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

Comparing libeio/eio.c (file contents):
Revision 1.67 by root, Sun Jun 5 20:25:00 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 */ 63/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
64/* intptr_t only comes form stdint.h, says idiot openbsd coder */ 64/* intptr_t only comes from stdint.h, says idiot openbsd coder */
65#if HAVE_STDINT_H 65#if HAVE_STDINT_H
66# include <stdint.h> 66# include <stdint.h>
67#endif 67#endif
68 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
69#ifndef EIO_FINISH 82#ifndef EIO_FINISH
70# 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
71#endif 84#endif
72 85
73#ifndef EIO_DESTROY 86#ifndef EIO_DESTROY
… …
76 89
77#ifndef EIO_FEED 90#ifndef EIO_FEED
78# 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)
79#endif 92#endif
80 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
81#ifdef _WIN32 105#ifdef _WIN32
82 106
83 /*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
84#else 233#else
85 234
86# include <sys/time.h> 235 #include <sys/time.h>
87# include <sys/select.h> 236 #include <sys/select.h>
88# 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
89# include <utime.h> 281# include <utime.h>
90# include <signal.h>
91# include <dirent.h>
92
93#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
94# include <sys/mman.h>
95#endif
96
97/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
98# if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
99# define _DIRENT_HAVE_D_TYPE /* sigh */
100# define D_INO(de) (de)->d_fileno
101# define D_NAMLEN(de) (de)->d_namlen
102# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
103# define D_INO(de) (de)->d_ino
104# endif 282#endif
105 283
106#ifdef _D_EXACT_NAMLEN 284#if HAVE_SYS_SYSCALL_H
107# undef D_NAMLEN 285# include <sys/syscall.h>
108# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
109#endif
110
111# ifdef _DIRENT_HAVE_D_TYPE
112# define D_TYPE(de) (de)->d_type
113# endif 286#endif
114 287
115# ifndef EIO_STRUCT_DIRENT 288#if HAVE_SYS_PRCTL_H
116# define EIO_STRUCT_DIRENT struct dirent 289# include <sys/prctl.h>
117# endif
118
119#endif 290#endif
120 291
121#if HAVE_SENDFILE 292#if HAVE_SENDFILE
122# if __linux 293# if __linux
123# include <sys/sendfile.h> 294# include <sys/sendfile.h>
… …
138#endif 309#endif
139#ifndef D_INO 310#ifndef D_INO
140# define D_INO(de) 0 311# define D_INO(de) 0
141#endif 312#endif
142#ifndef D_NAMLEN 313#ifndef D_NAMLEN
143# define D_NAMLEN(de) strlen ((de)->d_name) 314# define D_NAMLEN(entp) strlen (D_NAME (entp))
144#endif 315#endif
145 316
146/* used for struct dirent, AIX doesn't provide it */ 317/* used for struct dirent, AIX doesn't provide it */
147#ifndef NAME_MAX 318#ifndef NAME_MAX
148# define NAME_MAX 4096 319# define NAME_MAX 4096
… …
154#endif 325#endif
155 326
156/* buffer size for various temporary buffers */ 327/* buffer size for various temporary buffers */
157#define EIO_BUFSIZE 65536 328#define EIO_BUFSIZE 65536
158 329
159#define dBUF \ 330#define dBUF \
160 char *eio_buf; \
161 ETP_WORKER_LOCK (self); \
162 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 331 char *eio_buf = malloc (EIO_BUFSIZE); \
163 ETP_WORKER_UNLOCK (self); \
164 errno = ENOMEM; \ 332 errno = ENOMEM; \
165 if (!eio_buf) \ 333 if (!eio_buf) \
166 return -1; 334 return -1
167 335
168#define EIO_TICKS ((1000000 + 1023) >> 10) 336#define FUBd \
337 free (eio_buf)
169 338
170/*****************************************************************************/ 339/*****************************************************************************/
171 340
172#if __GNUC__ >= 3 341struct etp_tmpbuf;
173# 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
174#else 349#else
175# 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);
176#endif 352#endif
177 353
178#define expect_false(expr) expect ((expr) != 0, 0) 354struct eio_pwd
179#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};
180 362
181/*****************************************************************************/ 363/*****************************************************************************/
182 364
183#define ETP_PRI_MIN EIO_PRI_MIN 365#define ETP_PRI_MIN EIO_PRI_MIN
184#define ETP_PRI_MAX EIO_PRI_MAX 366#define ETP_PRI_MAX EIO_PRI_MAX
185 367
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
186struct etp_worker; 378struct etp_worker;
187
188#define ETP_REQ eio_req 379#define ETP_REQ eio_req
189#define ETP_DESTROY(req) eio_destroy (req) 380#define ETP_DESTROY(req) eio_destroy (req)
190static int eio_finish (eio_req *req); 381static int eio_finish (eio_req *req);
191#define ETP_FINISH(req) eio_finish (req) 382#define ETP_FINISH(req) eio_finish (req)
192static void eio_execute (struct etp_worker *self, eio_req *req); 383static void eio_execute (struct etp_worker *self, eio_req *req);
193#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) 384#define ETP_EXECUTE(wrk,req) eio_execute (wrk, req)
194 385
195#define ETP_WORKER_CLEAR(req) \ 386#include "etp.c"
196 if (wrk->dbuf) \
197 { \
198 free (wrk->dbuf); \
199 wrk->dbuf = 0; \
200 } \
201 \
202 if (wrk->dirp) \
203 { \
204 closedir (wrk->dirp); \
205 wrk->dirp = 0; \
206 }
207 387
208#define ETP_WORKER_COMMON \ 388static struct etp_pool eio_pool;
209 void *dbuf; \ 389#define EIO_POOL (&eio_pool)
210 DIR *dirp;
211 390
212/*****************************************************************************/ 391/*****************************************************************************/
213 392
214#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
215
216/* calculate time difference in ~1/EIO_TICKS of a second */
217static int tvdiff (struct timeval *tv1, struct timeval *tv2)
218{
219 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
220 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
221}
222
223static unsigned int started, idle, wanted = 4;
224
225static void (*want_poll_cb) (void);
226static void (*done_poll_cb) (void);
227
228static unsigned int max_poll_time; /* reslock */
229static unsigned int max_poll_reqs; /* reslock */
230
231static volatile unsigned int nreqs; /* reqlock */
232static volatile unsigned int nready; /* reqlock */
233static volatile unsigned int npending; /* reqlock */
234static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
235static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
236
237static xmutex_t wrklock;
238static xmutex_t reslock;
239static xmutex_t reqlock;
240static xcond_t reqwait;
241
242#if !HAVE_PREADWRITE
243/*
244 * make our pread/pwrite emulation safe against themselves, but not against
245 * normal read/write by using a mutex. slows down execution a lot,
246 * but that's your problem, not mine.
