
1863int ecb_cold 1734int ecb_cold
1864eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1735eio_init (void (*want_poll)(void), void (*done_poll)(void))
1865{ 1736{
1866 return etp_init (want_poll, done_poll); 1737 eio_want_poll_cb = want_poll;
1738 eio_done_poll_cb = done_poll;
1739
1740 return etp_init (EIO_POOL, 0, 0, 0);
1867} 1741}
1868 1742
1869ecb_inline void 1743ecb_inline void
1870eio_api_destroy (eio_req *req) 1744eio_api_destroy (eio_req *req)
1871{ 1745{
1872 free (req); 1746 free (req);
1873} 1747}
1874 1748
1875#define REQ(rtype) \ 1749#define REQ(rtype) \
1876 eio_req *req; \ 1750 eio_req *req; \
1877 \ 1751 \
1878 req = (eio_req *)calloc (1, sizeof *req); \ 1752 req = (eio_req *)calloc (1, sizeof *req); \
1879 if (!req) \ 1753 if (!req) \
1880 return 0; \ 1754 return 0; \
1894 { \ 1768 { \
1895 eio_api_destroy (req); \ 1769 eio_api_destroy (req); \
1896 return 0; \ 1770 return 0; \
1897 } 1771 }
1898 1772
1773#define SINGLEDOT(ptr) (0[(char *)(ptr)] == '.' && !1[(char *)(ptr)])
1774
1899static void 1775static void
1900eio_execute (etp_worker *self, eio_req *req) 1776eio_execute (etp_worker *self, eio_req *req)
1901{ 1777{
1778#if HAVE_AT
1779 int dirfd;
1780#else
1781 const char *path;
1782#endif
1783
1902 if (ecb_expect_false (EIO_CANCELLED (req))) 1784 if (ecb_expect_false (EIO_CANCELLED (req)))
1903 { 1785 {
1904 req->result = -1; 1786 req->result = -1;
1905 req->errorno = ECANCELED; 1787 req->errorno = ECANCELED;
1906 return; 1788 return;
1907 } 1789 }
1908 1790
1791 if (ecb_expect_false (req->wd == EIO_INVALID_WD))
1792 {
1793 req->result = -1;
1794 req->errorno = ENOENT;
1795 return;
1796 }
1797
1798 if (req->type >= EIO_OPEN)
1799 {
1800 #if HAVE_AT
1801 dirfd = WD2FD (req->wd);
1802 #else
1803 path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1804 #endif
1805 }
1806
1909 switch (req->type) 1807 switch (req->type)
1910 { 1808 {
1809 case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
1810 req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
1811 break;
1812 case EIO_WD_CLOSE: req->result = 0;
1813 eio_wd_close_sync (req->wd); break;
1814
1815 case EIO_SEEK: eio__lseek (req); break;
1911 case EIO_READ: ALLOC (req->size); 1816 case EIO_READ: ALLOC (req->size);
1912 req->result = req->offs >= 0 1817 req->result = req->offs >= 0
1913 ? pread (req->int1, req->ptr2, req->size, req->offs) 1818 ? pread (req->int1, req->ptr2, req->size, req->offs)
1914 : read (req->int1, req->ptr2, req->size); break; 1819 : read (req->int1, req->ptr2, req->size); break;
1915 case EIO_WRITE: req->result = req->offs >= 0 1820 case EIO_WRITE: req->result = req->offs >= 0
1916 ? pwrite (req->int1, req->ptr2, req->size, req->offs) 1821 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1917 : write (req->int1, req->ptr2, req->size); break; 1822 : write (req->int1, req->ptr2, req->size); break;
1918 1823
1824 case EIO_FCNTL: req->result = fcntl (req->int1, (int) req->int2, req->ptr2); break;
1825 case EIO_IOCTL: req->result = ioctl (req->int1, (unsigned long)req->int2, req->ptr2); break;
1826
1919 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break; 1827 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1920 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break; 1828 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
1829
1830#if HAVE_AT
1921 1831
1922 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1832 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1923 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 1833 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break;
