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

Comparing libeio/eio.c (file contents):
Revision 1.73 by root, Fri Jun 17 18:55:42 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 */
65/* intptr_t only comes form stdint.h, says idiot openbsd coder */ 59/* intptr_t only comes from stdint.h, says idiot openbsd coder */
66#if HAVE_STDINT_H 60#if HAVE_STDINT_H
67# include <stdint.h> 61# include <stdint.h>
68#endif 62#endif
69 63
64#ifndef ECANCELED
65# define ECANCELED EDOM
66#endif
67#ifndef ELOOP
68# define ELOOP EDOM
69#endif
70
71#if !defined(ENOTSOCK) && defined(WSAENOTSOCK)
72# define ENOTSOCK WSAENOTSOCK
73#endif
74
75static void eio_destroy (eio_req *req);
76
70#ifndef EIO_FINISH 77#ifndef EIO_FINISH
71# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0 78# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
72#endif 79#endif
73 80
74#ifndef EIO_DESTROY 81#ifndef EIO_DESTROY
… …
77 84
78#ifndef EIO_FEED 85#ifndef EIO_FEED
79# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0) 86# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)
80#endif 87#endif
81 88
89#ifndef EIO_FD_TO_WIN32_HANDLE
90# define EIO_FD_TO_WIN32_HANDLE(fd) _get_osfhandle (fd)
91#endif
92#ifndef EIO_WIN32_HANDLE_TO_FD
93# define EIO_WIN32_HANDLE_TO_FD(handle) _open_osfhandle (handle, 0)
94#endif
95
96#define EIO_ERRNO(errval,retval) ((errno = errval), retval)
97
98#define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1)
99
82#ifdef _WIN32 100#ifdef _WIN32
83 101
84 /*doh*/ 102 #undef PAGESIZE
103 #define PAGESIZE 4096 /* GetSystemInfo? */
104
105 /* TODO: look at how perl does stat (non-sloppy), unlink (ro-files), utime, link */
106
107 #ifdef EIO_STRUCT_STATI64
108 /* look at perl's non-sloppy stat */
109 #define stat(path,buf) _stati64 (path,buf)
110 #define fstat(fd,buf) _fstati64 (fd,buf)
111 #endif
112 #define lstat(path,buf) stat (path,buf)
113 #define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
114 #define mkdir(path,mode) _mkdir (path)
115 #define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
116
117 #define chmod(path,mode) _chmod (path, mode)
118 #define dup(fd) _dup (fd)
119 #define dup2(fd1,fd2) _dup2 (fd1, fd2)
120 #define pipe(fds) _pipe (fds, 4096, O_BINARY)
121
122 #define fcntl(fd,cmd,arg) EIO_ENOSYS ()
123 #define ioctl(fd,cmd,arg) EIO_ENOSYS ()
124 #define fchmod(fd,mode) EIO_ENOSYS ()
125 #define chown(path,uid,gid) EIO_ENOSYS ()
126 #define fchown(fd,uid,gid) EIO_ENOSYS ()
127 #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
128 #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
129 #define mknod(path,mode,dev) EIO_ENOSYS ()
130 #define sync() EIO_ENOSYS ()
131 #define readlink(path,buf,s) EIO_ENOSYS ()
132 #define statvfs(path,buf) EIO_ENOSYS ()
133 #define fstatvfs(fd,buf) EIO_ENOSYS ()
134
135 #define pread(fd,buf,count,offset) eio__pread (fd, buf, count, offset)
136 #define pwrite(fd,buf,count,offset) eio__pwrite (fd, buf, count, offset)
137
138 #if __GNUC__
139 typedef long long eio_off_t; /* signed for compatibility to msvc */
140 #else
141 typedef __int64 eio_off_t; /* unsigned not supported by msvc */
142 #endif
143
144 static eio_ssize_t
145 eio__pread (int fd, void *buf, eio_ssize_t count, eio_off_t offset)
146 {
147 OVERLAPPED o = { 0 };
148 DWORD got;
149
150 o.Offset = offset;
151 o.OffsetHigh = offset >> 32;
152
153 return ReadFile ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd), buf, count, &got, &o)
154 ? got : -1;
155 }
156
157 static eio_ssize_t
158 eio__pwrite (int fd, void *buf, eio_ssize_t count, eio_off_t offset)
159 {
160 OVERLAPPED o = { 0 };
161 DWORD got;
162
163 o.Offset = offset;
164 o.OffsetHigh = offset >> 32;
165
166 return WriteFile ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd), buf, count, &got, &o)
167 ? got : -1;
168 }
169
170 /* rename() uses MoveFile, which fails to overwrite */
171 #define rename(old,neu) eio__rename (old, neu)
172
173 static int
174 eio__rename (const char *old, const char *neu)
175 {
176 if (MoveFileEx (old, neu, MOVEFILE_REPLACE_EXISTING))
177 return 0;
178
179 /* should steal _dosmaperr */
180 switch (GetLastError ())
181 {
182 case ERROR_FILE_NOT_FOUND:
183 case ERROR_PATH_NOT_FOUND:
184 case ERROR_INVALID_DRIVE:
185 case ERROR_NO_MORE_FILES:
186 case ERROR_BAD_NETPATH:
187 case ERROR_BAD_NET_NAME:
188 case ERROR_BAD_PATHNAME:
189 case ERROR_FILENAME_EXCED_RANGE:
190 errno = ENOENT;
191 break;
192
193 default:
194 errno = EACCES;
195 break;
196 }
197
198 return -1;
199 }
200
201 /* we could even stat and see if it exists */
202 static int
203 symlink (const char *old, const char *neu)
204 {
205 #if WINVER >= 0x0600
206 int flags;
207
208 /* This tries out all combinations of SYMBOLIC_LINK_FLAG_DIRECTORY
209 * and SYMBOLIC_LINK_FLAG_ALLOW_UNPRIVILEGED_CREATE,
210 * with directory first.
211 */
212 for (flags = 3; flags >= 0; --flags)
213 if (CreateSymbolicLink (neu, old, flags))
214 return 0;
215 #endif
216
217 return EIO_ERRNO (ENOENT, -1);
218 }
219
220 /* POSIX API only, causing trouble for win32 apps */
221 #define CreateHardLink(neu,old,flags) 0 /* not really creating hardlink, still using relative paths? */
222 #define CreateSymbolicLink(neu,old,flags) 0 /* vista+ only */
223
224 struct statvfs
225 {
226 int dummy;
227 };
228
229 #define DT_DIR EIO_DT_DIR
230 #define DT_REG EIO_DT_REG
231 #define D_NAME(entp) entp.cFileName
232 #define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
233
85#else 234#else
86 235
87# include <sys/time.h> 236 #include <sys/time.h>
88# include <sys/select.h> 237 #include <sys/select.h>
89# include <unistd.h> 238 #include <unistd.h>
239 #include <signal.h>
240 #include <dirent.h>
241
242 #ifdef ANDROID
243 #include <sys/vfs.h>
244 #define statvfs statfs
245 #define fstatvfs fstatfs
246 #include <asm/page.h> /* supposedly limits.h does #define PAGESIZE PAGESIZE */
247 #else
248 #include <sys/statvfs.h>
249 #endif
250
251 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
252 #include <sys/mman.h>
253 #endif
254
255 #define D_NAME(entp) entp->d_name
256
257 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
258 #if __FreeBSD__ || __NetBSD__ || __OpenBSD__
259 #define _DIRENT_HAVE_D_TYPE /* sigh */
260 #define D_INO(de) (de)->d_fileno
261 #define D_NAMLEN(de) (de)->d_namlen
262 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
263 #define D_INO(de) (de)->d_ino
264 #endif
265
266 #ifdef _D_EXACT_NAMLEN
267 #undef D_NAMLEN
268 #define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
269 #endif
270
271 #ifdef _DIRENT_HAVE_D_TYPE
272 #define D_TYPE(de) (de)->d_type
273 #endif
274
275 #ifndef EIO_STRUCT_DIRENT
276 #define EIO_STRUCT_DIRENT struct dirent
277 #endif
278
279#endif
280
281#if HAVE_UTIMES
90# include <utime.h> 282# include <utime.h>
91# include <signal.h>
92# include <dirent.h>
93