247 */
248static xmutex_t preadwritelock = X_MUTEX_INIT;
249#endif
250
251typedef struct etp_worker
252{
253 /* locked by wrklock */
254 struct etp_worker *prev, *next;
255
256 xthread_t tid;
257
258 /* locked by reslock, reqlock or wrklock */
259 ETP_REQ *req; /* currently processed request */
260
261 ETP_WORKER_COMMON
262} etp_worker;
263
264static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */
265
266#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
267#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
268
269/* worker threads management */
270
271static void etp_worker_clear (etp_worker *wrk)
272{
273 ETP_WORKER_CLEAR (wrk);
274}
275
276static void etp_worker_free (etp_worker *wrk)
277{
278 wrk->next->prev = wrk->prev;
279 wrk->prev->next = wrk->next;
280
281 free (wrk);
282}
283
284static unsigned int etp_nreqs (void)
285{
286 int retval;
287 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
288 retval = nreqs;
289 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
290 return retval;
291}
292
293static unsigned int etp_nready (void)
294{
295 unsigned int retval;
296
297 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
298 retval = nready;
299 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
300
301 return retval;
302}
303
304static unsigned int etp_npending (void)
305{
306 unsigned int retval;
307
308 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
309 retval = npending;
310 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
311
312 return retval;
313}
314
315static unsigned int etp_nthreads (void)
316{
317 unsigned int retval;
318
319 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
320 retval = started;
321 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
322
323 return retval;
324}
325
326/*
327 * a somewhat faster data structure might be nice, but
328 * with 8 priorities this actually needs <20 insns
329 * per shift, the most expensive operation.
330 */
331typedef struct {
332 ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */
333 int size;
334} etp_reqq;
335
336static etp_reqq req_queue;
337static etp_reqq res_queue;
338
339static int reqq_push (etp_reqq *q, ETP_REQ *req)
340{
341 int pri = req->pri;
342 req->next = 0;
343
344 if (q->qe[pri])
345 {
346 q->qe[pri]->next = req;
347 q->qe[pri] = req;
348 }
349 else
350 q->qe[pri] = q->qs[pri] = req;
351
352 return q->size++;
353}
354
355static ETP_REQ *reqq_shift (etp_reqq *q)
356{
357 int pri;
358
359 if (!q->size)
360 return 0;
361
362 --q->size;
363
364 for (pri = ETP_NUM_PRI; pri--; )
365 {
366 eio_req *req = q->qs[pri];
367
368 if (req)
369 {
370 if (!(q->qs[pri] = (eio_req *)req->next))
371 q->qe[pri] = 0;
372
373 return req;
374 }
375 }
376
377 abort ();
378}
379
380static void etp_thread_init (void)
381{
382 X_MUTEX_CREATE (wrklock);
383 X_MUTEX_CREATE (reslock);
384 X_MUTEX_CREATE (reqlock);
385 X_COND_CREATE (reqwait);
386}
387
388static void etp_atfork_prepare (void)
389{
390 X_LOCK (wrklock);
391 X_LOCK (reqlock);
392 X_LOCK (reslock);
393#if !HAVE_PREADWRITE
394 X_LOCK (preadwritelock);
395#endif
396}
397
398static void etp_atfork_parent (void)
399{
400#if !HAVE_PREADWRITE
401 X_UNLOCK (preadwritelock);
402#endif
403 X_UNLOCK (reslock);
404 X_UNLOCK (reqlock);
405 X_UNLOCK (wrklock);
406}
407
408static void etp_atfork_child (void)
409{
410 ETP_REQ *prv;
411
412 while ((prv = reqq_shift (&req_queue)))
413 ETP_DESTROY (prv);
414
415 while ((prv = reqq_shift (&res_queue)))
416 ETP_DESTROY (prv);
417
418 while (wrk_first.next != &wrk_first)
419 {
420 etp_worker *wrk = wrk_first.next;
421
422 if (wrk->req)
423 ETP_DESTROY (wrk->req);
424
425 etp_worker_clear (wrk);
426 etp_worker_free (wrk);
427 }
428
429 started = 0;
430 idle = 0;
431 nreqs = 0;
432 nready = 0;
433 npending = 0;
434
435 etp_thread_init ();
436}
437
438static void 393static void
439etp_once_init (void)
440{
441 etp_thread_init ();
442 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
443}
444
445static int
446etp_init (void (*want_poll)(void), void (*done_poll)(void))
447{
448 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
449
450 pthread_once (&doinit, etp_once_init);
451
452 want_poll_cb = want_poll;
453 done_poll_cb = done_poll;
454
455 return 0;
456}
457
458X_THREAD_PROC (etp_proc);
459
460static void etp_start_thread (void)
461{
462 etp_worker *wrk = calloc (1, sizeof (etp_worker));
463
464 /*TODO*/
465 assert (("unable to allocate worker thread data", wrk));
466
467 X_LOCK (wrklock);
468
469 if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
470 {
471 wrk->prev = &wrk_first;
472 wrk->next = wrk_first.next;
473 wrk_first.next->prev = wrk;
474 wrk_first.next = wrk;
475 ++started;
476 }
477 else
478 free (wrk);
479
480 X_UNLOCK (wrklock);
481}
482
483static void etp_maybe_start_thread (void)
484{
485 if (expect_true (etp_nthreads () >= wanted))
486 return;
487
488 /* todo: maybe use idle here, but might be less exact */
489 if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
490 return;
491
492 etp_start_thread ();
493}
494
495static void etp_end_thread (void)
496{
497 eio_req *req = calloc (1, sizeof (eio_req));
498
499 req->type = -1;
500 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
501
502 X_LOCK (reqlock);
503 reqq_push (&req_queue, req);
504 X_COND_SIGNAL (reqwait);
505 X_UNLOCK (reqlock);
506
507 X_LOCK (wrklock);
508 --started;
509 X_UNLOCK (wrklock);
510}
511
512static int etp_poll (void)
513{
514 unsigned int maxreqs;
515 unsigned int maxtime;
516 struct timeval tv_start, tv_now;
517
518 X_LOCK (reslock);
519 maxreqs = max_poll_reqs;
520 maxtime = max_poll_time;
521 X_UNLOCK (reslock);
522
523 if (maxtime)
524 gettimeofday (&tv_start, 0);
525
526 for (;;)
527 {
528 ETP_REQ *req;
529
530 etp_maybe_start_thread ();
531
532 X_LOCK (reslock);
533 req = reqq_shift (&res_queue);
534
535 if (req)
536 {
537 --npending;
538
539 if (!res_queue.size && done_poll_cb)
540 done_poll_cb ();
541 }
542
543 X_UNLOCK (reslock);
544
545 if (!req)
546 return 0;
547
548 X_LOCK (reqlock);
549 --nreqs;
550 X_UNLOCK (reqlock);
551
552 if (expect_false (req->type == EIO_GROUP && req->size))
553 {
554 req->int1 = 1; /* mark request as delayed */
555 continue;
556 }
557 else
558 {
559 int res = ETP_FINISH (req);
560 if (expect_false (res))
561 return res;
562 }
563
564 if (expect_false (maxreqs && !--maxreqs))
565 break;
566
567 if (maxtime)
568 {
569 gettimeofday (&tv_now, 0);
570
571 if (tvdiff (&tv_start, &tv_now) >= maxtime)
572 break;
573 }
574 }
575
576 errno = EAGAIN;
577 return -1;
578}
579
580static void etp_cancel (ETP_REQ *req)
581{
582 X_LOCK (wrklock);
583 req->flags |= EIO_FLAG_CANCELLED;
584 X_UNLOCK (wrklock);
585
586 eio_grp_cancel (req);
587}
588
589static void etp_submit (ETP_REQ *req)
590{
591 req->pri -= ETP_PRI_MIN;
592
593 if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
594 if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
595
596 if (expect_false (req->type == EIO_GROUP))
597 {
598 /* I hope this is worth it :/ */
599 X_LOCK (reqlock);
600 ++nreqs;
601 X_UNLOCK (reqlock);
602
603 X_LOCK (reslock);
604
605 ++npending;
606
607 if (!reqq_push (&res_queue, req) && want_poll_cb)
608 want_poll_cb ();
609
610 X_UNLOCK (reslock);
611 }
612 else
613 {
614 X_LOCK (reqlock);
615 ++nreqs;
616 ++nready;
617 reqq_push (&req_queue, req);
618 X_COND_SIGNAL (reqwait);
