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

Comparing libeio/eio.c (file contents):
Revision 1.27 by root, Wed Oct 22 18:15:36 2008 UTC vs.
Revision 1.161 by root, Fri Feb 16 21:20:52 2024 UTC

1/* 1/*
2 * libeio implementation 2 * libeio implementation
3 * 3 *
4 * Copyright (c) 2007,2008 Marc Alexander Lehmann <libeio@schmorp.de> 4 * Copyright (c) 2007,2008,2009,2010,2011,2012,2013,2016,2017,2018,2019 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,
… …
35 * and other provisions required by the GPL. If you do not delete the 35 * and other provisions required by the GPL. If you do not delete the
36 * provisions above, a recipient may use your version of this file under 36 * provisions above, a recipient may use your version of this file under
37 * either the BSD or the GPL. 37 * either the BSD or the GPL.
38 */ 38 */
39 39
40#ifndef _WIN32
41# include "config.h"
42#endif
43
40#include "eio.h" 44#include "eio.h"
41#include "xthread.h" 45#include "ecb.h"
42 46
43#include <errno.h> 47#include <errno.h>
44#include <stddef.h> 48#include <stddef.h>
45#include <stdlib.h> 49#include <stdlib.h>
46#include <string.h> 50#include <string.h>
… …
49#include <sys/stat.h> 53#include <sys/stat.h>
50#include <limits.h> 54#include <limits.h>
51#include <fcntl.h> 55#include <fcntl.h>
52#include <assert.h> 56#include <assert.h>
53 57
58/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
59/* intptr_t only comes from stdint.h, says idiot openbsd coder */
60#if HAVE_STDINT_H
61# include <stdint.h>
62#endif
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
54#ifndef EIO_FINISH 77#ifndef EIO_FINISH
55# 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
56#endif 79#endif
57 80
58#ifndef EIO_DESTROY 81#ifndef EIO_DESTROY
… …
61 84
62#ifndef EIO_FEED 85#ifndef EIO_FEED
63# 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)
64#endif 87#endif
65 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
66#ifdef _WIN32 100#ifdef _WIN32
67 101
68 /*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)
69 233
70#else 234#else
71 235
72# include "config.h" 236 #include <sys/ioctl.h>
73# include <sys/time.h> 237 #include <sys/time.h>
74# include <sys/select.h> 238 #include <sys/select.h>
75# include <unistd.h> 239 #include <unistd.h>
240 #include <signal.h>
241 #include <dirent.h>
242
243 #ifdef ANDROID
244 #include <sys/vfs.h>
245 #define statvfs statfs
246 #define fstatvfs fstatfs
247 #include <asm/page.h> /* supposedly limits.h does #define PAGESIZE PAGESIZE */
248 #else
249 #include <sys/statvfs.h>
250 #endif
251
252 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
253 #include <sys/mman.h>
254 #endif
255
256 #define D_NAME(entp) entp->d_name
257
258 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
259 #if __FreeBSD__ || __NetBSD__ || __OpenBSD__
260 #define _DIRENT_HAVE_D_TYPE /* sigh */
261 #define D_INO(de) (de)->d_fileno
262 #define D_NAMLEN(de) (de)->d_namlen
263 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
264 #define D_INO(de) (de)->d_ino
265 #endif
266
267 #ifdef _D_EXACT_NAMLEN
268 #undef D_NAMLEN
269 #define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
270 #endif
271
272 #ifdef _DIRENT_HAVE_D_TYPE
273 #define D_TYPE(de) (de)->d_type
274 #endif
275
276 #ifndef EIO_STRUCT_DIRENT
277 #define EIO_STRUCT_DIRENT struct dirent
278 #endif
279
280#endif
281
282#if HAVE_UTIMES
76# include <utime.h> 283# include <utime.h>
77# include <signal.h>
78# include <dirent.h>
79
80# ifndef EIO_STRUCT_DIRENT
81# define EIO_STRUCT_DIRENT struct dirent
82# endif 284#endif
83 285
286#if HAVE_SYS_SYSCALL_H
287# include <sys/syscall.h>
84#endif 288#endif
85 289
86#if HAVE_SENDFILE 290#if HAVE_SENDFILE
87# if __linux 291# if __linux
88# include <sys/sendfile.h> 292# include <sys/sendfile.h>
89# elif __freebsd 293# elif __FreeBSD__ || defined __APPLE__
90# include <sys/socket.h> 294# include <sys/socket.h>
91# include <sys/uio.h> 295# include <sys/uio.h>
92# elif __hpux 296# elif __hpux
93# include <sys/socket.h> 297# include <sys/socket.h>
94# elif __solaris /* not yet */ 298# elif __solaris
95# include <sys/sendfile.h> 299# include <sys/sendfile.h>
96# else 300# else
97# error sendfile support requested but not available 301# error sendfile support requested but not available
98# endif 302# endif
99#endif 303#endif
100 304
101/* number of seconds after which an idle threads exit */ 305#if HAVE_RENAMEAT2
102#define IDLE_TIMEOUT 10 306# include <sys/syscall.h>
307# include <linux/fs.h>
308#endif
309
310#ifndef D_TYPE
311# define D_TYPE(de) 0
312#endif
313#ifndef D_INO
314# define D_INO(de) 0
315#endif
316#ifndef D_NAMLEN
317# define D_NAMLEN(entp) strlen (D_NAME (entp))
318#endif
103 319
104/* used for struct dirent, AIX doesn't provide it */ 320/* used for struct dirent, AIX doesn't provide it */
105#ifndef NAME_MAX 321#ifndef NAME_MAX
106# define NAME_MAX 4096 322# define NAME_MAX 4096
107#endif 323#endif
108 324
325/* used for readlink etc. */
326#ifndef PATH_MAX
327# define PATH_MAX 0
328#endif
329
330#ifndef O_CLOEXEC
331 #define O_CLOEXEC 0
332#endif
333
334#ifndef O_NONBLOCK
335 #define O_NONBLOCK 0
336#endif
337
338#ifndef O_SEARCH
339 #define O_SEARCH O_RDONLY
340#endif
341
342#ifndef O_DIRECTORY
343 #define O_DIRECTORY 0
344#endif
345
346#ifndef EIO_PATH_MIN
347# define EIO_PATH_MIN 8160
348#endif
349
350#if defined(O_PATH)
351# define EIO_O_PATH O_PATH
352#elif defined(O_SEARCH)
353# define EIO_O_PATH O_SEARCH
354#else
355# define EIO_O_PATH 0
356#endif
357
358#define EIO_PATH_MAX (PATH_MAX <= EIO_PATH_MIN ? EIO_PATH_MIN : PATH_MAX)
359
109/* buffer size for various temporary buffers */ 360/* buffer size for various temporary buffers */
110#define EIO_BUFSIZE 65536 361#define EIO_BUFSIZE 65536
111 362
112#define dBUF \ 363#define dBUF \
113 char *eio_buf; \
114 ETP_WORKER_LOCK (self); \
115 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 364 char *eio_buf = malloc (EIO_BUFSIZE); \
116 ETP_WORKER_UNLOCK (self); \
117 errno = ENOMEM; \ 365 errno = ENOMEM; \