1924 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1834 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1835 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, AT_SYMLINK_NOFOLLOW); break;
1836 case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break;
1837 case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break;
1838 case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break;
1839 case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break;
1840
1841 case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
1842 case EIO_RMDIR: /* complications arise because "." cannot be removed, so we might have to expand */
1843 req->result = req->wd && SINGLEDOT (req->ptr1)
1844 ? rmdir (req->wd->str)
1845 : unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break;
1846 case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); break;
1847 case EIO_RENAME: /* complications arise because "." cannot be renamed, so we might have to expand */
1848 req->result = req->wd && SINGLEDOT (req->ptr1)
1849 ? rename (req->wd->str, req->ptr2)
1850 : renameat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2); break;
1851 case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2, 0); break;
1852 case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
1853 case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1854 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1855 req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1856 case EIO_READLINK: ALLOC (EIO_PATH_MAX);
1857 req->result = readlinkat (dirfd, req->ptr1, req->ptr2, EIO_PATH_MAX);
1858 if (req->result == EIO_PATH_MAX)
1859 {
1860 req->result = -1;
1861 errno = ENAMETOOLONG;
1862 }
1863 break;
1864 case EIO_UTIME:
1865 case EIO_FUTIME:
1866 {
1867 struct timespec ts[2];
1868 struct timespec *times;
1869
1870 if (req->nv1 != -1. || req->nv2 != -1.)
1871 {
1872 ts[0].tv_sec = req->nv1;
1873 ts[0].tv_nsec = (req->nv1 - ts[0].tv_sec) * 1e9;
1874 ts[1].tv_sec = req->nv2;
1875 ts[1].tv_nsec = (req->nv2 - ts[1].tv_sec) * 1e9;
1876
1877 times = ts;
1878 }
1879 else
1880 times = 0;
1881
1882 req->result = req->type == EIO_FUTIME
1883 ? futimens (req->int1, times)
1884 : utimensat (dirfd, req->ptr1, times, 0);
1885 }
1886 break;
1887
1888#else
1889
1890 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1891 req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
1892 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1925 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 1893 req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
1894 case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
1895 case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
1896 case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
1897 case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
1898
1899 case EIO_UNLINK: req->result = unlink (path ); break;
1900 case EIO_RMDIR: req->result = rmdir (path ); break;
1901 case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
1902 case EIO_RENAME: req->result = rename (path , req->ptr2); break;
1903 case EIO_LINK: req->result = link (path , req->ptr2); break;
1904 case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
1905 case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
1906 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1907 req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
1908 case EIO_READLINK: ALLOC (EIO_PATH_MAX);
1909 req->result = readlink (path, req->ptr2, EIO_PATH_MAX);
1910 if (req->result == EIO_PATH_MAX)
1911 {
1912 req->result = -1;
1913 errno = ENAMETOOLONG;
1914 }
1915 break;
1916
1917 case EIO_UTIME:
1918 case EIO_FUTIME:
1919 {
1920 struct timeval tv[2];
1921 struct timeval *times;
1922
1923 if (req->nv1 != -1. || req->nv2 != -1.)