619 X_UNLOCK (reqlock);
620
621 etp_maybe_start_thread ();
622 }
623}
624
625static void etp_set_max_poll_time (double nseconds)
626{
627 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
628 max_poll_time = nseconds * EIO_TICKS;
629 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
630}
631
632static void etp_set_max_poll_reqs (unsigned int maxreqs)
633{
634 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
635 max_poll_reqs = maxreqs;
636 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
637}
638
639static void etp_set_max_idle (unsigned int nthreads)
640{
641 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
642 max_idle = nthreads;
643 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
644}
645
646static void etp_set_idle_timeout (unsigned int seconds)
647{
648 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
649 idle_timeout = seconds;
650 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
651}
652
653static void etp_set_min_parallel (unsigned int nthreads)
654{
655 if (wanted < nthreads)
656 wanted = nthreads;
657}
658
659static void etp_set_max_parallel (unsigned int nthreads)
660{
661 if (wanted > nthreads)
662 wanted = nthreads;
663
664 while (started > wanted)
665 etp_end_thread ();
666}
667
668/*****************************************************************************/
669
670static void grp_try_feed (eio_req *grp) 394grp_try_feed (eio_req *grp)
671{ 395{
672 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 396 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
673 { 397 {
674 grp->flags &= ~EIO_FLAG_GROUPADD; 398 grp->flags &= ~ETP_FLAG_GROUPADD;
675 399
676 EIO_FEED (grp); 400 EIO_FEED (grp);
677 401
678 /* stop if no progress has been made */ 402 /* stop if no progress has been made */
679 if (!(grp->flags & EIO_FLAG_GROUPADD)) 403 if (!(grp->flags & ETP_FLAG_GROUPADD))
680 { 404 {
681 grp->feed = 0; 405 grp->feed = 0;
682 break; 406 break;
683 } 407 }
684 } 408 }
685} 409}
686 410
411static int
687static int grp_dec (eio_req *grp) 412grp_dec (eio_req *grp)
688{ 413{
689 --grp->size; 414 --grp->size;
690 415
691 /* call feeder, if applicable */ 416 /* call feeder, if applicable */
692 grp_try_feed (grp); 417 grp_try_feed (grp);
693 418
694 /* finish, if done */ 419 /* finish, if done */
695 if (!grp->size && grp->int1) 420 if (!grp->size && grp->flags & ETP_FLAG_DELAYED)
696 return eio_finish (grp); 421 return eio_finish (grp);
697 else 422 else
698 return 0; 423 return 0;
699} 424}
700 425
426static void
701void eio_destroy (eio_req *req) 427eio_destroy (eio_req *req)
702{ 428{
703 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); 429 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
704 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); 430 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
705 431
706 EIO_DESTROY (req); 432 EIO_DESTROY (req);
707} 433}
708 434
435static int
709static int eio_finish (eio_req *req) 436eio_finish (eio_req *req)
710{ 437{
711 int res = EIO_FINISH (req); 438 int res = EIO_FINISH (req);
712 439
713 if (req->grp) 440 if (req->grp)
714 { 441 {
… …
722 if (grp->grp_first == req) 449 if (grp->grp_first == req)
723 grp->grp_first = req->grp_next; 450 grp->grp_first = req->grp_next;
724 451
725 res2 = grp_dec (grp); 452 res2 = grp_dec (grp);
726 453
727 if (!res && res2) 454 if (!res)
728 res = res2; 455 res = res2;
729 } 456 }
730 457
731 eio_destroy (req); 458 eio_destroy (req);
732 459
733 return res; 460 return res;
734} 461}
735 462
463void
736void eio_grp_cancel (eio_req *grp) 464eio_grp_cancel (eio_req *grp)
737{ 465{
738 for (grp = grp->grp_first; grp; grp = grp->grp_next) 466 etp_grp_cancel (EIO_POOL, grp);
739 eio_cancel (grp);
740} 467}
741 468
469void
742void eio_cancel (eio_req *req) 470eio_cancel (eio_req *req)
743{ 471{
744 etp_cancel (req); 472 etp_cancel (EIO_POOL, req);
745} 473}
746 474
475void
747void eio_submit (eio_req *req) 476eio_submit (eio_req *req)
748{ 477{
749 etp_submit (req); 478 etp_submit (EIO_POOL, req);
750} 479}
751 480
752unsigned int eio_nreqs (void) 481unsigned int
482eio_nreqs (void)
753{ 483{
754 return etp_nreqs (); 484 return etp_nreqs (EIO_POOL);
755} 485}
756 486
757unsigned int eio_nready (void) 487unsigned int
488eio_nready (void)
758{ 489{
759 return etp_nready (); 490 return etp_nready (EIO_POOL);
760} 491}
761 492
762unsigned int eio_npending (void) 493unsigned int
494eio_npending (void)
763{ 495{
764 return etp_npending (); 496 return etp_npending (EIO_POOL);
765} 497}
766 498
767unsigned int eio_nthreads (void) 499unsigned int ecb_cold
500eio_nthreads (void)
768{ 501{
769 return etp_nthreads (); 502 return etp_nthreads (EIO_POOL);
770} 503}
771 504
505void ecb_cold
772void eio_set_max_poll_time (double nseconds) 506eio_set_max_poll_time (double nseconds)
773{ 507{
774 etp_set_max_poll_time (nseconds); 508 etp_set_max_poll_time (EIO_POOL, nseconds);
775} 509}
776 510
511void ecb_cold
777void eio_set_max_poll_reqs (unsigned int maxreqs) 512eio_set_max_poll_reqs (unsigned int maxreqs)
778{ 513{
779 etp_set_max_poll_reqs (maxreqs); 514 etp_set_max_poll_reqs (EIO_POOL, maxreqs);
780} 515}
781 516
517void ecb_cold
782void eio_set_max_idle (unsigned int nthreads) 518eio_set_max_idle (unsigned int nthreads)
783{ 519{
784 etp_set_max_idle (nthreads); 520 etp_set_max_idle (EIO_POOL, nthreads);
785} 521}
786 522
523void ecb_cold
787void eio_set_idle_timeout (unsigned int seconds) 524eio_set_idle_timeout (unsigned int seconds)
788{ 525{
789 etp_set_idle_timeout (seconds); 526 etp_set_idle_timeout (EIO_POOL, seconds);
790} 527}
791 528
529void ecb_cold
792void eio_set_min_parallel (unsigned int nthreads) 530eio_set_min_parallel (unsigned int nthreads)
793{ 531{
794 etp_set_min_parallel (nthreads); 532 etp_set_min_parallel (EIO_POOL, nthreads);
795} 533}
796 534
535void ecb_cold
797void eio_set_max_parallel (unsigned int nthreads) 536eio_set_max_parallel (unsigned int nthreads)
798{ 537{
799 etp_set_max_parallel (nthreads); 538 etp_set_max_parallel (EIO_POOL, nthreads);
800} 539}
801 540
802int eio_poll (void) 541int eio_poll (void)
803{ 542{
804 return etp_poll (); 543 return etp_poll (EIO_POOL);
805} 544}
806 545
807/*****************************************************************************/ 546/*****************************************************************************/
808/* work around various missing functions */ 547/* work around various missing functions */
809
810#if !HAVE_PREADWRITE
811# undef pread
812# undef pwrite
813# define pread eio__pread
814# define pwrite eio__pwrite
815
816static ssize_t
817eio__pread (int fd, void *buf, size_t count, off_t offset)
818{
819 ssize_t res;
820 off_t ooffset;
821
822 X_LOCK (preadwritelock);
823 ooffset = lseek (fd, 0, SEEK_CUR);
824 lseek (fd, offset, SEEK_SET);
825 res = read (fd, buf, count);
826 lseek (fd, ooffset, SEEK_SET);
827 X_UNLOCK (preadwritelock);
828
829 return res;
830}
831
832static ssize_t
833eio__pwrite (int fd, void *buf, size_t count, off_t offset)
834{
835 ssize_t res;
836 off_t ooffset;
837
838 X_LOCK (preadwritelock);
839 ooffset = lseek (fd, 0, SEEK_CUR);
840 lseek (fd, offset, SEEK_SET);
841 res = write (fd, buf, count);
842 lseek (fd, ooffset, SEEK_SET);
843 X_UNLOCK (preadwritelock);
844
845 return res;
846}
847#endif
848 548
849#ifndef HAVE_UTIMES 549#ifndef HAVE_UTIMES
850 550
851# undef utimes 551# undef utimes
852# define utimes(path,times) eio__utimes (path, times) 552# define utimes(path,times) eio__utimes (path, times)
… …
872#ifndef HAVE_FUTIMES 572#ifndef HAVE_FUTIMES
873 573
874# undef futimes 574# undef futimes
875# define futimes(fd,times) eio__futimes (fd, times) 575# define futimes(fd,times) eio__futimes (fd, times)
876 576
577static int
877static int eio__futimes (int fd, const struct timeval tv[2]) 578eio__futimes (int fd, const struct timeval tv[2])
878{ 579{
879 errno = ENOSYS; 580 errno = ENOSYS;
880 return -1; 581 return -1;
881} 582}
882 583
… …
885#if !HAVE_FDATASYNC 586#if !HAVE_FDATASYNC
886# undef fdatasync 587# undef fdatasync
887# define fdatasync(fd) fsync (fd) 588# define fdatasync(fd) fsync (fd)
888#endif 589#endif
889 590
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
890/* sync_file_range always needs emulation */ 608/* sync_file_range always needs emulation */
891int 609static int
892eio__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)
893{ 611{
894#if HAVE_SYNC_FILE_RANGE 612#if HAVE_SYNC_FILE_RANGE
895 int res; 613 int res;
896 614
… …
913 /* 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
914 * call fdatasync, as that matches the expectation of its users best */ 632 * call fdatasync, as that matches the expectation of its users best */
915 return fdatasync (fd); 633 return fdatasync (fd);
916} 634}
917 635
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
918#if !HAVE_READAHEAD 646#if !HAVE_READAHEAD
919# undef readahead 647# undef readahead
920# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 648# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
921 649
922static ssize_t 650static eio_ssize_t
923eio__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)
924{ 652{
925 size_t todo = count; 653 size_t todo = count;
926 dBUF; 654 dBUF;
927 655
… …
932 pread (fd, eio_buf, len, offset); 660 pread (fd, eio_buf, len, offset);
933 offset += len; 661 offset += len;
934 todo -= len; 662 todo -= len;
935 } 663 }
936 664
937 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 */
938 return count; 669 return 0;
939} 670}
940 671
941#endif 672#endif
942 673
943/* sendfile always needs emulation */ 674/* sendfile always needs emulation */
944static ssize_t 675static eio_ssize_t
945eio__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)
946{ 677{
678 eio_ssize_t written = 0;
947 ssize_t res; 679 eio_ssize_t res;
948 680
949 if (!count) 681 if (!count)
950 return 0; 682 return 0;
951 683
684 for (;;)
685 {
686#ifdef __APPLE__
687# undef HAVE_SENDFILE /* broken, as everything on os x */
688#endif
952#if HAVE_SENDFILE 689#if HAVE_SENDFILE
953# if __linux 690# if __linux
691 off_t soffset = offset;
954 res = sendfile (ofd, ifd, &offset, count); 692 res = sendfile (ofd, ifd, &soffset, count);
955 693
956# elif __FreeBSD__ 694# elif __FreeBSD__
957 /* 695 /*
958 * 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
959 * wasted on making it similar to other I/O functions. 697 * wasted on making it similar to other I/O functions.
960 */ 698 */
961 {
962 off_t sbytes; 699 off_t sbytes;
963 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 700 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
964 701
965 #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 */
966 /* freebsd' sendfile will return 0 on success */ 703 /* freebsd' sendfile will return 0 on success */
967 /* 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 */
968 /* not on e.g. EIO or EPIPE - sounds broken */ 705 /* not on e.g. EIO or EPIPE - sounds broken */
969 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0) 706 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
970 res = sbytes; 707 res = sbytes;
971 #endif 708 #endif
972 709
973 /* according to source inspection, this is correct, and useful behaviour */ 710 /* according to source inspection, this is correct, and useful behaviour */
974 if (sbytes) 711 if (sbytes)
975 res = sbytes; 712 res = sbytes;
976 }
977 713
978# elif defined (__APPLE__) 714# elif defined __APPLE__
979
980 {
981 off_t sbytes = count; 715 off_t sbytes = count;
982 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0); 716 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
983 717
984 /* according to the manpage, sbytes is always valid */ 718 /* according to the manpage, sbytes is always valid */
985 if (sbytes) 719 if (sbytes)
986 res = sbytes; 720 res = sbytes;
987 }
988 721
989# elif __hpux 722# elif __hpux
990 res = sendfile (ofd, ifd, offset, count, 0, 0); 723 res = sendfile (ofd, ifd, offset, count, 0, 0);
991 724
992# elif __solaris 725# elif __solaris
993 {
994 struct sendfilevec vec; 726 struct sendfilevec vec;
995 size_t sbytes; 727 size_t sbytes;
996 728
997 vec.sfv_fd = ifd; 729 vec.sfv_fd = ifd;
998 vec.sfv_flag = 0; 730 vec.sfv_flag = 0;
999 vec.sfv_off = offset; 731 vec.sfv_off = offset;
1000 vec.sfv_len = count; 732 vec.sfv_len = count;
1001 733
1002 res = sendfilev (ofd, &vec, 1, &sbytes); 734 res = sendfilev (ofd, &vec, 1, &sbytes);
1003 735
1004 if (res < 0 && sbytes) 736 if (res < 0 && sbytes)
1005 res = sbytes; 737 res = sbytes;
1006 }
1007 738
1008# endif 739# endif
1009 740
1010#elif defined (_WIN32) 741#elif defined (_WIN32) && 0
1011
1012 /* 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 */
1013 { 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 */
1014 HANDLE h = TO_SOCKET (ifd); 746 HANDLE h = TO_SOCKET (ifd);
1015 SetFilePointer (h, offset, 0, FILE_BEGIN); 747 SetFilePointer (h, offset, 0, FILE_BEGIN);
1016 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0); 748 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1017 }
1018 749
1019#else 750#else
1020 res = -1; 751 res = EIO_ENOSYS ();
1021 errno = ENOSYS;
1022#endif 752#endif
1023 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
1024 if (res < 0 777 if (res < 0
1025 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 778 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1026 /* BSDs */ 779 /* BSDs */
1027#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... */
1028 || errno == ENOTSUP 781 || errno == ENOTSUP
1029#endif 782#endif
783#ifdef EOPNOTSUPP /* windows */
1030 || errno == EOPNOTSUPP /* BSDs */ 784 || errno == EOPNOTSUPP /* BSDs */
785#endif
1031#if __solaris 786#if __solaris
1032 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 787 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
1033#endif 788#endif
1034 ) 789 )
1035 ) 790 )
… …
1039 794
1040 res = 0; 795 res = 0;
1041 796
1042 while (count) 797 while (count)
1043 { 798 {
1044 ssize_t cnt; 799 eio_ssize_t cnt;
1045 800
1046 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);
1047 802
1048 if (cnt <= 0) 803 if (cnt <= 0)
1049 { 804 {
… …
1061 816
1062 offset += cnt; 817 offset += cnt;
1063 res += cnt; 818 res += cnt;
1064 count -= cnt; 819 count -= cnt;
1065 } 820 }
821
822 FUBd;
1066 } 823 }
1067 824
1068 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
1069} 1133}
1070 1134
1071static signed char 1135static signed char
1072eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) 1136eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1073{ 1137{
… …
1081 1145
1082#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 */
1083#define EIO_SORT_FAST 60 /* when to only use insertion sort */ 1147#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1084 1148
1085static void 1149static void
1086eio_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)
1087{ 1151{
1088 unsigned char bits [9 + sizeof (ino_t) * 8]; 1152 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1089 unsigned char *bit = bits; 1153 unsigned char *bit = bits;
1090 1154
1091 assert (CHAR_BIT == 8); 1155 assert (CHAR_BIT == 8);
1092 assert (sizeof (eio_dirent) * 8 < 256); 1156 assert (sizeof (eio_dirent) * 8 < 256);
1093 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */ 1157 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
… …
1095 1159
1096 if (size <= EIO_SORT_FAST) 1160 if (size <= EIO_SORT_FAST)
1097 return; 1161 return;
1098 1162
1099 /* 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 */
1100 /* 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 */
1101 /* inode_bits must contain all inodes ORed together */ 1165 /* inode_bits must contain all inodes ORed together */
1102 /* 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 */
1103 { 1167 {
1104 ino_t endianness; 1168 eio_ino_t endianness;
1105 int i, j; 1169 int i, j;
1106 1170
1107 /* 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" */
1108 for (i = 0; i < sizeof (ino_t); ++i) 1172 for (i = 0; i < sizeof (eio_ino_t); ++i)
1109 ((unsigned char *)&endianness)[i] = i; 1173 ((unsigned char *)&endianness)[i] = i;
1110 1174
1111 *bit++ = 0; 1175 *bit++ = 0;
1112 1176
1113 for (i = 0; i < sizeof (ino_t); ++i) 1177 for (i = 0; i < sizeof (eio_ino_t); ++i)
1114 { 1178 {
1115 /* shifting off the byte offsets out of "endianness" */ 1179 /* shifting off the byte offsets out of "endianness" */
1116 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8; 1180 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1117 endianness >>= 8; 1181 endianness >>= 8;
1118 1182
1119 for (j = 0; j < 8; ++j) 1183 for (j = 0; j < 8; ++j)
1120 if (inode_bits & (((ino_t)1) << (i * 8 + j))) 1184 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1121 *bit++ = offs + j; 1185 *bit++ = offs + j;
1122 } 1186 }
1123 1187
1124 for (j = 0; j < 8; ++j) 1188 for (j = 0; j < 8; ++j)
1125 if (score_bits & (1 << j)) 1189 if (score_bits & (1 << j))
1126 *bit++ = offsetof (eio_dirent, score) * 8 + j; 1190 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1127 } 1191 }
1128 1192
1129 /* now actually do the sorting (a variant of MSD radix sort) */ 1193 /* now actually do the sorting (a variant of MSD radix sort) */
1130 { 1194 {
1131 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base; 1195 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1132 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end; 1196 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1133 unsigned char *bit_stk [9 + sizeof (ino_t) * 8]; 1197 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1134 int stk_idx = 0; 1198 int stk_idx = 0;
1135 1199
1136 base_stk [stk_idx] = dents; 1200 base_stk [stk_idx] = dents;
1137 end_stk [stk_idx] = dents + size; 1201 end_stk [stk_idx] = dents + size;
1138 bit_stk [stk_idx] = bit - 1; 1202 bit_stk [stk_idx] = bit - 1;
… …
1217 } 1281 }
1218 } 1282 }
1219} 1283}
1220 1284
1221static void 1285static void
1222eio_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)
1223{ 1287{
1224 if (size <= 1) 1288 if (size <= 1)
1225 return; /* our insertion sort relies on size > 0 */ 1289 return; /* our insertion sort relies on size > 0 */
1226 1290
1227 /* 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 */
… …
1235 1299
1236/* read a full directory */ 1300/* read a full directory */
1237static void 1301static void
1238eio__scandir (eio_req *req, etp_worker *self) 1302eio__scandir (eio_req *req, etp_worker *self)
1239{ 1303{
1240 DIR *dirp;
1241 EIO_STRUCT_DIRENT *entp;
1242 char *name, *names; 1304 char *name, *names;
1243 int namesalloc = 4096; 1305 int namesalloc = 4096 - sizeof (void *) * 4;
1244 int namesoffs = 0; 1306 int namesoffs = 0;
1245 int flags = req->int1; 1307 int flags = req->int1;
1246 eio_dirent *dents = 0; 1308 eio_dirent *dents = 0;
1247 int dentalloc = 128; 1309 int dentalloc = 128;
1248 int dentoffs = 0; 1310 int dentoffs = 0;
1249 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
1250 1319
1251 req->result = -1; 1320 req->result = -1;
1252 1321
1253 if (!(flags & EIO_READDIR_DENTS)) 1322 if (!(flags & EIO_READDIR_DENTS))
1254 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); 1323 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1255 1324
1256 X_LOCK (wrklock); 1325#ifdef _WIN32
1257 /* 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
1258 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);
1259 1396
1260 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1397 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1261 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; 1398 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1262 req->ptr2 = names = malloc (namesalloc); 1399 req->ptr2 = names = malloc (namesalloc);
1263 X_UNLOCK (wrklock);
1264 1400
1265 if (dirp && names && (!flags || dents)) 1401 if (!names || (flags && !dents))
1402 return;
1403
1266 for (;;) 1404 for (;;)
1267 { 1405 {
1406 int done;
1407
1408#ifdef _WIN32
1409 done = !dirp;
1410#else
1268 errno = 0; 1411 errno = 0;
1269 entp = readdir (dirp); 1412 entp = readdir (dirp);
1413 done = !entp;
1414#endif
1270 1415
1271 if (!entp) 1416 if (done)
1272 { 1417 {
1418#ifndef _WIN32
1419 int old_errno = errno;
1420 closedir (dirp);
1421 errno = old_errno;
1422
1273 if (errno) 1423 if (errno)
1274 break; 1424 break;
1425#endif
1275 1426
1276 /* sort etc. */ 1427 /* sort etc. */
1277 req->int1 = flags; 1428 req->int1 = flags;
1278 req->result = dentoffs; 1429 req->result = dentoffs;
1279 1430
1280 if (flags & EIO_READDIR_STAT_ORDER) 1431 if (flags & EIO_READDIR_STAT_ORDER)
1281 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);
1282 else if (flags & EIO_READDIR_DIRS_FIRST) 1433 else if (flags & EIO_READDIR_DIRS_FIRST)
1283 if (flags & EIO_READDIR_FOUND_UNKNOWN) 1434 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1284 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 */
1285 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))
1286 { 1487 {
1287 /* 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
1288 eio_dirent *oth = dents + dentoffs; 1495 ent = dents + dentoffs;
1289 eio_dirent *dir = dents;
1290 1496
1291 /* 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 */
1292 /* by walking from both sides and swapping if necessary */ 1498 ent->namelen = len - 1;
1293 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)
1294 { 1559 {
1295 if (dir->type == EIO_DT_DIR) 1560 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1296 ++dir; 1561 ent->score = 1;
1297 else if ((--oth)->type == EIO_DT_DIR) 1562 else if (!strchr (name, '.')) /* absence of dots indicate likely dirs */
1298 { 1563 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1299 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1300
1301 ++dir;
1302 }
1303 } 1564 }
1304 1565 else if (ent->type == EIO_DT_DIR)
1305 /* now sort the dirs only (dirs all have the same score) */ 1566 ent->score = 0;
1306 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1307 } 1567 }
1308
1309 break;
1310 }
1311
1312 /* now add the entry to our list(s) */
1313 name = entp->d_name;
1314
1315 /* skip . and .. entries */