94#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
95# include <sys/mman.h>
96#endif
97
98/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
99# if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
100# define _DIRENT_HAVE_D_TYPE /* sigh */
101# define D_INO(de) (de)->d_fileno
102# define D_NAMLEN(de) (de)->d_namlen
103# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
104# define D_INO(de) (de)->d_ino
105# endif 283#endif
106 284
107#ifdef _D_EXACT_NAMLEN 285#if HAVE_SYS_SYSCALL_H
108# undef D_NAMLEN 286# include <sys/syscall.h>
109# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
110#endif
111
112# ifdef _DIRENT_HAVE_D_TYPE
113# define D_TYPE(de) (de)->d_type
114# endif
115
116# ifndef EIO_STRUCT_DIRENT
117# define EIO_STRUCT_DIRENT struct dirent
118# endif
119
120#endif 287#endif
121 288
122#if HAVE_SENDFILE 289#if HAVE_SENDFILE
123# if __linux 290# if __linux
124# include <sys/sendfile.h> 291# include <sys/sendfile.h>
… …
139#endif 306#endif
140#ifndef D_INO 307#ifndef D_INO
141# define D_INO(de) 0 308# define D_INO(de) 0
142#endif 309#endif
143#ifndef D_NAMLEN 310#ifndef D_NAMLEN
144# define D_NAMLEN(de) strlen ((de)->d_name) 311# define D_NAMLEN(entp) strlen (D_NAME (entp))
145#endif 312#endif
146 313
147/* used for struct dirent, AIX doesn't provide it */ 314/* used for struct dirent, AIX doesn't provide it */
148#ifndef NAME_MAX 315#ifndef NAME_MAX
149# define NAME_MAX 4096 316# define NAME_MAX 4096
150#endif 317#endif
151 318
152/* used for readlink etc. */ 319/* used for readlink etc. */
153#ifndef PATH_MAX 320#ifndef PATH_MAX
154# define PATH_MAX 4096 321# define PATH_MAX 0
155#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)
156 329
157/* buffer size for various temporary buffers */ 330/* buffer size for various temporary buffers */
158#define EIO_BUFSIZE 65536 331#define EIO_BUFSIZE 65536
159 332
160#define dBUF \ 333#define dBUF \
161 char *eio_buf; \
162 ETP_WORKER_LOCK (self); \
163 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 334 char *eio_buf = malloc (EIO_BUFSIZE); \
164 ETP_WORKER_UNLOCK (self); \
165 errno = ENOMEM; \ 335 errno = ENOMEM; \
166 if (!eio_buf) \ 336 if (!eio_buf) \
167 return -1; 337 return -1
168 338
169#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/*****************************************************************************/
170 367
171#define ETP_PRI_MIN EIO_PRI_MIN 368#define ETP_PRI_MIN EIO_PRI_MIN
172#define ETP_PRI_MAX EIO_PRI_MAX 369#define ETP_PRI_MAX EIO_PRI_MAX
173 370
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
174struct etp_worker; 381struct etp_worker;
175
176#define ETP_REQ eio_req 382#define ETP_REQ eio_req
177#define ETP_DESTROY(req) eio_destroy (req) 383#define ETP_DESTROY(req) eio_destroy (req)
178static int eio_finish (eio_req *req); 384static int eio_finish (eio_req *req);
179#define ETP_FINISH(req) eio_finish (req) 385#define ETP_FINISH(req) eio_finish (req)
180static void eio_execute (struct etp_worker *self, eio_req *req); 386static void eio_execute (struct etp_worker *self, eio_req *req);
181#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) 387#define ETP_EXECUTE(wrk,req) eio_execute (wrk, req)
182 388
183#define ETP_WORKER_CLEAR(req) \ 389#include "etp.c"
184 if (wrk->dbuf) \
185 { \
186 free (wrk->dbuf); \
187 wrk->dbuf = 0; \
188 } \
189 \
190 if (wrk->dirp) \
191 { \
192 closedir (wrk->dirp); \
193 wrk->dirp = 0; \
194 }
195 390
196#define ETP_WORKER_COMMON \ 391static struct etp_pool eio_pool;
197 void *dbuf; \ 392#define EIO_POOL (&eio_pool)
198 DIR *dirp;
199
200/*****************************************************************************/
201
202#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
203
204/* calculate time difference in ~1/EIO_TICKS of a second */
205ecb_inline int
206tvdiff (struct timeval *tv1, struct timeval *tv2)
207{
208 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
209 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
210}
211
212static unsigned int started, idle, wanted = 4;
213
214static void (*want_poll_cb) (void);
215static void (*done_poll_cb) (void);
216
217static unsigned int max_poll_time; /* reslock */
218static unsigned int max_poll_reqs; /* reslock */
219
220static volatile unsigned int nreqs; /* reqlock */
221static volatile unsigned int nready; /* reqlock */
222static volatile unsigned int npending; /* reqlock */
223static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
224static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
225
226static xmutex_t wrklock;
227static xmutex_t reslock;
228static xmutex_t reqlock;
229static xcond_t reqwait;
230
231#if !HAVE_PREADWRITE
232/*
233 * make our pread/pwrite emulation safe against themselves, but not against
234 * normal read/write by using a mutex. slows down execution a lot,
235 * but that's your problem, not mine.
236 */
237static xmutex_t preadwritelock = X_MUTEX_INIT;
238#endif
239
240typedef struct etp_worker
241{
242 /* locked by wrklock */
243 struct etp_worker *prev, *next;
244
245 xthread_t tid;
246
247 /* locked by reslock, reqlock or wrklock */
248 ETP_REQ *req; /* currently processed request */
249
250 ETP_WORKER_COMMON
251} etp_worker;
252
253static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */
254
255#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
256#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
257
258/* worker threads management */
259
260static void ecb_cold
261etp_worker_clear (etp_worker *wrk)
262{
263 ETP_WORKER_CLEAR (wrk);
264}
265
266static void ecb_cold
267etp_worker_free (etp_worker *wrk)
268{
269 wrk->next->prev = wrk->prev;
270 wrk->prev->next = wrk->next;
271
272 free (wrk);
273}
274
275static unsigned int
276etp_nreqs (void)
277{
278 int retval;
279 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
280 retval = nreqs;
281 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
282 return retval;
283}
284
285static unsigned int
286etp_nready (void)
287{
288 unsigned int retval;
289
290 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
291 retval = nready;
292 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
293
294 return retval;
295}
296
297static unsigned int
298etp_npending (void)
299{
300 unsigned int retval;
301
302 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
303 retval = npending;
304 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
305
306 return retval;
307}
308
309static unsigned int
310etp_nthreads (void)
311{
312 unsigned int retval;
313
314 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
315 retval = started;
316 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
317
318 return retval;
319}
320
321/*
322 * a somewhat faster data structure might be nice, but
323 * with 8 priorities this actually needs <20 insns
324 * per shift, the most expensive operation.