1924 {
1925 tv[0].tv_sec = req->nv1;
1926 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
1927 tv[1].tv_sec = req->nv2;
1928 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
1929
1930 times = tv;
1931 }
1932 else
1933 times = 0;
1934
1935 req->result = req->type == EIO_FUTIME
1936 ? futimes (req->int1, times)
1937 : utimes (req->ptr1, times);
1938 }
1939 break;
1940
1941#endif
1942
1943 case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
1944 {
1945 ALLOC (req->result);
1946 memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
1947 }
1948 break;
1949
1926 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1950 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1927 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; 1951 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1928 1952
1929 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1930 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1931 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS)); 1953 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1932 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break; 1954 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1933 1955
1934 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1935 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; 1956 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1936 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1937 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break; 1957 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1938 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1939 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break; 1958 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1940 1959
1941 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
1942 case EIO_CLOSE: req->result = close (req->int1); break; 1960 case EIO_CLOSE: req->result = close (req->int1); break;
1943 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break; 1961 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1944 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
1945 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1946 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1947 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1948 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1949 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1950 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1951
1952 case EIO_REALPATH: eio__realpath (req, self); break;
1953
1954 case EIO_READLINK: ALLOC (PATH_MAX);
1955 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1956
1957 case EIO_SYNC: req->result = 0; sync (); break; 1962 case EIO_SYNC: req->result = 0; sync (); break;
1958 case EIO_FSYNC: req->result = fsync (req->int1); break; 1963 case EIO_FSYNC: req->result = fsync (req->int1); break;
1959 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 1964 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1965 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
1966 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1960 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break; 1967 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1961 case EIO_MTOUCH: req->result = eio__mtouch (req); break; 1968 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1962 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break; 1969 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1963 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break; 1970 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1964 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break; 1971 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
1965 1972
1966 case EIO_READDIR: eio__scandir (req, self); break; 1973 case EIO_READDIR: eio__scandir (req, self); break;
1967 1974
1968 case EIO_BUSY: 1975 case EIO_BUSY:
1969#ifdef _WIN32 1976#ifdef _WIN32
1978 req->result = select (0, 0, 0, 0, &tv); 1985 req->result = select (0, 0, 0, 0, &tv);
1979 } 1986 }
1980#endif 1987#endif
1981 break; 1988 break;
1982 1989
1983 case EIO_UTIME: 1990#if 0
1984 case EIO_FUTIME:
1985 {
1986 struct timeval tv[2];
1987 struct timeval *times;
1988
1989 if (req->nv1 != -1. || req->nv2 != -1.)
1990 {
1991 tv[0].tv_sec = req->nv1;
1992 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
1993 tv[1].tv_sec = req->nv2;
1994 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
1995
1996 times = tv;
1997 }
1998 else
1999 times = 0;
2000
2001 req->result = req->type == EIO_FUTIME
2002 ? futimes (req->int1, times)
2003 : utimes (req->ptr1, times);
2004 }
2005 break;
2006
2007 case EIO_GROUP: 1991 case EIO_GROUP:
2008 abort (); /* handled in eio_request */ 1992 abort (); /* handled in eio_request */
1993#endif
2009 1994
2010 case EIO_NOP: 1995 case EIO_NOP:
2011 req->result = 0; 1996 req->result = 0;
2012 break; 1997 break;
2013 1998
2014 case EIO_CUSTOM: 1999 case EIO_CUSTOM:
2015 req->feed (req); 2000 req->feed (req);
2016 break; 2001 break;
2017 2002
2018 default: 2003 default:
2019 errno = ENOSYS;
2020 req->result = -1; 2004 req->result = EIO_ENOSYS ();
2021 break; 2005 break;
2022 } 2006 }
2023 2007
2024 req->errorno = errno; 2008 req->errorno = errno;
2025} 2009}
2026 2010
2027#ifndef EIO_NO_WRAPPERS 2011#ifndef EIO_NO_WRAPPERS
2028 2012
2013eio_req *eio_wd_open (const char *path, int pri, eio_cb cb, void *data)
2014{
2015 REQ (EIO_WD_OPEN); PATH; SEND;
2016}
2017
2018eio_req *eio_wd_close (eio_wd wd, int pri, eio_cb cb, void *data)
2019{
2020 REQ (EIO_WD_CLOSE); req->wd = wd; SEND;
2021}
2022
2029eio_req *eio_nop (int pri, eio_cb cb, void *data) 2023eio_req *eio_nop (int pri, eio_cb cb, void *data)
2030{ 2024{
2031 REQ (EIO_NOP); SEND; 2025 REQ (EIO_NOP); SEND;
2032} 2026}
2033 2027
2049eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2043eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2050{ 2044{
