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

Comparing libeio/eio.c (file contents):
Revision 1.65 by root, Mon May 30 12:56:50 2011 UTC vs.
Revision 1.139 by root, Thu Jun 25 18:14:19 2015 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 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,
… …
40#ifndef _WIN32 40#ifndef _WIN32
41# include "config.h" 41# include "config.h"
42#endif 42#endif
43 43
44#include "eio.h" 44#include "eio.h"
45#include "ecb.h"
45 46
46#ifdef EIO_STACKSIZE 47#ifdef EIO_STACKSIZE
47# define XTHREAD_STACKSIZE EIO_STACKSIZE 48# define X_STACKSIZE EIO_STACKSIZE
48#endif 49#endif
49#include "xthread.h" 50#include "xthread.h"
50 51
51#include <errno.h> 52#include <errno.h>
52#include <stddef.h> 53#include <stddef.h>
53#include <stdlib.h> 54#include <stdlib.h>
54#include <string.h> 55#include <string.h>
55#include <errno.h> 56#include <errno.h>
56#include <sys/types.h> 57#include <sys/types.h>
57#include <sys/stat.h> 58#include <sys/stat.h>
58#include <sys/statvfs.h>
59#include <limits.h> 59#include <limits.h>
60#include <fcntl.h> 60#include <fcntl.h>
61#include <assert.h> 61#include <assert.h>
62 62
63/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
64/* intptr_t only comes from stdint.h, says idiot openbsd coder */
65#if HAVE_STDINT_H
66# include <stdint.h>
67#endif
68
69#ifndef ECANCELED
70# define ECANCELED EDOM
71#endif
72#ifndef ELOOP
73# define ELOOP EDOM
74#endif
75
76#if !defined(ENOTSOCK) && defined(WSAENOTSOCK)
77# define ENOTSOCK WSAENOTSOCK
78#endif
79
80static void eio_destroy (eio_req *req);
81
63#ifndef EIO_FINISH 82#ifndef EIO_FINISH
64# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0 83# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
65#endif 84#endif
66 85
67#ifndef EIO_DESTROY 86#ifndef EIO_DESTROY
… …
70 89
71#ifndef EIO_FEED 90#ifndef EIO_FEED
72# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0) 91# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)
73#endif 92#endif
74 93
94#ifndef EIO_FD_TO_WIN32_HANDLE
95# define EIO_FD_TO_WIN32_HANDLE(fd) _get_osfhandle (fd)
96#endif
97#ifndef EIO_WIN32_HANDLE_TO_FD
98# define EIO_WIN32_HANDLE_TO_FD(handle) _open_osfhandle (handle, 0)
99#endif
100
101#define EIO_ERRNO(errval,retval) ((errno = errval), retval)
102
103#define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1)
104
75#ifdef _WIN32 105#ifdef _WIN32
76 106
77 /*doh*/ 107 #undef PAGESIZE
108 #define PAGESIZE 4096 /* GetSystemInfo? */
109
110 /* TODO: look at how perl does stat (non-sloppy), unlink (ro-files), utime, link */
111
112 #ifdef EIO_STRUCT_STATI64
113 /* look at perl's non-sloppy stat */
114 #define stat(path,buf) _stati64 (path,buf)
115 #define fstat(fd,buf) _fstati64 (fd,buf)
116 #endif
117 #define lstat(path,buf) stat (path,buf)
118 #define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
119 #define mkdir(path,mode) _mkdir (path)
120 #define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
121
122 #define chmod(path,mode) _chmod (path, mode)
123 #define dup(fd) _dup (fd)
124 #define dup2(fd1,fd2) _dup2 (fd1, fd2)
125 #define pipe(fds) _pipe (fds, 4096, O_BINARY)
126
127 #define fchmod(fd,mode) EIO_ENOSYS ()
128 #define chown(path,uid,gid) EIO_ENOSYS ()
129 #define fchown(fd,uid,gid) EIO_ENOSYS ()
130 #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
131 #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
132 #define mknod(path,mode,dev) EIO_ENOSYS ()
133 #define sync() EIO_ENOSYS ()
134 #define readlink(path,buf,s) EIO_ENOSYS ()
135 #define statvfs(path,buf) EIO_ENOSYS ()
136 #define fstatvfs(fd,buf) EIO_ENOSYS ()
137
138 #define pread(fd,buf,count,offset) eio__pread (fd, buf, count, offset)
139 #define pwrite(fd,buf,count,offset) eio__pwrite (fd, buf, count, offset)
140
141 #if __GNUC__
142 typedef long long eio_off_t; /* signed for compatibility to msvc */
143 #else
144 typedef __int64 eio_off_t; /* unsigned not supported by msvc */
145 #endif
146
147 static eio_ssize_t
148 eio__pread (int fd, void *buf, eio_ssize_t count, eio_off_t offset)
149 {
150 OVERLAPPED o = { 0 };
151 DWORD got;
152
153 o.Offset = offset;
154 o.OffsetHigh = offset >> 32;
155
156 return ReadFile ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd), buf, count, &got, &o)
157 ? got : -1;
158 }
159
160 static eio_ssize_t
161 eio__pwrite (int fd, void *buf, eio_ssize_t count, eio_off_t offset)
162 {
163 OVERLAPPED o = { 0 };
164 DWORD got;
165
166 o.Offset = offset;
167 o.OffsetHigh = offset >> 32;
168
169 return WriteFile ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd), buf, count, &got, &o)
170 ? got : -1;
171 }
172
173 /* rename() uses MoveFile, which fails to overwrite */
174 #define rename(old,neu) eio__rename (old, neu)
175
176 static int
177 eio__rename (const char *old, const char *neu)
178 {
179 if (MoveFileEx (old, neu, MOVEFILE_REPLACE_EXISTING))
180 return 0;
181
182 /* should steal _dosmaperr */
183 switch (GetLastError ())
184 {
185 case ERROR_FILE_NOT_FOUND:
186 case ERROR_PATH_NOT_FOUND:
187 case ERROR_INVALID_DRIVE:
188 case ERROR_NO_MORE_FILES:
189 case ERROR_BAD_NETPATH:
190 case ERROR_BAD_NET_NAME:
191 case ERROR_BAD_PATHNAME:
192 case ERROR_FILENAME_EXCED_RANGE:
193 errno = ENOENT;
194 break;
195
196 default:
197 errno = EACCES;
198 break;
199 }
200
201 return -1;
202 }
203
204 /* we could even stat and see if it exists */
205 static int
206 symlink (const char *old, const char *neu)
207 {
208 #if WINVER >= 0x0600
209 if (CreateSymbolicLink (neu, old, 1))
210 return 0;
211
212 if (CreateSymbolicLink (neu, old, 0))
213 return 0;
214 #endif
215
216 return EIO_ERRNO (ENOENT, -1);
217 }
218
219 /* POSIX API only, causing trouble for win32 apps */
220 #define CreateHardLink(neu,old,flags) 0 /* not really creating hardlink, still using relative paths? */
221 #define CreateSymbolicLink(neu,old,flags) 0 /* vista+ only */
222
223 struct statvfs
224 {
225 int dummy;
226 };
227
228 #define DT_DIR EIO_DT_DIR
229 #define DT_REG EIO_DT_REG
230 #define D_NAME(entp) entp.cFileName
231 #define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
232
78#else 233#else
79 234
80# include <sys/time.h> 235 #include <sys/time.h>
81# include <sys/select.h> 236 #include <sys/select.h>
82# include <unistd.h> 237 #include <unistd.h>
238 #include <signal.h>
239 #include <dirent.h>
240
241 #ifdef ANDROID
242 #include <sys/vfs.h>
243 #define statvfs statfs
244 #define fstatvfs fstatfs
245 #include <asm/page.h> /* supposedly limits.h does #define PAGESIZE PAGESIZE */
246 #else
247 #include <sys/statvfs.h>
248 #endif
249
250 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
251 #include <sys/mman.h>
252 #endif
253
254 #define D_NAME(entp) entp->d_name
255
256 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
257 #if __FreeBSD__ || __NetBSD__ || __OpenBSD__
258 #define _DIRENT_HAVE_D_TYPE /* sigh */
259 #define D_INO(de) (de)->d_fileno
260 #define D_NAMLEN(de) (de)->d_namlen
261 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
262 #define D_INO(de) (de)->d_ino
263 #endif
264
265 #ifdef _D_EXACT_NAMLEN
266 #undef D_NAMLEN
267 #define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
268 #endif
269
270 #ifdef _DIRENT_HAVE_D_TYPE
271 #define D_TYPE(de) (de)->d_type
272 #endif
273
274 #ifndef EIO_STRUCT_DIRENT
275 #define EIO_STRUCT_DIRENT struct dirent
276 #endif
277
278#endif
279
280#if HAVE_UTIMES
83# include <utime.h> 281# include <utime.h>
84# include <signal.h>
85# include <dirent.h>
86
87#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
88# include <sys/mman.h>
89#endif
90
91/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
92# if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
93# define _DIRENT_HAVE_D_TYPE /* sigh */
94# define D_INO(de) (de)->d_fileno
95# define D_NAMLEN(de) (de)->d_namlen
96# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
97# define D_INO(de) (de)->d_ino
98# endif 282#endif
99 283
100#ifdef _D_EXACT_NAMLEN 284#if HAVE_SYS_SYSCALL_H
101# undef D_NAMLEN 285# include <sys/syscall.h>
102# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
103#endif
104
105# ifdef _DIRENT_HAVE_D_TYPE
106# define D_TYPE(de) (de)->d_type
107# endif 286#endif
108 287
109# ifndef EIO_STRUCT_DIRENT 288#if HAVE_SYS_PRCTL_H
110# define EIO_STRUCT_DIRENT struct dirent 289# include <sys/prctl.h>
111# endif
112
113#endif 290#endif
114 291
115#if HAVE_SENDFILE 292#if HAVE_SENDFILE
116# if __linux 293# if __linux
117# include <sys/sendfile.h> 294# include <sys/sendfile.h>
… …
132#endif 309#endif
133#ifndef D_INO 310#ifndef D_INO
134# define D_INO(de) 0 311# define D_INO(de) 0
135#endif 312#endif
136#ifndef D_NAMLEN 313#ifndef D_NAMLEN
137# define D_NAMLEN(de) strlen ((de)->d_name) 314# define D_NAMLEN(entp) strlen (D_NAME (entp))
138#endif 315#endif
139 316
140/* used for struct dirent, AIX doesn't provide it */ 317/* used for struct dirent, AIX doesn't provide it */
141#ifndef NAME_MAX 318#ifndef NAME_MAX
142# define NAME_MAX 4096 319# define NAME_MAX 4096
… …
148#endif 325#endif
149 326
150/* buffer size for various temporary buffers */ 327/* buffer size for various temporary buffers */
151#define EIO_BUFSIZE 65536 328#define EIO_BUFSIZE 65536
152 329
153#define dBUF \ 330#define dBUF \
154 char *eio_buf; \
155 ETP_WORKER_LOCK (self); \
156 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 331 char *eio_buf = malloc (EIO_BUFSIZE); \
157 ETP_WORKER_UNLOCK (self); \
158 errno = ENOMEM; \ 332 errno = ENOMEM; \
159 if (!eio_buf) \ 333 if (!eio_buf) \
160 return -1; 334 return -1
161 335
162#define EIO_TICKS ((1000000 + 1023) >> 10) 336#define FUBd \
337 free (eio_buf)
163 338
164/*****************************************************************************/ 339/*****************************************************************************/
165 340
166#if __GNUC__ >= 3 341struct etp_tmpbuf;
167# define expect(expr,value) __builtin_expect ((expr),(value)) 342
343#if _POSIX_VERSION >= 200809L
344 #define HAVE_AT 1
345 #define WD2FD(wd) ((wd) ? (wd)->fd : AT_FDCWD)
346 #ifndef O_SEARCH
347 #define O_SEARCH O_RDONLY
348 #endif
168#else 349#else
169# define expect(expr,value) (expr) 350 #define HAVE_AT 0
351 static const char *wd_expand (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path);
170#endif 352#endif
171 353
172#define expect_false(expr) expect ((expr) != 0, 0) 354struct eio_pwd
173#define expect_true(expr) expect ((expr) != 0, 1) 355{
356#if HAVE_AT
357 int fd;
358#endif
359 int len;
360 char str[1]; /* actually, a 0-terminated canonical path */
361};
174 362
175/*****************************************************************************/ 363/*****************************************************************************/
176 364
177#define ETP_PRI_MIN EIO_PRI_MIN 365#define ETP_PRI_MIN EIO_PRI_MIN
178#define ETP_PRI_MAX EIO_PRI_MAX 366#define ETP_PRI_MAX EIO_PRI_MAX
179 367
368#define ETP_TYPE_QUIT -1
369#define ETP_TYPE_GROUP EIO_GROUP
370
371static void eio_nop_callback (void) { }
372static void (*eio_want_poll_cb)(void) = eio_nop_callback;
373static void (*eio_done_poll_cb)(void) = eio_nop_callback;
374
375#define ETP_WANT_POLL(pool) eio_want_poll_cb ()
376#define ETP_DONE_POLL(pool) eio_done_poll_cb ()
377
180struct etp_worker; 378struct etp_worker;
181
182#define ETP_REQ eio_req 379#define ETP_REQ eio_req
183#define ETP_DESTROY(req) eio_destroy (req) 380#define ETP_DESTROY(req) eio_destroy (req)
184static int eio_finish (eio_req *req); 381static int eio_finish (eio_req *req);
185#define ETP_FINISH(req) eio_finish (req) 382#define ETP_FINISH(req) eio_finish (req)
186static void eio_execute (struct etp_worker *self, eio_req *req); 383static void eio_execute (struct etp_worker *self, eio_req *req);
187#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) 384#define ETP_EXECUTE(wrk,req) eio_execute (wrk, req)
188 385
189#define ETP_WORKER_CLEAR(req) \ 386#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 }
201 387
202#define ETP_WORKER_COMMON \ 388static struct etp_pool eio_pool;
203 void *dbuf; \ 389#define EIO_POOL (&eio_pool)
204 DIR *dirp;
205 390
206/*****************************************************************************/ 391/*****************************************************************************/
207 392
208#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
209
210/* calculate time difference in ~1/EIO_TICKS of a second */
211static int tvdiff (struct timeval *tv1, struct timeval *tv2)
212{
213 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
214 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
215}
216
217static unsigned int started, idle, wanted = 4;
218
219static void (*want_poll_cb) (void);
220static void (*done_poll_cb) (void);
221
222static unsigned int max_poll_time; /* reslock */
223static unsigned int max_poll_reqs; /* reslock */
224
225static volatile unsigned int nreqs; /* reqlock */
226static volatile unsigned int nready; /* reqlock */
227static volatile unsigned int npending; /* reqlock */
228static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
229static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
230
231static xmutex_t wrklock;
232static xmutex_t reslock;
233static xmutex_t reqlock;
234static xcond_t reqwait;
235
236#if !HAVE_PREADWRITE
237/*
238 * make our pread/pwrite emulation safe against themselves, but not against
239 * normal read/write by using a mutex. slows down execution a lot,
240 * but that's your problem, not mine.
241 */
242static xmutex_t preadwritelock = X_MUTEX_INIT;
243#endif
244
245typedef struct etp_worker
246{
247 /* locked by wrklock */
248 struct etp_worker *prev, *next;
249
250 xthread_t tid;
251
252 /* locked by reslock, reqlock or wrklock */
253 ETP_REQ *req; /* currently processed request */
254
255 ETP_WORKER_COMMON
256} etp_worker;
257
258static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */
259
260#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
261#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
262
263/* worker threads management */
264
265static void etp_worker_clear (etp_worker *wrk)
266{
267 ETP_WORKER_CLEAR (wrk);
268}
269
270static void etp_worker_free (etp_worker *wrk)
271{
272 wrk->next->prev = wrk->prev;
273 wrk->prev->next = wrk->next;
274
275 free (wrk);
276}
277
278static unsigned int etp_nreqs (void)
279{
280 int retval;
281 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
282 retval = nreqs;
283 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
284 return retval;
285}
286
287static unsigned int etp_nready (void)
288{
289 unsigned int retval;
290
291 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
292 retval = nready;
293 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
294
295 return retval;
296}
297
298static unsigned int etp_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 etp_nthreads (void)
310{
311 unsigned int retval;
312
313 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
314 retval = started;
315 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
316
317 return retval;
318}
319
320/*
321 * a somewhat faster data structure might be nice, but
322 * with 8 priorities this actually needs <20 insns
323 * per shift, the most expensive operation.
324 */
325typedef struct {
326 ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */
327 int size;
328} etp_reqq;
329
330static etp_reqq req_queue;
331static etp_reqq res_queue;
332
333static int reqq_push (etp_reqq *q, ETP_REQ *req)
334{
335 int pri = req->pri;
336 req->next = 0;
337
338 if (q->qe[pri])
339 {
340 q->qe[pri]->next = req;
341 q->qe[pri] = req;
342 }
343 else
344 q->qe[pri] = q->qs[pri] = req;
345
346 return q->size++;
347}
348
349static ETP_REQ *reqq_shift (etp_reqq *q)
350{
351 int pri;
352
353 if (!q->size)
354 return 0;
355
356 --q->size;
357
358 for (pri = ETP_NUM_PRI; pri--; )
359 {
360 eio_req *req = q->qs[pri];
361
362 if (req)
363 {
364 if (!(q->qs[pri] = (eio_req *)req->next))
365 q->qe[pri] = 0;
366
367 return req;
368 }
369 }
370
371 abort ();
372}
373
374static void etp_thread_init (void)
375{
376 X_MUTEX_CREATE (wrklock);
377 X_MUTEX_CREATE (reslock);
378 X_MUTEX_CREATE (reqlock);
379 X_COND_CREATE (reqwait);
380}
381
382static void etp_atfork_prepare (void)
383{
384 X_LOCK (wrklock);
385 X_LOCK (reqlock);
386 X_LOCK (reslock);
387#if !HAVE_PREADWRITE
388 X_LOCK (preadwritelock);
389#endif
390}
391
392static void etp_atfork_parent (void)
393{
394#if !HAVE_PREADWRITE
395 X_UNLOCK (preadwritelock);
396#endif
397 X_UNLOCK (reslock);
398 X_UNLOCK (reqlock);
399 X_UNLOCK (wrklock);
400}
401
402static void etp_atfork_child (void)
403{
404 ETP_REQ *prv;
405
406 while ((prv = reqq_shift (&req_queue)))
407 ETP_DESTROY (prv);
408
409 while ((prv = reqq_shift (&res_queue)))
410 ETP_DESTROY (prv);
411
412 while (wrk_first.next != &wrk_first)
413 {
414 etp_worker *wrk = wrk_first.next;
415
416 if (wrk->req)
417 ETP_DESTROY (wrk->req);
418
419 etp_worker_clear (wrk);
420 etp_worker_free (wrk);
421 }
422
423 started = 0;
424 idle = 0;
425 nreqs = 0;
426 nready = 0;
427 npending = 0;
428
429 etp_thread_init ();
430}
431
432static void 393static void
433etp_once_init (void)
434{
435 etp_thread_init ();
436 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
437}
438
439static int
440etp_init (void (*want_poll)(void), void (*done_poll)(void))
441{
442 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
443
444 pthread_once (&doinit, etp_once_init);
445
446 want_poll_cb = want_poll;
447 done_poll_cb = done_poll;
448
449 return 0;
450}
451
452X_THREAD_PROC (etp_proc);
453
454static void etp_start_thread (void)
455{
456 etp_worker *wrk = calloc (1, sizeof (etp_worker));
457
458 /*TODO*/
459 assert (("unable to allocate worker thread data", wrk));
460
461 X_LOCK (wrklock);
462
463 if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
464 {
465 wrk->prev = &wrk_first;
466 wrk->next = wrk_first.next;
467 wrk_first.next->prev = wrk;
468 wrk_first.next = wrk;
469 ++started;
470 }
471 else
472 free (wrk);
473
474 X_UNLOCK (wrklock);
475}
476
477static void etp_maybe_start_thread (void)
478{
479 if (expect_true (etp_nthreads () >= wanted))
480 return;
481
482 /* todo: maybe use idle here, but might be less exact */
483 if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
484 return;
485
486 etp_start_thread ();
487}
488
489static void etp_end_thread (void)
490{
491 eio_req *req = calloc (1, sizeof (eio_req));
492
493 req->type = -1;
494 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
495
496 X_LOCK (reqlock);
497 reqq_push (&req_queue, req);
498 X_COND_SIGNAL (reqwait);
499 X_UNLOCK (reqlock);
500
501 X_LOCK (wrklock);
502 --started;
503 X_UNLOCK (wrklock);
504}
505
506static int etp_poll (void)
507{
508 unsigned int maxreqs;
509 unsigned int maxtime;
510 struct timeval tv_start, tv_now;
511
512 X_LOCK (reslock);
513 maxreqs = max_poll_reqs;
514 maxtime = max_poll_time;
515 X_UNLOCK (reslock);
516
517 if (maxtime)
518 gettimeofday (&tv_start, 0);
519
520 for (;;)
521 {
522 ETP_REQ *req;
523
524 etp_maybe_start_thread ();
525
526 X_LOCK (reslock);
527 req = reqq_shift (&res_queue);
528
529 if (req)
530 {
531 --npending;
532
533 if (!res_queue.size && done_poll_cb)
534 done_poll_cb ();
535 }
536
537 X_UNLOCK (reslock);
538
539 if (!req)
540 return 0;
541
542 X_LOCK (reqlock);
543 --nreqs;
544 X_UNLOCK (reqlock);
545
546 if (expect_false (req->type == EIO_GROUP && req->size))
547 {
548 req->int1 = 1; /* mark request as delayed */
549 continue;
550 }
551 else
552 {
553 int res = ETP_FINISH (req);
554 if (expect_false (res))
555 return res;
556 }
557
558 if (expect_false (maxreqs && !--maxreqs))
559 break;
560
561 if (maxtime)
562 {
563 gettimeofday (&tv_now, 0);
564
565 if (tvdiff (&tv_start, &tv_now) >= maxtime)
566 break;
567 }
568 }
569
570 errno = EAGAIN;
571 return -1;
572}
573
574static void etp_cancel (ETP_REQ *req)
575{
576 X_LOCK (wrklock);
577 req->flags |= EIO_FLAG_CANCELLED;
578 X_UNLOCK (wrklock);
579
580 eio_grp_cancel (req);
581}
582
583static void etp_submit (ETP_REQ *req)
584{
585 req->pri -= ETP_PRI_MIN;
586
587 if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
588 if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
589
590 if (expect_false (req->type == EIO_GROUP))
591 {
592 /* I hope this is worth it :/ */
593 X_LOCK (reqlock);
594 ++nreqs;
595 X_UNLOCK (reqlock);
596
597 X_LOCK (reslock);
598
599 ++npending;
600
601 if (!reqq_push (&res_queue, req) && want_poll_cb)
602 want_poll_cb ();
603
604 X_UNLOCK (reslock);
605 }
606 else
607 {
608 X_LOCK (reqlock);
609 ++nreqs;
610 ++nready;
611 reqq_push (&req_queue, req);
612 X_COND_SIGNAL (reqwait);
613 X_UNLOCK (reqlock);
614
615 etp_maybe_start_thread ();
616 }
617}
618
619static void etp_set_max_poll_time (double nseconds)
620{
621 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
622 max_poll_time = nseconds * EIO_TICKS;
623 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
624}
625
626static void etp_set_max_poll_reqs (unsigned int maxreqs)
627{
628 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
629 max_poll_reqs = maxreqs;
630 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
631}
632
633static void etp_set_max_idle (unsigned int nthreads)
634{
635 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
636 max_idle = nthreads;
637 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
638}
639
640static void etp_set_idle_timeout (unsigned int seconds)
641{
642 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
643 idle_timeout = seconds;
644 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
645}
646
647static void etp_set_min_parallel (unsigned int nthreads)
648{
649 if (wanted < nthreads)
650 wanted = nthreads;
651}
652
653static void etp_set_max_parallel (unsigned int nthreads)
654{
655 if (wanted > nthreads)
656 wanted = nthreads;
657
658 while (started > wanted)
659 etp_end_thread ();
660}
661
662/*****************************************************************************/
663
664static void grp_try_feed (eio_req *grp) 394grp_try_feed (eio_req *grp)
665{ 395{
666 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 396 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
667 { 397 {
668 grp->flags &= ~EIO_FLAG_GROUPADD; 398 grp->flags &= ~ETP_FLAG_GROUPADD;
669 399
670 EIO_FEED (grp); 400 EIO_FEED (grp);
671 401
672 /* stop if no progress has been made */ 402 /* stop if no progress has been made */
673 if (!(grp->flags & EIO_FLAG_GROUPADD)) 403 if (!(grp->flags & ETP_FLAG_GROUPADD))
674 { 404 {
675 grp->feed = 0; 405 grp->feed = 0;
676 break; 406 break;
677 } 407 }
678 } 408 }
679} 409}
680 410
411static int
681static int grp_dec (eio_req *grp) 412grp_dec (eio_req *grp)
682{ 413{
683 --grp->size; 414 --grp->size;
684 415
685 /* call feeder, if applicable */ 416 /* call feeder, if applicable */
686 grp_try_feed (grp); 417 grp_try_feed (grp);
687 418
688 /* finish, if done */ 419 /* finish, if done */
689 if (!grp->size && grp->int1) 420 if (!grp->size && grp->flags & ETP_FLAG_DELAYED)
690 return eio_finish (grp); 421 return eio_finish (grp);
691 else 422 else
692 return 0; 423 return 0;
693} 424}
694 425
426static void
695void eio_destroy (eio_req *req) 427eio_destroy (eio_req *req)
696{ 428{
697 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); 429 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
698 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); 430 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
699 431
700 EIO_DESTROY (req); 432 EIO_DESTROY (req);
701} 433}
702 434
435static int
703static int eio_finish (eio_req *req) 436eio_finish (eio_req *req)
704{ 437{
705 int res = EIO_FINISH (req); 438 int res = EIO_FINISH (req);
706 439
707 if (req->grp) 440 if (req->grp)
708 { 441 {
… …
716 if (grp->grp_first == req) 449 if (grp->grp_first == req)
717 grp->grp_first = req->grp_next; 450 grp->grp_first = req->grp_next;
718 451
719 res2 = grp_dec (grp); 452 res2 = grp_dec (grp);
720 453
721 if (!res && res2) 454 if (!res)
722 res = res2; 455 res = res2;
723 } 456 }
724 457
725 eio_destroy (req); 458 eio_destroy (req);
726 459
727 return res; 460 return res;
728} 461}
729 462
463void
730void eio_grp_cancel (eio_req *grp) 464eio_grp_cancel (eio_req *grp)
731{ 465{
732 for (grp = grp->grp_first; grp; grp = grp->grp_next) 466 etp_grp_cancel (EIO_POOL, grp);
733 eio_cancel (grp);
734} 467}
735 468
469void
736void eio_cancel (eio_req *req) 470eio_cancel (eio_req *req)
737{ 471{
738 etp_cancel (req); 472 etp_cancel (EIO_POOL, req);
739} 473}
740 474
475void
741void eio_submit (eio_req *req) 476eio_submit (eio_req *req)
742{ 477{
743 etp_submit (req); 478 etp_submit (EIO_POOL, req);
744} 479}
745 480
746unsigned int eio_nreqs (void) 481unsigned int
482eio_nreqs (void)
747{ 483{
748 return etp_nreqs (); 484 return etp_nreqs (EIO_POOL);
749} 485}
750 486
751unsigned int eio_nready (void) 487unsigned int
488eio_nready (void)
752{ 489{
753 return etp_nready (); 490 return etp_nready (EIO_POOL);
754} 491}
755 492
756unsigned int eio_npending (void) 493unsigned int
494eio_npending (void)
757{ 495{
758 return etp_npending (); 496 return etp_npending (EIO_POOL);
759} 497}
760 498
761unsigned int eio_nthreads (void) 499unsigned int ecb_cold
500eio_nthreads (void)
762{ 501{
763 return etp_nthreads (); 502 return etp_nthreads (EIO_POOL);
764} 503}
765 504
505void ecb_cold
766void eio_set_max_poll_time (double nseconds) 506eio_set_max_poll_time (double nseconds)
767{ 507{
768 etp_set_max_poll_time (nseconds); 508 etp_set_max_poll_time (EIO_POOL, nseconds);
769} 509}
770 510
511void ecb_cold
771void eio_set_max_poll_reqs (unsigned int maxreqs) 512eio_set_max_poll_reqs (unsigned int maxreqs)
772{ 513{
773 etp_set_max_poll_reqs (maxreqs); 514 etp_set_max_poll_reqs (EIO_POOL, maxreqs);
774} 515}
775 516
517void ecb_cold
776void eio_set_max_idle (unsigned int nthreads) 518eio_set_max_idle (unsigned int nthreads)
777{ 519{
778 etp_set_max_idle (nthreads); 520 etp_set_max_idle (EIO_POOL, nthreads);
779} 521}
780 522
523void ecb_cold
781void eio_set_idle_timeout (unsigned int seconds) 524eio_set_idle_timeout (unsigned int seconds)
782{ 525{
783 etp_set_idle_timeout (seconds); 526 etp_set_idle_timeout (EIO_POOL, seconds);
784} 527}
785 528
529void ecb_cold
786void eio_set_min_parallel (unsigned int nthreads) 530eio_set_min_parallel (unsigned int nthreads)
787{ 531{
788 etp_set_min_parallel (nthreads); 532 etp_set_min_parallel (EIO_POOL, nthreads);
789} 533}
790 534
535void ecb_cold
791void eio_set_max_parallel (unsigned int nthreads) 536eio_set_max_parallel (unsigned int nthreads)
792{ 537{
793 etp_set_max_parallel (nthreads); 538 etp_set_max_parallel (EIO_POOL, nthreads);
794} 539}
795 540
796int eio_poll (void) 541int eio_poll (void)
797{ 542{
798 return etp_poll (); 543 return etp_poll (EIO_POOL);
799} 544}
800 545
801/*****************************************************************************/ 546/*****************************************************************************/
802/* work around various missing functions */ 547/* work around various missing functions */
803
804#if !HAVE_PREADWRITE
805# undef pread
806# undef pwrite
807# define pread eio__pread
808# define pwrite eio__pwrite
809
810static ssize_t
811eio__pread (int fd, void *buf, size_t count, off_t offset)
812{
813 ssize_t res;
814 off_t ooffset;
815
816 X_LOCK (preadwritelock);
817 ooffset = lseek (fd, 0, SEEK_CUR);
818 lseek (fd, offset, SEEK_SET);
819 res = read (fd, buf, count);
820 lseek (fd, ooffset, SEEK_SET);
821 X_UNLOCK (preadwritelock);
822
823 return res;
824}
825
826static ssize_t
827eio__pwrite (int fd, void *buf, size_t count, off_t offset)
828{
829 ssize_t res;
830 off_t ooffset;
831
832 X_LOCK (preadwritelock);
833 ooffset = lseek (fd, 0, SEEK_CUR);
834 lseek (fd, offset, SEEK_SET);
835 res = write (fd, buf, count);
836 lseek (fd, ooffset, SEEK_SET);
837 X_UNLOCK (preadwritelock);
838
839 return res;
840}
841#endif
842 548
843#ifndef HAVE_UTIMES 549#ifndef HAVE_UTIMES
844 550
845# undef utimes 551# undef utimes
846# define utimes(path,times) eio__utimes (path, times) 552# define utimes(path,times) eio__utimes (path, times)
… …
866#ifndef HAVE_FUTIMES 572#ifndef HAVE_FUTIMES
867 573
868# undef futimes 574# undef futimes
869# define futimes(fd,times) eio__futimes (fd, times) 575# define futimes(fd,times) eio__futimes (fd, times)
870 576
577static int
871static int eio__futimes (int fd, const struct timeval tv[2]) 578eio__futimes (int fd, const struct timeval tv[2])
872{ 579{
873 errno = ENOSYS; 580 errno = ENOSYS;
874 return -1; 581 return -1;
875} 582}
876 583
… …
879#if !HAVE_FDATASYNC 586#if !HAVE_FDATASYNC
880# undef fdatasync 587# undef fdatasync
881# define fdatasync(fd) fsync (fd) 588# define fdatasync(fd) fsync (fd)
882#endif 589#endif
883 590
591static int
592eio__syncfs (int fd)
593{
594 int res;
595
596#if HAVE_SYS_SYNCFS
597 res = (int)syscall (__NR_syncfs, (int)(fd));
598#else
599 res = EIO_ENOSYS ();
600#endif
601
602 if (res < 0 && errno == ENOSYS && fd >= 0)
603 sync ();
604
605 return res;
606}
607
884/* sync_file_range always needs emulation */ 608/* sync_file_range always needs emulation */
885int 609static int
886eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags) 610eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
887{ 611{
888#if HAVE_SYNC_FILE_RANGE 612#if HAVE_SYNC_FILE_RANGE
889 int res; 613 int res;
890 614
… …
907 /* even though we could play tricks with the flags, it's better to always 631 /* even though we could play tricks with the flags, it's better to always
908 * call fdatasync, as that matches the expectation of its users best */ 632 * call fdatasync, as that matches the expectation of its users best */
909 return fdatasync (fd); 633 return fdatasync (fd);
910} 634}
911 635
636static int
637eio__fallocate (int fd, int mode, off_t offset, size_t len)
638{
639#if HAVE_LINUX_FALLOCATE
640 return fallocate (fd, mode, offset, len);
641#else
642 return EIO_ENOSYS ();
643#endif
644}
645
912#if !HAVE_READAHEAD 646#if !HAVE_READAHEAD
913# undef readahead 647# undef readahead
914# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 648# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
915 649
916static ssize_t 650static eio_ssize_t
917eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) 651eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
918{ 652{
919 size_t todo = count; 653 size_t todo = count;
920 dBUF; 654 dBUF;
921 655
… …
926 pread (fd, eio_buf, len, offset); 660 pread (fd, eio_buf, len, offset);
927 offset += len; 661 offset += len;
928 todo -= len; 662 todo -= len;
929 } 663 }
930 664
931 errno = 0; 665 FUBd;
666
667 /* linux's readahead basically only fails for EBADF or EINVAL (not mmappable) */
668 /* but not for e.g. EIO or eof, so we also never fail */
932 return count; 669 return 0;
933} 670}
934 671
935#endif 672#endif
936 673
937/* sendfile always needs emulation */ 674/* sendfile always needs emulation */
938static ssize_t 675static eio_ssize_t
939eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) 676eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
940{ 677{
678 eio_ssize_t written = 0;
941 ssize_t res; 679 eio_ssize_t res;
942 680
943 if (!count) 681 if (!count)
944 return 0; 682 return 0;
945 683
684 for (;;)
685 {
686#ifdef __APPLE__
687# undef HAVE_SENDFILE /* broken, as everything on os x */
688#endif
946#if HAVE_SENDFILE 689#if HAVE_SENDFILE
947# if __linux 690# if __linux
691 off_t soffset = offset;
948 res = sendfile (ofd, ifd, &offset, count); 692 res = sendfile (ofd, ifd, &soffset, count);
949 693
950# elif __FreeBSD__ 694# elif __FreeBSD__
951 /* 695 /*
952 * Of course, the freebsd sendfile is a dire hack with no thoughts 696 * Of course, the freebsd sendfile is a dire hack with no thoughts
953 * wasted on making it similar to other I/O functions. 697 * wasted on making it similar to other I/O functions.
954 */ 698 */
955 {
956 off_t sbytes; 699 off_t sbytes;
957 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 700 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
958 701
959 #if 0 /* according to the manpage, this is correct, but broken behaviour */ 702 #if 0 /* according to the manpage, this is correct, but broken behaviour */
960 /* freebsd' sendfile will return 0 on success */ 703 /* freebsd' sendfile will return 0 on success */
961 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */ 704 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
962 /* not on e.g. EIO or EPIPE - sounds broken */ 705 /* not on e.g. EIO or EPIPE - sounds broken */
963 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0) 706 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
964 res = sbytes; 707 res = sbytes;
965 #endif 708 #endif
966 709
967 /* according to source inspection, this is correct, and useful behaviour */ 710 /* according to source inspection, this is correct, and useful behaviour */
968 if (sbytes) 711 if (sbytes)
969 res = sbytes; 712 res = sbytes;
970 }
971 713
972# elif defined (__APPLE__) 714# elif defined __APPLE__
973
974 {
975 off_t sbytes = count; 715 off_t sbytes = count;
976 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0); 716 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
977 717
978 /* according to the manpage, sbytes is always valid */ 718 /* according to the manpage, sbytes is always valid */
979 if (sbytes) 719 if (sbytes)
980 res = sbytes; 720 res = sbytes;
981 }
982 721
983# elif __hpux 722# elif __hpux
984 res = sendfile (ofd, ifd, offset, count, 0, 0); 723 res = sendfile (ofd, ifd, offset, count, 0, 0);
985 724
986# elif __solaris 725# elif __solaris
987 {
988 struct sendfilevec vec; 726 struct sendfilevec vec;
989 size_t sbytes; 727 size_t sbytes;
990 728
991 vec.sfv_fd = ifd; 729 vec.sfv_fd = ifd;
992 vec.sfv_flag = 0; 730 vec.sfv_flag = 0;
993 vec.sfv_off = offset; 731 vec.sfv_off = offset;
994 vec.sfv_len = count; 732 vec.sfv_len = count;
995 733
996 res = sendfilev (ofd, &vec, 1, &sbytes); 734 res = sendfilev (ofd, &vec, 1, &sbytes);
997 735
998 if (res < 0 && sbytes) 736 if (res < 0 && sbytes)
999 res = sbytes; 737 res = sbytes;
1000 }
1001 738
1002# endif 739# endif
1003 740
1004#elif defined (_WIN32) 741#elif defined (_WIN32) && 0
1005
1006 /* does not work, just for documentation of what would need to be done */ 742 /* does not work, just for documentation of what would need to be done */
1007 { 743 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
744 /* libeio guarantees that the file offset does not change, and windows */
745 /* has no way to get an independent handle to the same file description */
1008 HANDLE h = TO_SOCKET (ifd); 746 HANDLE h = TO_SOCKET (ifd);
1009 SetFilePointer (h, offset, 0, FILE_BEGIN); 747 SetFilePointer (h, offset, 0, FILE_BEGIN);
1010 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0); 748 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1011 }
1012 749
1013#else 750#else
1014 res = -1; 751 res = EIO_ENOSYS ();
1015 errno = ENOSYS;
1016#endif 752#endif
1017 753
754 /* we assume sendfile can copy at least 128mb in one go */
755 if (res <= 128 * 1024 * 1024)
756 {
757 if (res > 0)
758 written += res;
759
760 if (written)
761 return written;
762
763 break;
764 }
765 else
766 {
767 /* if we requested more, then probably the kernel was lazy */
768 written += res;
769 offset += res;
770 count -= res;
771
772 if (!count)
773 return written;
774 }
775 }
776
1018 if (res < 0 777 if (res < 0
1019 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 778 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1020 /* BSDs */ 779 /* BSDs */
1021#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */ 780#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1022 || errno == ENOTSUP 781 || errno == ENOTSUP
1023#endif 782#endif
783#ifdef EOPNOTSUPP /* windows */
1024 || errno == EOPNOTSUPP /* BSDs */ 784 || errno == EOPNOTSUPP /* BSDs */
785#endif
1025#if __solaris 786#if __solaris
1026 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 787 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
1027#endif 788#endif
1028 ) 789 )
1029 ) 790 )
… …
1033 794
1034 res = 0; 795 res = 0;
1035 796
1036 while (count) 797 while (count)
1037 { 798 {
1038 ssize_t cnt; 799 eio_ssize_t cnt;
1039 800
1040 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 801 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
1041 802
1042 if (cnt <= 0) 803 if (cnt <= 0)
1043 { 804 {
… …
1055 816
1056 offset += cnt; 817 offset += cnt;
1057 res += cnt; 818 res += cnt;
1058 count -= cnt; 819 count -= cnt;
1059 } 820 }
821
822 FUBd;
1060 } 823 }
1061 824
1062 return res; 825 return res;
826}
827
828#ifdef PAGESIZE
829# define eio_pagesize() PAGESIZE
830#else
831static intptr_t
832eio_pagesize (void)
833{
834 static intptr_t page;
835
836 if (!page)
837 page = sysconf (_SC_PAGESIZE);
838
839 return page;
840}
841#endif
842
843static void
844eio_page_align (void **addr, size_t *length)
845{
846 intptr_t mask = eio_pagesize () - 1;
847
848 /* round down addr */
849 intptr_t adj = mask & (intptr_t)*addr;
850
851 *addr = (void *)((intptr_t)*addr - adj);
852 *length += adj;
853
854 /* round up length */
855 *length = (*length + mask) & ~mask;
856}
857
858#if !_POSIX_MEMLOCK
859# define eio__mlockall(a) EIO_ENOSYS ()
860#else
861
862static int
863eio__mlockall (int flags)
864{
865 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
866 extern int mallopt (int, int);
867 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
868 #endif
869
870 if (EIO_MCL_CURRENT != MCL_CURRENT
871 || EIO_MCL_FUTURE != MCL_FUTURE)
872 {
873 flags = 0
874 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
875 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
876 }
877
878 return mlockall (flags);
879}
880#endif
881
882#if !_POSIX_MEMLOCK_RANGE
883# define eio__mlock(a,b) EIO_ENOSYS ()
884#else
885
886static int
887eio__mlock (void *addr, size_t length)
888{
889 eio_page_align (&addr, &length);
890
891 return mlock (addr, length);
892}
893
894#endif
895
896#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
897# define eio__msync(a,b,c) EIO_ENOSYS ()
898#else
899
900static int
901eio__msync (void *mem, size_t len, int flags)
902{
903 eio_page_align (&mem, &len);
904
905 if (EIO_MS_ASYNC != MS_SYNC
906 || EIO_MS_INVALIDATE != MS_INVALIDATE
907 || EIO_MS_SYNC != MS_SYNC)
908 {
909 flags = 0
910 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
911 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
912 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
913 }
914
915 return msync (mem, len, flags);
916}
917
918#endif
919
920static int
921eio__mtouch (eio_req *req)
922{
923 void *mem = req->ptr2;
924 size_t len = req->size;
925 int flags = req->int1;
926
927 eio_page_align (&mem, &len);
928
929 {
930 intptr_t addr = (intptr_t)mem;
931 intptr_t end = addr + len;
932 intptr_t page = eio_pagesize ();
933
934 if (addr < end)
935 if (flags & EIO_MT_MODIFY) /* modify */
936 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
937 else
938 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
939 }
940
941 return 0;
942}
943
944/*****************************************************************************/
945/* requests implemented outside eio_execute, because they are so large */
946
947static void
948eio__lseek (eio_req *req)
949{
950 /* this usually gets optimised away completely, or your compiler sucks, */
951 /* or the whence constants really are not 0, 1, 2 */
952 int whence = req->int2 == EIO_SEEK_SET ? SEEK_SET
953 : req->int2 == EIO_SEEK_CUR ? SEEK_CUR
954 : req->int2 == EIO_SEEK_END ? SEEK_END
955 : req->int2;
956
957 req->offs = lseek (req->int1, req->offs, whence);
958 req->result = req->offs == (off_t)-1 ? -1 : 0;
959}
960
961/* result will always end up in tmpbuf, there is always space for adding a 0-byte */
962static int
963eio__realpath (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
964{
965 char *res;
966 const char *rel = path;
967 char *tmp1, *tmp2;
968#if SYMLOOP_MAX > 32
969 int symlinks = SYMLOOP_MAX;
970#else
971 int symlinks = 32;
972#endif
973
974 errno = EINVAL;
975 if (!rel)
976 return -1;
977
978 errno = ENOENT;
979 if (!*rel)
980 return -1;
981
982 res = etp_tmpbuf_get (tmpbuf, PATH_MAX * 3);
983#ifdef _WIN32
984 if (_access (rel, 4) != 0)
985 return -1;
986
987 symlinks = GetFullPathName (rel, PATH_MAX * 3, res, 0);
988
989 errno = ENAMETOOLONG;
990 if (symlinks >= PATH_MAX * 3)
991 return -1;
992
993 errno = EIO;
994 if (symlinks <= 0)
995 return -1;
996
997 return symlinks;
998
999#else
1000 tmp1 = res + PATH_MAX;
1001 tmp2 = tmp1 + PATH_MAX;
1002
1003#if 0 /* disabled, the musl way to do things is just too racy */
1004#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1005 /* on linux we may be able to ask the kernel */
1006 {
1007 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1008
1009 if (fd >= 0)
1010 {
1011 sprintf (tmp1, "/proc/self/fd/%d", fd);
1012 req->result = readlink (tmp1, res, PATH_MAX);
1013 /* here we should probably stat the open file and the disk file, to make sure they still match */
1014 close (fd);
1015
1016 if (req->result > 0)
1017 goto done;
1018 }
1019 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1020 return -1;
1021 }
1022#endif
1023#endif
1024
1025 if (*rel != '/')
1026 {
1027 int len;
1028
1029 errno = ENOENT;
1030 if (wd == EIO_INVALID_WD)
1031 return -1;
1032
1033 if (wd == EIO_CWD)
1034 {
1035 if (!getcwd (res, PATH_MAX))
1036 return -1;
1037
1038 len = strlen (res);
1039 }
1040 else
1041 memcpy (res, wd->str, len = wd->len);
1042
1043 if (res [1]) /* only use if not / */
1044 res += len;
1045 }
1046
1047 while (*rel)
1048 {
1049 eio_ssize_t len, linklen;
1050 const char *beg = rel;
1051
1052 while (*rel && *rel != '/')
1053 ++rel;
1054
1055 len = rel - beg;
1056
1057 if (!len) /* skip slashes */
1058 {
1059 ++rel;
1060 continue;
1061 }
1062
1063 if (beg [0] == '.')