325 */
326typedef struct {
327 ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */
328 int size;
329} etp_reqq;
330
331static etp_reqq req_queue;
332static etp_reqq res_queue;
333
334static int ecb_noinline
335reqq_push (etp_reqq *q, ETP_REQ *req)
336{
337 int pri = req->pri;
338 req->next = 0;
339
340 if (q->qe[pri])
341 {
342 q->qe[pri]->next = req;
343 q->qe[pri] = req;
344 }
345 else
346 q->qe[pri] = q->qs[pri] = req;
347
348 return q->size++;
349}
350
351static ETP_REQ * ecb_noinline
352reqq_shift (etp_reqq *q)
353{
354 int pri;
355
356 if (!q->size)
357 return 0;
358
359 --q->size;
360
361 for (pri = ETP_NUM_PRI; pri--; )
362 {
363 eio_req *req = q->qs[pri];
364
365 if (req)
366 {
367 if (!(q->qs[pri] = (eio_req *)req->next))
368 q->qe[pri] = 0;
369
370 return req;
371 }
372 }
373
374 abort ();
375}
376
377static void ecb_cold
378etp_thread_init (void)
379{
380 X_MUTEX_CREATE (wrklock);
381 X_MUTEX_CREATE (reslock);
382 X_MUTEX_CREATE (reqlock);
383 X_COND_CREATE (reqwait);
384}
385
386static void ecb_cold
387etp_atfork_prepare (void)
388{
389 X_LOCK (wrklock);
390 X_LOCK (reqlock);
391 X_LOCK (reslock);
392#if !HAVE_PREADWRITE
393 X_LOCK (preadwritelock);
394#endif
395}
396
397static void ecb_cold
398etp_atfork_parent (void)
399{
400#if !HAVE_PREADWRITE
401 X_UNLOCK (preadwritelock);
402#endif
403 X_UNLOCK (reslock);
404 X_UNLOCK (reqlock);
405 X_UNLOCK (wrklock);
406}
407
408static void ecb_cold
409etp_atfork_child (void)
410{
411 ETP_REQ *prv;
412
413 while ((prv = reqq_shift (&req_queue)))
414 ETP_DESTROY (prv);
415
416 while ((prv = reqq_shift (&res_queue)))
417 ETP_DESTROY (prv);
418
419 while (wrk_first.next != &wrk_first)
420 {
421 etp_worker *wrk = wrk_first.next;
422
423 if (wrk->req)
424 ETP_DESTROY (wrk->req);
425
426 etp_worker_clear (wrk);
427 etp_worker_free (wrk);
428 }
429
430 started = 0;
431 idle = 0;
432 nreqs = 0;
433 nready = 0;
434 npending = 0;
435
436 etp_thread_init ();
437}
438
439static void ecb_cold
440etp_once_init (void)
441{
442 etp_thread_init ();
443 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
444}
445
446static int ecb_cold
447etp_init (void (*want_poll)(void), void (*done_poll)(void))
448{
449 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
450
451 pthread_once (&doinit, etp_once_init);
452
453 want_poll_cb = want_poll;
454 done_poll_cb = done_poll;
455
456 return 0;
457}
458
459X_THREAD_PROC (etp_proc);
460
461static void ecb_cold
462etp_start_thread (void)
463{
464 etp_worker *wrk = calloc (1, sizeof (etp_worker));
465
466 /*TODO*/
467 assert (("unable to allocate worker thread data", wrk));
468
469 X_LOCK (wrklock);
470
471 if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
472 {
473 wrk->prev = &wrk_first;
474 wrk->next = wrk_first.next;
475 wrk_first.next->prev = wrk;
476 wrk_first.next = wrk;
477 ++started;
478 }
479 else
480 free (wrk);
481
482 X_UNLOCK (wrklock);
483}
484
485static void
486etp_maybe_start_thread (void)
487{
488 if (ecb_expect_true (etp_nthreads () >= wanted))
489 return;
490
491 /* todo: maybe use idle here, but might be less exact */
492 if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
493 return;
494
495 etp_start_thread ();
496}
497
498static void ecb_cold
499etp_end_thread (void)
500{
501 eio_req *req = calloc (1, sizeof (eio_req));
502
503 req->type = -1;
504 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
505
506 X_LOCK (reqlock);
507 reqq_push (&req_queue, req);
508 X_COND_SIGNAL (reqwait);
509 X_UNLOCK (reqlock);
510
511 X_LOCK (wrklock);
512 --started;
513 X_UNLOCK (wrklock);
514}
515
516static int
517etp_poll (void)
518{
519 unsigned int maxreqs;
520 unsigned int maxtime;
521 struct timeval tv_start, tv_now;
522
523 X_LOCK (reslock);
524 maxreqs = max_poll_reqs;
525 maxtime = max_poll_time;
526 X_UNLOCK (reslock);
527
528 if (maxtime)
529 gettimeofday (&tv_start, 0);
530
531 for (;;)
532 {
533 ETP_REQ *req;
534
535 etp_maybe_start_thread ();
536
537 X_LOCK (reslock);
538 req = reqq_shift (&res_queue);
539
540 if (req)
541 {
542 --npending;
543
544 if (!res_queue.size && done_poll_cb)
545 done_poll_cb ();
546 }
547
548 X_UNLOCK (reslock);
549
550 if (!req)
551 return 0;
552
553 X_LOCK (reqlock);
554 --nreqs;
555 X_UNLOCK (reqlock);
556
557 if (ecb_expect_false (req->type == EIO_GROUP && req->size))
558 {
559 req->int1 = 1; /* mark request as delayed */
560 continue;
561 }
562 else
563 {
564 int res = ETP_FINISH (req);
565 if (ecb_expect_false (res))
566 return res;
567 }
568
569 if (ecb_expect_false (maxreqs && !--maxreqs))
570 break;
571
572 if (maxtime)
573 {
574 gettimeofday (&tv_now, 0);
575
576 if (tvdiff (&tv_start, &tv_now) >= maxtime)
577 break;
578 }
579 }
580
581 errno = EAGAIN;
582 return -1;
583}
584
585static void
586etp_cancel (ETP_REQ *req)
587{
588 X_LOCK (wrklock);
589 req->flags |= EIO_FLAG_CANCELLED;
590 X_UNLOCK (wrklock);
591
592 eio_grp_cancel (req);
593}
594
595static void
596etp_submit (ETP_REQ *req)
597{
598 req->pri -= ETP_PRI_MIN;
599
600 if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
601 if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
602
603 if (ecb_expect_false (req->type == EIO_GROUP))
604 {
605 /* I hope this is worth it :/ */
606 X_LOCK (reqlock);
607 ++nreqs;
608 X_UNLOCK (reqlock);
609
610 X_LOCK (reslock);
611
612 ++npending;
613
614 if (!reqq_push (&res_queue, req) && want_poll_cb)
615 want_poll_cb ();
616
617 X_UNLOCK (reslock);
618 }
619 else
620 {
621 X_LOCK (reqlock);
622 ++nreqs;
623 ++nready;
624 reqq_push (&req_queue, req);
625 X_COND_SIGNAL (reqwait);
626 X_UNLOCK (reqlock);
627
628 etp_maybe_start_thread ();
629 }
630}
631
632static void ecb_cold
633etp_set_max_poll_time (double nseconds)
634{
635 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
636 max_poll_time = nseconds * EIO_TICKS;
637 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
638}
639
640static void ecb_cold
641etp_set_max_poll_reqs (unsigned int maxreqs)
642{
643 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
644 max_poll_reqs = maxreqs;
645 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
646}
647
648static void ecb_cold
649etp_set_max_idle (unsigned int nthreads)
650{
651 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
652 max_idle = nthreads;
653 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
654}
655
656static void ecb_cold
657etp_set_idle_timeout (unsigned int seconds)
658{
659 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
660 idle_timeout = seconds;
661 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
662}
663
664static void ecb_cold
665etp_set_min_parallel (unsigned int nthreads)
666{
667 if (wanted < nthreads)
668 wanted = nthreads;
669}
670
671static void ecb_cold
672etp_set_max_parallel (unsigned int nthreads)
673{
674 if (wanted > nthreads)
675 wanted = nthreads;
676
677 while (started > wanted)
678 etp_end_thread ();
679}
680 393
681/*****************************************************************************/ 394/*****************************************************************************/
682 395
683static void 396static void
684grp_try_feed (eio_req *grp) 397grp_try_feed (eio_req *grp)
685{ 398{
686 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 399 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
687 { 400 {
688 grp->flags &= ~EIO_FLAG_GROUPADD; 401 grp->flags &= ~ETP_FLAG_GROUPADD;
689 402
690 EIO_FEED (grp); 403 EIO_FEED (grp);
691 404
692 /* stop if no progress has been made */ 405 /* stop if no progress has been made */
693 if (!(grp->flags & EIO_FLAG_GROUPADD)) 406 if (!(grp->flags & ETP_FLAG_GROUPADD))
694 { 407 {
695 grp->feed = 0; 408 grp->feed = 0;
696 break; 409 break;
697 } 410 }
698 } 411 }
… …
705 418
706 /* call feeder, if applicable */ 419 /* call feeder, if applicable */
707 grp_try_feed (grp); 420 grp_try_feed (grp);
708 421
709 /* finish, if done */ 422 /* finish, if done */
710 if (!grp->size && grp->int1) 423 if (!grp->size && grp->flags & ETP_FLAG_DELAYED)
711 return eio_finish (grp); 424 return eio_finish (grp);
712 else 425 else
713 return 0; 426 return 0;
714} 427}
715 428
716void 429static void
717eio_destroy (eio_req *req) 430eio_destroy (eio_req *req)
718{ 431{
719 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); 432 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
720 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); 433 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
721 434
… …
739 if (grp->grp_first == req) 452 if (grp->grp_first == req)
740 grp->grp_first = req->grp_next; 453 grp->grp_first = req->grp_next;
741 454
742 res2 = grp_dec (grp); 455 res2 = grp_dec (grp);
743 456
744 if (!res && res2) 457 if (!res)
745 res = res2; 458 res = res2;
746 } 459 }
747 460
748 eio_destroy (req); 461 eio_destroy (req);