118 if (!eio_buf) \ 366 if (!eio_buf) \
119 return -1; 367 return -1
120 368
121#define EIO_TICKS ((1000000 + 1023) >> 10) 369#define FUBd \
370 free (eio_buf)
122 371
123/*****************************************************************************/ 372/*****************************************************************************/
124 373
125#if __GNUC__ >= 3 374struct etp_tmpbuf;
126# define expect(expr,value) __builtin_expect ((expr),(value)) 375
376#if _POSIX_VERSION >= 200809L
377 #define HAVE_AT 1
378 #define WD2FD(wd) ((wd) ? (wd)->fd : AT_FDCWD)
127#else 379#else
128# define expect(expr,value) (expr) 380 #define HAVE_AT 0
381 static const char *wd_expand (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path);
129#endif 382#endif
130 383
131#define expect_false(expr) expect ((expr) != 0, 0) 384struct eio_pwd
132#define expect_true(expr) expect ((expr) != 0, 1) 385{
386#if HAVE_AT
387 int fd;
388#endif
389 int len;
390 char str[1]; /* actually, a 0-terminated canonical path */
391};
133 392
134/*****************************************************************************/ 393/*****************************************************************************/
135 394
136#define ETP_PRI_MIN EIO_PRI_MIN 395#define ETP_PRI_MIN EIO_PRI_MIN
137#define ETP_PRI_MAX EIO_PRI_MAX 396#define ETP_PRI_MAX EIO_PRI_MAX
138 397
398#define ETP_TYPE_QUIT -1
399#define ETP_TYPE_GROUP EIO_GROUP
400
401static void eio_nop_callback (void) { }
402static void (*eio_want_poll_cb)(void) = eio_nop_callback;
403static void (*eio_done_poll_cb)(void) = eio_nop_callback;
404
405#define ETP_WANT_POLL(pool) eio_want_poll_cb ()
406#define ETP_DONE_POLL(pool) eio_done_poll_cb ()
407
139struct etp_worker; 408struct etp_worker;
140
141#define ETP_REQ eio_req 409#define ETP_REQ eio_req
142#define ETP_DESTROY(req) eio_destroy (req) 410#define ETP_DESTROY(req) eio_destroy (req)
143static int eio_finish (eio_req *req); 411static int eio_finish (eio_req *req);
144#define ETP_FINISH(req) eio_finish (req) 412#define ETP_FINISH(req) eio_finish (req)
145static void eio_execute (struct etp_worker *self, eio_req *req); 413static void eio_execute (struct etp_worker *self, eio_req *req);
146#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) 414#define ETP_EXECUTE(wrk,req) eio_execute (wrk, req)
147 415
148#define ETP_WORKER_CLEAR(req) \ 416#include "etp.c"
149 if (wrk->dbuf) \ 417
150 { \ 418static struct etp_pool eio_pool;
151 free (wrk->dbuf); \ 419#define EIO_POOL (&eio_pool)
152 wrk->dbuf = 0; \
153 } \
154 \
155 if (wrk->dirp) \
156 { \
157 closedir (wrk->dirp); \
158 wrk->dirp = 0; \
159 }
160#define ETP_WORKER_COMMON \
161 void *dbuf; \
162 DIR *dirp;
163 420
164/*****************************************************************************/ 421/*****************************************************************************/
165 422
166#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
167
168/* calculcate time difference in ~1/EIO_TICKS of a second */
169static int tvdiff (struct timeval *tv1, struct timeval *tv2)
170{
171 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
172 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
173}
174
175static unsigned int started, idle, wanted = 4;
176
177static void (*want_poll_cb) (void);
178static void (*done_poll_cb) (void);
179
180static unsigned int max_poll_time; /* reslock */
181static unsigned int max_poll_reqs; /* reslock */
182
183static volatile unsigned int nreqs; /* reqlock */
184static volatile unsigned int nready; /* reqlock */
185static volatile unsigned int npending; /* reqlock */
186static volatile unsigned int max_idle = 4;
187
188static mutex_t wrklock = X_MUTEX_INIT;
189static mutex_t reslock = X_MUTEX_INIT;
190static mutex_t reqlock = X_MUTEX_INIT;
191static cond_t reqwait = X_COND_INIT;
192
193#if !HAVE_PREADWRITE
194/*
195 * make our pread/pwrite emulation safe against themselves, but not against
196 * normal read/write by using a mutex. slows down execution a lot,
197 * but that's your problem, not mine.
198 */
199static mutex_t preadwritelock = X_MUTEX_INIT;
200#endif
201
202typedef struct etp_worker
203{
204 /* locked by wrklock */
205 struct etp_worker *prev, *next;
206
207 thread_t tid;
208
209 /* locked by reslock, reqlock or wrklock */
210 ETP_REQ *req; /* currently processed request */
211
212 ETP_WORKER_COMMON
213} etp_worker;
214
215static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */
216
217#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
218#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
219
220/* worker threads management */
221
222static void etp_worker_clear (etp_worker *wrk)
223{
224 ETP_WORKER_CLEAR (wrk);
225}
226
227static void etp_worker_free (etp_worker *wrk)
228{
229 wrk->next->prev = wrk->prev;
230 wrk->prev->next = wrk->next;
231
232 free (wrk);
233}
234
235static unsigned int etp_nreqs (void)
236{
237 int retval;
238 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
239 retval = nreqs;
240 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
241 return retval;
242}
243
244static unsigned int etp_nready (void)
245{
246 unsigned int retval;
247
248 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
249 retval = nready;
250 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
251
252 return retval;
253}
254
255static unsigned int etp_npending (void)
256{
257 unsigned int retval;
258
259 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
260 retval = npending;
261 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
262
263 return retval;
264}
265
266static unsigned int etp_nthreads (void)
267{
268 unsigned int retval;
269
270 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
271 retval = started;
272 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
273
274 return retval;
275}
276
277/*
278 * a somewhat faster data structure might be nice, but
279 * with 8 priorities this actually needs <20 insns
280 * per shift, the most expensive operation.