1064 {
1065 if (len == 1)
1066 continue; /* . - nop */
1067
1068 if (beg [1] == '.' && len == 2)
1069 {
1070 /* .. - back up one component, if possible */
1071
1072 while (res != tmpbuf->ptr)
1073 if (*--res == '/')
1074 break;
1075
1076 continue;
1077 }
1078 }
1079
1080 errno = ENAMETOOLONG;
1081 if (res + 1 + len + 1 >= tmp1)
1082 return -1;
1083
1084 /* copy one component */
1085 *res = '/';
1086 memcpy (res + 1, beg, len);
1087
1088 /* zero-terminate, for readlink */
1089 res [len + 1] = 0;
1090
1091 /* now check if it's a symlink */
1092 linklen = readlink (tmpbuf->ptr, tmp1, PATH_MAX);
1093
1094 if (linklen < 0)
1095 {
1096 if (errno != EINVAL)
1097 return -1;
1098
1099 /* it's a normal directory. hopefully */
1100 res += len + 1;
1101 }
1102 else
1103 {
1104 /* yay, it was a symlink - build new path in tmp2 */
1105 int rellen = strlen (rel);
1106
1107 errno = ENAMETOOLONG;
1108 if (linklen + 1 + rellen >= PATH_MAX)
1109 return -1;
1110
1111 errno = ELOOP;
1112 if (!--symlinks)
1113 return -1;
1114
1115 if (*tmp1 == '/')
1116 res = tmpbuf->ptr; /* symlink resolves to an absolute path */
1117
1118 /* we need to be careful, as rel might point into tmp2 already */
1119 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1120 tmp2 [linklen] = '/';
1121 memcpy (tmp2, tmp1, linklen);
1122
1123 rel = tmp2;
1124 }
1125 }
1126
1127 /* special case for the lone root path */
1128 if (res == tmpbuf->ptr)
1129 *res++ = '/';
1130
1131 return res - (char *)tmpbuf->ptr;
1132#endif
1063} 1133}
1064 1134
1065static signed char 1135static signed char
1066eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) 1136eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1067{ 1137{
… …
1075 1145
1076#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */ 1146#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1077#define EIO_SORT_FAST 60 /* when to only use insertion sort */ 1147#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1078 1148
1079static void 1149static void
1080eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1150eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1081{ 1151{
1082 unsigned char bits [9 + sizeof (ino_t) * 8]; 1152 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1083 unsigned char *bit = bits; 1153 unsigned char *bit = bits;
1084 1154
1085 assert (CHAR_BIT == 8); 1155 assert (CHAR_BIT == 8);
1086 assert (sizeof (eio_dirent) * 8 < 256); 1156 assert (sizeof (eio_dirent) * 8 < 256);
1087 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */ 1157 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
… …
1089 1159
1090 if (size <= EIO_SORT_FAST) 1160 if (size <= EIO_SORT_FAST)
1091 return; 1161 return;
1092 1162
1093 /* first prepare an array of bits to test in our radix sort */ 1163 /* first prepare an array of bits to test in our radix sort */
1094 /* try to take endianness into account, as well as differences in ino_t sizes */ 1164 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1095 /* inode_bits must contain all inodes ORed together */ 1165 /* inode_bits must contain all inodes ORed together */
1096 /* which is used to skip bits that are 0 everywhere, which is very common */ 1166 /* which is used to skip bits that are 0 everywhere, which is very common */
1097 { 1167 {
1098 ino_t endianness; 1168 eio_ino_t endianness;
1099 int i, j; 1169 int i, j;
1100 1170
1101 /* we store the byte offset of byte n into byte n of "endianness" */ 1171 /* we store the byte offset of byte n into byte n of "endianness" */
1102 for (i = 0; i < sizeof (ino_t); ++i) 1172 for (i = 0; i < sizeof (eio_ino_t); ++i)
1103 ((unsigned char *)&endianness)[i] = i; 1173 ((unsigned char *)&endianness)[i] = i;
1104 1174
1105 *bit++ = 0; 1175 *bit++ = 0;
1106 1176
1107 for (i = 0; i < sizeof (ino_t); ++i) 1177 for (i = 0; i < sizeof (eio_ino_t); ++i)
1108 { 1178 {
1109 /* shifting off the byte offsets out of "endianness" */ 1179 /* shifting off the byte offsets out of "endianness" */
1110 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8; 1180 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1111 endianness >>= 8; 1181 endianness >>= 8;
1112 1182
1113 for (j = 0; j < 8; ++j) 1183 for (j = 0; j < 8; ++j)
1114 if (inode_bits & (((ino_t)1) << (i * 8 + j))) 1184 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1115 *bit++ = offs + j; 1185 *bit++ = offs + j;
1116 } 1186 }
1117 1187
1118 for (j = 0; j < 8; ++j) 1188 for (j = 0; j < 8; ++j)
1119 if (score_bits & (1 << j)) 1189 if (score_bits & (1 << j))
1120 *bit++ = offsetof (eio_dirent, score) * 8 + j; 1190 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1121 } 1191 }
1122 1192
1123 /* now actually do the sorting (a variant of MSD radix sort) */ 1193 /* now actually do the sorting (a variant of MSD radix sort) */
1124 { 1194 {
1125 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base; 1195 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1126 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end; 1196 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1127 unsigned char *bit_stk [9 + sizeof (ino_t) * 8]; 1197 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1128 int stk_idx = 0; 1198 int stk_idx = 0;
1129 1199
1130 base_stk [stk_idx] = dents; 1200 base_stk [stk_idx] = dents;
1131 end_stk [stk_idx] = dents + size; 1201 end_stk [stk_idx] = dents + size;
1132 bit_stk [stk_idx] = bit - 1; 1202 bit_stk [stk_idx] = bit - 1;
… …
1211 } 1281 }
1212 } 1282 }
1213} 1283}
1214 1284
1215static void 1285static void
1216eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1286eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1217{ 1287{
1218 if (size <= 1) 1288 if (size <= 1)
1219 return; /* our insertion sort relies on size > 0 */ 1289 return; /* our insertion sort relies on size > 0 */
1220 1290
1221 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */ 1291 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
… …
1229 1299
1230/* read a full directory */ 1300/* read a full directory */
1231static void 1301static void
1232eio__scandir (eio_req *req, etp_worker *self) 1302eio__scandir (eio_req *req, etp_worker *self)
1233{ 1303{
1234 DIR *dirp;
1235 EIO_STRUCT_DIRENT *entp;
1236 char *name, *names; 1304 char *name, *names;
1237 int namesalloc = 4096; 1305 int namesalloc = 4096 - sizeof (void *) * 4;
1238 int namesoffs = 0; 1306 int namesoffs = 0;
1239 int flags = req->int1; 1307 int flags = req->int1;
1240 eio_dirent *dents = 0; 1308 eio_dirent *dents = 0;
1241 int dentalloc = 128; 1309 int dentalloc = 128;
1242 int dentoffs = 0; 1310 int dentoffs = 0;
1243 ino_t inode_bits = 0; 1311 eio_ino_t inode_bits = 0;
1312#ifdef _WIN32
1313 HANDLE dirp;
1314 WIN32_FIND_DATA entp;
1315#else
1316 DIR *dirp;
1317 EIO_STRUCT_DIRENT *entp;
1318#endif
1244 1319
1245 req->result = -1; 1320 req->result = -1;
1246 1321
1247 if (!(flags & EIO_READDIR_DENTS)) 1322 if (!(flags & EIO_READDIR_DENTS))
1248 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); 1323 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1249 1324
1250 X_LOCK (wrklock); 1325#ifdef _WIN32
1251 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 1326 {
1327 int len = strlen ((const char *)req->ptr1);
1328 char *path = malloc (MAX_PATH);
1329 const char *fmt;
1330 const char *reqpath = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1331
1332 if (!len)
1333 fmt = "./*";
1334 else if (reqpath[len - 1] == '/' || reqpath[len - 1] == '\\')
1335 fmt = "%s*";
1336 else
1337 fmt = "%s/*";
1338
1339 _snprintf (path, MAX_PATH, fmt, reqpath);
1340 dirp = FindFirstFile (path, &entp);
1341 free (path);
1342
1343 if (dirp == INVALID_HANDLE_VALUE)
1344 {
1345 /* should steal _dosmaperr */
1346 switch (GetLastError ())
1347 {
1348 case ERROR_FILE_NOT_FOUND:
1349 req->result = 0;
1350 break;
1351
1352 case ERROR_INVALID_NAME:
1353 case ERROR_PATH_NOT_FOUND:
1354 case ERROR_NO_MORE_FILES:
1355 errno = ENOENT;
1356 break;
1357
1358 case ERROR_NOT_ENOUGH_MEMORY:
1359 errno = ENOMEM;
1360 break;
1361
1362 default:
1363 errno = EINVAL;
1364 break;
1365 }
1366
1367 return;
1368 }
1369 }
1370#else
1371 #if HAVE_AT
1372 if (req->wd)
1373 {
1374 int fd = openat (WD2FD (req->wd), req->ptr1, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1375
1376 if (fd < 0)
1377 return;
1378
1379 dirp = fdopendir (fd);
1380
1381 if (!dirp)
1382 close (fd);
1383 }
1384 else
1252 self->dirp = dirp = opendir (req->ptr1); 1385 dirp = opendir (req->ptr1);
1386 #else
1387 dirp = opendir (wd_expand (&self->tmpbuf, req->wd, req->ptr1));
1388 #endif
1389
1390 if (!dirp)
1391 return;
1392#endif
1393
1394 if (req->flags & EIO_FLAG_PTR1_FREE)
1395 free (req->ptr1);
1253 1396
1254 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1397 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1255 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; 1398 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1256 req->ptr2 = names = malloc (namesalloc); 1399 req->ptr2 = names = malloc (namesalloc);
1257 X_UNLOCK (wrklock);
1258 1400
1259 if (dirp && names && (!flags || dents)) 1401 if (!names || (flags && !dents))
1402 return;
1403
1260 for (;;) 1404 for (;;)
1261 { 1405 {
1406 int done;
1407
1408#ifdef _WIN32
1409 done = !dirp;
1410#else
1262 errno = 0; 1411 errno = 0;
1263 entp = readdir (dirp); 1412 entp = readdir (dirp);
1413 done = !entp;
1414#endif
1264 1415
1265 if (!entp) 1416 if (done)
1266 { 1417 {
1418#ifndef _WIN32
1419 int old_errno = errno;
1420 closedir (dirp);
1421 errno = old_errno;
1422
1267 if (errno) 1423 if (errno)
1268 break; 1424 break;
1425#endif
1269 1426
1270 /* sort etc. */ 1427 /* sort etc. */
1271 req->int1 = flags; 1428 req->int1 = flags;
1272 req->result = dentoffs; 1429 req->result = dentoffs;
1273 1430
1274 if (flags & EIO_READDIR_STAT_ORDER) 1431 if (flags & EIO_READDIR_STAT_ORDER)