1316 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1317 {
1318 int len = D_NAMLEN (entp) + 1;
1319
1320 while (expect_false (namesoffs + len > namesalloc))
1321 {
1322 namesalloc *= 2;
1323 X_LOCK (wrklock);
1324 req->ptr2 = names = realloc (names, namesalloc);
1325 X_UNLOCK (wrklock);
1326
1327 if (!names)
1328 break;
1329 } 1568 }
1330 1569
1331 memcpy (names + namesoffs, name, len);
1332
1333 if (dents)
1334 {
1335 struct eio_dirent *ent;
1336
1337 if (expect_false (dentoffs == dentalloc))
1338 {
1339 dentalloc *= 2;
1340 X_LOCK (wrklock);
1341 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1342 X_UNLOCK (wrklock);
1343
1344 if (!dents)
1345 break;
1346 }
1347
1348 ent = dents + dentoffs;
1349
1350 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1351 ent->namelen = len - 1;
1352 ent->inode = D_INO (entp);
1353
1354 inode_bits |= ent->inode;
1355
1356 switch (D_TYPE (entp))
1357 {
1358 default:
1359 ent->type = EIO_DT_UNKNOWN;
1360 flags |= EIO_READDIR_FOUND_UNKNOWN;
1361 break;
1362
1363 #ifdef DT_FIFO
1364 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1365 #endif
1366 #ifdef DT_CHR
1367 case DT_CHR: ent->type = EIO_DT_CHR; break;
1368 #endif
1369 #ifdef DT_MPC
1370 case DT_MPC: ent->type = EIO_DT_MPC; break;
1371 #endif
1372 #ifdef DT_DIR
1373 case DT_DIR: ent->type = EIO_DT_DIR; break;
1374 #endif
1375 #ifdef DT_NAM
1376 case DT_NAM: ent->type = EIO_DT_NAM; break;
1377 #endif
1378 #ifdef DT_BLK
1379 case DT_BLK: ent->type = EIO_DT_BLK; break;
1380 #endif
1381 #ifdef DT_MPB
1382 case DT_MPB: ent->type = EIO_DT_MPB; break;
1383 #endif
1384 #ifdef DT_REG
1385 case DT_REG: ent->type = EIO_DT_REG; break;
1386 #endif
1387 #ifdef DT_NWK
1388 case DT_NWK: ent->type = EIO_DT_NWK; break;
1389 #endif
1390 #ifdef DT_CMP
1391 case DT_CMP: ent->type = EIO_DT_CMP; break;
1392 #endif
1393 #ifdef DT_LNK
1394 case DT_LNK: ent->type = EIO_DT_LNK; break;
1395 #endif
1396 #ifdef DT_SOCK
1397 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1398 #endif
1399 #ifdef DT_DOOR
1400 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1401 #endif
1402 #ifdef DT_WHT
1403 case DT_WHT: ent->type = EIO_DT_WHT; break;
1404 #endif
1405 }
1406
1407 ent->score = 7;
1408
1409 if (flags & EIO_READDIR_DIRS_FIRST)
1410 {
1411 if (ent->type == EIO_DT_UNKNOWN)
1412 {
1413 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1414 ent->score = 1;
1415 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1416 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1417 }
1418 else if (ent->type == EIO_DT_DIR)
1419 ent->score = 0;
1420 }
1421 }
1422
1423 namesoffs += len; 1570 namesoffs += len;
1424 ++dentoffs; 1571 ++dentoffs;
1425 } 1572 }
1426 1573
1427 if (EIO_CANCELLED (req)) 1574 if (EIO_CANCELLED (req))
1428 { 1575 {
1429 errno = ECANCELED; 1576 errno = ECANCELED;
1430 break; 1577 break;
1431 } 1578 }
1579
1580#ifdef _WIN32
1581 if (!FindNextFile (dirp, &entp))
1582 {
1583 FindClose (dirp);
1584 dirp = 0;
1432 } 1585 }
1586#endif
1587 }
1433} 1588}
1434 1589
1435#ifdef PAGESIZE 1590/*****************************************************************************/
1436# define eio_pagesize() PAGESIZE 1591/* working directory stuff */
1437#else 1592/* various deficiencies in the posix 2008 api force us to */
1438static intptr_t 1593/* keep the absolute path in string form at all times */
1439eio_pagesize (void) 1594/* fuck yeah. */
1440{
1441 static intptr_t page;
1442 1595
1443 if (!page) 1596#if !HAVE_AT
1444 page = sysconf (_SC_PAGESIZE);
1445 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
1446 return page; 1652 return res;
1447} 1653}
1448#endif
1449 1654
1450static void 1655eio_wd
1451eio_page_align (void **addr, size_t *length) 1656eio_wd_open_sync (eio_wd wd, const char *path)
1452{ 1657{
1453 intptr_t mask = eio_pagesize () - 1; 1658 struct etp_tmpbuf tmpbuf = { };
1659 wd = eio__wd_open_sync (&tmpbuf, wd, path);
1660 free (tmpbuf.ptr);
1454 1661
1455 /* round down addr */ 1662 return wd;
1456 intptr_t adj = mask & (intptr_t)*addr;
1457
1458 *addr = (void *)((intptr_t)*addr - adj);
1459 *length += adj;
1460
1461 /* round up length */
1462 *length = (*length + mask) & ~mask;
1463} 1663}
1464 1664
1465#if !_POSIX_MEMLOCK 1665void
1466# define eio__mlockall(a) ((errno = ENOSYS), -1) 1666eio_wd_close_sync (eio_wd wd)
1467#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 */
1468 1680
1469static int 1681static int
1470eio__mlockall (int flags) 1682eio__truncateat (int dirfd, const char *path, off_t length)
1471{ 1683{
1472 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7 1684 int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
1473 extern int mallopt (int, int); 1685 int res;
1474 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1475 #endif
1476 1686
1477 if (EIO_MCL_CURRENT != MCL_CURRENT 1687 if (fd < 0)
1478 || EIO_MCL_FUTURE != MCL_FUTURE) 1688 return fd;
1479 {
1480 flags = 0
1481 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1482 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1483 }
1484 1689
1485 return mlockall (flags); 1690 res = ftruncate (fd, length);
1691 close (fd);
1692 return res;
1486} 1693}
1487#endif
1488
1489#if !_POSIX_MEMLOCK_RANGE
1490# define eio__mlock(a,b) ((errno = ENOSYS), -1)
1491#else
1492 1694
1493static int 1695static int
1494eio__mlock (void *addr, size_t length) 1696eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
1495{ 1697{
1496 eio_page_align (&addr, &length); 1698 int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
1699 int res;
1497 1700
1498 return mlock (addr, length); 1701 if (fd < 0)
1499} 1702 return fd;
1500 1703
1501#endif 1704 res = fstatvfs (fd, buf);
1502 1705 close (fd);
1503#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1504# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
1505#else
1506
1507int
1508eio__msync (void *mem, size_t len, int flags)
1509{
1510 eio_page_align (&mem, &len);
1511
1512 if (EIO_MS_ASYNC != MS_SYNC
1513 || EIO_MS_INVALIDATE != MS_INVALIDATE
1514 || EIO_MS_SYNC != MS_SYNC)
1515 {
1516 flags = 0
1517 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1518 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1519 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1520 }
1521
1522 return msync (mem, len, flags);
1523}
1524
1525#endif
1526
1527int
1528eio__mtouch (eio_req *req)
1529{
1530 void *mem = req->ptr2;
1531 size_t len = req->size;
1532 int flags = req->int1;
1533
1534 eio_page_align (&mem, &len);
1535
1536 {
1537 intptr_t addr = (intptr_t)mem;
1538 intptr_t end = addr + len;
1539 intptr_t page = eio_pagesize ();
1540
1541 if (addr < end)
1542 if (flags & EIO_MT_MODIFY) /* modify */
1543 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1544 else
1545 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1546 }
1547
1548 return 0; 1706 return res;
1707
1549} 1708}
1709
1710#endif
1550 1711
1551/*****************************************************************************/ 1712/*****************************************************************************/
1552 1713
1553#define ALLOC(len) \ 1714#define ALLOC(len) \
1554 if (!req->ptr2) \ 1715 if (!req->ptr2) \
1555 { \ 1716 { \
1556 X_LOCK (wrklock); \ 1717 X_LOCK (EIO_POOL->wrklock); \
1557 req->flags |= EIO_FLAG_PTR2_FREE; \ 1718 req->flags |= EIO_FLAG_PTR2_FREE; \
1558 X_UNLOCK (wrklock); \ 1719 X_UNLOCK (EIO_POOL->wrklock); \
1559 req->ptr2 = malloc (len); \ 1720 req->ptr2 = malloc (len); \
1560 if (!req->ptr2) \ 1721 if (!req->ptr2) \
1561 { \ 1722 { \
1562 errno = ENOMEM; \ 1723 errno = ENOMEM; \
1563 req->result = -1; \ 1724 req->result = -1; \
1564 break; \ 1725 break; \
1565 } \ 1726 } \
1566 } 1727 }
1567 1728
1568X_THREAD_PROC (etp_proc)
1569{
1570 ETP_REQ *req;
1571 struct timespec ts;
1572 etp_worker *self = (etp_worker *)thr_arg;
1573
1574 /* try to distribute timeouts somewhat randomly */
1575 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1576
1577 for (;;)
1578 {
1579 X_LOCK (reqlock);
1580
1581 for (;;)
1582 {
1583 self->req = req = reqq_shift (&req_queue);
1584
1585 if (req)
1586 break;
1587
1588 ++idle;
1589
1590 ts.tv_sec = time (0) + idle_timeout;
1591 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
1592 {
1593 if (idle > max_idle)
1594 {
1595 --idle;
1596 X_UNLOCK (reqlock);
1597 X_LOCK (wrklock);
1598 --started;
1599 X_UNLOCK (wrklock);
1600 goto quit;
1601 }
1602
1603 /* we are allowed to idle, so do so without any timeout */
1604 X_COND_WAIT (reqwait, reqlock);
1605 }
1606
1607 --idle;
1608 }
1609
1610 --nready;
1611
1612 X_UNLOCK (reqlock);
1613
1614 if (req->type < 0)
1615 goto quit;
1616
1617 if (!EIO_CANCELLED (req))
1618 ETP_EXECUTE (self, req);
1619
1620 X_LOCK (reslock);
1621
1622 ++npending;
1623
1624 if (!reqq_push (&res_queue, req) && want_poll_cb)
1625 want_poll_cb ();
1626
1627 self->req = 0;
1628 etp_worker_clear (self);
1629
1630 X_UNLOCK (reslock);
1631 }
1632
1633quit:
1634 X_LOCK (wrklock);
1635 etp_worker_free (self);
1636 X_UNLOCK (wrklock);
1637
1638 return 0;
1639}
1640
1641/*****************************************************************************/ 1729/*****************************************************************************/
1642 1730
1731int ecb_cold
1643int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1732eio_init (void (*want_poll)(void), void (*done_poll)(void))
1644{ 1733{
1645 return etp_init (want_poll, done_poll); 1734 eio_want_poll_cb = want_poll;
1646} 1735 eio_done_poll_cb = done_poll;
1647 1736
1737 return etp_init (EIO_POOL, 0, 0, 0);
1738}
1739
1740ecb_inline void
1648static void eio_api_destroy (eio_req *req) 1741eio_api_destroy (eio_req *req)
1649{ 1742{
1650 free (req); 1743 free (req);
1651} 1744}
1652 1745
1653#define REQ(rtype) \ 1746#define REQ(rtype) \
1654 eio_req *req; \ 1747 eio_req *req; \
1655 \ 1748 \
1656 req = (eio_req *)calloc (1, sizeof *req); \ 1749 req = (eio_req *)calloc (1, sizeof *req); \
1657 if (!req) \ 1750 if (!req) \
1658 return 0; \ 1751 return 0; \
… …
1672 { \ 1765 { \
1673 eio_api_destroy (req); \ 1766 eio_api_destroy (req); \
1674 return 0; \ 1767 return 0; \
1675 } 1768 }
1676 1769
1770#define SINGLEDOT(ptr) (0[(char *)(ptr)] == '.' && !1[(char *)(ptr)])
1771
1772static void
1677static void eio_execute (etp_worker *self, eio_req *req) 1773eio_execute (etp_worker *self, eio_req *req)
1678{ 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
1679 switch (req->type) 1804 switch (req->type)
1680 { 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;
1681 case EIO_READ: ALLOC (req->size); 1813 case EIO_READ: ALLOC (req->size);
1682 req->result = req->offs >= 0 1814 req->result = req->offs >= 0
1683 ? pread (req->int1, req->ptr2, req->size, req->offs) 1815 ? pread (req->int1, req->ptr2, req->size, req->offs)
1684 : read (req->int1, req->ptr2, req->size); break; 1816 : read (req->int1, req->ptr2, req->size); break;
1685 case EIO_WRITE: req->result = req->offs >= 0 1817 case EIO_WRITE: req->result = req->offs >= 0
1686 ? pwrite (req->int1, req->ptr2, req->size, req->offs) 1818 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1687 : write (req->int1, req->ptr2, req->size); break; 1819 : write (req->int1, req->ptr2, req->size); break;
1688 1820
1689 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;
1690 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
1691 1825
1692 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1826 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1693 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;
1694 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));
1695 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
1696 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1932 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1697 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; 1933 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1698 1934
1699 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1700 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1701 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS)); 1935 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1702 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break; 1936 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1703 1937
1704 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1705 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;
1706 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1707 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;
1708 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1709 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break; 1940 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1710 1941
1711 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
1712 case EIO_CLOSE: req->result = close (req->int1); break; 1942 case EIO_CLOSE: req->result = close (req->int1); break;
1713 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break; 1943 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1714 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
1715 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1716 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1717 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1718 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1719 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1720 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1721
1722 case EIO_READLINK: ALLOC (PATH_MAX);
1723 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1724
1725 case EIO_SYNC: req->result = 0; sync (); break; 1944 case EIO_SYNC: req->result = 0; sync (); break;
1726 case EIO_FSYNC: req->result = fsync (req->int1); break; 1945 case EIO_FSYNC: req->result = fsync (req->int1); break;
1727 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;
1728 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;
1729 case EIO_MTOUCH: req->result = eio__mtouch (req); break; 1950 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1730 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;
1731 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break; 1952 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1732 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;
1733 1954
1734 case EIO_READDIR: eio__scandir (req, self); break; 1955 case EIO_READDIR: eio__scandir (req, self); break;
1735 1956
1736 case EIO_BUSY: 1957 case EIO_BUSY:
1737#ifdef _WIN32 1958#ifdef _WIN32
… …
1746 req->result = select (0, 0, 0, 0, &tv); 1967 req->result = select (0, 0, 0, 0, &tv);
1747 } 1968 }
1748#endif 1969#endif
1749 break; 1970 break;
1750 1971
1751 case EIO_UTIME: 1972#if 0
1752 case EIO_FUTIME:
1753 {
1754 struct timeval tv[2];
1755 struct timeval *times;
1756
1757 if (req->nv1 != -1. || req->nv2 != -1.)