749 462
… …
751} 464}
752 465
753void 466void
754eio_grp_cancel (eio_req *grp) 467eio_grp_cancel (eio_req *grp)
755{ 468{
756 for (grp = grp->grp_first; grp; grp = grp->grp_next) 469 etp_grp_cancel (EIO_POOL, grp);
757 eio_cancel (grp);
758} 470}
759 471
760void 472void
761eio_cancel (eio_req *req) 473eio_cancel (eio_req *req)
762{ 474{
763 etp_cancel (req); 475 etp_cancel (EIO_POOL, req);
764} 476}
765 477
766void 478void
767eio_submit (eio_req *req) 479eio_submit (eio_req *req)
768{ 480{
769 etp_submit (req); 481 etp_submit (EIO_POOL, req);
770} 482}
771 483
772unsigned int 484unsigned int
773eio_nreqs (void) 485eio_nreqs (void)
774{ 486{
775 return etp_nreqs (); 487 return etp_nreqs (EIO_POOL);
776} 488}
777 489
778unsigned int 490unsigned int
779eio_nready (void) 491eio_nready (void)
780{ 492{
781 return etp_nready (); 493 return etp_nready (EIO_POOL);
782} 494}
783 495
784unsigned int 496unsigned int
785eio_npending (void) 497eio_npending (void)
786{ 498{
787 return etp_npending (); 499 return etp_npending (EIO_POOL);
788} 500}
789 501
790unsigned int ecb_cold 502unsigned int ecb_cold
791eio_nthreads (void) 503eio_nthreads (void)
792{ 504{
793 return etp_nthreads (); 505 return etp_nthreads (EIO_POOL);
794} 506}
795 507
796void ecb_cold 508void ecb_cold
797eio_set_max_poll_time (double nseconds) 509eio_set_max_poll_time (double nseconds)
798{ 510{
799 etp_set_max_poll_time (nseconds); 511 etp_set_max_poll_time (EIO_POOL, nseconds);
800} 512}
801 513
802void ecb_cold 514void ecb_cold
803eio_set_max_poll_reqs (unsigned int maxreqs) 515eio_set_max_poll_reqs (unsigned int maxreqs)
804{ 516{
805 etp_set_max_poll_reqs (maxreqs); 517 etp_set_max_poll_reqs (EIO_POOL, maxreqs);
806} 518}
807 519
808void ecb_cold 520void ecb_cold
809eio_set_max_idle (unsigned int nthreads) 521eio_set_max_idle (unsigned int nthreads)
810{ 522{
811 etp_set_max_idle (nthreads); 523 etp_set_max_idle (EIO_POOL, nthreads);
812} 524}
813 525
814void ecb_cold 526void ecb_cold
815eio_set_idle_timeout (unsigned int seconds) 527eio_set_idle_timeout (unsigned int seconds)
816{ 528{
817 etp_set_idle_timeout (seconds); 529 etp_set_idle_timeout (EIO_POOL, seconds);
818} 530}
819 531
820void ecb_cold 532void ecb_cold
821eio_set_min_parallel (unsigned int nthreads) 533eio_set_min_parallel (unsigned int nthreads)
822{ 534{
823 etp_set_min_parallel (nthreads); 535 etp_set_min_parallel (EIO_POOL, nthreads);
824} 536}
825 537
826void ecb_cold 538void ecb_cold
827eio_set_max_parallel (unsigned int nthreads) 539eio_set_max_parallel (unsigned int nthreads)
828{ 540{
829 etp_set_max_parallel (nthreads); 541 etp_set_max_parallel (EIO_POOL, nthreads);
830} 542}
831 543
832int eio_poll (void) 544int eio_poll (void)
833{ 545{
834 return etp_poll (); 546 return etp_poll (EIO_POOL);
835} 547}
836 548
837/*****************************************************************************/ 549/*****************************************************************************/
838/* work around various missing functions */ 550/* work around various missing functions */
839
840#if !HAVE_PREADWRITE
841# undef pread
842# undef pwrite
843# define pread eio__pread
844# define pwrite eio__pwrite
845
846static ssize_t
847eio__pread (int fd, void *buf, size_t count, off_t offset)
848{
849 ssize_t res;
850 off_t ooffset;
851
852 X_LOCK (preadwritelock);
853 ooffset = lseek (fd, 0, SEEK_CUR);
854 lseek (fd, offset, SEEK_SET);
855 res = read (fd, buf, count);
856 lseek (fd, ooffset, SEEK_SET);
857 X_UNLOCK (preadwritelock);
858
859 return res;
860}
861
862static ssize_t
863eio__pwrite (int fd, void *buf, size_t count, off_t offset)
864{
865 ssize_t res;
866 off_t ooffset;
867
868 X_LOCK (preadwritelock);
869 ooffset = lseek (fd, 0, SEEK_CUR);
870 lseek (fd, offset, SEEK_SET);
871 res = write (fd, buf, count);
872 lseek (fd, ooffset, SEEK_SET);
873 X_UNLOCK (preadwritelock);
874
875 return res;
876}
877#endif
878 551
879#ifndef HAVE_UTIMES 552#ifndef HAVE_UTIMES
880 553
881# undef utimes 554# undef utimes
882# define utimes(path,times) eio__utimes (path, times) 555# define utimes(path,times) eio__utimes (path, times)
… …
915 588
916#if !HAVE_FDATASYNC 589#if !HAVE_FDATASYNC
917# undef fdatasync 590# undef fdatasync
918# define fdatasync(fd) fsync (fd) 591# define fdatasync(fd) fsync (fd)
919#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}
920 610
921/* sync_file_range always needs emulation */ 611/* sync_file_range always needs emulation */
922static int 612static int
923eio__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)
924{ 614{
… …
944 /* 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
945 * call fdatasync, as that matches the expectation of its users best */ 635 * call fdatasync, as that matches the expectation of its users best */
946 return fdatasync (fd); 636 return fdatasync (fd);
947} 637}
948 638
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
949#if !HAVE_READAHEAD 649#if !HAVE_READAHEAD
950# undef readahead 650# undef readahead
951# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 651# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
952 652
953static ssize_t 653static eio_ssize_t
954eio__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)
955{ 655{
956 size_t todo = count; 656 size_t todo = count;
957 dBUF; 657 dBUF;
958 658
… …
963 pread (fd, eio_buf, len, offset); 663 pread (fd, eio_buf, len, offset);
964 offset += len; 664 offset += len;
965 todo -= len; 665 todo -= len;
966 } 666 }
967 667
968 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 */
969 return count; 672 return 0;
970} 673}
971 674
972#endif 675#endif
973 676
974/* sendfile always needs emulation */ 677/* sendfile always needs emulation */
975static ssize_t 678static eio_ssize_t
976eio__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)
977{ 680{
978 ssize_t written = 0; 681 eio_ssize_t written = 0;
979 ssize_t res; 682 eio_ssize_t res;
980 683
981 if (!count) 684 if (!count)
982 return 0; 685 return 0;
983 686
984 for (;;) 687 for (;;)
985 { 688 {
689#ifdef __APPLE__
690# undef HAVE_SENDFILE /* broken, as everything on os x */
691#endif
986#if HAVE_SENDFILE 692#if HAVE_SENDFILE
987# if __linux 693# if __linux
988 off_t soffset = offset; 694 off_t soffset = offset;
989 res = sendfile (ofd, ifd, &soffset, count); 695 res = sendfile (ofd, ifd, &soffset, count);
990 696
… …
1006 712
1007 /* according to source inspection, this is correct, and useful behaviour */ 713 /* according to source inspection, this is correct, and useful behaviour */
1008 if (sbytes) 714 if (sbytes)
1009 res = sbytes; 715 res = sbytes;
1010 716
1011# elif defined (__APPLE__) 717# elif defined __APPLE__
1012 off_t sbytes = count; 718 off_t sbytes = count;
1013 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0); 719 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
1014 720
1015 /* according to the manpage, sbytes is always valid */ 721 /* according to the manpage, sbytes is always valid */
1016 if (sbytes) 722 if (sbytes)
… …
1033 if (res < 0 && sbytes) 739 if (res < 0 && sbytes)
1034 res = sbytes; 740 res = sbytes;
1035 741
1036# endif 742# endif
1037 743
1038#elif defined (_WIN32) 744#elif defined (_WIN32) && 0
1039 /* 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 */
1040 /* 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, */
1041 /* libeio guarantees that the file offset does not change, and windows */ 747 /* libeio guarantees that the file offset does not change, and windows */
1042 /* has no way to get an independent handle to the same file description */ 748 /* has no way to get an independent handle to the same file description */
1043 HANDLE h = TO_SOCKET (ifd); 749 HANDLE h = TO_SOCKET (ifd);
1044 SetFilePointer (h, offset, 0, FILE_BEGIN); 750 SetFilePointer (h, offset, 0, FILE_BEGIN);
1045 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0); 751 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1046 752
1047#else 753#else
1048 res = -1; 754 res = EIO_ENOSYS ();
1049 errno = ENOSYS;
1050#endif 755#endif
1051 756
1052 /* we assume sendfile can copy at least 128mb in one go */ 757 /* we assume sendfile can copy at least 128mb in one go */
1053 if (res <= 128 * 1024 * 1024) 758 if (res <= 128 * 1024 * 1024)
1054 { 759 {
… …
1076 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 781 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1077 /* BSDs */ 782 /* BSDs */
1078#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... */
1079 || errno == ENOTSUP 784 || errno == ENOTSUP
1080#endif 785#endif
786#ifdef EOPNOTSUPP /* windows */
1081 || errno == EOPNOTSUPP /* BSDs */ 787 || errno == EOPNOTSUPP /* BSDs */
788#endif
1082#if __solaris 789#if __solaris
1083 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 790 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
1084#endif 791#endif
1085 ) 792 )
1086 ) 793 )
… …
1090 797
1091 res = 0; 798 res = 0;
1092 799
1093 while (count) 800 while (count)
1094 { 801 {
1095 ssize_t cnt; 802 eio_ssize_t cnt;
1096 803
1097 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);
1098 805
1099 if (cnt <= 0) 806 if (cnt <= 0)
1100 { 807 {
… …
1112 819
1113 offset += cnt; 820 offset += cnt;
1114 res += cnt; 821 res += cnt;
1115 count -= cnt; 822 count -= cnt;
1116 } 823 }
824
825 FUBd;
1117 } 826 }
1118 827
1119 return res; 828 return res;
829}
830
831#ifdef PAGESIZE
832# define eio_pagesize() PAGESIZE
833#else
834static intptr_t
835eio_pagesize (void)
836{
837 static intptr_t page;
838
839 if (!page)
840 page = sysconf (_SC_PAGESIZE);
841
842 return page;
843}
844#endif
845
846static void
847eio_page_align (void **addr, size_t *length)
848{
849 intptr_t mask = eio_pagesize () - 1;
850
851 /* round down addr */
852 intptr_t adj = mask & (intptr_t)*addr;
853
854 *addr = (void *)((intptr_t)*addr - adj);
855 *length += adj;
856
857 /* round up length */
858 *length = (*length + mask) & ~mask;
859}
860
861#if !_POSIX_MEMLOCK
862# define eio__mlockall(a) EIO_ENOSYS ()
863#else
864
865static int
866eio__mlockall (int flags)
867{
868 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
869 extern int mallopt (int, int);
870 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
871 #endif
872
873 if (EIO_MCL_CURRENT != MCL_CURRENT
874 || EIO_MCL_FUTURE != MCL_FUTURE)
875 {
876 flags = 0
877 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
878 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
879 }
880
881 return mlockall (flags);
882}
883#endif
884
885#if !_POSIX_MEMLOCK_RANGE
886# define eio__mlock(a,b) EIO_ENOSYS ()
887#else
888
889static int
890eio__mlock (void *addr, size_t length)
891{
892 eio_page_align (&addr, &length);
893
894 return mlock (addr, length);
895}
896
897#endif
898
899#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
900# define eio__msync(a,b,c) EIO_ENOSYS ()
901#else
902
903static int
904eio__msync (void *mem, size_t len, int flags)
905{
906 eio_page_align (&mem, &len);
907
908 if (EIO_MS_ASYNC != MS_SYNC
909 || EIO_MS_INVALIDATE != MS_INVALIDATE
910 || EIO_MS_SYNC != MS_SYNC)
911 {
912 flags = 0
913 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
914 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
915 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
916 }
917
918 return msync (mem, len, flags);
919}
920
921#endif
922
923static int
924eio__mtouch (eio_req *req)
925{
926 void *mem = req->ptr2;
927 size_t len = req->size;
928 int flags = req->int1;
929
930 eio_page_align (&mem, &len);
931
932 {
933 intptr_t addr = (intptr_t)mem;
934 intptr_t end = addr + len;
935 intptr_t page = eio_pagesize ();
936
937 if (addr < end)
938 if (flags & EIO_MT_MODIFY) /* modify */
939 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
940 else
941 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
942 }
943
944 return 0;
945}
946
947/*****************************************************************************/
948/* requests implemented outside eio_execute, because they are so large */
949
950static void
951eio__lseek (eio_req *req)
952{
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{
968 char *res;
969 const char *rel = path;
970 char *tmp1, *tmp2;
971#if SYMLOOP_MAX > 32
972 int symlinks = SYMLOOP_MAX;
973#else
974 int symlinks = 32;
975#endif
976
977 errno = EINVAL;
978 if (!rel)
979 return -1;
980
981 errno = ENOENT;
982 if (!*rel)
983 return -1;
984
985 res = etp_tmpbuf_get (tmpbuf, EIO_PATH_MAX * 3);
986#ifdef _WIN32
987 if (_access (rel, 4) != 0)
988 return -1;
989
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
1003 tmp1 = res + EIO_PATH_MAX;
1004 tmp2 = tmp1 + EIO_PATH_MAX;
1005
1006#if 0 /* disabled, the musl way to do things is just too racy */
1007#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1008 /* on linux we may be able to ask the kernel */
1009 {
1010 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1011
1012 if (fd >= 0)
1013 {
1014 sprintf (tmp1, "/proc/self/fd/%d", fd);
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 */
1017 close (fd);
1018
1019 if (req->result > 0)
1020 goto done;
1021 }
1022 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1023 return -1;
1024 }
1025#endif
1026#endif
1027
1028 if (*rel != '/')
1029 {
1030 int len;
1031
1032 errno = ENOENT;
1033 if (wd == EIO_INVALID_WD)
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);
1045
1046 if (res [1]) /* only use if not / */
1047 res += len;
1048 }
1049
1050 while (*rel)
1051 {
1052 eio_ssize_t len, linklen;
1053 const char *beg = rel;
1054
1055 while (*rel && *rel != '/')
1056 ++rel;
1057
1058 len = rel - beg;
1059
1060 if (!len) /* skip slashes */
1061 {
1062 ++rel;
1063 continue;
1064 }
1065
1066 if (beg [0] == '.')
1067 {
1068 if (len == 1)
1069 continue; /* . - nop */
1070
1071 if (beg [1] == '.' && len == 2)
1072 {
1073 /* .. - back up one component, if possible */
1074
1075 while (res != tmpbuf->ptr)
1076 if (*--res == '/')
1077 break;
1078
1079 continue;
1080 }
1081 }
1082
1083 errno = ENAMETOOLONG;
1084 if (res + 1 + len + 1 >= tmp1)
1085 return -1;
1086
1087 /* copy one component */
1088 *res = '/';
1089 memcpy (res + 1, beg, len);
1090
1091 /* zero-terminate, for readlink */
1092 res [len + 1] = 0;
1093
1094 /* now check if it's a symlink */
1095 linklen = readlink (tmpbuf->ptr, tmp1, EIO_PATH_MAX);
1096
1097 if (linklen < 0)
1098 {
1099 if (errno != EINVAL)
1100 return -1;
1101
1102 /* it's a normal directory. hopefully */
1103 res += len + 1;
1104 }
1105 else
1106 {
1107 /* yay, it was a symlink - build new path in tmp2 */
1108 int rellen = strlen (rel);
1109
1110 errno = ENAMETOOLONG;
1111 if (linklen + 1 + rellen >= EIO_PATH_MAX) /* also catch linklen >= EIO_PATH_MAX */
1112 return -1;
1113
1114 errno = ELOOP;
1115 if (!--symlinks)
1116 return -1;
1117
1118 if (*tmp1 == '/')
1119 res = tmpbuf->ptr; /* symlink resolves to an absolute path */
1120
1121 /* we need to be careful, as rel might point into tmp2 already */
1122 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1123 tmp2 [linklen] = '/';
1124 memcpy (tmp2, tmp1, linklen);
1125
1126 rel = tmp2;
1127 }
1128 }
1129
1130 /* special case for the lone root path */
1131 if (res == tmpbuf->ptr)
1132 *res++ = '/';
1133
1134 return res - (char *)tmpbuf->ptr;
1135#endif
1120} 1136}
1121 1137
1122static signed char 1138static signed char
1123eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) 1139eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1124{ 1140{
… …
1132 1148
1133#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 */
1134#define EIO_SORT_FAST 60 /* when to only use insertion sort */ 1150#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1135 1151
1136static void 1152static void
1137eio_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)
1138{ 1154{
1139 unsigned char bits [9 + sizeof (ino_t) * 8]; 1155 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1140 unsigned char *bit = bits; 1156 unsigned char *bit = bits;
1141 1157
1142 assert (CHAR_BIT == 8); 1158 assert (CHAR_BIT == 8);
1143 assert (sizeof (eio_dirent) * 8 < 256); 1159 assert (sizeof (eio_dirent) * 8 < 256);
1144 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */ 1160 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
… …
1146 1162
1147 if (size <= EIO_SORT_FAST) 1163 if (size <= EIO_SORT_FAST)
1148 return; 1164 return;
1149 1165
1150 /* 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 */
1151 /* 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 */
1152 /* inode_bits must contain all inodes ORed together */ 1168 /* inode_bits must contain all inodes ORed together */
1153 /* 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 */
1154 { 1170 {
1155 ino_t endianness; 1171 eio_ino_t endianness;
1156 int i, j; 1172 int i, j;
1157 1173
1158 /* 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" */
1159 for (i = 0; i < sizeof (ino_t); ++i) 1175 for (i = 0; i < sizeof (eio_ino_t); ++i)
1160 ((unsigned char *)&endianness)[i] = i; 1176 ((unsigned char *)&endianness)[i] = i;
1161 1177
1162 *bit++ = 0; 1178 *bit++ = 0;
1163 1179
1164 for (i = 0; i < sizeof (ino_t); ++i) 1180 for (i = 0; i < sizeof (eio_ino_t); ++i)
1165 { 1181 {
1166 /* shifting off the byte offsets out of "endianness" */ 1182 /* shifting off the byte offsets out of "endianness" */
1167 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8; 1183 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1168 endianness >>= 8; 1184 endianness >>= 8;
1169 1185
1170 for (j = 0; j < 8; ++j) 1186 for (j = 0; j < 8; ++j)
1171 if (inode_bits & (((ino_t)1) << (i * 8 + j))) 1187 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1172 *bit++ = offs + j; 1188 *bit++ = offs + j;
1173 } 1189 }
1174 1190
1175 for (j = 0; j < 8; ++j) 1191 for (j = 0; j < 8; ++j)
1176 if (score_bits & (1 << j)) 1192 if (score_bits & (1 << j))
1177 *bit++ = offsetof (eio_dirent, score) * 8 + j; 1193 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1178 } 1194 }
1179 1195
1180 /* now actually do the sorting (a variant of MSD radix sort) */ 1196 /* now actually do the sorting (a variant of MSD radix sort) */
1181 { 1197 {
1182 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base; 1198 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1183 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end; 1199 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1184 unsigned char *bit_stk [9 + sizeof (ino_t) * 8]; 1200 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1185 int stk_idx = 0; 1201 int stk_idx = 0;
1186 1202
1187 base_stk [stk_idx] = dents; 1203 base_stk [stk_idx] = dents;
1188 end_stk [stk_idx] = dents + size; 1204 end_stk [stk_idx] = dents + size;
1189 bit_stk [stk_idx] = bit - 1; 1205 bit_stk [stk_idx] = bit - 1;
… …
1268 } 1284 }
1269 } 1285 }
1270} 1286}
1271 1287
1272static void 1288static void
1273eio_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)
1274{ 1290{
1275 if (size <= 1) 1291 if (size <= 1)
1276 return; /* our insertion sort relies on size > 0 */ 1292 return; /* our insertion sort relies on size > 0 */
1277 1293
1278 /* 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 */
… …
1286 1302
1287/* read a full directory */ 1303/* read a full directory */
1288static void 1304static void
1289eio__scandir (eio_req *req, etp_worker *self) 1305eio__scandir (eio_req *req, etp_worker *self)
1290{ 1306{
1291 DIR *dirp;
1292 EIO_STRUCT_DIRENT *entp;
1293 char *name, *names; 1307 char *name, *names;
1294 int namesalloc = 4096; 1308 int namesalloc = 4096 - sizeof (void *) * 4;
1295 int namesoffs = 0; 1309 int namesoffs = 0;
1296 int flags = req->int1; 1310 int flags = req->int1;
1297 eio_dirent *dents = 0; 1311 eio_dirent *dents = 0;
1298 int dentalloc = 128; 1312 int dentalloc = 128;
1299 int dentoffs = 0; 1313 int dentoffs = 0;
1300 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
1301 1322
1302 req->result = -1; 1323 req->result = -1;
1303 1324
1304 if (!(flags & EIO_READDIR_DENTS)) 1325 if (!(flags & EIO_READDIR_DENTS))
1305 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); 1326 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1306 1327
1307 X_LOCK (wrklock); 1328#ifdef _WIN32
1308 /* 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
1309 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
1396
1397 if (req->flags & EIO_FLAG_PTR1_FREE)
1398 free (req->ptr1);
1310 1399
1311 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1400 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1312 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; 1401 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1313 req->ptr2 = names = malloc (namesalloc); 1402 req->ptr2 = names = malloc (namesalloc);
1314 X_UNLOCK (wrklock);
1315 1403
1316 if (dirp && names && (!flags || dents)) 1404 if (!names || (flags && !dents))
1405 return;
1406
1317 for (;;) 1407 for (;;)
1318 { 1408 {
1409 int done;
1410
1411#ifdef _WIN32
1412 done = !dirp;
1413#else
1319 errno = 0; 1414 errno = 0;
1320 entp = readdir (dirp); 1415 entp = readdir (dirp);
1416 done = !entp;
1417#endif
1321 1418
1322 if (!entp) 1419 if (done)
1323 { 1420 {
1421#ifndef _WIN32
1422 int old_errno = errno;
1423 closedir (dirp);
1424 errno = old_errno;
1425
1324 if (errno) 1426 if (errno)
1325 break; 1427 break;
1428#endif
1326 1429
1327 /* sort etc. */ 1430 /* sort etc. */
1328 req->int1 = flags; 1431 req->int1 = flags;
1329 req->result = dentoffs; 1432 req->result = dentoffs;
1330 1433
1331 if (flags & EIO_READDIR_STAT_ORDER) 1434 if (flags & EIO_READDIR_STAT_ORDER)
1332 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);
1333 else if (flags & EIO_READDIR_DIRS_FIRST) 1436 else if (flags & EIO_READDIR_DIRS_FIRST)
1334 if (flags & EIO_READDIR_FOUND_UNKNOWN) 1437 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1335 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 */
1336 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))
1337 { 1490 {
1338 /* 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
1339 eio_dirent *oth = dents + dentoffs; 1498 ent = dents + dentoffs;
1340 eio_dirent *dir = dents;
1341 1499
1342 /* 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 */
1343 /* by walking from both sides and swapping if necessary */ 1501 ent->namelen = len - 1;
1344 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)
1345 { 1562 {
1346 if (dir->type == EIO_DT_DIR) 1563 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1347 ++dir; 1564 ent->score = 1;
1348 else if ((--oth)->type == EIO_DT_DIR) 1565 else if (!strchr (name, '.')) /* absence of dots indicate likely dirs */
1349 { 1566 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1350 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1351
1352 ++dir;
1353 }
1354 } 1567 }
1355 1568 else if (ent->type == EIO_DT_DIR)
1356 /* now sort the dirs only (dirs all have the same score) */ 1569 ent->score = 0;
1357 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1358 } 1570 }
1359
1360 break;
1361 }
1362
1363 /* now add the entry to our list(s) */
1364 name = entp->d_name;
1365
1366 /* skip . and .. entries */
1367 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1368 {
1369 int len = D_NAMLEN (entp) + 1;
1370
1371 while (ecb_expect_false (namesoffs + len > namesalloc))
1372 {
1373 namesalloc *= 2;
1374 X_LOCK (wrklock);
1375 req->ptr2 = names = realloc (names, namesalloc);
1376 X_UNLOCK (wrklock);
1377
1378 if (!names)
1379 break;
1380 } 1571 }
1381 1572
1382 memcpy (names + namesoffs, name, len);
1383
1384 if (dents)
1385 {
1386 struct eio_dirent *ent;
1387
1388 if (ecb_expect_false (dentoffs == dentalloc))
1389 {
1390 dentalloc *= 2;
1391 X_LOCK (wrklock);
1392 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1393 X_UNLOCK (wrklock);
1394
1395 if (!dents)
1396 break;
1397 }
1398
1399 ent = dents + dentoffs;
1400
1401 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1402 ent->namelen = len - 1;
1403 ent->inode = D_INO (entp);
1404
1405 inode_bits |= ent->inode;
1406
1407 switch (D_TYPE (entp))
1408 {
1409 default:
1410 ent->type = EIO_DT_UNKNOWN;
1411 flags |= EIO_READDIR_FOUND_UNKNOWN;
1412 break;
1413
1414 #ifdef DT_FIFO
1415 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1416 #endif
1417 #ifdef DT_CHR
1418 case DT_CHR: ent->type = EIO_DT_CHR; break;
1419 #endif
1420 #ifdef DT_MPC
1421 case DT_MPC: ent->type = EIO_DT_MPC; break;
1422 #endif
1423 #ifdef DT_DIR
1424 case DT_DIR: ent->type = EIO_DT_DIR; break;
1425 #endif
1426 #ifdef DT_NAM
1427 case DT_NAM: ent->type = EIO_DT_NAM; break;
1428 #endif
1429 #ifdef DT_BLK
1430 case DT_BLK: ent->type = EIO_DT_BLK; break;
1431 #endif
1432 #ifdef DT_MPB
1433 case DT_MPB: ent->type = EIO_DT_MPB; break;
1434 #endif
1435 #ifdef DT_REG
1436 case DT_REG: ent->type = EIO_DT_REG; break;
1437 #endif
1438 #ifdef DT_NWK
1439 case DT_NWK: ent->type = EIO_DT_NWK; break;
1440 #endif
1441 #ifdef DT_CMP
1442 case DT_CMP: ent->type = EIO_DT_CMP; break;
1443 #endif
1444 #ifdef DT_LNK
1445 case DT_LNK: ent->type = EIO_DT_LNK; break;
1446 #endif
1447 #ifdef DT_SOCK
1448 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1449 #endif
1450 #ifdef DT_DOOR
1451 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1452 #endif
1453 #ifdef DT_WHT
1454 case DT_WHT: ent->type = EIO_DT_WHT; break;
1455 #endif
1456 }
1457
1458 ent->score = 7;
1459
1460 if (flags & EIO_READDIR_DIRS_FIRST)
1461 {
1462 if (ent->type == EIO_DT_UNKNOWN)
1463 {
1464 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1465 ent->score = 1;
1466 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1467 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1468 }
1469 else if (ent->type == EIO_DT_DIR)
1470 ent->score = 0;
1471 }
1472 }
1473
1474 namesoffs += len; 1573 namesoffs += len;
1475 ++dentoffs; 1574 ++dentoffs;
1476 } 1575 }
1477 1576
1478 if (EIO_CANCELLED (req)) 1577 if (EIO_CANCELLED (req))
1479 { 1578 {
1480 errno = ECANCELED; 1579 errno = ECANCELED;
1481 break; 1580 break;
1482 } 1581 }
1582
1583#ifdef _WIN32
1584 if (!FindNextFile (dirp, &entp))
1585 {
1586 FindClose (dirp);
1587 dirp = 0;
1483 } 1588 }
1589#endif
1590 }
1484} 1591}
1485 1592
1486#ifdef PAGESIZE 1593/*****************************************************************************/
1487# define eio_pagesize() PAGESIZE 1594/* working directory stuff */