2051 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2045 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2052} 2046}
2053 2047
2048eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2049{
2050 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2051}
2052
2053eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
2054{
2055 REQ (EIO_SYNCFS); req->int1 = fd; SEND;
2056}
2057
2058eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2059{
2060 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2061}
2062
2054eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2063eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2055{ 2064{
2056 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2065 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2057} 2066}
2058 2067
2064eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data) 2073eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
2065{ 2074{
2066 REQ (EIO_MLOCKALL); req->int1 = flags; SEND; 2075 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2067} 2076}
2068 2077
2069eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) 2078eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2070{ 2079{
2071 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; 2080 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
2072}
2073
2074eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2075{
2076 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2077} 2081}
2078 2082
2079eio_req *eio_close (int fd, int pri, eio_cb cb, void *data) 2083eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
2080{ 2084{
2081 REQ (EIO_CLOSE); req->int1 = fd; SEND; 2085 REQ (EIO_CLOSE); req->int1 = fd; SEND;
2084eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data) 2088eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
2085{ 2089{
2086 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND; 2090 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
2087} 2091}
2088 2092
2093eio_req *eio_seek (int fd, off_t offset, int whence, int pri, eio_cb cb, void *data)
2094{
2095 REQ (EIO_SEEK); req->int1 = fd; req->offs = offset; req->int2 = whence; SEND;
2096}
2097
2089eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data) 2098eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
2090{ 2099{
2091 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND; 2100 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
2092} 2101}
2093 2102
2094eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data) 2103eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
2095{ 2104{
2096 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND; 2105 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
2097} 2106}
2098 2107
2108eio_req *eio_fcntl (int fd, int cmd, void *arg, int pri, eio_cb cb, void *data)
2109{
2110 REQ (EIO_IOCTL); req->int1 = fd; req->int2 = cmd; req->ptr2 = arg; SEND;
2111}
2112
2113eio_req *eio_ioctl (int fd, unsigned long request, void *buf, int pri, eio_cb cb, void *data)
2114{
2115 REQ (EIO_IOCTL); req->int1 = fd; req->int2 = request; req->ptr2 = buf; SEND;
2116}
2117
2099eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) 2118eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
2100{ 2119{
2101 REQ (EIO_FSTAT); req->int1 = fd; SEND; 2120 REQ (EIO_FSTAT); req->int1 = fd; SEND;
2102} 2121}
2103 2122
2119eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) 2138eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
2120{ 2139{
2121 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2140 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
2122} 2141}
2123 2142
2124eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2143eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
2125{ 2144{
2126 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2145 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
2127} 2146}
2128 2147
2129eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) 2148eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
2149eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data) 2168eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
2150{ 2169{
2151 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2170 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
2152} 2171}
2153 2172
2154eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2173eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
2155{ 2174{
2156 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2175 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
2157} 2176}
2158 2177
2159eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) 2178eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
2289void 2308void
2290eio_grp_add (eio_req *grp, eio_req *req) 2309eio_grp_add (eio_req *grp, eio_req *req)
2291{ 2310{
2292 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2311 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
2293 2312
2294 grp->flags |= EIO_FLAG_GROUPADD; 2313 grp->flags |= ETP_FLAG_GROUPADD;
2295 2314
2296 ++grp->size; 2315 ++grp->size;
2297 req->grp = grp; 2316 req->grp = grp;
2298 2317
2299 req->grp_prev = 0; 2318 req->grp_prev = 0;
2306} 2325}
2307 2326
2308/*****************************************************************************/ 2327/*****************************************************************************/
2309/* misc garbage */ 2328/* misc garbage */
2310 2329
2311ssize_t 2330eio_ssize_t
2312eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2331eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2313{ 2332{
2314 etp_worker wrk;
2315 ssize_t ret;
2316
2317 wrk.dbuf = 0;
2318
2319 ret = eio__sendfile (ofd, ifd, offset, count, &wrk); 2333 return eio__sendfile (ofd, ifd, offset, count);
2320
2321 if (wrk.dbuf)
2322 free (wrk.dbuf);
2323
2324 return ret;
2325} 2334}
2326 2335

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines