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

Comparing libeio/eio.c (file contents):
Revision 1.70 by root, Fri Jun 10 06:55:10 2011 UTC vs.
Revision 1.143 by root, Sat Dec 3 16:33:46 2016 UTC

1/* 1/*
2 * libeio implementation 2 * libeio implementation
3 * 3 *
4 * Copyright (c) 2007,2008,2009,2010,2011 Marc Alexander Lehmann <libeio@schmorp.de> 4 * Copyright (c) 2007,2008,2009,2010,2011,2012,2013,2016 Marc Alexander Lehmann <libeio@schmorp.de>
5 * All rights reserved. 5 * All rights reserved.
6 * 6 *
7 * Redistribution and use in source and binary forms, with or without modifica- 7 * Redistribution and use in source and binary forms, with or without modifica-
8 * tion, are permitted provided that the following conditions are met: 8 * tion, are permitted provided that the following conditions are met:
9 * 9 *
10 * 1. Redistributions of source code must retain the above copyright notice, 10 * 1. Redistributions of source code must retain the above copyright notice,
11 * this list of conditions and the following disclaimer. 11 * this list of conditions and the following disclaimer.
12 * 12 *
13 * 2. Redistributions in binary form must reproduce the above copyright 13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the 14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution. 15 * documentation and/or other materials provided with the distribution.
16 * 16 *
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR IMPLIED 17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR IMPLIED
18 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MER- 18 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MER-
19 * CHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO 19 * CHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO
20 * EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPE- 20 * EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPE-
21 * CIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, 21 * CIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
… …
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
46#ifdef EIO_STACKSIZE
47# define XTHREAD_STACKSIZE EIO_STACKSIZE
48#endif
49#include "xthread.h" 45#include "ecb.h"
50 46
51#include <errno.h> 47#include <errno.h>
52#include <stddef.h> 48#include <stddef.h>
53#include <stdlib.h> 49#include <stdlib.h>
54#include <string.h> 50#include <string.h>
55#include <errno.h> 51#include <errno.h>
56#include <sys/types.h> 52#include <sys/types.h>
57#include <sys/stat.h> 53#include <sys/stat.h>
58#include <sys/statvfs.h>
59#include <limits.h> 54#include <limits.h>
60#include <fcntl.h> 55#include <fcntl.h>
61#include <assert.h> 56#include <assert.h>
62 57
63/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */ 58/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
64/* intptr_t only comes form stdint.h, says idiot openbsd coder */ 59/* intptr_t only comes from stdint.h, says idiot openbsd coder */
65#if HAVE_STDINT_H 60#if HAVE_STDINT_H
66# include <stdint.h> 61# include <stdint.h>
67#endif 62#endif
68 63
64#ifndef ECANCELED
65# define ECANCELED EDOM
66#endif
67#ifndef ELOOP
68# define ELOOP EDOM
69#endif
70
71#if !defined(ENOTSOCK) && defined(WSAENOTSOCK)
72# define ENOTSOCK WSAENOTSOCK
73#endif
74
75static void eio_destroy (eio_req *req);
76
69#ifndef EIO_FINISH 77#ifndef EIO_FINISH
70# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0 78# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
71#endif 79#endif
72 80
73#ifndef EIO_DESTROY 81#ifndef EIO_DESTROY
… …
76 84
77#ifndef EIO_FEED 85#ifndef EIO_FEED
78# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0) 86# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)
79#endif 87#endif
80 88
89#ifndef EIO_FD_TO_WIN32_HANDLE
90# define EIO_FD_TO_WIN32_HANDLE(fd) _get_osfhandle (fd)
91#endif
92#ifndef EIO_WIN32_HANDLE_TO_FD
93# define EIO_WIN32_HANDLE_TO_FD(handle) _open_osfhandle (handle, 0)
94#endif
95
96#define EIO_ERRNO(errval,retval) ((errno = errval), retval)
97
98#define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1)
99
81#ifdef _WIN32 100#ifdef _WIN32
82 101
83 /*doh*/ 102 #undef PAGESIZE
103 #define PAGESIZE 4096 /* GetSystemInfo? */
104
105 /* TODO: look at how perl does stat (non-sloppy), unlink (ro-files), utime, link */
106
107 #ifdef EIO_STRUCT_STATI64
108 /* look at perl's non-sloppy stat */
109 #define stat(path,buf) _stati64 (path,buf)
110 #define fstat(fd,buf) _fstati64 (fd,buf)
111 #endif
112 #define lstat(path,buf) stat (path,buf)
113 #define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
114 #define mkdir(path,mode) _mkdir (path)
115 #define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
116
117 #define chmod(path,mode) _chmod (path, mode)
118 #define dup(fd) _dup (fd)
119 #define dup2(fd1,fd2) _dup2 (fd1, fd2)
120 #define pipe(fds) _pipe (fds, 4096, O_BINARY)
121
122 #define fcntl(fd,cmd,arg) EIO_ENOSYS ()
123 #define ioctl(fd,cmd,arg) EIO_ENOSYS ()
124 #define fchmod(fd,mode) EIO_ENOSYS ()
125 #define chown(path,uid,gid) EIO_ENOSYS ()
126 #define fchown(fd,uid,gid) EIO_ENOSYS ()
127 #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
128 #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
129 #define mknod(path,mode,dev) EIO_ENOSYS ()
130 #define sync() EIO_ENOSYS ()
131 #define readlink(path,buf,s) EIO_ENOSYS ()
132 #define statvfs(path,buf) EIO_ENOSYS ()
133 #define fstatvfs(fd,buf) EIO_ENOSYS ()
134
135 #define pread(fd,buf,count,offset) eio__pread (fd, buf, count, offset)
136 #define pwrite(fd,buf,count,offset) eio__pwrite (fd, buf, count, offset)
137
138 #if __GNUC__
139 typedef long long eio_off_t; /* signed for compatibility to msvc */
140 #else
141 typedef __int64 eio_off_t; /* unsigned not supported by msvc */
142 #endif
143
144 static eio_ssize_t
145 eio__pread (int fd, void *buf, eio_ssize_t count, eio_off_t offset)
146 {
147 OVERLAPPED o = { 0 };
148 DWORD got;
149
150 o.Offset = offset;
151 o.OffsetHigh = offset >> 32;
152
153 return ReadFile ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd), buf, count, &got, &o)
154 ? got : -1;
155 }
156
157 static eio_ssize_t
158 eio__pwrite (int fd, void *buf, eio_ssize_t count, eio_off_t offset)
159 {
160 OVERLAPPED o = { 0 };
161 DWORD got;
162
163 o.Offset = offset;
164 o.OffsetHigh = offset >> 32;
165
166 return WriteFile ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd), buf, count, &got, &o)
167 ? got : -1;
168 }
169
170 /* rename() uses MoveFile, which fails to overwrite */
171 #define rename(old,neu) eio__rename (old, neu)
172
173 static int
174 eio__rename (const char *old, const char *neu)
175 {
176 if (MoveFileEx (old, neu, MOVEFILE_REPLACE_EXISTING))
177 return 0;
178
179 /* should steal _dosmaperr */
180 switch (GetLastError ())
181 {
182 case ERROR_FILE_NOT_FOUND:
183 case ERROR_PATH_NOT_FOUND:
184 case ERROR_INVALID_DRIVE:
185 case ERROR_NO_MORE_FILES:
186 case ERROR_BAD_NETPATH:
187 case ERROR_BAD_NET_NAME:
188 case ERROR_BAD_PATHNAME:
189 case ERROR_FILENAME_EXCED_RANGE:
190 errno = ENOENT;
191 break;
192
193 default:
194 errno = EACCES;
195 break;
196 }
197
198 return -1;
199 }
200
201 /* we could even stat and see if it exists */
202 static int
203 symlink (const char *old, const char *neu)
204 {
205 #if WINVER >= 0x0600
206 int flags;
207
208 /* This tries out all combinations of SYMBOLIC_LINK_FLAG_DIRECTORY
209 * and SYMBOLIC_LINK_FLAG_ALLOW_UNPRIVILEGED_CREATE,
210 * with directory first.