1275 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits); 1432 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1276 else if (flags & EIO_READDIR_DIRS_FIRST) 1433 else if (flags & EIO_READDIR_DIRS_FIRST)
1277 if (flags & EIO_READDIR_FOUND_UNKNOWN) 1434 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1278 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */ 1435 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1279 else 1436 else
1437 {
1438 /* in this case, all is known, and we just put dirs first and sort them */
1439 eio_dirent *oth = dents + dentoffs;
1440 eio_dirent *dir = dents;
1441
1442 /* now partition dirs to the front, and non-dirs to the back */
1443 /* by walking from both sides and swapping if necessary */
1444 while (oth > dir)
1445 {
1446 if (dir->type == EIO_DT_DIR)
1447 ++dir;
1448 else if ((--oth)->type == EIO_DT_DIR)
1449 {
1450 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1451
1452 ++dir;
1453 }
1454 }
1455
1456 /* now sort the dirs only (dirs all have the same score) */
1457 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1458 }
1459
1460 break;
1461 }
1462
1463 /* now add the entry to our list(s) */
1464 name = D_NAME (entp);
1465
1466 /* skip . and .. entries */
1467 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1468 {
1469 int len = D_NAMLEN (entp) + 1;
1470
1471 while (ecb_expect_false (namesoffs + len > namesalloc))
1472 {
1473 namesalloc *= 2;
1474 req->ptr2 = names = realloc (names, namesalloc);
1475
1476 if (!names)
1477 break;
1478 }
1479
1480 memcpy (names + namesoffs, name, len);
1481
1482 if (dents)
1483 {
1484 struct eio_dirent *ent;
1485
1486 if (ecb_expect_false (dentoffs == dentalloc))
1280 { 1487 {
1281 /* in this case, all is known, and we just put dirs first and sort them */ 1488 dentalloc *= 2;
1489 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1490
1491 if (!dents)
1492 break;
1493 }
1494
1282 eio_dirent *oth = dents + dentoffs; 1495 ent = dents + dentoffs;
1283 eio_dirent *dir = dents;
1284 1496
1285 /* now partition dirs to the front, and non-dirs to the back */ 1497 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1286 /* by walking from both sides and swapping if necessary */ 1498 ent->namelen = len - 1;
1287 while (oth > dir) 1499 ent->inode = D_INO (entp);
1500
1501 inode_bits |= ent->inode;
1502
1503 switch (D_TYPE (entp))
1504 {
1505 default:
1506 ent->type = EIO_DT_UNKNOWN;
1507 flags |= EIO_READDIR_FOUND_UNKNOWN;
1508 break;
1509
1510 #ifdef DT_FIFO
1511 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1512 #endif
1513 #ifdef DT_CHR
1514 case DT_CHR: ent->type = EIO_DT_CHR; break;
1515 #endif
1516 #ifdef DT_MPC
1517 case DT_MPC: ent->type = EIO_DT_MPC; break;
1518 #endif
1519 #ifdef DT_DIR
1520 case DT_DIR: ent->type = EIO_DT_DIR; break;
1521 #endif
1522 #ifdef DT_NAM
1523 case DT_NAM: ent->type = EIO_DT_NAM; break;
1524 #endif
1525 #ifdef DT_BLK
1526 case DT_BLK: ent->type = EIO_DT_BLK; break;
1527 #endif
1528 #ifdef DT_MPB
1529 case DT_MPB: ent->type = EIO_DT_MPB; break;
1530 #endif
1531 #ifdef DT_REG
1532 case DT_REG: ent->type = EIO_DT_REG; break;
1533 #endif
1534 #ifdef DT_NWK
1535 case DT_NWK: ent->type = EIO_DT_NWK; break;
1536 #endif
1537 #ifdef DT_CMP
1538 case DT_CMP: ent->type = EIO_DT_CMP; break;
1539 #endif
1540 #ifdef DT_LNK
1541 case DT_LNK: ent->type = EIO_DT_LNK; break;
1542 #endif
1543 #ifdef DT_SOCK
1544 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1545 #endif
1546 #ifdef DT_DOOR
1547 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1548 #endif
1549 #ifdef DT_WHT
1550 case DT_WHT: ent->type = EIO_DT_WHT; break;
1551 #endif
1552 }
1553
1554 ent->score = 7;
1555
1556 if (flags & EIO_READDIR_DIRS_FIRST)
1557 {
1558 if (ent->type == EIO_DT_UNKNOWN)
1288 { 1559 {
1289 if (dir->type == EIO_DT_DIR) 1560 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1290 ++dir; 1561 ent->score = 1;
1291 else if ((--oth)->type == EIO_DT_DIR) 1562 else if (!strchr (name, '.')) /* absence of dots indicate likely dirs */
1292 { 1563 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1293 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1294
1295 ++dir;
1296 }
1297 } 1564 }
1298 1565 else if (ent->type == EIO_DT_DIR)
1299 /* now sort the dirs only (dirs all have the same score) */ 1566 ent->score = 0;
1300 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1301 } 1567 }
1302
1303 break;
1304 }
1305
1306 /* now add the entry to our list(s) */
1307 name = entp->d_name;
1308
1309 /* skip . and .. entries */
1310 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1311 {
1312 int len = D_NAMLEN (entp) + 1;
1313
1314 while (expect_false (namesoffs + len > namesalloc))
1315 {
1316 namesalloc *= 2;
1317 X_LOCK (wrklock);
1318 req->ptr2 = names = realloc (names, namesalloc);
1319 X_UNLOCK (wrklock);
1320
1321 if (!names)
1322 break;
1323 } 1568 }
1324 1569
1325 memcpy (names + namesoffs, name, len);
1326
1327 if (dents)
1328 {
1329 struct eio_dirent *ent;
1330
1331 if (expect_false (dentoffs == dentalloc))
1332 {
1333 dentalloc *= 2;
1334 X_LOCK (wrklock);
1335 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1336 X_UNLOCK (wrklock);
1337
1338 if (!dents)
1339 break;
1340 }
1341
1342 ent = dents + dentoffs;
1343
1344 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1345 ent->namelen = len - 1;
1346 ent->inode = D_INO (entp);
1347
1348 inode_bits |= ent->inode;
1349
1350 switch (D_TYPE (entp))
1351 {
1352 default:
1353 ent->type = EIO_DT_UNKNOWN;
1354 flags |= EIO_READDIR_FOUND_UNKNOWN;
1355 break;
1356
1357 #ifdef DT_FIFO
1358 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1359 #endif
1360 #ifdef DT_CHR
1361 case DT_CHR: ent->type = EIO_DT_CHR; break;
1362 #endif
1363 #ifdef DT_MPC
1364 case DT_MPC: ent->type = EIO_DT_MPC; break;
1365 #endif
1366 #ifdef DT_DIR
1367 case DT_DIR: ent->type = EIO_DT_DIR; break;
1368 #endif
1369 #ifdef DT_NAM
1370 case DT_NAM: ent->type = EIO_DT_NAM; break;
1371 #endif
1372 #ifdef DT_BLK
1373 case DT_BLK: ent->type = EIO_DT_BLK; break;
1374 #endif
1375 #ifdef DT_MPB
1376 case DT_MPB: ent->type = EIO_DT_MPB; break;
1377 #endif
1378 #ifdef DT_REG
1379 case DT_REG: ent->type = EIO_DT_REG; break;
1380 #endif
1381 #ifdef DT_NWK
1382 case DT_NWK: ent->type = EIO_DT_NWK; break;
1383 #endif
1384 #ifdef DT_CMP
1385 case DT_CMP: ent->type = EIO_DT_CMP; break;
1386 #endif
1387 #ifdef DT_LNK
1388 case DT_LNK: ent->type = EIO_DT_LNK; break;
1389 #endif
1390 #ifdef DT_SOCK
1391 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1392 #endif
1393 #ifdef DT_DOOR
1394 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1395 #endif
1396 #ifdef DT_WHT
1397 case DT_WHT: ent->type = EIO_DT_WHT; break;
1398 #endif
1399 }
1400
1401 ent->score = 7;
1402
1403 if (flags & EIO_READDIR_DIRS_FIRST)
1404 {
1405 if (ent->type == EIO_DT_UNKNOWN)
1406 {
1407 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1408 ent->score = 1;
1409 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1410 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1411 }
1412 else if (ent->type == EIO_DT_DIR)
1413 ent->score = 0;
1414 }
1415 }
1416
1417 namesoffs += len; 1570 namesoffs += len;
1418 ++dentoffs; 1571 ++dentoffs;
1419 } 1572 }
1420 1573
1421 if (EIO_CANCELLED (req)) 1574 if (EIO_CANCELLED (req))
1422 { 1575 {
1423 errno = ECANCELED; 1576 errno = ECANCELED;
1424 break; 1577 break;
1425 } 1578 }
1579
1580#ifdef _WIN32
1581 if (!FindNextFile (dirp, &entp))
1582 {
1583 FindClose (dirp);
1584 dirp = 0;
1426 } 1585 }
1586#endif
1587 }
1427} 1588}
1428 1589
1429#ifdef PAGESIZE 1590/*****************************************************************************/
1430# define eio_pagesize() PAGESIZE 1591/* working directory stuff */
1431#else 1592/* various deficiencies in the posix 2008 api force us to */
1432static intptr_t 1593/* keep the absolute path in string form at all times */
1433eio_pagesize (void) 1594/* fuck yeah. */
1434{
1435 static intptr_t page;
1436 1595
1437 if (!page) 1596#if !HAVE_AT
1438 page = sysconf (_SC_PAGESIZE);
1439 1597
1598/* a bit like realpath, but usually faster because it doesn'T have to return */
1599/* an absolute or canonical path */
1600static const char *
1601wd_expand (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
1602{
1603 if (!wd || *path == '/')
1604 return path;
1605
1606 if (path [0] == '.' && !path [1])
1607 return wd->str;
1608
1609 {
1610 int l1 = wd->len;
1611 int l2 = strlen (path);
1612
1613 char *res = etp_tmpbuf_get (tmpbuf, l1 + l2 + 2);
1614
1615 memcpy (res, wd->str, l1);
1616 res [l1] = '/';
1617 memcpy (res + l1 + 1, path, l2 + 1);
1618
1619 return res;
1620 }
1621}
1622
1623#endif
1624
1625static eio_wd
1626eio__wd_open_sync (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
1627{
1628 int fd;
1629 eio_wd res;
1630 int len = eio__realpath (tmpbuf, wd, path);
1631
1632 if (len < 0)
1633 return EIO_INVALID_WD;
1634
1635#if HAVE_AT
1636 fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1637
1638 if (fd < 0)
1639 return EIO_INVALID_WD;
1640#endif
1641
1642 res = malloc (sizeof (*res) + len); /* one extra 0-byte */
1643
1644#if HAVE_AT
1645 res->fd = fd;
1646#endif
1647
1648 res->len = len;
1649 memcpy (res->str, tmpbuf->ptr, len);
1650 res->str [len] = 0;
1651
1440 return page; 1652 return res;
1441} 1653}
1442#endif
1443 1654
1444static void 1655eio_wd
1445eio_page_align (void **addr, size_t *length) 1656eio_wd_open_sync (eio_wd wd, const char *path)
1446{ 1657{
1447 intptr_t mask = eio_pagesize () - 1; 1658 struct etp_tmpbuf tmpbuf = { };
1659 wd = eio__wd_open_sync (&tmpbuf, wd, path);
1660 free (tmpbuf.ptr);
1448 1661
1449 /* round down addr */ 1662 return wd;
1450 intptr_t adj = mask & (intptr_t)*addr;
1451
1452 *addr = (void *)((intptr_t)*addr - adj);
1453 *length += adj;
1454
1455 /* round up length */
1456 *length = (*length + mask) & ~mask;
1457} 1663}
1458 1664
1459#if !_POSIX_MEMLOCK 1665void
1460# define eio__mlockall(a) ((errno = ENOSYS), -1) 1666eio_wd_close_sync (eio_wd wd)
1461#else 1667{
1668 if (wd != EIO_INVALID_WD && wd != EIO_CWD)
1669 {
1670 #if HAVE_AT
1671 close (wd->fd);
1672 #endif
1673 free (wd);
1674 }
1675}
1676
1677#if HAVE_AT
1678
1679/* they forgot these */
1462 1680
1463static int 1681static int
1464eio__mlockall (int flags) 1682eio__truncateat (int dirfd, const char *path, off_t length)
1465{ 1683{
1466 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7 1684 int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
1467 extern int mallopt (int, int); 1685 int res;
1468 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1469 #endif
1470 1686
1471 if (EIO_MCL_CURRENT != MCL_CURRENT 1687 if (fd < 0)
1472 || EIO_MCL_FUTURE != MCL_FUTURE) 1688 return fd;
1473 {
1474 flags = 0
1475 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1476 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1477 }
1478 1689
1479 return mlockall (flags); 1690 res = ftruncate (fd, length);
1691 close (fd);
1692 return res;
1480} 1693}
1481#endif
1482
1483#if !_POSIX_MEMLOCK_RANGE
1484# define eio__mlock(a,b) ((errno = ENOSYS), -1)
1485#else
1486 1694
1487static int 1695static int
1488eio__mlock (void *addr, size_t length) 1696eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
1489{ 1697{
1490 eio_page_align (&addr, &length); 1698 int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
1699 int res;
1491 1700
1492 return mlock (addr, length); 1701 if (fd < 0)
1493} 1702 return fd;
1494 1703
1495#endif 1704 res = fstatvfs (fd, buf);
1496 1705 close (fd);
1497#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1498# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
1499#else
1500
1501int
1502eio__msync (void *mem, size_t len, int flags)
1503{
1504 eio_page_align (&mem, &len);
1505
1506 if (EIO_MS_ASYNC != MS_SYNC
1507 || EIO_MS_INVALIDATE != MS_INVALIDATE
1508 || EIO_MS_SYNC != MS_SYNC)
1509 {
1510 flags = 0
1511 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1512 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1513 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1514 }
1515
1516 return msync (mem, len, flags);
1517}
1518
1519#endif
1520
1521int
1522eio__mtouch (void *mem, size_t len, int flags)
1523{
1524 eio_page_align (&mem, &len);
1525
1526 {
1527 intptr_t addr = (intptr_t)mem;
1528 intptr_t end = addr + len;
1529 intptr_t page = eio_pagesize ();
1530
1531 if (addr < end)
1532 if (flags & EIO_MT_MODIFY) /* modify */
1533 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
1534 else
1535 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
1536 }
1537
1538 return 0; 1706 return res;
1707
1539} 1708}
1709
1710#endif
1540 1711
1541/*****************************************************************************/ 1712/*****************************************************************************/
1542 1713
1543#define ALLOC(len) \ 1714#define ALLOC(len) \
1544 if (!req->ptr2) \ 1715 if (!req->ptr2) \
1545 { \ 1716 { \
1546 X_LOCK (wrklock); \ 1717 X_LOCK (EIO_POOL->wrklock); \
1547 req->flags |= EIO_FLAG_PTR2_FREE; \ 1718 req->flags |= EIO_FLAG_PTR2_FREE; \
1548 X_UNLOCK (wrklock); \ 1719 X_UNLOCK (EIO_POOL->wrklock); \
1549 req->ptr2 = malloc (len); \ 1720 req->ptr2 = malloc (len); \
1550 if (!req->ptr2) \ 1721 if (!req->ptr2) \
1551 { \ 1722 { \
1552 errno = ENOMEM; \ 1723 errno = ENOMEM; \
1553 req->result = -1; \ 1724 req->result = -1; \
1554 break; \ 1725 break; \
1555 } \ 1726 } \
1556 } 1727 }
1557 1728
1558X_THREAD_PROC (etp_proc)
1559{
1560 ETP_REQ *req;
1561 struct timespec ts;
1562 etp_worker *self = (etp_worker *)thr_arg;
1563
1564 /* try to distribute timeouts somewhat randomly */
1565 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1566
1567 for (;;)
1568 {
1569 X_LOCK (reqlock);
1570
1571 for (;;)
1572 {
1573 self->req = req = reqq_shift (&req_queue);
1574
1575 if (req)
1576 break;
1577
1578 ++idle;
1579
1580 ts.tv_sec = time (0) + idle_timeout;
1581 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
1582 {
1583 if (idle > max_idle)
1584 {
1585 --idle;
1586 X_UNLOCK (reqlock);
1587 X_LOCK (wrklock);
1588 --started;
1589 X_UNLOCK (wrklock);
1590 goto quit;
1591 }
1592
1593 /* we are allowed to idle, so do so without any timeout */
1594 X_COND_WAIT (reqwait, reqlock);
1595 }
1596
1597 --idle;
1598 }
1599
1600 --nready;
1601
1602 X_UNLOCK (reqlock);
1603
1604 if (req->type < 0)
1605 goto quit;
1606
1607 if (!EIO_CANCELLED (req))
1608 ETP_EXECUTE (self, req);
1609
1610 X_LOCK (reslock);
1611
1612 ++npending;
1613
1614 if (!reqq_push (&res_queue, req) && want_poll_cb)
1615 want_poll_cb ();
1616
1617 self->req = 0;
1618 etp_worker_clear (self);
1619
1620 X_UNLOCK (reslock);
1621 }
1622
1623quit:
1624 X_LOCK (wrklock);
1625 etp_worker_free (self);
1626 X_UNLOCK (wrklock);
1627
1628 return 0;
1629}
1630
1631/*****************************************************************************/ 1729/*****************************************************************************/
1632 1730
1731int ecb_cold
1633int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1732eio_init (void (*want_poll)(void), void (*done_poll)(void))
1634{ 1733{
1635 return etp_init (want_poll, done_poll); 1734 eio_want_poll_cb = want_poll;
1636} 1735 eio_done_poll_cb = done_poll;
1637 1736
1737 return etp_init (EIO_POOL, 0, 0, 0);
1738}
1739
1740ecb_inline void
1638static void eio_api_destroy (eio_req *req) 1741eio_api_destroy (eio_req *req)
1639{ 1742{
1640 free (req); 1743 free (req);
1641} 1744}
1642 1745
1643#define REQ(rtype) \ 1746#define REQ(rtype) \
1644 eio_req *req; \ 1747 eio_req *req; \
1645 \ 1748 \
1646 req = (eio_req *)calloc (1, sizeof *req); \ 1749 req = (eio_req *)calloc (1, sizeof *req); \
1647 if (!req) \ 1750 if (!req) \
1648 return 0; \ 1751 return 0; \
… …
1662 { \ 1765 { \
1663 eio_api_destroy (req); \ 1766 eio_api_destroy (req); \
1664 return 0; \ 1767 return 0; \
1665 } 1768 }
1666 1769
1770#define SINGLEDOT(ptr) (0[(char *)(ptr)] == '.' && !1[(char *)(ptr)])
1771
1772static void
1667static void eio_execute (etp_worker *self, eio_req *req) 1773eio_execute (etp_worker *self, eio_req *req)
1668{ 1774{
1775#if HAVE_AT
1776 int dirfd;
1777#else
1778 const char *path;
1779#endif
1780
1781 if (ecb_expect_false (EIO_CANCELLED (req)))
1782 {
1783 req->result = -1;
1784 req->errorno = ECANCELED;
1785 return;
1786 }
1787
1788 if (ecb_expect_false (req->wd == EIO_INVALID_WD))
1789 {
1790 req->result = -1;
1791 req->errorno = ENOENT;
1792 return;
1793 }
1794
1795 if (req->type >= EIO_OPEN)
1796 {
1797 #if HAVE_AT
1798 dirfd = WD2FD (req->wd);
1799 #else
1800 path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1801 #endif
1802 }
1803
1669 switch (req->type) 1804 switch (req->type)
1670 { 1805 {
1806 case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
1807 req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
1808 break;
1809 case EIO_WD_CLOSE: req->result = 0;
1810 eio_wd_close_sync (req->wd); break;
1811
1812 case EIO_SEEK: eio__lseek (req); break;
1671 case EIO_READ: ALLOC (req->size); 1813 case EIO_READ: ALLOC (req->size);
1672 req->result = req->offs >= 0 1814 req->result = req->offs >= 0
1673 ? pread (req->int1, req->ptr2, req->size, req->offs) 1815 ? pread (req->int1, req->ptr2, req->size, req->offs)
1674 : read (req->int1, req->ptr2, req->size); break; 1816 : read (req->int1, req->ptr2, req->size); break;
1675 case EIO_WRITE: req->result = req->offs >= 0 1817 case EIO_WRITE: req->result = req->offs >= 0
1676 ? pwrite (req->int1, req->ptr2, req->size, req->offs) 1818 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1677 : write (req->int1, req->ptr2, req->size); break; 1819 : write (req->int1, req->ptr2, req->size); break;
1678 1820
1679 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break; 1821 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1680 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break; 1822 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
1823
1824#if HAVE_AT
1681 1825
1682 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1826 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1683 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 1827 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break;
1684 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1828 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1829 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, AT_SYMLINK_NOFOLLOW); break;
1830 case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break;
1831 case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break;
1832 case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break;
1833 case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break;
1834
1835 case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
1836 case EIO_RMDIR: /* complications arise because "." cannot be removed, so we might have to expand */
1837 req->result = req->wd && SINGLEDOT (req->ptr1)
1838 ? rmdir (req->wd->str)
1839 : unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break;
1840 case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); break;
1841 case EIO_RENAME: /* complications arise because "." cannot be renamed, so we might have to expand */
1842 req->result = req->wd && SINGLEDOT (req->ptr1)
1843 ? rename (req->wd->str, req->ptr2)
1844 : renameat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2); break;
1845 case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2, 0); break;
1846 case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
1847 case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1848 case EIO_READLINK: ALLOC (PATH_MAX);
1849 req->result = readlinkat (dirfd, req->ptr1, req->ptr2, PATH_MAX); break;
1850 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1851 req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1852 case EIO_UTIME:
1853 case EIO_FUTIME:
1854 {
1855 struct timespec ts[2];
1856 struct timespec *times;
1857
1858 if (req->nv1 != -1. || req->nv2 != -1.)