1488#else 1595/* various deficiencies in the posix 2008 api force us to */
1489static intptr_t 1596/* keep the absolute path in string form at all times */
1490eio_pagesize (void) 1597/* fuck yeah. */
1491{
1492 static intptr_t page;
1493 1598
1494 if (!page) 1599#if !HAVE_AT
1495 page = sysconf (_SC_PAGESIZE);
1496 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
1497 return page; 1655 return res;
1498} 1656}
1499#endif
1500 1657
1501static void 1658eio_wd
1502eio_page_align (void **addr, size_t *length) 1659eio_wd_open_sync (eio_wd wd, const char *path)
1503{ 1660{
1504 intptr_t mask = eio_pagesize () - 1; 1661 struct etp_tmpbuf tmpbuf = { };
1662 wd = eio__wd_open_sync (&tmpbuf, wd, path);
1663 free (tmpbuf.ptr);
1505 1664
1506 /* round down addr */ 1665 return wd;
1507 intptr_t adj = mask & (intptr_t)*addr;
1508
1509 *addr = (void *)((intptr_t)*addr - adj);
1510 *length += adj;
1511
1512 /* round up length */
1513 *length = (*length + mask) & ~mask;
1514} 1666}
1515 1667
1516#if !_POSIX_MEMLOCK 1668void
1517# define eio__mlockall(a) ((errno = ENOSYS), -1) 1669eio_wd_close_sync (eio_wd wd)
1518#else 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 */
1519 1683
1520static int 1684static int
1521eio__mlockall (int flags) 1685eio__truncateat (int dirfd, const char *path, off_t length)
1522{ 1686{
1523 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7 1687 int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
1524 extern int mallopt (int, int); 1688 int res;
1525 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1526 #endif
1527 1689
1528 if (EIO_MCL_CURRENT != MCL_CURRENT 1690 if (fd < 0)
1529 || EIO_MCL_FUTURE != MCL_FUTURE) 1691 return fd;
1530 {
1531 flags = 0
1532 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1533 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1534 }
1535 1692
1536 return mlockall (flags); 1693 res = ftruncate (fd, length);
1694 close (fd);
1695 return res;
1537} 1696}
1538#endif
1539
1540#if !_POSIX_MEMLOCK_RANGE
1541# define eio__mlock(a,b) ((errno = ENOSYS), -1)
1542#else
1543 1697
1544static int 1698static int
1545eio__mlock (void *addr, size_t length) 1699eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
1546{ 1700{
1547 eio_page_align (&addr, &length); 1701 int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
1702 int res;
1548 1703
1549 return mlock (addr, length); 1704 if (fd < 0)
1550} 1705 return fd;
1551 1706
1552#endif 1707 res = fstatvfs (fd, buf);
1553 1708 close (fd);
1554#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1555# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
1556#else
1557
1558static int
1559eio__msync (void *mem, size_t len, int flags)
1560{
1561 eio_page_align (&mem, &len);
1562
1563 if (EIO_MS_ASYNC != MS_SYNC
1564 || EIO_MS_INVALIDATE != MS_INVALIDATE
1565 || EIO_MS_SYNC != MS_SYNC)
1566 {
1567 flags = 0
1568 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1569 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1570 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1571 }
1572
1573 return msync (mem, len, flags);
1574}
1575
1576#endif
1577
1578static int
1579eio__mtouch (eio_req *req)
1580{
1581 void *mem = req->ptr2;
1582 size_t len = req->size;
1583 int flags = req->int1;
1584
1585 eio_page_align (&mem, &len);
1586
1587 {
1588 intptr_t addr = (intptr_t)mem;
1589 intptr_t end = addr + len;
1590 intptr_t page = eio_pagesize ();
1591
1592 if (addr < end)
1593 if (flags & EIO_MT_MODIFY) /* modify */
1594 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1595 else
1596 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1597 }
1598
1599 return 0; 1709 return res;
1710
1600} 1711}
1712
1713#endif
1601 1714
1602/*****************************************************************************/ 1715/*****************************************************************************/
1603 1716
1604#define ALLOC(len) \ 1717#define ALLOC(len) \
1605 if (!req->ptr2) \ 1718 if (!req->ptr2) \
1606 { \ 1719 { \
1607 X_LOCK (wrklock); \ 1720 X_LOCK (EIO_POOL->wrklock); \
1608 req->flags |= EIO_FLAG_PTR2_FREE; \ 1721 req->flags |= EIO_FLAG_PTR2_FREE; \
1609 X_UNLOCK (wrklock); \ 1722 X_UNLOCK (EIO_POOL->wrklock); \
1610 req->ptr2 = malloc (len); \ 1723 req->ptr2 = malloc (len); \
1611 if (!req->ptr2) \ 1724 if (!req->ptr2) \
1612 { \ 1725 { \
1613 errno = ENOMEM; \ 1726 errno = ENOMEM; \
1614 req->result = -1; \ 1727 req->result = -1; \
1615 break; \ 1728 break; \
1616 } \ 1729 } \
1617 } 1730 }
1618 1731
1619X_THREAD_PROC (etp_proc)
1620{
1621 ETP_REQ *req;
1622 struct timespec ts;
1623 etp_worker *self = (etp_worker *)thr_arg;
1624
1625 /* try to distribute timeouts somewhat randomly */
1626 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1627
1628 for (;;)
1629 {
1630 X_LOCK (reqlock);
1631
1632 for (;;)
1633 {
1634 self->req = req = reqq_shift (&req_queue);
1635
1636 if (req)
1637 break;
1638
1639 ++idle;
1640
1641 ts.tv_sec = time (0) + idle_timeout;
1642 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
1643 {
1644 if (idle > max_idle)
1645 {
1646 --idle;
1647 X_UNLOCK (reqlock);
1648 X_LOCK (wrklock);
1649 --started;
1650 X_UNLOCK (wrklock);
1651 goto quit;
1652 }
1653
1654 /* we are allowed to idle, so do so without any timeout */
1655 X_COND_WAIT (reqwait, reqlock);
1656 }
1657
1658 --idle;
1659 }
1660
1661 --nready;
1662
1663 X_UNLOCK (reqlock);
1664
1665 if (req->type < 0)
1666 goto quit;
1667
1668 if (!EIO_CANCELLED (req))
1669 ETP_EXECUTE (self, req);
1670
1671 X_LOCK (reslock);
1672
1673 ++npending;
1674
1675 if (!reqq_push (&res_queue, req) && want_poll_cb)
1676 want_poll_cb ();
1677
1678 self->req = 0;
1679 etp_worker_clear (self);
1680
1681 X_UNLOCK (reslock);
1682 }
1683
1684quit:
1685 X_LOCK (wrklock);
1686 etp_worker_free (self);
1687 X_UNLOCK (wrklock);
1688
1689 return 0;
1690}
1691
1692/*****************************************************************************/ 1732/*****************************************************************************/
1693 1733
1694int ecb_cold 1734int ecb_cold
1695eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1735eio_init (void (*want_poll)(void), void (*done_poll)(void))
1696{ 1736{
1697 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);
1698} 1741}
1699 1742
1700ecb_inline void 1743ecb_inline void
1701eio_api_destroy (eio_req *req) 1744eio_api_destroy (eio_req *req)
1702{ 1745{
1703 free (req); 1746 free (req);
1704} 1747}
1705 1748
1706#define REQ(rtype) \ 1749#define REQ(rtype) \
1707 eio_req *req; \ 1750 eio_req *req; \
1708 \ 1751 \
1709 req = (eio_req *)calloc (1, sizeof *req); \ 1752 req = (eio_req *)calloc (1, sizeof *req); \
1710 if (!req) \ 1753 if (!req) \
1711 return 0; \ 1754 return 0; \
… …
1725 { \ 1768 { \
1726 eio_api_destroy (req); \ 1769 eio_api_destroy (req); \
1727 return 0; \ 1770 return 0; \
1728 } 1771 }
1729 1772
1773#define SINGLEDOT(ptr) (0[(char *)(ptr)] == '.' && !1[(char *)(ptr)])
1774
1730static void 1775static void
1731eio_execute (etp_worker *self, eio_req *req) 1776eio_execute (etp_worker *self, eio_req *req)
1732{ 1777{
1778#if HAVE_AT
1779 int dirfd;
1780#else
1781 const char *path;
1782#endif
1783
1784 if (ecb_expect_false (EIO_CANCELLED (req)))
1785 {
1786 req->result = -1;
1787 req->errorno = ECANCELED;
1788 return;
1789 }
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
1733 switch (req->type) 1807 switch (req->type)
1734 { 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;
1735 case EIO_READ: ALLOC (req->size); 1816 case EIO_READ: ALLOC (req->size);
1736 req->result = req->offs >= 0 1817 req->result = req->offs >= 0
1737 ? pread (req->int1, req->ptr2, req->size, req->offs) 1818 ? pread (req->int1, req->ptr2, req->size, req->offs)
1738 : read (req->int1, req->ptr2, req->size); break; 1819 : read (req->int1, req->ptr2, req->size); break;
1739 case EIO_WRITE: req->result = req->offs >= 0 1820 case EIO_WRITE: req->result = req->offs >= 0
1740 ? pwrite (req->int1, req->ptr2, req->size, req->offs) 1821 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1741 : write (req->int1, req->ptr2, req->size); break; 1822 : write (req->int1, req->ptr2, req->size); break;
1742 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
1743 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;
1744 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
1745 1831
1746 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1832 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1747 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;
1748 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));
1749 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
1750 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1950 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1751 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; 1951 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1752 1952
1753 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1754 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1755 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS)); 1953 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1756 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break; 1954 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1757 1955
1758 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1759 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;
1760 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1761 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;
1762 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1763 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break; 1958 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1764 1959
1765 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
1766 case EIO_CLOSE: req->result = close (req->int1); break; 1960 case EIO_CLOSE: req->result = close (req->int1); break;
1767 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break; 1961 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1768 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
1769 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1770 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1771 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1772 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1773 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1774 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1775
1776 case EIO_READLINK: ALLOC (PATH_MAX);
1777 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1778
1779 case EIO_SYNC: req->result = 0; sync (); break; 1962 case EIO_SYNC: req->result = 0; sync (); break;
1780 case EIO_FSYNC: req->result = fsync (req->int1); break; 1963 case EIO_FSYNC: req->result = fsync (req->int1); break;
1781 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;
1782 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;
1783 case EIO_MTOUCH: req->result = eio__mtouch (req); break; 1968 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1784 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;
1785 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break; 1970 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1786 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;
1787 1972
1788 case EIO_READDIR: eio__scandir (req, self); break; 1973 case EIO_READDIR: eio__scandir (req, self); break;
1789 1974
1790 case EIO_BUSY: 1975 case EIO_BUSY:
1791#ifdef _WIN32 1976#ifdef _WIN32
… …
1800 req->result = select (0, 0, 0, 0, &tv); 1985 req->result = select (0, 0, 0, 0, &tv);
1801 } 1986 }
1802#endif 1987#endif
1803 break; 1988 break;
1804 1989
1805 case EIO_UTIME: 1990#if 0
1806 case EIO_FUTIME:
1807 {
1808 struct timeval tv[2];
1809 struct timeval *times;
1810
1811 if (req->nv1 != -1. || req->nv2 != -1.)
1812 {
1813 tv[0].tv_sec = req->nv1;
1814 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
1815 tv[1].tv_sec = req->nv2;
1816 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
1817
1818 times = tv;
1819 }
1820 else
1821 times = 0;
1822
1823 req->result = req->type == EIO_FUTIME
1824 ? futimes (req->int1, times)
1825 : utimes (req->ptr1, times);
1826 }
1827 break;
1828
1829 case EIO_GROUP: 1991 case EIO_GROUP:
1830 abort (); /* handled in eio_request */ 1992 abort (); /* handled in eio_request */
1993#endif
1831 1994
1832 case EIO_NOP: 1995 case EIO_NOP:
1833 req->result = 0; 1996 req->result = 0;
1834 break; 1997 break;
1835 1998
1836 case EIO_CUSTOM: 1999 case EIO_CUSTOM:
1837 req->feed (req); 2000 req->feed (req);
1838 break; 2001 break;
1839 2002
1840 default: 2003 default:
1841 errno = ENOSYS;
1842 req->result = -1; 2004 req->result = EIO_ENOSYS ();
1843 break; 2005 break;
1844 } 2006 }
1845 2007
1846 req->errorno = errno; 2008 req->errorno = errno;
1847} 2009}
1848 2010
1849#ifndef EIO_NO_WRAPPERS 2011#ifndef EIO_NO_WRAPPERS
1850 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
1851eio_req *eio_nop (int pri, eio_cb cb, void *data) 2023eio_req *eio_nop (int pri, eio_cb cb, void *data)
1852{ 2024{
1853 REQ (EIO_NOP); SEND; 2025 REQ (EIO_NOP); SEND;
1854} 2026}
1855 2027
… …
1871eio_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)
1872{ 2044{
1873 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;
1874} 2046}
1875 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
1876eio_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)
1877{ 2064{
1878 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;
1879} 2066}
1880 2067
… …
1886eio_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)
1887{ 2074{
1888 REQ (EIO_MLOCKALL); req->int1 = flags; SEND; 2075 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
1889} 2076}
1890 2077
1891eio_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)
1892{ 2079{
1893 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;
1894}
1895
1896eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1897{
1898 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1899} 2081}
1900 2082
1901eio_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)
1902{ 2084{
1903 REQ (EIO_CLOSE); req->int1 = fd; SEND; 2085 REQ (EIO_CLOSE); req->int1 = fd; SEND;
… …
1906eio_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)
1907{ 2089{
1908 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;
1909} 2091}
1910 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
1911eio_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)
1912{ 2099{
1913 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;
1914} 2101}
1915 2102
1916eio_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)
1917{ 2104{
1918 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;
1919} 2106}
1920 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
1921eio_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)
1922{ 2119{
1923 REQ (EIO_FSTAT); req->int1 = fd; SEND; 2120 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1924} 2121}
1925 2122
… …
1941eio_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)
1942{ 2139{
1943 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2140 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1944} 2141}
1945 2142
1946eio_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)
1947{ 2144{
1948 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;
1949} 2146}
1950 2147
1951eio_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)
… …
1971eio_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)
1972{ 2169{
1973 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2170 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1974} 2171}
1975 2172
1976eio_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)
1977{ 2174{
1978 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;
1979} 2176}
1980 2177
1981eio_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)
… …
1995} 2192}
1996 2193
1997eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data) 2194eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1998{ 2195{
1999 return eio__1path (EIO_READLINK, path, pri, cb, data); 2196 return eio__1path (EIO_READLINK, path, pri, cb, data);
2197}
2198
2199eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2200{
2201 return eio__1path (EIO_REALPATH, path, pri, cb, data);
2000} 2202}
2001 2203
2002eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) 2204eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
2003{ 2205{
2004 return eio__1path (EIO_STAT, path, pri, cb, data); 2206 return eio__1path (EIO_STAT, path, pri, cb, data);
… …
2106void 2308void
2107eio_grp_add (eio_req *grp, eio_req *req) 2309eio_grp_add (eio_req *grp, eio_req *req)
2108{ 2310{
2109 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));
2110 2312
2111 grp->flags |= EIO_FLAG_GROUPADD; 2313 grp->flags |= ETP_FLAG_GROUPADD;
2112 2314
2113 ++grp->size; 2315 ++grp->size;
2114 req->grp = grp; 2316 req->grp = grp;
2115 2317
2116 req->grp_prev = 0; 2318 req->grp_prev = 0;
… …
2123} 2325}
2124 2326
2125/*****************************************************************************/ 2327/*****************************************************************************/
2126/* misc garbage */ 2328/* misc garbage */
2127 2329
2128ssize_t 2330eio_ssize_t
2129eio_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)
2130{ 2332{
2131 etp_worker wrk;
2132 ssize_t ret;
2133
2134 wrk.dbuf = 0;
2135
2136 ret = eio__sendfile (ofd, ifd, offset, count, &wrk); 2333 return eio__sendfile (ofd, ifd, offset, count);
2137
2138 if (wrk.dbuf)
2139 free (wrk.dbuf);
2140
2141 return ret;
2142} 2334}
2143 2335

Diff Legend

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