281 */
282typedef struct {
283 ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */
284 int size;
285} etp_reqq;
286
287static etp_reqq req_queue;
288static etp_reqq res_queue;
289
290static int reqq_push (etp_reqq *q, ETP_REQ *req)
291{
292 int pri = req->pri;
293 req->next = 0;
294
295 if (q->qe[pri])
296 {
297 q->qe[pri]->next = req;
298 q->qe[pri] = req;
299 }
300 else
301 q->qe[pri] = q->qs[pri] = req;
302
303 return q->size++;
304}
305
306static ETP_REQ *reqq_shift (etp_reqq *q)
307{
308 int pri;
309
310 if (!q->size)
311 return 0;
312
313 --q->size;
314
315 for (pri = ETP_NUM_PRI; pri--; )
316 {
317 eio_req *req = q->qs[pri];
318
319 if (req)
320 {
321 if (!(q->qs[pri] = (eio_req *)req->next))
322 q->qe[pri] = 0;
323
324 return req;
325 }
326 }
327
328 abort ();
329}
330
331static void etp_atfork_prepare (void)
332{
333 X_LOCK (wrklock);
334 X_LOCK (reqlock);
335 X_LOCK (reslock);
336#if !HAVE_PREADWRITE
337 X_LOCK (preadwritelock);
338#endif
339}
340
341static void etp_atfork_parent (void)
342{
343#if !HAVE_PREADWRITE
344 X_UNLOCK (preadwritelock);
345#endif
346 X_UNLOCK (reslock);
347 X_UNLOCK (reqlock);
348 X_UNLOCK (wrklock);
349}
350
351static void etp_atfork_child (void)
352{
353 ETP_REQ *prv;
354
355 while ((prv = reqq_shift (&req_queue)))
356 ETP_DESTROY (prv);
357
358 while ((prv = reqq_shift (&res_queue)))
359 ETP_DESTROY (prv);
360
361 while (wrk_first.next != &wrk_first)
362 {
363 etp_worker *wrk = wrk_first.next;
364
365 if (wrk->req)
366 ETP_DESTROY (wrk->req);
367
368 etp_worker_clear (wrk);
369 etp_worker_free (wrk);
370 }
371
372 started = 0;
373 idle = 0;
374 nreqs = 0;
375 nready = 0;
376 npending = 0;
377
378 etp_atfork_parent ();
379}
380
381static void 423static void
382etp_once_init (void)
383{
384 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
385}
386
387static int
388etp_init (void (*want_poll)(void), void (*done_poll)(void))
389{
390 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
391
392 pthread_once (&doinit, etp_once_init);
393
394 want_poll_cb = want_poll;
395 done_poll_cb = done_poll;
396
397 return 0;
398}
399
400X_THREAD_PROC (etp_proc);
401
402static void etp_start_thread (void)
403{
404 etp_worker *wrk = calloc (1, sizeof (etp_worker));
405
406 /*TODO*/
407 assert (("unable to allocate worker thread data", wrk));
408
409 X_LOCK (wrklock);
410
411 if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
412 {
413 wrk->prev = &wrk_first;
414 wrk->next = wrk_first.next;
415 wrk_first.next->prev = wrk;
416 wrk_first.next = wrk;
417 ++started;
418 }
419 else
420 free (wrk);
421
422 X_UNLOCK (wrklock);
423}
424
425static void etp_maybe_start_thread (void)
426{
427 if (expect_true (etp_nthreads () >= wanted))
428 return;
429
430 /* todo: maybe use idle here, but might be less exact */
431 if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
432 return;
433
434 etp_start_thread ();
435}
436
437static void etp_end_thread (void)
438{
439 eio_req *req = calloc (1, sizeof (eio_req));
440
441 req->type = -1;
442 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
443
444 X_LOCK (reqlock);
445 reqq_push (&req_queue, req);
446 X_COND_SIGNAL (reqwait);
447 X_UNLOCK (reqlock);
448
449 X_LOCK (wrklock);
450 --started;
451 X_UNLOCK (wrklock);
452}
453
454static int etp_poll (void)
455{
456 unsigned int maxreqs;
457 unsigned int maxtime;
458 struct timeval tv_start, tv_now;
459
460 X_LOCK (reslock);
461 maxreqs = max_poll_reqs;
462 maxtime = max_poll_time;
463 X_UNLOCK (reslock);
464
465 if (maxtime)
466 gettimeofday (&tv_start, 0);
467
468 for (;;)
469 {
470 ETP_REQ *req;
471
472 etp_maybe_start_thread ();
473
474 X_LOCK (reslock);
475 req = reqq_shift (&res_queue);
476
477 if (req)
478 {
479 --npending;
480
481 if (!res_queue.size && done_poll_cb)
482 done_poll_cb ();
483 }
484
485 X_UNLOCK (reslock);
486
487 if (!req)
488 return 0;
489
490 X_LOCK (reqlock);
491 --nreqs;
492 X_UNLOCK (reqlock);
493
494 if (expect_false (req->type == EIO_GROUP && req->size))
495 {
496 req->int1 = 1; /* mark request as delayed */
497 continue;
498 }
499 else
500 {
501 int res = ETP_FINISH (req);
502 if (expect_false (res))
503 return res;
504 }
505
506 if (expect_false (maxreqs && !--maxreqs))
507 break;
508
509 if (maxtime)
510 {
511 gettimeofday (&tv_now, 0);
512
513 if (tvdiff (&tv_start, &tv_now) >= maxtime)
514 break;
515 }
516 }
517
518 errno = EAGAIN;
519 return -1;
520}
521
522static void etp_cancel (ETP_REQ *req)
523{
524 X_LOCK (wrklock);
525 req->flags |= EIO_FLAG_CANCELLED;
526 X_UNLOCK (wrklock);
527
528 eio_grp_cancel (req);
529}
530
531static void etp_submit (ETP_REQ *req)
532{
533 req->pri -= ETP_PRI_MIN;
534
535 if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
536 if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
537
538 if (expect_false (req->type == EIO_GROUP))
539 {
540 /* I hope this is worth it :/ */
541 X_LOCK (reqlock);
542 ++nreqs;
543 X_UNLOCK (reqlock);
544
545 X_LOCK (reslock);
546
547 ++npending;
548
549 if (!reqq_push (&res_queue, req) && want_poll_cb)
550 want_poll_cb ();
551
552 X_UNLOCK (reslock);
553 }
554 else
555 {
556 X_LOCK (reqlock);
557 ++nreqs;
558 ++nready;
559 reqq_push (&req_queue, req);
560 X_COND_SIGNAL (reqwait);
561 X_UNLOCK (reqlock);
562
563 etp_maybe_start_thread ();
564 }
565}
566
567static void etp_set_max_poll_time (double nseconds)
568{
569 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
570 max_poll_time = nseconds;
571 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
572}
573
574static void etp_set_max_poll_reqs (unsigned int maxreqs)
575{
576 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
577 max_poll_reqs = maxreqs;
578 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
579}
580
581static void etp_set_max_idle (unsigned int nthreads)
582{
583 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
584 max_idle = nthreads <= 0 ? 1 : nthreads;
585 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
586}
587
588static void etp_set_min_parallel (unsigned int nthreads)
589{
590 if (wanted < nthreads)
591 wanted = nthreads;
592}
593
594static void etp_set_max_parallel (unsigned int nthreads)
595{
596 if (wanted > nthreads)
597 wanted = nthreads;
598
599 while (started > wanted)
600 etp_end_thread ();
601}
602
603/*****************************************************************************/
604
605static void grp_try_feed (eio_req *grp) 424grp_try_feed (eio_req *grp)
606{ 425{
607 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 426 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
608 { 427 {
609 grp->flags &= ~EIO_FLAG_GROUPADD; 428 grp->flags &= ~ETP_FLAG_GROUPADD;
610 429
611 EIO_FEED (grp); 430 EIO_FEED (grp);
612 431
613 /* stop if no progress has been made */ 432 /* stop if no progress has been made */
614 if (!(grp->flags & EIO_FLAG_GROUPADD)) 433 if (!(grp->flags & ETP_FLAG_GROUPADD))
615 { 434 {
616 grp->feed = 0; 435 grp->feed = 0;
617 break; 436 break;
618 } 437 }
619 } 438 }
620} 439}
621 440
441static int
622static int grp_dec (eio_req *grp) 442grp_dec (eio_req *grp)
623{ 443{
624 --grp->size; 444 --grp->size;
625 445
626 /* call feeder, if applicable */ 446 /* call feeder, if applicable */
627 grp_try_feed (grp); 447 grp_try_feed (grp);
628 448
629 /* finish, if done */ 449 /* finish, if done */
630 if (!grp->size && grp->int1) 450 if (!grp->size && grp->flags & ETP_FLAG_DELAYED)
631 return eio_finish (grp); 451 return eio_finish (grp);
632 else 452 else
633 return 0; 453 return 0;
634} 454}
635 455
456static void
636void eio_destroy (eio_req *req) 457eio_destroy (eio_req *req)
637{ 458{
638 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); 459 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
639 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); 460 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
640 461
641 EIO_DESTROY (req); 462 EIO_DESTROY (req);
642} 463}
643 464
465static int
644static int eio_finish (eio_req *req) 466eio_finish (eio_req *req)
645{ 467{
646 int res = EIO_FINISH (req); 468 int res = EIO_FINISH (req);
647 469
648 if (req->grp) 470 if (req->grp)
649 { 471 {
… …
657 if (grp->grp_first == req) 479 if (grp->grp_first == req)
658 grp->grp_first = req->grp_next; 480 grp->grp_first = req->grp_next;
659 481
660 res2 = grp_dec (grp); 482 res2 = grp_dec (grp);
661 483
662 if (!res && res2) 484 if (!res)
663 res = res2; 485 res = res2;
664 } 486 }
665 487
666 eio_destroy (req); 488 eio_destroy (req);
667 489
668 return res; 490 return res;
669} 491}
670 492
493void
671void eio_grp_cancel (eio_req *grp) 494eio_grp_cancel (eio_req *grp)
672{ 495{
673 for (grp = grp->grp_first; grp; grp = grp->grp_next) 496 etp_grp_cancel (EIO_POOL, grp);
674 eio_cancel (grp);
675} 497}
676 498
499void
677void eio_cancel (eio_req *req) 500eio_cancel (eio_req *req)
678{ 501{
679 etp_cancel (req); 502 etp_cancel (EIO_POOL, req);
680} 503}
681 504
505void
682void eio_submit (eio_req *req) 506eio_submit (eio_req *req)
683{ 507{
684 etp_submit (req); 508 etp_submit (EIO_POOL, req);
685} 509}
686 510
687unsigned int eio_nreqs (void) 511unsigned int
512eio_nreqs (void)
688{ 513{
689 return etp_nreqs (); 514 return etp_nreqs (EIO_POOL);
690} 515}
691 516
692unsigned int eio_nready (void) 517unsigned int
518eio_nready (void)
693{ 519{
694 return etp_nready (); 520 return etp_nready (EIO_POOL);
695} 521}
696 522
697unsigned int eio_npending (void) 523unsigned int
524eio_npending (void)
698{ 525{
699 return etp_npending (); 526 return etp_npending (EIO_POOL);
700} 527}
701 528
702unsigned int eio_nthreads (void) 529unsigned int ecb_cold
530eio_nthreads (void)
703{ 531{
704 return etp_nthreads (); 532 return etp_nthreads (EIO_POOL);
705} 533}
706 534
535void ecb_cold
707void eio_set_max_poll_time (double nseconds) 536eio_set_max_poll_time (double nseconds)
708{ 537{
709 etp_set_max_poll_time (nseconds); 538 etp_set_max_poll_time (EIO_POOL, nseconds);
710} 539}
711 540
541void ecb_cold
712void eio_set_max_poll_reqs (unsigned int maxreqs) 542eio_set_max_poll_reqs (unsigned int maxreqs)
713{ 543{
714 etp_set_max_poll_reqs (maxreqs); 544 etp_set_max_poll_reqs (EIO_POOL, maxreqs);
715} 545}
716 546
547void ecb_cold
717void eio_set_max_idle (unsigned int nthreads) 548eio_set_max_idle (unsigned int nthreads)
718{ 549{
719 etp_set_max_idle (nthreads); 550 etp_set_max_idle (EIO_POOL, nthreads);
720} 551}
721 552
553void ecb_cold
554eio_set_idle_timeout (unsigned int seconds)
555{
556 etp_set_idle_timeout (EIO_POOL, seconds);
557}
558
559void ecb_cold
722void eio_set_min_parallel (unsigned int nthreads) 560eio_set_min_parallel (unsigned int nthreads)
723{ 561{
724 etp_set_min_parallel (nthreads); 562 etp_set_min_parallel (EIO_POOL, nthreads);
725} 563}
726 564
565void ecb_cold
727void eio_set_max_parallel (unsigned int nthreads) 566eio_set_max_parallel (unsigned int nthreads)
728{ 567{
729 etp_set_max_parallel (nthreads); 568 etp_set_max_parallel (EIO_POOL, nthreads);
730} 569}
731 570
732int eio_poll (void) 571int eio_poll (void)
733{ 572{
734 return etp_poll (); 573 return etp_poll (EIO_POOL);
735} 574}
736 575
737/*****************************************************************************/ 576/*****************************************************************************/
738/* work around various missing functions */ 577/* work around various missing functions */
739 578
740#if !HAVE_PREADWRITE 579#if HAVE_POSIX_CLOSE && !__linux
741# undef pread 580# define eio__close(fd) posix_close (fd, 0)
742# undef pwrite 581#else
743# define pread eio__pread 582# define eio__close(fd) close (fd)
744# define pwrite eio__pwrite
745
746static ssize_t
747eio__pread (int fd, void *buf, size_t count, off_t offset)
748{
749 ssize_t res;
750 off_t ooffset;
751
752 X_LOCK (preadwritelock);
753 ooffset = lseek (fd, 0, SEEK_CUR);
754 lseek (fd, offset, SEEK_SET);
755 res = read (fd, buf, count);
756 lseek (fd, ooffset, SEEK_SET);
757 X_UNLOCK (preadwritelock);
758
759 return res;
760}
761
762static ssize_t
763eio__pwrite (int fd, void *buf, size_t count, off_t offset)
764{
765 ssize_t res;
766 off_t ooffset;
767
768 X_LOCK (preadwritelock);
769 ooffset = lseek (fd, 0, SEEK_CUR);
770 lseek (fd, offset, SEEK_SET);
771 res = write (fd, buf, count);
772 lseek (fd, offset, SEEK_SET);
773 X_UNLOCK (preadwritelock);
774
775 return res;
776}
777#endif 583#endif
778 584
585/* close() without disturbing errno */
586static void
587silent_close (int fd)
588{
589 int saved_errno = errno;
590 eio__close (fd);
591 errno = saved_errno;
592}
593
779#ifndef HAVE_FUTIMES 594#ifndef HAVE_UTIMES
780 595
781# undef utimes 596# undef utimes
782# undef futimes
783# define utimes(path,times) eio__utimes (path, times) 597# define utimes(path,times) eio__utimes (path, times)
784# define futimes(fd,times) eio__futimes (fd, times)
785 598
786static int 599static int
787eio__utimes (const char *filename, const struct timeval times[2]) 600eio__utimes (const char *filename, const struct timeval times[2])
788{ 601{
789 if (times) 602 if (times)
… …
797 } 610 }
798 else 611 else
799 return utime (filename, 0); 612 return utime (filename, 0);
800} 613}
801 614
615#endif
616
617#ifndef HAVE_FUTIMES
618
619# undef futimes
620# define futimes(fd,times) eio__futimes (fd, times)
621
622static int
802static int eio__futimes (int fd, const struct timeval tv[2]) 623eio__futimes (int fd, const struct timeval tv[2])
803{ 624{
804 errno = ENOSYS; 625 errno = ENOSYS;
805 return -1; 626 return -1;
806} 627}
807 628
… …
810#if !HAVE_FDATASYNC 631#if !HAVE_FDATASYNC
811# undef fdatasync 632# undef fdatasync
812# define fdatasync(fd) fsync (fd) 633# define fdatasync(fd) fsync (fd)
813#endif 634#endif
814 635
636static int
637eio__syncfs (int fd)
638{
639 int res;
640
641#if HAVE_SYS_SYNCFS
642 res = (int)syscall (__NR_syncfs, (int)(fd));
643#else
644 res = EIO_ENOSYS ();
645#endif
646
647 if (res < 0 && errno == ENOSYS && fd >= 0)
648 sync ();
649
650 return res;
651}
652
815/* sync_file_range always needs emulation */ 653/* sync_file_range always needs emulation */
816int 654static int
817eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags) 655eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
818{ 656{
819#if HAVE_SYNC_FILE_RANGE 657#if HAVE_SYNC_FILE_RANGE
820 int res; 658 int res;
821 659
… …
829 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0); 667 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
830 } 668 }
831 669
832 res = sync_file_range (fd, offset, nbytes, flags); 670 res = sync_file_range (fd, offset, nbytes, flags);
833 671
834 if (res != ENOSYS) 672 if (!res || errno != ENOSYS)
835 return res; 673 return res;
836#endif 674#endif
837 675
838 /* even though we could play tricks with the flags, it's better to always 676 /* even though we could play tricks with the flags, it's better to always
839 * call fdatasync, as thta matches the expectation of it's users best */ 677 * call fdatasync, as that matches the expectation of its users best */
840 return fdatasync (fd); 678 return fdatasync (fd);
679}
680
681static int
682eio__fallocate (int fd, int mode, off_t offset, size_t len)
683{
684#if HAVE_LINUX_FALLOCATE
685 return fallocate (fd, mode, offset, len);
686#else
687 return EIO_ENOSYS ();
688#endif
841} 689}
842 690
843#if !HAVE_READAHEAD 691#if !HAVE_READAHEAD
844# undef readahead 692# undef readahead
845# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 693# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
846 694
847static ssize_t 695static eio_ssize_t
848eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) 696eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
849{ 697{
850 size_t todo = count; 698 size_t todo = count;
851 dBUF; 699 dBUF;
852 700
… …
857 pread (fd, eio_buf, len, offset); 705 pread (fd, eio_buf, len, offset);
858 offset += len; 706 offset += len;
859 todo -= len; 707 todo -= len;
860 } 708 }
861 709
862 errno = 0; 710 FUBd;
711
712 /* linux's readahead basically only fails for EBADF or EINVAL (not mmappable) */
713 /* but not for e.g. EIO or eof, so we also never fail */
863 return count; 714 return 0;
864} 715}
865 716
866#endif 717#endif
867 718
868/* sendfile always needs emulation */ 719/* sendfile always needs emulation */
869static ssize_t 720static eio_ssize_t
870eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) 721eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
871{ 722{
723 eio_ssize_t written = 0;
872 ssize_t res; 724 eio_ssize_t res;
873 725
874 if (!count) 726 if (!count)
875 return 0; 727 return 0;
876 728
729 for (;;)
730 {
731#ifdef __APPLE__
732# undef HAVE_SENDFILE /* broken, as everything on os x */
733#endif
877#if HAVE_SENDFILE 734#if HAVE_SENDFILE
878# if __linux 735# if __linux
736 off_t soffset = offset;
879 res = sendfile (ofd, ifd, &offset, count); 737 res = sendfile (ofd, ifd, &soffset, count);
880 738
881# elif __freebsd 739# elif __FreeBSD__
882 /* 740 /*
883 * Of course, the freebsd sendfile is a dire hack with no thoughts 741 * Of course, the freebsd sendfile is a dire hack with no thoughts
884 * wasted on making it similar to other I/O functions. 742 * wasted on making it similar to other I/O functions.