211 */
212 for (flags = 3; flags >= 0; --flags)
213 if (CreateSymbolicLink (neu, old, flags))
214 return 0;
215 #endif
216
217 return EIO_ERRNO (ENOENT, -1);
218 }
219
220 /* POSIX API only, causing trouble for win32 apps */
221 #define CreateHardLink(neu,old,flags) 0 /* not really creating hardlink, still using relative paths? */
222 #define CreateSymbolicLink(neu,old,flags) 0 /* vista+ only */
223
224 struct statvfs
225 {
226 int dummy;
227 };
228
229 #define DT_DIR EIO_DT_DIR
230 #define DT_REG EIO_DT_REG
231 #define D_NAME(entp) entp.cFileName
232 #define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
233
84#else 234#else
85 235
86# include <sys/time.h> 236 #include <sys/time.h>
87# include <sys/select.h> 237 #include <sys/select.h>
88# include <unistd.h> 238 #include <unistd.h>
239 #include <signal.h>
240 #include <dirent.h>
241
242 #ifdef ANDROID
243 #include <sys/vfs.h>
244 #define statvfs statfs
245 #define fstatvfs fstatfs
246 #include <asm/page.h> /* supposedly limits.h does #define PAGESIZE PAGESIZE */
247 #else
248 #include <sys/statvfs.h>
249 #endif
250
251 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
252 #include <sys/mman.h>
253 #endif
254
255 #define D_NAME(entp) entp->d_name
256
257 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
258 #if __FreeBSD__ || __NetBSD__ || __OpenBSD__
259 #define _DIRENT_HAVE_D_TYPE /* sigh */
260 #define D_INO(de) (de)->d_fileno
261 #define D_NAMLEN(de) (de)->d_namlen
262 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
263 #define D_INO(de) (de)->d_ino
264 #endif
265
266 #ifdef _D_EXACT_NAMLEN
267 #undef D_NAMLEN
268 #define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
269 #endif
270
271 #ifdef _DIRENT_HAVE_D_TYPE
272 #define D_TYPE(de) (de)->d_type
273 #endif
274
275 #ifndef EIO_STRUCT_DIRENT
276 #define EIO_STRUCT_DIRENT struct dirent
277 #endif
278
279#endif
280
281#if HAVE_UTIMES
89# include <utime.h> 282# 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 283#endif
105 284
106#ifdef _D_EXACT_NAMLEN 285#if HAVE_SYS_SYSCALL_H
107# undef D_NAMLEN 286# 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
114
115# ifndef EIO_STRUCT_DIRENT
116# define EIO_STRUCT_DIRENT struct dirent
117# endif
118
119#endif 287#endif
120 288
121#if HAVE_SENDFILE 289#if HAVE_SENDFILE
122# if __linux 290# if __linux
123# include <sys/sendfile.h> 291# include <sys/sendfile.h>
… …
138#endif 306#endif
139#ifndef D_INO 307#ifndef D_INO
140# define D_INO(de) 0 308# define D_INO(de) 0
141#endif 309#endif
142#ifndef D_NAMLEN 310#ifndef D_NAMLEN
143# define D_NAMLEN(de) strlen ((de)->d_name) 311# define D_NAMLEN(entp) strlen (D_NAME (entp))
144#endif 312#endif
145 313
146/* used for struct dirent, AIX doesn't provide it */ 314/* used for struct dirent, AIX doesn't provide it */
147#ifndef NAME_MAX 315#ifndef NAME_MAX
148# define NAME_MAX 4096 316# define NAME_MAX 4096
… …
154#endif 322#endif
155 323
156/* buffer size for various temporary buffers */ 324/* buffer size for various temporary buffers */
157#define EIO_BUFSIZE 65536 325#define EIO_BUFSIZE 65536
158 326
159#define dBUF \ 327#define dBUF \
160 char *eio_buf; \
161 ETP_WORKER_LOCK (self); \
162 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 328 char *eio_buf = malloc (EIO_BUFSIZE); \
163 ETP_WORKER_UNLOCK (self); \
164 errno = ENOMEM; \ 329 errno = ENOMEM; \
165 if (!eio_buf) \ 330 if (!eio_buf) \
166 return -1; 331 return -1
167 332
168#define EIO_TICKS ((1000000 + 1023) >> 10) 333#define FUBd \
334 free (eio_buf)
169 335
170/*****************************************************************************/ 336/*****************************************************************************/
171 337
172#if __GNUC__ >= 3 338struct etp_tmpbuf;
173# define expect(expr,value) __builtin_expect ((expr),(value)) 339
340#if _POSIX_VERSION >= 200809L
341 #define HAVE_AT 1
342 #define WD2FD(wd) ((wd) ? (wd)->fd : AT_FDCWD)
343 #ifndef O_SEARCH
344 #define O_SEARCH O_RDONLY
345 #endif
174#else 346#else
175# define expect(expr,value) (expr) 347 #define HAVE_AT 0
348 static const char *wd_expand (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path);
176#endif 349#endif
177 350
178#define expect_false(expr) expect ((expr) != 0, 0) 351struct eio_pwd
179#define expect_true(expr) expect ((expr) != 0, 1) 352{
353#if HAVE_AT
354 int fd;
355#endif
356 int len;
357 char str[1]; /* actually, a 0-terminated canonical path */
358};
180 359
181/*****************************************************************************/ 360/*****************************************************************************/
182 361
183#define ETP_PRI_MIN EIO_PRI_MIN 362#define ETP_PRI_MIN EIO_PRI_MIN
184#define ETP_PRI_MAX EIO_PRI_MAX 363#define ETP_PRI_MAX EIO_PRI_MAX
185 364
365#define ETP_TYPE_QUIT -1
366#define ETP_TYPE_GROUP EIO_GROUP
367
368static void eio_nop_callback (void) { }
369static void (*eio_want_poll_cb)(void) = eio_nop_callback;
370static void (*eio_done_poll_cb)(void) = eio_nop_callback;
371
372#define ETP_WANT_POLL(pool) eio_want_poll_cb ()
373#define ETP_DONE_POLL(pool) eio_done_poll_cb ()
374
186struct etp_worker; 375struct etp_worker;
187
188#define ETP_REQ eio_req 376#define ETP_REQ eio_req
189#define ETP_DESTROY(req) eio_destroy (req) 377#define ETP_DESTROY(req) eio_destroy (req)
190static int eio_finish (eio_req *req); 378static int eio_finish (eio_req *req);
191#define ETP_FINISH(req) eio_finish (req) 379#define ETP_FINISH(req) eio_finish (req)
192static void eio_execute (struct etp_worker *self, eio_req *req); 380static void eio_execute (struct etp_worker *self, eio_req *req);
193#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) 381#define ETP_EXECUTE(wrk,req) eio_execute (wrk, req)
194 382
195#define ETP_WORKER_CLEAR(req) \ 383#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 }
207 384
208#define ETP_WORKER_COMMON \ 385static struct etp_pool eio_pool;
209 void *dbuf; \ 386#define EIO_POOL (&eio_pool)
210 DIR *dirp;
211 387
212/*****************************************************************************/ 388/*****************************************************************************/
213 389
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 390static 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) 391grp_try_feed (eio_req *grp)
671{ 392{
672 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 393 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
673 { 394 {
674 grp->flags &= ~EIO_FLAG_GROUPADD; 395 grp->flags &= ~ETP_FLAG_GROUPADD;
675 396
676 EIO_FEED (grp); 397 EIO_FEED (grp);
677 398
678 /* stop if no progress has been made */ 399 /* stop if no progress has been made */
679 if (!(grp->flags & EIO_FLAG_GROUPADD)) 400 if (!(grp->flags & ETP_FLAG_GROUPADD))
680 { 401 {
681 grp->feed = 0; 402 grp->feed = 0;
682 break; 403 break;
683 } 404 }
684 } 405 }
685} 406}
686 407
408static int
687static int grp_dec (eio_req *grp) 409grp_dec (eio_req *grp)
688{ 410{
689 --grp->size; 411 --grp->size;
690 412
691 /* call feeder, if applicable */ 413 /* call feeder, if applicable */
692 grp_try_feed (grp); 414 grp_try_feed (grp);
693 415
694 /* finish, if done */ 416 /* finish, if done */
695 if (!grp->size && grp->int1) 417 if (!grp->size && grp->flags & ETP_FLAG_DELAYED)
696 return eio_finish (grp); 418 return eio_finish (grp);
697 else 419 else
698 return 0; 420 return 0;
699} 421}
700 422
423static void
701void eio_destroy (eio_req *req) 424eio_destroy (eio_req *req)
702{ 425{
703 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); 426 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
704 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); 427 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
705 428
706 EIO_DESTROY (req); 429 EIO_DESTROY (req);
707} 430}
708 431
432static int
709static int eio_finish (eio_req *req) 433eio_finish (eio_req *req)
710{ 434{
711 int res = EIO_FINISH (req); 435 int res = EIO_FINISH (req);
712 436
713 if (req->grp) 437 if (req->grp)
714 { 438 {
… …
722 if (grp->grp_first == req) 446 if (grp->grp_first == req)
723 grp->grp_first = req->grp_next; 447 grp->grp_first = req->grp_next;
724 448
725 res2 = grp_dec (grp); 449 res2 = grp_dec (grp);
726 450
727 if (!res && res2) 451 if (!res)
728 res = res2; 452 res = res2;
729 } 453 }
730 454
731 eio_destroy (req); 455 eio_destroy (req);
732 456
733 return res; 457 return res;
734} 458}
735 459
460void
736void eio_grp_cancel (eio_req *grp) 461eio_grp_cancel (eio_req *grp)
737{ 462{
738 for (grp = grp->grp_first; grp; grp = grp->grp_next) 463 etp_grp_cancel (EIO_POOL, grp);
739 eio_cancel (grp);
740} 464}
741 465
466void
742void eio_cancel (eio_req *req) 467eio_cancel (eio_req *req)
743{ 468{
744 etp_cancel (req); 469 etp_cancel (EIO_POOL, req);
745} 470}
746 471
472void
747void eio_submit (eio_req *req) 473eio_submit (eio_req *req)
748{ 474{
749 etp_submit (req); 475 etp_submit (EIO_POOL, req);
750} 476}
751 477
752unsigned int eio_nreqs (void) 478unsigned int
479eio_nreqs (void)
753{ 480{
754 return etp_nreqs (); 481 return etp_nreqs (EIO_POOL);
755} 482}
756 483
757unsigned int eio_nready (void) 484unsigned int
485eio_nready (void)
758{ 486{
759 return etp_nready (); 487 return etp_nready (EIO_POOL);
760} 488}
761 489
762unsigned int eio_npending (void) 490unsigned int
491eio_npending (void)
763{ 492{
764 return etp_npending (); 493 return etp_npending (EIO_POOL);
765} 494}
766 495
767unsigned int eio_nthreads (void) 496unsigned int ecb_cold
497eio_nthreads (void)
768{ 498{
769 return etp_nthreads (); 499 return etp_nthreads (EIO_POOL);
770} 500}
771 501
502void ecb_cold
772void eio_set_max_poll_time (double nseconds) 503eio_set_max_poll_time (double nseconds)
773{ 504{
774 etp_set_max_poll_time (nseconds); 505 etp_set_max_poll_time (EIO_POOL, nseconds);
775} 506}
776 507
508void ecb_cold
777void eio_set_max_poll_reqs (unsigned int maxreqs) 509eio_set_max_poll_reqs (unsigned int maxreqs)
778{ 510{
779 etp_set_max_poll_reqs (maxreqs); 511 etp_set_max_poll_reqs (EIO_POOL, maxreqs);
780} 512}
781 513
514void ecb_cold
782void eio_set_max_idle (unsigned int nthreads) 515eio_set_max_idle (unsigned int nthreads)
783{ 516{
784 etp_set_max_idle (nthreads); 517 etp_set_max_idle (EIO_POOL, nthreads);
785} 518}
786 519
520void ecb_cold
787void eio_set_idle_timeout (unsigned int seconds) 521eio_set_idle_timeout (unsigned int seconds)
788{ 522{
789 etp_set_idle_timeout (seconds); 523 etp_set_idle_timeout (EIO_POOL, seconds);
790} 524}
791 525
526void ecb_cold
792void eio_set_min_parallel (unsigned int nthreads) 527eio_set_min_parallel (unsigned int nthreads)
793{ 528{
794 etp_set_min_parallel (nthreads); 529 etp_set_min_parallel (EIO_POOL, nthreads);
795} 530}
796 531
532void ecb_cold
797void eio_set_max_parallel (unsigned int nthreads) 533eio_set_max_parallel (unsigned int nthreads)
798{ 534{
799 etp_set_max_parallel (nthreads); 535 etp_set_max_parallel (EIO_POOL, nthreads);
800} 536}
801 537
802int eio_poll (void) 538int eio_poll (void)
803{ 539{
804 return etp_poll (); 540 return etp_poll (EIO_POOL);
805} 541}
806 542
807/*****************************************************************************/ 543/*****************************************************************************/
808/* work around various missing functions */ 544/* 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
848 545
849#ifndef HAVE_UTIMES 546#ifndef HAVE_UTIMES
850 547
851# undef utimes 548# undef utimes
852# define utimes(path,times) eio__utimes (path, times) 549# define utimes(path,times) eio__utimes (path, times)
… …
872#ifndef HAVE_FUTIMES 569#ifndef HAVE_FUTIMES
873 570
874# undef futimes 571# undef futimes
875# define futimes(fd,times) eio__futimes (fd, times) 572# define futimes(fd,times) eio__futimes (fd, times)
876 573
574static int
877static int eio__futimes (int fd, const struct timeval tv[2]) 575eio__futimes (int fd, const struct timeval tv[2])
878{ 576{
879 errno = ENOSYS; 577 errno = ENOSYS;
880 return -1; 578 return -1;
881} 579}
882 580
… …
885#if !HAVE_FDATASYNC 583#if !HAVE_FDATASYNC
886# undef fdatasync 584# undef fdatasync
887# define fdatasync(fd) fsync (fd) 585# define fdatasync(fd) fsync (fd)
888#endif 586#endif
889 587
588static int
589eio__syncfs (int fd)
590{
591 int res;
592
593#if HAVE_SYS_SYNCFS
594 res = (int)syscall (__NR_syncfs, (int)(fd));
595#else
596 res = EIO_ENOSYS ();
597#endif
598
599 if (res < 0 && errno == ENOSYS && fd >= 0)
600 sync ();
601
602 return res;
603}
604
890/* sync_file_range always needs emulation */ 605/* sync_file_range always needs emulation */
891int 606static int
892eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags) 607eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
893{ 608{
894#if HAVE_SYNC_FILE_RANGE 609#if HAVE_SYNC_FILE_RANGE
895 int res; 610 int res;
896 611
… …
913 /* even though we could play tricks with the flags, it's better to always 628 /* 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 */ 629 * call fdatasync, as that matches the expectation of its users best */
915 return fdatasync (fd); 630 return fdatasync (fd);
916} 631}
917 632
633static int
634eio__fallocate (int fd, int mode, off_t offset, size_t len)
635{
636#if HAVE_LINUX_FALLOCATE
637 return fallocate (fd, mode, offset, len);
638#else
639 return EIO_ENOSYS ();
640#endif
641}
642
918#if !HAVE_READAHEAD 643#if !HAVE_READAHEAD
919# undef readahead 644# undef readahead
920# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 645# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
921 646
922static ssize_t 647static eio_ssize_t
923eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) 648eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
924{ 649{
925 size_t todo = count; 650 size_t todo = count;
926 dBUF; 651 dBUF;
927 652
… …
932 pread (fd, eio_buf, len, offset); 657 pread (fd, eio_buf, len, offset);
933 offset += len; 658 offset += len;
934 todo -= len; 659 todo -= len;
935 } 660 }
936 661
937 errno = 0; 662 FUBd;
663
664 /* linux's readahead basically only fails for EBADF or EINVAL (not mmappable) */
665 /* but not for e.g. EIO or eof, so we also never fail */
938 return count; 666 return 0;
939} 667}
940 668
941#endif 669#endif
942 670
943/* sendfile always needs emulation */ 671/* sendfile always needs emulation */
944static ssize_t 672static eio_ssize_t
945eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) 673eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
946{ 674{
947 ssize_t written = 0; 675 eio_ssize_t written = 0;
948 ssize_t res; 676 eio_ssize_t res;
949 677
950 if (!count) 678 if (!count)
951 return 0; 679 return 0;
952 680
953 for (;;) 681 for (;;)
954 { 682 {
683#ifdef __APPLE__
684# undef HAVE_SENDFILE /* broken, as everything on os x */
685#endif
955#if HAVE_SENDFILE 686#if HAVE_SENDFILE
956# if __linux 687# if __linux
957 off_t soffset = offset; 688 off_t soffset = offset;
958 res = sendfile (ofd, ifd, &soffset, count); 689 res = sendfile (ofd, ifd, &soffset, count);
959 690
… …
975 706
976 /* according to source inspection, this is correct, and useful behaviour */ 707 /* according to source inspection, this is correct, and useful behaviour */
977 if (sbytes) 708 if (sbytes)
978 res = sbytes; 709 res = sbytes;
979 710
980# elif defined (__APPLE__) 711# elif defined __APPLE__
981 off_t sbytes = count; 712 off_t sbytes = count;
982 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0); 713 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
983 714
984 /* according to the manpage, sbytes is always valid */ 715 /* according to the manpage, sbytes is always valid */
985 if (sbytes) 716 if (sbytes)
… …
1002 if (res < 0 && sbytes) 733 if (res < 0 && sbytes)
1003 res = sbytes; 734 res = sbytes;
1004 735
1005# endif 736# endif
1006 737
1007#elif defined (_WIN32) 738#elif defined (_WIN32) && 0
1008 /* does not work, just for documentation of what would need to be done */ 739 /* does not work, just for documentation of what would need to be done */
1009 /* actually, cannot be done like this, as TransmitFile changes the file offset, */ 740 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
1010 /* libeio guarantees that the file offset does not change, and windows */ 741 /* libeio guarantees that the file offset does not change, and windows */
1011 /* has no way to get an independent handle to the same file description */ 742 /* has no way to get an independent handle to the same file description */
1012 HANDLE h = TO_SOCKET (ifd); 743 HANDLE h = TO_SOCKET (ifd);
1013 SetFilePointer (h, offset, 0, FILE_BEGIN); 744 SetFilePointer (h, offset, 0, FILE_BEGIN);
1014 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0); 745 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1015 746
1016#else 747#else
1017 res = -1; 748 res = EIO_ENOSYS ();
1018 errno = ENOSYS;
1019#endif 749#endif
1020 750
1021 /* we assume sendfile can copy at least 128mb in one go */ 751 /* we assume sendfile can copy at least 128mb in one go */
1022 if (res <= 128 * 1024 * 1024) 752 if (res <= 128 * 1024 * 1024)
1023 { 753 {
… …
1045 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 775 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1046 /* BSDs */ 776 /* BSDs */
1047#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */ 777#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1048 || errno == ENOTSUP 778 || errno == ENOTSUP
1049#endif 779#endif
780#ifdef EOPNOTSUPP /* windows */
1050 || errno == EOPNOTSUPP /* BSDs */ 781 || errno == EOPNOTSUPP /* BSDs */
782#endif
1051#if __solaris 783#if __solaris
1052 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 784 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
1053#endif 785#endif
1054 ) 786 )
1055 ) 787 )
… …
1059 791
1060 res = 0; 792 res = 0;
1061 793
1062 while (count) 794 while (count)
1063 { 795 {
1064 ssize_t cnt; 796 eio_ssize_t cnt;
1065 797
1066 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 798 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
1067 799
1068 if (cnt <= 0) 800 if (cnt <= 0)
1069 { 801 {
… …
1081 813
1082 offset += cnt; 814 offset += cnt;
1083 res += cnt; 815 res += cnt;
1084 count -= cnt; 816 count -= cnt;
1085 } 817 }
818
819 FUBd;
1086 } 820 }
1087 821
1088 return res; 822 return res;
823}
824
825#ifdef PAGESIZE
826# define eio_pagesize() PAGESIZE
827#else
828static intptr_t
829eio_pagesize (void)
830{
831 static intptr_t page;
832
833 if (!page)
834 page = sysconf (_SC_PAGESIZE);
835
836 return page;
837}
838#endif
839
840static void
841eio_page_align (void **addr, size_t *length)
842{
843 intptr_t mask = eio_pagesize () - 1;
844
845 /* round down addr */
846 intptr_t adj = mask & (intptr_t)*addr;
847
848 *addr = (void *)((intptr_t)*addr - adj);
849 *length += adj;
850
851 /* round up length */
852 *length = (*length + mask) & ~mask;
853}
854
855#if !_POSIX_MEMLOCK
856# define eio__mlockall(a) EIO_ENOSYS ()
857#else
858
859static int
860eio__mlockall (int flags)
861{
862 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
863 extern int mallopt (int, int);
864 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
865 #endif
866
867 if (EIO_MCL_CURRENT != MCL_CURRENT
868 || EIO_MCL_FUTURE != MCL_FUTURE)
869 {
870 flags = 0
871 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
872 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
873 }
874
875 return mlockall (flags);
876}
877#endif
878
879#if !_POSIX_MEMLOCK_RANGE
880# define eio__mlock(a,b) EIO_ENOSYS ()
881#else
882
883static int
884eio__mlock (void *addr, size_t length)
885{
886 eio_page_align (&addr, &length);
887
888 return mlock (addr, length);
889}
890
891#endif
892
893#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
894# define eio__msync(a,b,c) EIO_ENOSYS ()
895#else
896
897static int
898eio__msync (void *mem, size_t len, int flags)
899{
900 eio_page_align (&mem, &len);
901
902 if (EIO_MS_ASYNC != MS_SYNC
903 || EIO_MS_INVALIDATE != MS_INVALIDATE
904 || EIO_MS_SYNC != MS_SYNC)
905 {
906 flags = 0
907 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
908 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
909 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
910 }
911
912 return msync (mem, len, flags);
913}
914
915#endif
916
917static int
918eio__mtouch (eio_req *req)
919{
920 void *mem = req->ptr2;
921 size_t len = req->size;
922 int flags = req->int1;
923
924 eio_page_align (&mem, &len);
925
926 {
927 intptr_t addr = (intptr_t)mem;
928 intptr_t end = addr + len;
929 intptr_t page = eio_pagesize ();
930
931 if (addr < end)
932 if (flags & EIO_MT_MODIFY) /* modify */
933 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
934 else
935 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
936 }
937
938 return 0;
939}
940
941/*****************************************************************************/
942/* requests implemented outside eio_execute, because they are so large */
943
944static void
945eio__lseek (eio_req *req)
946{
947 /* this usually gets optimised away completely, or your compiler sucks, */
948 /* or the whence constants really are not 0, 1, 2 */
949 int whence = req->int2 == EIO_SEEK_SET ? SEEK_SET
950 : req->int2 == EIO_SEEK_CUR ? SEEK_CUR
951 : req->int2 == EIO_SEEK_END ? SEEK_END
952 : req->int2;
953
954 req->offs = lseek (req->int1, req->offs, whence);
955 req->result = req->offs == (off_t)-1 ? -1 : 0;
956}
957
958/* result will always end up in tmpbuf, there is always space for adding a 0-byte */
959static int
960eio__realpath (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
961{
962 char *res;
963 const char *rel = path;
964 char *tmp1, *tmp2;
965#if SYMLOOP_MAX > 32
966 int symlinks = SYMLOOP_MAX;
967#else
968 int symlinks = 32;
969#endif
970
971 errno = EINVAL;
972 if (!rel)
973 return -1;
974
975 errno = ENOENT;
976 if (!*rel)
977 return -1;
978
979 res = etp_tmpbuf_get (tmpbuf, PATH_MAX * 3);
980#ifdef _WIN32
981 if (_access (rel, 4) != 0)
982 return -1;
983
984 symlinks = GetFullPathName (rel, PATH_MAX * 3, res, 0);
985
986 errno = ENAMETOOLONG;
987 if (symlinks >= PATH_MAX * 3)
988 return -1;
989
990 errno = EIO;
991 if (symlinks <= 0)
992 return -1;
993
994 return symlinks;
995
996#else
997 tmp1 = res + PATH_MAX;
998 tmp2 = tmp1 + PATH_MAX;
999
1000#if 0 /* disabled, the musl way to do things is just too racy */
1001#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1002 /* on linux we may be able to ask the kernel */
1003 {
1004 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1005
1006 if (fd >= 0)
1007 {
1008 sprintf (tmp1, "/proc/self/fd/%d", fd);
1009 req->result = readlink (tmp1, res, PATH_MAX);
1010 /* here we should probably stat the open file and the disk file, to make sure they still match */
1011 close (fd);
1012
1013 if (req->result > 0)
1014 goto done;
1015 }
1016 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1017 return -1;
1018 }
1019#endif
1020#endif
1021
1022 if (*rel != '/')
1023 {
1024 int len;
1025
1026 errno = ENOENT;
1027 if (wd == EIO_INVALID_WD)
1028 return -1;
1029
1030 if (wd == EIO_CWD)
1031 {
1032 if (!getcwd (res, PATH_MAX))
1033 return -1;
1034
1035 len = strlen (res);
1036 }
1037 else
1038 memcpy (res, wd->str, len = wd->len);
1039
1040 if (res [1]) /* only use if not / */
1041 res += len;
1042 }
1043
1044 while (*rel)
1045 {
1046 eio_ssize_t len, linklen;
1047 const char *beg = rel;
1048
1049 while (*rel && *rel != '/')
1050 ++rel;
1051
1052 len = rel - beg;
1053
1054 if (!len) /* skip slashes */
1055 {
1056 ++rel;
1057 continue;
1058 }
1059
1060 if (beg [0] == '.')
1061 {
1062 if (len == 1)
1063 continue; /* . - nop */
1064
1065 if (beg [1] == '.' && len == 2)
1066 {
1067 /* .. - back up one component, if possible */
1068
1069 while (res != tmpbuf->ptr)
1070 if (*--res == '/')
1071 break;
1072
1073 continue;
1074 }
1075 }
1076
1077 errno = ENAMETOOLONG;
1078 if (res + 1 + len + 1 >= tmp1)
1079 return -1;
1080
1081 /* copy one component */
1082 *res = '/';
1083 memcpy (res + 1, beg, len);
1084
1085 /* zero-terminate, for readlink */
1086 res [len + 1] = 0;
1087
1088 /* now check if it's a symlink */
1089 linklen = readlink (tmpbuf->ptr, tmp1, PATH_MAX);
1090
1091 if (linklen < 0)
1092 {
1093 if (errno != EINVAL)
1094 return -1;
1095
1096 /* it's a normal directory. hopefully */
1097 res += len + 1;
1098 }
1099 else
1100 {
1101 /* yay, it was a symlink - build new path in tmp2 */
1102 int rellen = strlen (rel);
1103
1104 errno = ENAMETOOLONG;
1105 if (linklen + 1 + rellen >= PATH_MAX)
1106 return -1;
1107
1108 errno = ELOOP;
1109 if (!--symlinks)
1110 return -1;
1111
1112 if (*tmp1 == '/')
1113 res = tmpbuf->ptr; /* symlink resolves to an absolute path */
1114
1115 /* we need to be careful, as rel might point into tmp2 already */
1116 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1117 tmp2 [linklen] = '/';
1118 memcpy (tmp2, tmp1, linklen);
1119
1120 rel = tmp2;
1121 }
1122 }
1123
1124 /* special case for the lone root path */
1125 if (res == tmpbuf->ptr)
1126 *res++ = '/';
1127
1128 return res - (char *)tmpbuf->ptr;
1129#endif
1089} 1130}
1090 1131
1091static signed char 1132static signed char
1092eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) 1133eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1093{ 1134{
… …
1101 1142
1102#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */ 1143#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1103#define EIO_SORT_FAST 60 /* when to only use insertion sort */ 1144#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1104 1145
1105static void 1146static void
1106eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1147eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1107{ 1148{
1108 unsigned char bits [9 + sizeof (ino_t) * 8]; 1149 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1109 unsigned char *bit = bits; 1150 unsigned char *bit = bits;
1110 1151
1111 assert (CHAR_BIT == 8); 1152 assert (CHAR_BIT == 8);
1112 assert (sizeof (eio_dirent) * 8 < 256); 1153 assert (sizeof (eio_dirent) * 8 < 256);
1113 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */ 1154 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
… …
1115 1156
1116 if (size <= EIO_SORT_FAST) 1157 if (size <= EIO_SORT_FAST)
1117 return; 1158 return;
1118 1159
1119 /* first prepare an array of bits to test in our radix sort */ 1160 /* first prepare an array of bits to test in our radix sort */
1120 /* try to take endianness into account, as well as differences in ino_t sizes */ 1161 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1121 /* inode_bits must contain all inodes ORed together */ 1162 /* inode_bits must contain all inodes ORed together */
1122 /* which is used to skip bits that are 0 everywhere, which is very common */ 1163 /* which is used to skip bits that are 0 everywhere, which is very common */
1123 { 1164 {
1124 ino_t endianness; 1165 eio_ino_t endianness;
1125 int i, j; 1166 int i, j;
1126 1167
1127 /* we store the byte offset of byte n into byte n of "endianness" */ 1168 /* we store the byte offset of byte n into byte n of "endianness" */
1128 for (i = 0; i < sizeof (ino_t); ++i) 1169 for (i = 0; i < sizeof (eio_ino_t); ++i)
1129 ((unsigned char *)&endianness)[i] = i; 1170 ((unsigned char *)&endianness)[i] = i;
1130 1171
1131 *bit++ = 0; 1172 *bit++ = 0;
1132 1173
1133 for (i = 0; i < sizeof (ino_t); ++i) 1174 for (i = 0; i < sizeof (eio_ino_t); ++i)
1134 { 1175 {
1135 /* shifting off the byte offsets out of "endianness" */ 1176 /* shifting off the byte offsets out of "endianness" */
1136 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8; 1177 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1137 endianness >>= 8; 1178 endianness >>= 8;
1138 1179
1139 for (j = 0; j < 8; ++j) 1180 for (j = 0; j < 8; ++j)
1140 if (inode_bits & (((ino_t)1) << (i * 8 + j))) 1181 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1141 *bit++ = offs + j; 1182 *bit++ = offs + j;
1142 } 1183 }
1143 1184
1144 for (j = 0; j < 8; ++j) 1185 for (j = 0; j < 8; ++j)
1145 if (score_bits & (1 << j)) 1186 if (score_bits & (1 << j))
1146 *bit++ = offsetof (eio_dirent, score) * 8 + j; 1187 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1147 } 1188 }
1148 1189
1149 /* now actually do the sorting (a variant of MSD radix sort) */ 1190 /* now actually do the sorting (a variant of MSD radix sort) */
1150 { 1191 {
1151 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base; 1192 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1152 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end; 1193 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1153 unsigned char *bit_stk [9 + sizeof (ino_t) * 8]; 1194 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1154 int stk_idx = 0; 1195 int stk_idx = 0;
1155 1196
1156 base_stk [stk_idx] = dents; 1197 base_stk [stk_idx] = dents;
1157 end_stk [stk_idx] = dents + size; 1198 end_stk [stk_idx] = dents + size;
1158 bit_stk [stk_idx] = bit - 1; 1199 bit_stk [stk_idx] = bit - 1;
… …
1237 } 1278 }
1238 } 1279 }
1239} 1280}
1240 1281
1241static void 1282static void
1242eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1283eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1243{ 1284{
1244 if (size <= 1) 1285 if (size <= 1)
1245 return; /* our insertion sort relies on size > 0 */ 1286 return; /* our insertion sort relies on size > 0 */
1246 1287
1247 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */ 1288 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
… …
1255 1296
1256/* read a full directory */ 1297/* read a full directory */
1257static void 1298static void
1258eio__scandir (eio_req *req, etp_worker *self) 1299eio__scandir (eio_req *req, etp_worker *self)
1259{ 1300{
1260 DIR *dirp;
1261 EIO_STRUCT_DIRENT *entp;
1262 char *name, *names; 1301 char *name, *names;
1263 int namesalloc = 4096; 1302 int namesalloc = 4096 - sizeof (void *) * 4;
1264 int namesoffs = 0; 1303 int namesoffs = 0;
1265 int flags = req->int1; 1304 int flags = req->int1;
1266 eio_dirent *dents = 0; 1305 eio_dirent *dents = 0;
1267 int dentalloc = 128; 1306 int dentalloc = 128;
1268 int dentoffs = 0; 1307 int dentoffs = 0;
1269 ino_t inode_bits = 0; 1308 eio_ino_t inode_bits = 0;
1309#ifdef _WIN32
1310 HANDLE dirp;
1311 WIN32_FIND_DATA entp;
1312#else
1313 DIR *dirp;
1314 EIO_STRUCT_DIRENT *entp;
1315#endif
1270 1316
1271 req->result = -1; 1317 req->result = -1;
1272 1318
1273 if (!(flags & EIO_READDIR_DENTS)) 1319 if (!(flags & EIO_READDIR_DENTS))
1274 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); 1320 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1275 1321
1276 X_LOCK (wrklock); 1322#ifdef _WIN32
1277 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 1323 {
1324 int len = strlen ((const char *)req->ptr1);
1325 char *path = malloc (MAX_PATH);
1326 const char *fmt;
1327 const char *reqpath = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1328
1329 if (!len)
1330 fmt = "./*";
1331 else if (reqpath[len - 1] == '/' || reqpath[len - 1] == '\\')
1332 fmt = "%s*";
1333 else
1334 fmt = "%s/*";
1335
1336 _snprintf (path, MAX_PATH, fmt, reqpath);
1337 dirp = FindFirstFile (path, &entp);
1338 free (path);
1339
1340 if (dirp == INVALID_HANDLE_VALUE)
1341 {
1342 /* should steal _dosmaperr */
1343 switch (GetLastError ())
1344 {
1345 case ERROR_FILE_NOT_FOUND:
1346 req->result = 0;
1347 break;
1348
1349 case ERROR_INVALID_NAME:
1350 case ERROR_PATH_NOT_FOUND:
1351 case ERROR_NO_MORE_FILES:
1352 errno = ENOENT;
1353 break;
1354
1355 case ERROR_NOT_ENOUGH_MEMORY:
1356 errno = ENOMEM;
1357 break;
1358
1359 default:
1360 errno = EINVAL;
1361 break;
1362 }
1363
1364 return;
1365 }
1366 }
1367#else
1368 #if HAVE_AT
1369 if (req->wd)
1370 {
1371 int fd = openat (WD2FD (req->wd), req->ptr1, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1372
1373 if (fd < 0)
1374 return;
1375
1376 dirp = fdopendir (fd);
1377
1378 if (!dirp)
1379 close (fd);
1380 }
1381 else
1278 self->dirp = dirp = opendir (req->ptr1); 1382 dirp = opendir (req->ptr1);
1383 #else
1384 dirp = opendir (wd_expand (&self->tmpbuf, req->wd, req->ptr1));
1385 #endif
1386
1387 if (!dirp)
1388 return;
1389#endif
1390
1391 if (req->flags & EIO_FLAG_PTR1_FREE)
1392 free (req->ptr1);
1279 1393
1280 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1394 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1281 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; 1395 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1282 req->ptr2 = names = malloc (namesalloc); 1396 req->ptr2 = names = malloc (namesalloc);
1283 X_UNLOCK (wrklock);
1284 1397
1285 if (dirp && names && (!flags || dents)) 1398 if (!names || (flags && !dents))
1399 return;
1400
1286 for (;;) 1401 for (;;)
1287 { 1402 {
1403 int done;
1404
1405#ifdef _WIN32
1406 done = !dirp;
1407#else
1288 errno = 0; 1408 errno = 0;
1289 entp = readdir (dirp); 1409 entp = readdir (dirp);
1410 done = !entp;
1411#endif
1290 1412
1291 if (!entp) 1413 if (done)
1292 { 1414 {
1415#ifndef _WIN32
1416 int old_errno = errno;
1417 closedir (dirp);
1418 errno = old_errno;
1419
1293 if (errno) 1420 if (errno)
1294 break; 1421 break;
1422#endif
1295 1423
1296 /* sort etc. */ 1424 /* sort etc. */
1297 req->int1 = flags; 1425 req->int1 = flags;
1298 req->result = dentoffs; 1426 req->result = dentoffs;
1299 1427
1300 if (flags & EIO_READDIR_STAT_ORDER) 1428 if (flags & EIO_READDIR_STAT_ORDER)
1301 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits); 1429 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1302 else if (flags & EIO_READDIR_DIRS_FIRST) 1430 else if (flags & EIO_READDIR_DIRS_FIRST)
1303 if (flags & EIO_READDIR_FOUND_UNKNOWN) 1431 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1304 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */ 1432 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1305 else 1433 else
1434 {
1435 /* in this case, all is known, and we just put dirs first and sort them */
1436 eio_dirent *oth = dents + dentoffs;
1437 eio_dirent *dir = dents;
1438
1439 /* now partition dirs to the front, and non-dirs to the back */
1440 /* by walking from both sides and swapping if necessary */
1441 while (oth > dir)
1442 {
1443 if (dir->type == EIO_DT_DIR)
1444 ++dir;
1445 else if ((--oth)->type == EIO_DT_DIR)
1446 {
1447 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1448
1449 ++dir;
1450 }
1451 }
1452
1453 /* now sort the dirs only (dirs all have the same score) */
1454 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1455 }
1456
1457 break;
1458 }
1459
1460 /* now add the entry to our list(s) */
1461 name = D_NAME (entp);
1462
1463 /* skip . and .. entries */
1464 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1465 {
1466 int len = D_NAMLEN (entp) + 1;
1467
1468 while (ecb_expect_false (namesoffs + len > namesalloc))
1469 {
1470 namesalloc *= 2;
1471 req->ptr2 = names = realloc (names, namesalloc);
1472
1473 if (!names)
1474 break;
1475 }
1476
1477 memcpy (names + namesoffs, name, len);
1478
1479 if (dents)
1480 {
1481 struct eio_dirent *ent;
1482
1483 if (ecb_expect_false (dentoffs == dentalloc))
1306 { 1484 {
1307 /* in this case, all is known, and we just put dirs first and sort them */ 1485 dentalloc *= 2;
1486 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1487
1488 if (!dents)
1489 break;
1490 }
1491
1308 eio_dirent *oth = dents + dentoffs; 1492 ent = dents + dentoffs;
1309 eio_dirent *dir = dents;
1310 1493
1311 /* now partition dirs to the front, and non-dirs to the back */ 1494 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1312 /* by walking from both sides and swapping if necessary */ 1495 ent->namelen = len - 1;
1313 while (oth > dir) 1496 ent->inode = D_INO (entp);
1497
1498 inode_bits |= ent->inode;
1499
1500 switch (D_TYPE (entp))
1501 {
1502 default:
1503 ent->type = EIO_DT_UNKNOWN;
1504 flags |= EIO_READDIR_FOUND_UNKNOWN;
1505 break;
1506
1507 #ifdef DT_FIFO
1508 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1509 #endif
1510 #ifdef DT_CHR
1511 case DT_CHR: ent->type = EIO_DT_CHR; break;
1512 #endif
1513 #ifdef DT_MPC
1514 case DT_MPC: ent->type = EIO_DT_MPC; break;
1515 #endif
1516 #ifdef DT_DIR
1517 case DT_DIR: ent->type = EIO_DT_DIR; break;
1518 #endif
1519 #ifdef DT_NAM
1520 case DT_NAM: ent->type = EIO_DT_NAM; break;
1521 #endif
1522 #ifdef DT_BLK
1523 case DT_BLK: ent->type = EIO_DT_BLK; break;
1524 #endif
1525 #ifdef DT_MPB
1526 case DT_MPB: ent->type = EIO_DT_MPB; break;
1527 #endif
1528 #ifdef DT_REG
1529 case DT_REG: ent->type = EIO_DT_REG; break;
1530 #endif
1531 #ifdef DT_NWK
1532 case DT_NWK: ent->type = EIO_DT_NWK; break;
1533 #endif
1534 #ifdef DT_CMP
1535 case DT_CMP: ent->type = EIO_DT_CMP; break;
1536 #endif
1537 #ifdef DT_LNK
1538 case DT_LNK: ent->type = EIO_DT_LNK; break;
1539 #endif
1540 #ifdef DT_SOCK
1541 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1542 #endif
1543 #ifdef DT_DOOR
1544 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1545 #endif
1546 #ifdef DT_WHT
1547 case DT_WHT: ent->type = EIO_DT_WHT; break;
1548 #endif
1549 }
1550
1551 ent->score = 7;
1552
1553 if (flags & EIO_READDIR_DIRS_FIRST)
1554 {
1555 if (ent->type == EIO_DT_UNKNOWN)
1314 { 1556 {
1315 if (dir->type == EIO_DT_DIR) 1557 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1316 ++dir; 1558 ent->score = 1;
1317 else if ((--oth)->type == EIO_DT_DIR) 1559 else if (!strchr (name, '.')) /* absence of dots indicate likely dirs */
1318 { 1560 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1319 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1320
1321 ++dir;
1322 }
1323 } 1561 }
1324 1562 else if (ent->type == EIO_DT_DIR)
1325 /* now sort the dirs only (dirs all have the same score) */ 1563 ent->score = 0;
1326 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1327 } 1564 }
1328
1329 break;
1330 }
1331
1332 /* now add the entry to our list(s) */
1333 name = entp->d_name;
1334
1335 /* skip . and .. entries */
1336 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1337 {
1338 int len = D_NAMLEN (entp) + 1;
1339
1340 while (expect_false (namesoffs + len > namesalloc))
1341 {
1342 namesalloc *= 2;
1343 X_LOCK (wrklock);
1344 req->ptr2 = names = realloc (names, namesalloc);
1345 X_UNLOCK (wrklock);
1346
1347 if (!names)
1348 break;
1349 } 1565 }
1350 1566
1351 memcpy (names + namesoffs, name, len);
1352
1353 if (dents)
1354 {
1355 struct eio_dirent *ent;
1356
1357 if (expect_false (dentoffs == dentalloc))
1358 {
1359 dentalloc *= 2;
1360 X_LOCK (wrklock);
1361 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1362 X_UNLOCK (wrklock);
1363
1364 if (!dents)
1365 break;
1366 }
1367
1368 ent = dents + dentoffs;
1369
1370 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1371 ent->namelen = len - 1;
1372 ent->inode = D_INO (entp);
1373
1374 inode_bits |= ent->inode;
1375
1376 switch (D_TYPE (entp))
1377 {
1378 default:
1379 ent->type = EIO_DT_UNKNOWN;
1380 flags |= EIO_READDIR_FOUND_UNKNOWN;
1381 break;
1382
1383 #ifdef DT_FIFO
1384 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1385 #endif
1386 #ifdef DT_CHR
1387 case DT_CHR: ent->type = EIO_DT_CHR; break;
1388 #endif
1389 #ifdef DT_MPC
1390 case DT_MPC: ent->type = EIO_DT_MPC; break;
1391 #endif
1392 #ifdef DT_DIR
1393 case DT_DIR: ent->type = EIO_DT_DIR; break;
1394 #endif
1395 #ifdef DT_NAM
1396 case DT_NAM: ent->type = EIO_DT_NAM; break;
1397 #endif
1398 #ifdef DT_BLK
1399 case DT_BLK: ent->type = EIO_DT_BLK; break;
1400 #endif
1401 #ifdef DT_MPB
1402 case DT_MPB: ent->type = EIO_DT_MPB; break;
1403 #endif
1404 #ifdef DT_REG
1405 case DT_REG: ent->type = EIO_DT_REG; break;
1406 #endif
1407 #ifdef DT_NWK
1408 case DT_NWK: ent->type = EIO_DT_NWK; break;
1409 #endif
1410 #ifdef DT_CMP
1411 case DT_CMP: ent->type = EIO_DT_CMP; break;
1412 #endif
1413 #ifdef DT_LNK
1414 case DT_LNK: ent->type = EIO_DT_LNK; break;
1415 #endif
1416 #ifdef DT_SOCK
1417 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1418 #endif
1419 #ifdef DT_DOOR
1420 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1421 #endif
1422 #ifdef DT_WHT
1423 case DT_WHT: ent->type = EIO_DT_WHT; break;
1424 #endif
1425 }
1426
1427 ent->score = 7;
1428
1429 if (flags & EIO_READDIR_DIRS_FIRST)
1430 {
1431 if (ent->type == EIO_DT_UNKNOWN)
1432 {
1433 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1434 ent->score = 1;
1435 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1436 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1437 }
1438 else if (ent->type == EIO_DT_DIR)
1439 ent->score = 0;
1440 }
1441 }
1442
1443 namesoffs += len; 1567 namesoffs += len;
1444 ++dentoffs; 1568 ++dentoffs;
1445 } 1569 }
1446 1570
1447 if (EIO_CANCELLED (req)) 1571 if (EIO_CANCELLED (req))
1448 { 1572 {
1449 errno = ECANCELED; 1573 errno = ECANCELED;
1450 break; 1574 break;
1451 } 1575 }
1576
1577#ifdef _WIN32
1578 if (!FindNextFile (dirp, &entp))
1579 {
1580 FindClose (dirp);
1581 dirp = 0;
1452 } 1582 }
1583#endif
1584 }
1453} 1585}
1454 1586
1455#ifdef PAGESIZE 1587/*****************************************************************************/
1456# define eio_pagesize() PAGESIZE 1588/* working directory stuff */
1457#else 1589/* various deficiencies in the posix 2008 api force us to */
1458static intptr_t 1590/* keep the absolute path in string form at all times */
1459eio_pagesize (void) 1591/* fuck yeah. */
1460{
1461 static intptr_t page;
1462 1592
1463 if (!page) 1593#if !HAVE_AT
1464 page = sysconf (_SC_PAGESIZE);
1465 1594
1595/* a bit like realpath, but usually faster because it doesn'T have to return */
1596/* an absolute or canonical path */
1597static const char *
1598wd_expand (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
1599{
1600 if (!wd || *path == '/')
1601 return path;
1602
1603 if (path [0] == '.' && !path [1])
1604 return wd->str;
1605
1606 {
1607 int l1 = wd->len;
1608 int l2 = strlen (path);
1609
1610 char *res = etp_tmpbuf_get (tmpbuf, l1 + l2 + 2);
1611
1612 memcpy (res, wd->str, l1);
1613 res [l1] = '/';
1614 memcpy (res + l1 + 1, path, l2 + 1);
1615
1616 return res;
1617 }
1618}
1619
1620#endif
1621
1622static eio_wd
1623eio__wd_open_sync (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
1624{
1625 int fd;
1626 eio_wd res;
1627 int len = eio__realpath (tmpbuf, wd, path);
1628
1629 if (len < 0)
1630 return EIO_INVALID_WD;
1631
1632#if HAVE_AT
1633 fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1634
1635 if (fd < 0)
1636 return EIO_INVALID_WD;
1637#endif
1638
1639 res = malloc (sizeof (*res) + len); /* one extra 0-byte */
1640
1641#if HAVE_AT
1642 res->fd = fd;
1643#endif
1644
1645 res->len = len;
1646 memcpy (res->str, tmpbuf->ptr, len);
1647 res->str [len] = 0;
1648
1466 return page; 1649 return res;
1467} 1650}
1468#endif
1469 1651
1470static void 1652eio_wd
1471eio_page_align (void **addr, size_t *length) 1653eio_wd_open_sync (eio_wd wd, const char *path)
1472{ 1654{
1473 intptr_t mask = eio_pagesize () - 1; 1655 struct etp_tmpbuf tmpbuf = { };
1656 wd = eio__wd_open_sync (&tmpbuf, wd, path);
1657 free (tmpbuf.ptr);
1474 1658
1475 /* round down addr */ 1659 return wd;
1476 intptr_t adj = mask & (intptr_t)*addr;
1477
1478 *addr = (void *)((intptr_t)*addr - adj);
1479 *length += adj;
1480
1481 /* round up length */
1482 *length = (*length + mask) & ~mask;
1483} 1660}
1484 1661
1485#if !_POSIX_MEMLOCK 1662void
1486# define eio__mlockall(a) ((errno = ENOSYS), -1) 1663eio_wd_close_sync (eio_wd wd)
1487#else 1664{
1665 if (wd != EIO_INVALID_WD && wd != EIO_CWD)
1666 {
1667 #if HAVE_AT
1668 close (wd->fd);
1669 #endif
1670 free (wd);
1671 }
1672}
1673
1674#if HAVE_AT
1675
1676/* they forgot these */
1488 1677
1489static int 1678static int
1490eio__mlockall (int flags) 1679eio__truncateat (int dirfd, const char *path, off_t length)
1491{ 1680{
1492 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7 1681 int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
1493 extern int mallopt (int, int); 1682 int res;
1494 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1495 #endif
1496 1683
1497 if (EIO_MCL_CURRENT != MCL_CURRENT 1684 if (fd < 0)
1498 || EIO_MCL_FUTURE != MCL_FUTURE) 1685 return fd;
1499 {
1500 flags = 0
1501 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1502 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1503 }
1504 1686
1505 return mlockall (flags); 1687 res = ftruncate (fd, length);
1688 close (fd);
1689 return res;
1506} 1690}
1507#endif
1508
1509#if !_POSIX_MEMLOCK_RANGE
1510# define eio__mlock(a,b) ((errno = ENOSYS), -1)
1511#else
1512 1691
1513static int 1692static int
1514eio__mlock (void *addr, size_t length) 1693eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
1515{ 1694{
1516 eio_page_align (&addr, &length); 1695 int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
1696 int res;
1517 1697
1518 return mlock (addr, length); 1698 if (fd < 0)
1519} 1699 return fd;
1520 1700
1521#endif 1701 res = fstatvfs (fd, buf);
1522 1702 close (fd);
1523#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1524# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
1525#else
1526
1527int
1528eio__msync (void *mem, size_t len, int flags)
1529{
1530 eio_page_align (&mem, &len);
1531
1532 if (EIO_MS_ASYNC != MS_SYNC
1533 || EIO_MS_INVALIDATE != MS_INVALIDATE
1534 || EIO_MS_SYNC != MS_SYNC)
1535 {
1536 flags = 0
1537 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1538 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1539 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1540 }
1541
1542 return msync (mem, len, flags);
1543}
1544
1545#endif
1546
1547int
1548eio__mtouch (eio_req *req)
1549{
1550 void *mem = req->ptr2;
1551 size_t len = req->size;
1552 int flags = req->int1;
1553
1554 eio_page_align (&mem, &len);
1555
1556 {
1557 intptr_t addr = (intptr_t)mem;
1558 intptr_t end = addr + len;
1559 intptr_t page = eio_pagesize ();
1560
1561 if (addr < end)
1562 if (flags & EIO_MT_MODIFY) /* modify */
1563 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1564 else
1565 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1566 }
1567
1568 return 0; 1703 return res;
1704
1569} 1705}
1706
1707#endif
1570 1708
1571/*****************************************************************************/ 1709/*****************************************************************************/
1572 1710
1573#define ALLOC(len) \ 1711#define ALLOC(len) \
1574 if (!req->ptr2) \ 1712 if (!req->ptr2) \
1575 { \ 1713 { \
1576 X_LOCK (wrklock); \ 1714 X_LOCK (EIO_POOL->wrklock); \
1577 req->flags |= EIO_FLAG_PTR2_FREE; \ 1715 req->flags |= EIO_FLAG_PTR2_FREE; \
1578 X_UNLOCK (wrklock); \ 1716 X_UNLOCK (EIO_POOL->wrklock); \
1579 req->ptr2 = malloc (len); \ 1717 req->ptr2 = malloc (len); \
1580 if (!req->ptr2) \ 1718 if (!req->ptr2) \
1581 { \ 1719 { \
1582 errno = ENOMEM; \ 1720 errno = ENOMEM; \
1583 req->result = -1; \ 1721 req->result = -1; \
1584 break; \ 1722 break; \
1585 } \ 1723 } \
1586 } 1724 }
1587 1725
1588X_THREAD_PROC (etp_proc)
1589{
1590 ETP_REQ *req;
1591 struct timespec ts;
1592 etp_worker *self = (etp_worker *)thr_arg;
1593
1594 /* try to distribute timeouts somewhat randomly */
1595 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1596
1597 for (;;)
1598 {
1599 X_LOCK (reqlock);
1600
1601 for (;;)
1602 {
1603 self->req = req = reqq_shift (&req_queue);
1604
1605 if (req)
1606 break;
1607
1608 ++idle;
1609
1610 ts.tv_sec = time (0) + idle_timeout;
1611 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
1612 {
1613 if (idle > max_idle)
1614 {
1615 --idle;
1616 X_UNLOCK (reqlock);
1617 X_LOCK (wrklock);
1618 --started;
1619 X_UNLOCK (wrklock);
1620 goto quit;
1621 }
1622
1623 /* we are allowed to idle, so do so without any timeout */
1624 X_COND_WAIT (reqwait, reqlock);
1625 }
1626
1627 --idle;
1628 }
1629
1630 --nready;
1631
1632 X_UNLOCK (reqlock);
1633
1634 if (req->type < 0)
1635 goto quit;
1636
1637 if (!EIO_CANCELLED (req))
1638 ETP_EXECUTE (self, req);
1639
1640 X_LOCK (reslock);
1641
1642 ++npending;
1643
1644 if (!reqq_push (&res_queue, req) && want_poll_cb)
1645 want_poll_cb ();
1646
1647 self->req = 0;
1648 etp_worker_clear (self);
1649
1650 X_UNLOCK (reslock);
1651 }
1652
1653quit:
1654 X_LOCK (wrklock);
1655 etp_worker_free (self);
1656 X_UNLOCK (wrklock);
1657
1658 return 0;
1659}
1660
1661/*****************************************************************************/ 1726/*****************************************************************************/
1662 1727
1728int ecb_cold
1663int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1729eio_init (void (*want_poll)(void), void (*done_poll)(void))
1664{ 1730{
1665 return etp_init (want_poll, done_poll); 1731 eio_want_poll_cb = want_poll;
1666} 1732 eio_done_poll_cb = done_poll;
1667 1733
1734 return etp_init (EIO_POOL, 0, 0, 0);
1735}
1736
1737ecb_inline void
1668static void eio_api_destroy (eio_req *req) 1738eio_api_destroy (eio_req *req)
1669{ 1739{
1670 free (req); 1740 free (req);
1671} 1741}
1672 1742
1673#define REQ(rtype) \ 1743#define REQ(rtype) \
1674 eio_req *req; \ 1744 eio_req *req; \
1675 \ 1745 \
1676 req = (eio_req *)calloc (1, sizeof *req); \ 1746 req = (eio_req *)calloc (1, sizeof *req); \
1677 if (!req) \ 1747 if (!req) \
1678 return 0; \ 1748 return 0; \
… …
1692 { \ 1762 { \
1693 eio_api_destroy (req); \ 1763 eio_api_destroy (req); \
1694 return 0; \ 1764 return 0; \
1695 } 1765 }
1696 1766
1767#define SINGLEDOT(ptr) (0[(char *)(ptr)] == '.' && !1[(char *)(ptr)])
1768
1769static void
1697static void eio_execute (etp_worker *self, eio_req *req) 1770eio_execute (etp_worker *self, eio_req *req)
1698{ 1771{
1772#if HAVE_AT
1773 int dirfd;
1774#else
1775 const char *path;
1776#endif
1777
1778 if (ecb_expect_false (EIO_CANCELLED (req)))
1779 {
1780 req->result = -1;
1781 req->errorno = ECANCELED;
1782 return;
1783 }
1784
1785 if (ecb_expect_false (req->wd == EIO_INVALID_WD))
1786 {
1787 req->result = -1;
1788 req->errorno = ENOENT;
1789 return;
1790 }
1791
1792 if (req->type >= EIO_OPEN)
1793 {
1794 #if HAVE_AT
1795 dirfd = WD2FD (req->wd);
1796 #else
1797 path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1798 #endif
1799 }
1800
1699 switch (req->type) 1801 switch (req->type)
1700 { 1802 {
1803 case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
1804 req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
1805 break;
1806 case EIO_WD_CLOSE: req->result = 0;
1807 eio_wd_close_sync (req->wd); break;
1808
1809 case EIO_SEEK: eio__lseek (req); break;
1701 case EIO_READ: ALLOC (req->size); 1810 case EIO_READ: ALLOC (req->size);
1702 req->result = req->offs >= 0 1811 req->result = req->offs >= 0
1703 ? pread (req->int1, req->ptr2, req->size, req->offs) 1812 ? pread (req->int1, req->ptr2, req->size, req->offs)
1704 : read (req->int1, req->ptr2, req->size); break; 1813 : read (req->int1, req->ptr2, req->size); break;
1705 case EIO_WRITE: req->result = req->offs >= 0 1814 case EIO_WRITE: req->result = req->offs >= 0
1706 ? pwrite (req->int1, req->ptr2, req->size, req->offs) 1815 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1707 : write (req->int1, req->ptr2, req->size); break; 1816 : write (req->int1, req->ptr2, req->size); break;
1708 1817
1818 case EIO_FCNTL: req->result = fcntl (req->int1, (int) req->int2, req->ptr2); break;
1819 case EIO_IOCTL: req->result = ioctl (req->int1, (unsigned long)req->int2, req->ptr2); break;
1820
1709 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;
1710 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
1711 1825
1712 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1826 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1713 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;
1714 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));
1715 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
1716 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1932 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1717 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; 1933 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1718 1934
1719 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1720 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1721 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS)); 1935 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1722 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break; 1936 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1723 1937
1724 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1725 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;
1726 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1727 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;
1728 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1729 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break; 1940 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1730 1941
1731 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
1732 case EIO_CLOSE: req->result = close (req->int1); break; 1942 case EIO_CLOSE: req->result = close (req->int1); break;
1733 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break; 1943 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1734 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
1735 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1736 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1737 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1738 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1739 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1740 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1741
1742 case EIO_READLINK: ALLOC (PATH_MAX);
1743 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1744
1745 case EIO_SYNC: req->result = 0; sync (); break; 1944 case EIO_SYNC: req->result = 0; sync (); break;
1746 case EIO_FSYNC: req->result = fsync (req->int1); break; 1945 case EIO_FSYNC: req->result = fsync (req->int1); break;
1747 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;
1748 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;
1749 case EIO_MTOUCH: req->result = eio__mtouch (req); break; 1950 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1750 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;
1751 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break; 1952 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1752 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;
1753 1954
1754 case EIO_READDIR: eio__scandir (req, self); break; 1955 case EIO_READDIR: eio__scandir (req, self); break;
1755 1956
1756 case EIO_BUSY: 1957 case EIO_BUSY:
1757#ifdef _WIN32 1958#ifdef _WIN32
… …
1766 req->result = select (0, 0, 0, 0, &tv); 1967 req->result = select (0, 0, 0, 0, &tv);
1767 } 1968 }
1768#endif 1969#endif
1769 break; 1970 break;
1770 1971
1771 case EIO_UTIME: 1972#if 0
1772 case EIO_FUTIME:
1773 {
1774 struct timeval tv[2];
1775 struct timeval *times;
1776
1777 if (req->nv1 != -1. || req->nv2 != -1.)