1758 {
1759 tv[0].tv_sec = req->nv1;
1760 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
1761 tv[1].tv_sec = req->nv2;
1762 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
1763
1764 times = tv;
1765 }
1766 else
1767 times = 0;
1768
1769 req->result = req->type == EIO_FUTIME
1770 ? futimes (req->int1, times)
1771 : utimes (req->ptr1, times);
1772 }
1773 break;
1774
1775 case EIO_GROUP: 1973 case EIO_GROUP:
1776 abort (); /* handled in eio_request */ 1974 abort (); /* handled in eio_request */
1975#endif
1777 1976
1778 case EIO_NOP: 1977 case EIO_NOP:
1779 req->result = 0; 1978 req->result = 0;
1780 break; 1979 break;
1781 1980
1782 case EIO_CUSTOM: 1981 case EIO_CUSTOM:
1783 ((void (*)(eio_req *))req->feed) (req); 1982 req->feed (req);
1784 break; 1983 break;
1785 1984
1786 default: 1985 default:
1787 errno = ENOSYS;
1788 req->result = -1; 1986 req->result = EIO_ENOSYS ();
1789 break; 1987 break;
1790 } 1988 }
1791 1989
1792 req->errorno = errno; 1990 req->errorno = errno;
1793} 1991}
1794 1992
1795#ifndef EIO_NO_WRAPPERS 1993#ifndef EIO_NO_WRAPPERS
1796 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
1797eio_req *eio_nop (int pri, eio_cb cb, void *data) 2005eio_req *eio_nop (int pri, eio_cb cb, void *data)
1798{ 2006{
1799 REQ (EIO_NOP); SEND; 2007 REQ (EIO_NOP); SEND;
1800} 2008}
1801 2009
… …
1817eio_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)
1818{ 2026{
1819 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;
1820} 2028}
1821 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
1822eio_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)
1823{ 2046{
1824 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;
1825} 2048}
1826 2049
… …
1832eio_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)
1833{ 2056{
1834 REQ (EIO_MLOCKALL); req->int1 = flags; SEND; 2057 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
1835} 2058}
1836 2059
1837eio_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)
1838{ 2061{
1839 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;
1840}
1841
1842eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1843{
1844 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1845} 2063}
1846 2064
1847eio_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)
1848{ 2066{
1849 REQ (EIO_CLOSE); req->int1 = fd; SEND; 2067 REQ (EIO_CLOSE); req->int1 = fd; SEND;
… …
1852eio_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)
1853{ 2071{
1854 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;
1855} 2073}
1856 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
1857eio_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)
1858{ 2081{
1859 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;
1860} 2083}
1861 2084
… …
1887eio_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)
1888{ 2111{
1889 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2112 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1890} 2113}
1891 2114
1892eio_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)
1893{ 2116{
1894 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;
1895} 2118}
1896 2119
1897eio_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)
… …
1917eio_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)
1918{ 2141{
1919 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2142 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1920} 2143}
1921 2144
1922eio_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)
1923{ 2146{
1924 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;
1925} 2148}
1926 2149
1927eio_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)
… …
1941} 2164}
1942 2165
1943eio_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)
1944{ 2167{
1945 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);
1946} 2174}
1947 2175
1948eio_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)
1949{ 2177{
1950 return eio__1path (EIO_STAT, path, pri, cb, data); 2178 return eio__1path (EIO_STAT, path, pri, cb, data);
… …
2009eio_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)
2010{ 2238{
2011 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2239 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
2012} 2240}
2013 2241
2014eio_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)
2015{ 2243{
2016 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND; 2244 REQ (EIO_CUSTOM); req->feed = execute; SEND;
2017} 2245}
2018 2246
2019#endif 2247#endif
2020 2248
2021eio_req *eio_grp (eio_cb cb, void *data) 2249eio_req *eio_grp (eio_cb cb, void *data)
… …
2030#undef SEND 2258#undef SEND
2031 2259
2032/*****************************************************************************/ 2260/*****************************************************************************/
2033/* grp functions */ 2261/* grp functions */
2034 2262
2263void
2035void 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)
2036{ 2265{
2037 grp->int2 = limit; 2266 grp->int2 = limit;
2038 grp->feed = feed; 2267 grp->feed = feed;
2039 2268
2040 grp_try_feed (grp); 2269 grp_try_feed (grp);
2041} 2270}
2042 2271
2272void
2043void eio_grp_limit (eio_req *grp, int limit) 2273eio_grp_limit (eio_req *grp, int limit)
2044{ 2274{
2045 grp->int2 = limit; 2275 grp->int2 = limit;
2046 2276
2047 grp_try_feed (grp); 2277 grp_try_feed (grp);
2048} 2278}
2049 2279
2280void
2050void eio_grp_add (eio_req *grp, eio_req *req) 2281eio_grp_add (eio_req *grp, eio_req *req)
2051{ 2282{
2052 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));
2053 2284
2054 grp->flags |= EIO_FLAG_GROUPADD; 2285 grp->flags |= ETP_FLAG_GROUPADD;
2055 2286
2056 ++grp->size; 2287 ++grp->size;
2057 req->grp = grp; 2288 req->grp = grp;
2058 2289
2059 req->grp_prev = 0; 2290 req->grp_prev = 0;
… …
2066} 2297}
2067 2298
2068/*****************************************************************************/ 2299/*****************************************************************************/
2069/* misc garbage */ 2300/* misc garbage */
2070 2301
2302eio_ssize_t
2071ssize_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)
2072{ 2304{
2073 etp_worker wrk;
2074 ssize_t ret;
2075
2076 wrk.dbuf = 0;
2077
2078 ret = eio__sendfile (ofd, ifd, offset, count, &wrk); 2305 return eio__sendfile (ofd, ifd, offset, count);
2079
2080 if (wrk.dbuf)
2081 free (wrk.dbuf);
2082
2083 return ret;
2084} 2306}
2085 2307

Diff Legend

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