885 */ 743 */
886 {
887 off_t sbytes; 744 off_t sbytes;
888 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 745 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
889 746
890 if (res < 0 && sbytes) 747 #if 0 /* according to the manpage, this is correct, but broken behaviour */
891 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 748 /* freebsd' sendfile will return 0 on success */
749 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
750 /* not on e.g. EIO or EPIPE - sounds broken */
751 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
892 res = sbytes; 752 res = sbytes;
893 } 753 #endif
754
755 /* according to source inspection, this is correct, and useful behaviour */
756 if (sbytes)
757 res = sbytes;
758
759# elif defined __APPLE__
760 off_t sbytes = count;
761 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
762
763 /* according to the manpage, sbytes is always valid */
764 if (sbytes)
765 res = sbytes;
894 766
895# elif __hpux 767# elif __hpux
896 res = sendfile (ofd, ifd, offset, count, 0, 0); 768 res = sendfile (ofd, ifd, offset, count, 0, 0);
897 769
898# elif __solaris 770# elif __solaris
899 {
900 struct sendfilevec vec; 771 struct sendfilevec vec;
901 size_t sbytes; 772 size_t sbytes;
902 773
903 vec.sfv_fd = ifd; 774 vec.sfv_fd = ifd;
904 vec.sfv_flag = 0; 775 vec.sfv_flag = 0;
905 vec.sfv_off = offset; 776 vec.sfv_off = offset;
906 vec.sfv_len = count; 777 vec.sfv_len = count;
907 778
908 res = sendfilev (ofd, &vec, 1, &sbytes); 779 res = sendfilev (ofd, &vec, 1, &sbytes);
909 780
910 if (res < 0 && sbytes) 781 if (res < 0 && sbytes)
911 res = sbytes; 782 res = sbytes;
912 }
913 783
914# endif 784# endif
785
786#elif defined (_WIN32) && 0
787 /* does not work, just for documentation of what would need to be done */
788 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
789 /* libeio guarantees that the file offset does not change, and windows */
790 /* has no way to get an independent handle to the same file description */
791 HANDLE h = TO_SOCKET (ifd);
792 SetFilePointer (h, offset, 0, FILE_BEGIN);
793 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
794
915#else 795#else
916 res = -1; 796 res = EIO_ENOSYS ();
917 errno = ENOSYS;
918#endif 797#endif
919 798
799 /* we assume sendfile can copy at least 128mb in one go */
800 if (res <= 128 * 1024 * 1024)
801 {
802 if (res > 0)
803 written += res;
804
805 if (written)
806 return written;
807
808 break;
809 }
810 else
811 {
812 /* if we requested more, then probably the kernel was lazy */
813 written += res;
814 offset += res;
815 count -= res;
816
817 if (!count)
818 return written;
819 }
820 }
821
920 if (res < 0 822 if (res < 0
921 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 823 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
824 /* BSDs */
825#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
826 || errno == ENOTSUP
827#endif
828#ifdef EOPNOTSUPP /* windows */
829 || errno == EOPNOTSUPP /* BSDs */
830#endif
922#if __solaris 831#if __solaris
923 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 832 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
924#endif 833#endif
925 ) 834 )
926 ) 835 )
… …
930 839
931 res = 0; 840 res = 0;
932 841
933 while (count) 842 while (count)
934 { 843 {
935 ssize_t cnt; 844 eio_ssize_t cnt;
936 845
937 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 846 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
938 847
939 if (cnt <= 0) 848 if (cnt <= 0)
940 { 849 {
… …
952 861
953 offset += cnt; 862 offset += cnt;
954 res += cnt; 863 res += cnt;
955 count -= cnt; 864 count -= cnt;
956 } 865 }
866
867 FUBd;
957 } 868 }
958 869
959 return res; 870 return res;
871}
872
873#ifdef PAGESIZE
874# define eio_pagesize() PAGESIZE
875#else
876static intptr_t
877eio_pagesize (void)
878{
879 static intptr_t page;
880
881 if (!page)
882 page = sysconf (_SC_PAGESIZE);
883
884 return page;
885}
886#endif
887
888static void
889eio_page_align (void **addr, size_t *length)
890{
891 intptr_t mask = eio_pagesize () - 1;
892
893 /* round down addr */
894 intptr_t adj = mask & (intptr_t)*addr;
895
896 *addr = (void *)((intptr_t)*addr - adj);
897 *length += adj;
898
899 /* round up length */
900 *length = (*length + mask) & ~mask;
901}
902
903#if !_POSIX_MEMLOCK
904# define eio__mlockall(a) EIO_ENOSYS ()
905#else
906
907static int
908eio__mlockall (int flags)
909{
910 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
911 extern int mallopt (int, int);
912 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
913 #endif
914
915 #ifndef MCL_ONFAULT
916 if (flags & EIO_MCL_ONFAULT)
917 return EIO_ERRNO (EINVAL, -1);
918 #define MCL_ONFAULT 4
919 #endif
920
921 if (EIO_MCL_CURRENT != MCL_CURRENT
922 || EIO_MCL_FUTURE != MCL_FUTURE
923 || EIO_MCL_ONFAULT != MCL_ONFAULT)
924 {
925 flags = 0
926 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
927 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0)
928 | (flags & EIO_MCL_ONFAULT ? MCL_ONFAULT : 0)
929 ;
930 }
931
932 return mlockall (flags);
933}
934#endif
935
936#if !_POSIX_MEMLOCK_RANGE
937# define eio__mlock(a,b) EIO_ENOSYS ()
938#else
939
940static int
941eio__mlock (void *addr, size_t length)
942{
943 eio_page_align (&addr, &length);
944
945 return mlock (addr, length);
946}
947
948#endif
949
950#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
951# define eio__msync(a,b,c) EIO_ENOSYS ()
952#else
953
954static int
955eio__msync (void *mem, size_t len, int flags)
956{
957 eio_page_align (&mem, &len);
958
959 if (EIO_MS_ASYNC != MS_SYNC
960 || EIO_MS_INVALIDATE != MS_INVALIDATE
961 || EIO_MS_SYNC != MS_SYNC)
962 {
963 flags = 0
964 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
965 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
966 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
967 }
968
969 return msync (mem, len, flags);
970}
971
972#endif
973
974static int
975eio__mtouch (eio_req *req)
976{
977 void *mem = req->ptr2;
978 size_t len = req->size;
979 int flags = req->int1;
980
981 eio_page_align (&mem, &len);
982
983 {
984 intptr_t addr = (intptr_t)mem;
985 intptr_t end = addr + len;
986 intptr_t page = eio_pagesize ();
987
988 if (addr < end)
989 if (flags & EIO_MT_MODIFY) /* modify */
990 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < end && !EIO_CANCELLED (req));
991 else
992 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < end && !EIO_CANCELLED (req));
993 }
994
995 return 0;
996}
997
998/*****************************************************************************/
999/* requests implemented outside eio_execute, because they are so large */
1000
1001static void
1002eio__lseek (eio_req *req)
1003{
1004 /* this usually gets optimised away completely, or your compiler sucks, */
1005 /* or the whence constants really are not 0, 1, 2 */
1006 int whence = req->int2 == EIO_SEEK_SET ? SEEK_SET
1007 : req->int2 == EIO_SEEK_CUR ? SEEK_CUR
1008 : req->int2 == EIO_SEEK_END ? SEEK_END
1009 : req->int2;
1010
1011 req->offs = lseek (req->int1, req->offs, whence);
1012 req->result = req->offs == (off_t)-1 ? -1 : 0;
1013}
1014
1015/* result will always end up in tmpbuf, there is always space for adding a 0-byte */
1016static int
1017eio__realpath (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
1018{
1019 char *res;
1020 const char *rel = path;
1021 char *tmp1, *tmp2;
1022#if SYMLOOP_MAX > 32
1023 int symlinks = SYMLOOP_MAX;
1024#else
1025 int symlinks = 32;
1026#endif
1027
1028 errno = EINVAL;
1029 if (!rel)
1030 return -1;
1031
1032 errno = ENOENT;
1033 if (!*rel)
1034 return -1;
1035
1036 res = etp_tmpbuf_get (tmpbuf, EIO_PATH_MAX * 3);
1037#ifdef _WIN32
1038 if (_access (rel, 4) != 0)
1039 return -1;
1040
1041 symlinks = GetFullPathName (rel, EIO_PATH_MAX * 3, res, 0);
1042
1043 errno = ENAMETOOLONG;
1044 if (symlinks >= EIO_PATH_MAX * 3)
1045 return -1;
1046
1047 errno = EIO;
1048 if (symlinks <= 0)
1049 return -1;
1050
1051 return symlinks;
1052
1053#else
1054 tmp1 = res + EIO_PATH_MAX;
1055 tmp2 = tmp1 + EIO_PATH_MAX;
1056
1057#if 0 /* disabled, the musl way to do things is just too racy */
1058#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1059 /* on linux we may be able to ask the kernel */
1060 {
1061 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1062
1063 if (fd >= 0)
1064 {
1065 sprintf (tmp1, "/proc/self/fd/%d", fd);
1066 req->result = readlink (tmp1, res, EIO_PATH_MAX);
1067 /* here we should probably stat the open file and the disk file, to make sure they still match */
1068 eio__close (fd);
1069
1070 if (req->result > 0)
1071 goto done;
1072 }
1073 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1074 return -1;
1075 }
1076#endif
1077#endif
1078
1079 if (*rel != '/')
1080 {
1081 int len;
1082
1083 errno = ENOENT;
1084 if (wd == EIO_INVALID_WD)
1085 return -1;
1086
1087 if (wd == EIO_CWD)
1088 {
1089 if (!getcwd (res, EIO_PATH_MAX))
1090 return -1;
1091
1092 len = strlen (res);
1093 }
1094 else
1095 memcpy (res, wd->str, len = wd->len);
1096
1097 if (res [1]) /* only use if not / */
1098 res += len;
1099 }
1100
1101 while (*rel)
1102 {
1103 eio_ssize_t len, linklen;
1104 const char *beg = rel;
1105
1106 while (*rel && *rel != '/')
1107 ++rel;
1108
1109 len = rel - beg;
1110
1111 if (!len) /* skip slashes */
1112 {
1113 ++rel;
1114 continue;
1115 }
1116
1117 if (beg [0] == '.')
1118 {
1119 if (len == 1)
1120 continue; /* . - nop */
1121
1122 if (beg [1] == '.' && len == 2)
1123 {
1124 /* .. - back up one component, if possible */
1125
1126 while (res != tmpbuf->ptr)
1127 if (*--res == '/')
1128 break;
1129
1130 continue;
1131 }
1132 }
1133
1134 errno = ENAMETOOLONG;
1135 if (res + 1 + len + 1 >= tmp1)
1136 return -1;
1137
1138 /* copy one component */
1139 *res = '/';
1140 memcpy (res + 1, beg, len);
1141
1142 /* zero-terminate, for readlink */
1143 res [len + 1] = 0;
1144
1145 /* now check if it's a symlink */
1146 linklen = readlink (tmpbuf->ptr, tmp1, EIO_PATH_MAX);
1147
1148 if (linklen < 0)
1149 {
1150 if (errno != EINVAL)
1151 return -1;
1152
1153 /* it's a normal directory. hopefully */
1154 res += len + 1;
1155 }
1156 else
1157 {
1158 /* yay, it was a symlink - build new path in tmp2 */
1159 int rellen = strlen (rel);
1160
1161 errno = ENAMETOOLONG;
1162 if (linklen + 1 + rellen >= EIO_PATH_MAX) /* also catch linklen >= EIO_PATH_MAX */
1163 return -1;
1164
1165 errno = ELOOP;
1166 if (!--symlinks)
1167 return -1;
1168
1169 if (*tmp1 == '/')
1170 res = tmpbuf->ptr; /* symlink resolves to an absolute path */
1171
1172 /* we need to be careful, as rel might point into tmp2 already */
1173 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1174 tmp2 [linklen] = '/';
1175 memcpy (tmp2, tmp1, linklen);
1176
1177 rel = tmp2;
1178 }
1179 }
1180
1181 /* special case for the lone root path */
1182 if (res == tmpbuf->ptr)
1183 *res++ = '/';
1184
1185 return res - (char *)tmpbuf->ptr;
1186#endif
1187}
1188
1189static signed char
1190eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1191{
1192 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1193 : a->inode < b->inode ? -1
1194 : a->inode > b->inode ? 1
1195 : 0;
1196}
1197
1198#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1199
1200#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1201#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1202
1203static void
1204eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1205{
1206 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1207 unsigned char *bit = bits;
1208
1209 assert (CHAR_BIT == 8);
1210 assert (sizeof (eio_dirent) * 8 < 256);
1211 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1212 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1213
1214 if (size <= EIO_SORT_FAST)
1215 return;
1216
1217 /* first prepare an array of bits to test in our radix sort */
1218 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1219 /* inode_bits must contain all inodes ORed together */
1220 /* which is used to skip bits that are 0 everywhere, which is very common */
1221 {
1222 eio_ino_t endianness;
1223 int i, j;
1224
1225 /* we store the byte offset of byte n into byte n of "endianness" */
1226 for (i = 0; i < sizeof (eio_ino_t); ++i)
1227 ((unsigned char *)&endianness)[i] = i;
1228
1229 *bit++ = 0;
1230