1859 {
1860 ts[0].tv_sec = req->nv1;
1861 ts[0].tv_nsec = (req->nv1 - ts[0].tv_sec) * 1e9;
1862 ts[1].tv_sec = req->nv2;
1863 ts[1].tv_nsec = (req->nv2 - ts[1].tv_sec) * 1e9;
1864
1865 times = ts;
1866 }
1867 else
1868 times = 0;
1869
1870 req->result = req->type == EIO_FUTIME
1871 ? futimens (req->int1, times)
1872 : utimensat (dirfd, req->ptr1, times, 0);
1873 }
1874 break;
1875
1876#else
1877
1878 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1879 req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
1880 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1685 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 1881 req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
1882 case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
1883 case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
1884 case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
1885 case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
1886
1887 case EIO_UNLINK: req->result = unlink (path ); break;
1888 case EIO_RMDIR: req->result = rmdir (path ); break;
1889 case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
1890 case EIO_RENAME: req->result = rename (path , req->ptr2); break;
1891 case EIO_LINK: req->result = link (path , req->ptr2); break;
1892 case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
1893 case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
1894 case EIO_READLINK: ALLOC (PATH_MAX);
1895 req->result = readlink (path, req->ptr2, PATH_MAX); break;
1896 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1897 req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
1898
1899 case EIO_UTIME:
1900 case EIO_FUTIME:
1901 {
1902 struct timeval tv[2];
1903 struct timeval *times;
1904
1905 if (req->nv1 != -1. || req->nv2 != -1.)
1906 {
1907 tv[0].tv_sec = req->nv1;
1908 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
1909 tv[1].tv_sec = req->nv2;
1910 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
1911
1912 times = tv;
1913 }
1914 else
1915 times = 0;
1916
1917 req->result = req->type == EIO_FUTIME
1918 ? futimes (req->int1, times)
1919 : utimes (req->ptr1, times);
1920 }
1921 break;
1922
1923#endif
1924
1925 case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
1926 {
1927 ALLOC (req->result);
1928 memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
1929 }
1930 break;
1931
1686 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1932 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1687 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; 1933 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1688 1934
1689 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1690 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1691 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS)); 1935 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1692 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break; 1936 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1693 1937
1694 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1695 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; 1938 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1696 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1697 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break; 1939 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1698 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1699 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break; 1940 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1700 1941
1701 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
1702 case EIO_CLOSE: req->result = close (req->int1); break; 1942 case EIO_CLOSE: req->result = close (req->int1); break;
1703 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break; 1943 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1704 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
1705 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1706 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1707 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1708 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1709 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1710 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1711
1712 case EIO_READLINK: ALLOC (PATH_MAX);
1713 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1714
1715 case EIO_SYNC: req->result = 0; sync (); break; 1944 case EIO_SYNC: req->result = 0; sync (); break;
1716 case EIO_FSYNC: req->result = fsync (req->int1); break; 1945 case EIO_FSYNC: req->result = fsync (req->int1); break;
1717 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 1946 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1947 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
1948 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1718 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break; 1949 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1719 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break; 1950 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1720 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break; 1951 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1721 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break; 1952 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1722 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break; 1953 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
1723 1954
1724 case EIO_READDIR: eio__scandir (req, self); break; 1955 case EIO_READDIR: eio__scandir (req, self); break;
1725 1956
1726 case EIO_BUSY: 1957 case EIO_BUSY:
1727#ifdef _WIN32 1958#ifdef _WIN32
… …
1736 req->result = select (0, 0, 0, 0, &tv); 1967 req->result = select (0, 0, 0, 0, &tv);
1737 } 1968 }
1738#endif 1969#endif
1739 break; 1970 break;
1740 1971
1741 case EIO_UTIME: 1972#if 0
1742 case EIO_FUTIME:
1743 {
1744 struct timeval tv[2];
1745 struct timeval *times;
1746
1747 if (req->nv1 != -1. || req->nv2 != -1.)
1748 {
1749 tv[0].tv_sec = req->nv1;
1750 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
1751 tv[1].tv_sec = req->nv2;
1752 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
1753
1754 times = tv;
1755 }
1756 else
1757 times = 0;
1758
1759 req->result = req->type == EIO_FUTIME
1760 ? futimes (req->int1, times)
1761 : utimes (req->ptr1, times);
1762 }
1763 break;
1764
1765 case EIO_GROUP: 1973 case EIO_GROUP:
1766 abort (); /* handled in eio_request */ 1974 abort (); /* handled in eio_request */
1975#endif
1767 1976
1768 case EIO_NOP: 1977 case EIO_NOP:
1769 req->result = 0; 1978 req->result = 0;
1770 break; 1979 break;
1771 1980
1772 case EIO_CUSTOM: 1981 case EIO_CUSTOM:
1773 ((void (*)(eio_req *))req->feed) (req); 1982 req->feed (req);
1774 break; 1983 break;
1775 1984
1776 default: 1985 default:
1777 errno = ENOSYS;
1778 req->result = -1; 1986 req->result = EIO_ENOSYS ();
1779 break; 1987 break;
1780 } 1988 }
1781 1989
1782 req->errorno = errno; 1990 req->errorno = errno;
1783} 1991}
1784 1992
1785#ifndef EIO_NO_WRAPPERS 1993#ifndef EIO_NO_WRAPPERS
1786 1994
1995eio_req *eio_wd_open (const char *path, int pri, eio_cb cb, void *data)
1996{
1997 REQ (EIO_WD_OPEN); PATH; SEND;
1998}
1999
2000eio_req *eio_wd_close (eio_wd wd, int pri, eio_cb cb, void *data)
2001{
2002 REQ (EIO_WD_CLOSE); req->wd = wd; SEND;
2003}
2004
1787eio_req *eio_nop (int pri, eio_cb cb, void *data) 2005eio_req *eio_nop (int pri, eio_cb cb, void *data)
1788{ 2006{
1789 REQ (EIO_NOP); SEND; 2007 REQ (EIO_NOP); SEND;
1790} 2008}
1791 2009
… …
1807eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2025eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1808{ 2026{
1809 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2027 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1810} 2028}
1811 2029
2030eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2031{
2032 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2033}
2034
2035eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
2036{
2037 REQ (EIO_SYNCFS); req->int1 = fd; SEND;
2038}
2039
2040eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2041{
2042 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2043}
2044
1812eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2045eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1813{ 2046{
1814 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2047 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1815} 2048}
1816 2049
… …
1822eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data) 2055eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
1823{ 2056{
1824 REQ (EIO_MLOCKALL); req->int1 = flags; SEND; 2057 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
1825} 2058}
1826 2059
1827eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) 2060eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
1828{ 2061{
1829 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; 2062 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
1830}
1831
1832eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1833{
1834 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1835} 2063}
1836 2064
1837eio_req *eio_close (int fd, int pri, eio_cb cb, void *data) 2065eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
1838{ 2066{
1839 REQ (EIO_CLOSE); req->int1 = fd; SEND; 2067 REQ (EIO_CLOSE); req->int1 = fd; SEND;
… …
1842eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data) 2070eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
1843{ 2071{
1844 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND; 2072 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
1845} 2073}
1846 2074
2075eio_req *eio_seek (int fd, off_t offset, int whence, int pri, eio_cb cb, void *data)
2076{
2077 REQ (EIO_SEEK); req->int1 = fd; req->offs = offset; req->int2 = whence; SEND;
2078}
2079
1847eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data) 2080eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1848{ 2081{
1849 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND; 2082 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1850} 2083}
1851 2084
… …
1877eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) 2110eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1878{ 2111{
1879 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2112 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1880} 2113}
1881 2114
1882eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2115eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1883{ 2116{
1884 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2117 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1885} 2118}
1886 2119
1887eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) 2120eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
… …
1907eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data) 2140eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1908{ 2141{
1909 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2142 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1910} 2143}
1911 2144
1912eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2145eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1913{ 2146{
1914 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2147 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1915} 2148}
1916 2149
1917eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) 2150eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
… …
1931} 2164}
1932 2165
1933eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data) 2166eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1934{ 2167{
1935 return eio__1path (EIO_READLINK, path, pri, cb, data); 2168 return eio__1path (EIO_READLINK, path, pri, cb, data);
2169}
2170
2171eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2172{
2173 return eio__1path (EIO_REALPATH, path, pri, cb, data);
1936} 2174}
1937 2175
1938eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) 2176eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1939{ 2177{
1940 return eio__1path (EIO_STAT, path, pri, cb, data); 2178 return eio__1path (EIO_STAT, path, pri, cb, data);
… …
1999eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2237eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2000{ 2238{
2001 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2239 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
2002} 2240}
2003 2241
2004eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) 2242eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
2005{ 2243{
2006 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND; 2244 REQ (EIO_CUSTOM); req->feed = execute; SEND;
2007} 2245}
2008 2246
2009#endif 2247#endif
2010 2248
2011eio_req *eio_grp (eio_cb cb, void *data) 2249eio_req *eio_grp (eio_cb cb, void *data)
… …
2020#undef SEND 2258#undef SEND
2021 2259
2022/*****************************************************************************/ 2260/*****************************************************************************/
2023/* grp functions */ 2261/* grp functions */
2024 2262
2263void
2025void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2264eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
2026{ 2265{
2027 grp->int2 = limit; 2266 grp->int2 = limit;
2028 grp->feed = feed; 2267 grp->feed = feed;
2029 2268
2030 grp_try_feed (grp); 2269 grp_try_feed (grp);
2031} 2270}
2032 2271
2272void
2033void eio_grp_limit (eio_req *grp, int limit) 2273eio_grp_limit (eio_req *grp, int limit)
2034{ 2274{
2035 grp->int2 = limit; 2275 grp->int2 = limit;
2036 2276
2037 grp_try_feed (grp); 2277 grp_try_feed (grp);
2038} 2278}
2039 2279
2280void
2040void eio_grp_add (eio_req *grp, eio_req *req) 2281eio_grp_add (eio_req *grp, eio_req *req)
2041{ 2282{
2042 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2283 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
2043 2284
2044 grp->flags |= EIO_FLAG_GROUPADD; 2285 grp->flags |= ETP_FLAG_GROUPADD;
2045 2286
2046 ++grp->size; 2287 ++grp->size;
2047 req->grp = grp; 2288 req->grp = grp;
2048 2289
2049 req->grp_prev = 0; 2290 req->grp_prev = 0;
… …
2056} 2297}
2057 2298
2058/*****************************************************************************/ 2299/*****************************************************************************/
2059/* misc garbage */ 2300/* misc garbage */
2060 2301
2302eio_ssize_t
2061ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2303eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2062{ 2304{
2063 etp_worker wrk;
2064 ssize_t ret;
2065
2066 wrk.dbuf = 0;
2067
2068 ret = eio__sendfile (ofd, ifd, offset, count, &wrk); 2305 return eio__sendfile (ofd, ifd, offset, count);
2069
2070 if (wrk.dbuf)
2071 free (wrk.dbuf);
2072
2073 return ret;
2074} 2306}
2075 2307

Diff Legend

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