1778 {
1779 tv[0].tv_sec = req->nv1;
1780 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
1781 tv[1].tv_sec = req->nv2;
1782 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
1783
1784 times = tv;
1785 }
1786 else
1787 times = 0;
1788
1789 req->result = req->type == EIO_FUTIME
1790 ? futimes (req->int1, times)
1791 : utimes (req->ptr1, times);
1792 }
1793 break;
1794
1795 case EIO_GROUP: 1973 case EIO_GROUP:
1796 abort (); /* handled in eio_request */ 1974 abort (); /* handled in eio_request */
1975#endif
1797 1976
1798 case EIO_NOP: 1977 case EIO_NOP:
1799 req->result = 0; 1978 req->result = 0;
1800 break; 1979 break;
1801 1980
1802 case EIO_CUSTOM: 1981 case EIO_CUSTOM:
1803 req->feed (req); 1982 req->feed (req);
1804 break; 1983 break;
1805 1984
1806 default: 1985 default:
1807 errno = ENOSYS;
1808 req->result = -1; 1986 req->result = EIO_ENOSYS ();
1809 break; 1987 break;
1810 } 1988 }
1811 1989
1812 req->errorno = errno; 1990 req->errorno = errno;
1813} 1991}
1814 1992
1815#ifndef EIO_NO_WRAPPERS 1993#ifndef EIO_NO_WRAPPERS
1816 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
1817eio_req *eio_nop (int pri, eio_cb cb, void *data) 2005eio_req *eio_nop (int pri, eio_cb cb, void *data)
1818{ 2006{
1819 REQ (EIO_NOP); SEND; 2007 REQ (EIO_NOP); SEND;
1820} 2008}
1821 2009
… …
1837eio_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)
1838{ 2026{
1839 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;
1840} 2028}
1841 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
1842eio_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)
1843{ 2046{
1844 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;
1845} 2048}
1846 2049
… …
1852eio_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)
1853{ 2056{
1854 REQ (EIO_MLOCKALL); req->int1 = flags; SEND; 2057 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
1855} 2058}
1856 2059
1857eio_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)
1858{ 2061{
1859 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;
1860}
1861
1862eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1863{
1864 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1865} 2063}
1866 2064
1867eio_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)
1868{ 2066{
1869 REQ (EIO_CLOSE); req->int1 = fd; SEND; 2067 REQ (EIO_CLOSE); req->int1 = fd; SEND;
… …
1872eio_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)
1873{ 2071{
1874 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;
1875} 2073}
1876 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
1877eio_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)
1878{ 2081{
1879 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;
1880} 2083}
1881 2084
1882eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data) 2085eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1883{ 2086{
1884 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND; 2087 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1885} 2088}
1886 2089
2090eio_req *eio_fcntl (int fd, int cmd, void *arg, int pri, eio_cb cb, void *data)
2091{
2092 REQ (EIO_IOCTL); req->int1 = fd; req->int2 = cmd; req->ptr2 = arg; SEND;
2093}
2094
2095eio_req *eio_ioctl (int fd, unsigned long request, void *buf, int pri, eio_cb cb, void *data)
2096{
2097 REQ (EIO_IOCTL); req->int1 = fd; req->int2 = request; req->ptr2 = buf; SEND;
2098}
2099
1887eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) 2100eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1888{ 2101{
1889 REQ (EIO_FSTAT); req->int1 = fd; SEND; 2102 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1890} 2103}
1891 2104
… …
1907eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) 2120eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1908{ 2121{
1909 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2122 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1910} 2123}
1911 2124
1912eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2125eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1913{ 2126{
1914 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2127 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1915} 2128}
1916 2129
1917eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) 2130eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
… …
1937eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data) 2150eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1938{ 2151{
1939 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2152 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1940} 2153}
1941 2154
1942eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2155eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1943{ 2156{
1944 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2157 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1945} 2158}
1946 2159
1947eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) 2160eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
… …
1961} 2174}
1962 2175
1963eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data) 2176eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1964{ 2177{
1965 return eio__1path (EIO_READLINK, path, pri, cb, data); 2178 return eio__1path (EIO_READLINK, path, pri, cb, data);
2179}
2180
2181eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2182{
2183 return eio__1path (EIO_REALPATH, path, pri, cb, data);
1966} 2184}
1967 2185
1968eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) 2186eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1969{ 2187{
1970 return eio__1path (EIO_STAT, path, pri, cb, data); 2188 return eio__1path (EIO_STAT, path, pri, cb, data);
… …
2029eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2247eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2030{ 2248{
2031 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2249 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
2032} 2250}
2033 2251
2034eio_req *eio_custom (void (*)(eio_req *) execute, int pri, eio_cb cb, void *data); 2252eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
2035{ 2253{
2036 REQ (EIO_CUSTOM); req->feed = execute; SEND; 2254 REQ (EIO_CUSTOM); req->feed = execute; SEND;
2037} 2255}
2038 2256
2039#endif 2257#endif
… …
2050#undef SEND 2268#undef SEND
2051 2269
2052/*****************************************************************************/ 2270/*****************************************************************************/
2053/* grp functions */ 2271/* grp functions */
2054 2272
2273void
2055void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2274eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
2056{ 2275{
2057 grp->int2 = limit; 2276 grp->int2 = limit;
2058 grp->feed = feed; 2277 grp->feed = feed;
2059 2278
2060 grp_try_feed (grp); 2279 grp_try_feed (grp);
2061} 2280}
2062 2281
2282void
2063void eio_grp_limit (eio_req *grp, int limit) 2283eio_grp_limit (eio_req *grp, int limit)
2064{ 2284{
2065 grp->int2 = limit; 2285 grp->int2 = limit;
2066 2286
2067 grp_try_feed (grp); 2287 grp_try_feed (grp);
2068} 2288}
2069 2289
2290void
2070void eio_grp_add (eio_req *grp, eio_req *req) 2291eio_grp_add (eio_req *grp, eio_req *req)
2071{ 2292{
2072 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2293 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
2073 2294
2074 grp->flags |= EIO_FLAG_GROUPADD; 2295 grp->flags |= ETP_FLAG_GROUPADD;
2075 2296
2076 ++grp->size; 2297 ++grp->size;
2077 req->grp = grp; 2298 req->grp = grp;
2078 2299
2079 req->grp_prev = 0; 2300 req->grp_prev = 0;
… …
2086} 2307}
2087 2308
2088/*****************************************************************************/ 2309/*****************************************************************************/
2089/* misc garbage */ 2310/* misc garbage */
2090 2311
2312eio_ssize_t
2091ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2313eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2092{ 2314{
2093 etp_worker wrk;
2094 ssize_t ret;
2095
2096 wrk.dbuf = 0;
2097
2098 ret = eio__sendfile (ofd, ifd, offset, count, &wrk); 2315 return eio__sendfile (ofd, ifd, offset, count);
2099
2100 if (wrk.dbuf)
2101 free (wrk.dbuf);
2102
2103 return ret;
2104} 2316}
2105 2317

Diff Legend

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