1231 for (i = 0; i < sizeof (eio_ino_t); ++i)
1232 {
1233 /* shifting off the byte offsets out of "endianness" */
1234 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1235 endianness >>= 8;
1236
1237 for (j = 0; j < 8; ++j)
1238 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1239 *bit++ = offs + j;
1240 }
1241
1242 for (j = 0; j < 8; ++j)
1243 if (score_bits & (1 << j))
1244 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1245 }
1246
1247 /* now actually do the sorting (a variant of MSD radix sort) */
1248 {
1249 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1250 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1251 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1252 int stk_idx = 0;
1253
1254 base_stk [stk_idx] = dents;
1255 end_stk [stk_idx] = dents + size;
1256 bit_stk [stk_idx] = bit - 1;
1257
1258 do
1259 {
1260 base = base_stk [stk_idx];
1261 end = end_stk [stk_idx];
1262 bit = bit_stk [stk_idx];
1263
1264 for (;;)
1265 {
1266 unsigned char O = *bit >> 3;
1267 unsigned char M = 1 << (*bit & 7);
1268
1269 eio_dirent *a = base;
1270 eio_dirent *b = end;
1271
1272 if (b - a < EIO_SORT_CUTOFF)
1273 break;
1274
1275 /* now bit-partition the array on the bit */
1276 /* this ugly asymmetric loop seems to perform much better than typical */
1277 /* partition algos found in the literature */
1278 do
1279 if (!(((unsigned char *)a)[O] & M))
1280 ++a;
1281 else if (!(((unsigned char *)--b)[O] & M))
1282 {
1283 eio_dirent tmp = *a; *a = *b; *b = tmp;
1284 ++a;
1285 }
1286 while (b > a);
1287
1288 /* next bit, or stop, if no bits left in this path */
1289 if (!*--bit)
1290 break;
1291
1292 base_stk [stk_idx] = a;
1293 end_stk [stk_idx] = end;
1294 bit_stk [stk_idx] = bit;
1295 ++stk_idx;
1296
1297 end = a;
1298 }
1299 }
1300 while (stk_idx--);
1301 }
1302}
1303
1304static void
1305eio_dent_insertion_sort (eio_dirent *dents, int size)
1306{
1307 /* first move the smallest element to the front, to act as a sentinel */
1308 {
1309 int i;
1310 eio_dirent *min = dents;
1311
1312 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1313 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1314 if (EIO_DENT_CMP (dents [i], <, *min))
1315 min = &dents [i];
1316
1317 /* swap elements 0 and j (minimum) */
1318 {
1319 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1320 }
1321 }
1322
1323 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1324 {
1325 eio_dirent *i, *j;
1326
1327 for (i = dents + 1; i < dents + size; ++i)
1328 {
1329 eio_dirent value = *i;
1330
1331 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1332 j [1] = j [0];
1333
1334 j [1] = value;
1335 }
1336 }
1337}
1338
1339static void
1340eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1341{
1342 if (size <= 1)
1343 return; /* our insertion sort relies on size > 0 */
1344
1345 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1346 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1347 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1348
1349 /* use an insertion sort at the end, or for small arrays, */
1350 /* as insertion sort is more efficient for small partitions */
1351 eio_dent_insertion_sort (dents, size);
960} 1352}
961 1353
962/* read a full directory */ 1354/* read a full directory */
963static void 1355static void
964eio__scandir (eio_req *req, etp_worker *self) 1356eio__scandir (eio_req *req, etp_worker *self)
965{ 1357{
1358 char *name, *names;
1359 int namesalloc = 4096 - sizeof (void *) * 4;
1360 int namesoffs = 0;
1361 int flags = req->int1;
1362 eio_dirent *dents = 0;
1363 int dentalloc = 128;
1364 int dentoffs = 0;
1365 eio_ino_t inode_bits = 0;
1366#ifdef _WIN32
1367 HANDLE dirp;
1368 WIN32_FIND_DATA entp;
1369#else
966 DIR *dirp; 1370 DIR *dirp;
967 EIO_STRUCT_DIRENT *entp; 1371 EIO_STRUCT_DIRENT *entp;
968 char *name, *names; 1372#endif
969 int memlen = 4096;
970 int memofs = 0;
971 int res = 0;
972 1373
973 X_LOCK (wrklock); 1374 req->result = -1;
974 /* the corresponding closedir is in ETP_WORKER_CLEAR */
975 self->dirp = dirp = opendir (req->ptr1);
976 req->flags |= EIO_FLAG_PTR2_FREE;
977 req->ptr2 = names = malloc (memlen);
978 X_UNLOCK (wrklock);
979 1375
980 if (dirp && names) 1376 if (!(flags & EIO_READDIR_DENTS))
981 for (;;) 1377 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1378
1379#ifdef _WIN32
1380 {
1381 int len = strlen ((const char *)req->ptr1);
1382 char *path = malloc (MAX_PATH);
1383 const char *fmt;
1384 const char *reqpath = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1385
1386 if (!len)
1387 fmt = "./*";
1388 else if (reqpath[len - 1] == '/' || reqpath[len - 1] == '\\')
1389 fmt = "%s*";
1390 else
1391 fmt = "%s/*";
1392
1393 _snprintf (path, MAX_PATH, fmt, reqpath);
1394 dirp = FindFirstFile (path, &entp);
1395 free (path);
1396
1397 if (dirp == INVALID_HANDLE_VALUE)
982 { 1398 {
1399 /* should steal _dosmaperr */
1400 switch (GetLastError ())
1401 {
1402 case ERROR_FILE_NOT_FOUND:
1403 req->result = 0;
1404 break;
1405
1406 case ERROR_INVALID_NAME:
1407 case ERROR_PATH_NOT_FOUND:
1408 case ERROR_NO_MORE_FILES:
1409 errno = ENOENT;
1410 break;
1411
1412 case ERROR_NOT_ENOUGH_MEMORY:
1413 errno = ENOMEM;
1414 break;
1415
1416 default:
1417 errno = EINVAL;
1418 break;
1419 }
1420
1421 return;
1422 }
1423 }
1424#else
1425 #if HAVE_AT
1426 {
1427 int fd = openat (WD2FD (req->wd), req->ptr1, O_CLOEXEC | O_RDONLY | O_DIRECTORY | O_NONBLOCK);
1428
1429 if (fd < 0)
1430 return;
1431
1432 dirp = fdopendir (fd);
1433
1434 if (!dirp)
1435 {
1436 silent_close (fd);
1437 return;
1438 }
1439 }
1440 #else
1441 dirp = opendir (wd_expand (&self->tmpbuf, req->wd, req->ptr1));
1442
1443 if (!dirp)
1444 return;
1445 #endif
1446#endif
1447
1448 if (req->flags & EIO_FLAG_PTR1_FREE)
1449 free (req->ptr1);
1450
1451 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1452 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1453 req->ptr2 = names = malloc (namesalloc);
1454
1455 if (!names || (flags && !dents))
1456 return;
1457
1458 for (;;)
1459 {
1460 int done;
1461
1462#ifdef _WIN32
1463 done = !dirp;
1464#else
983 errno = 0; 1465 errno = 0;
984 entp = readdir (dirp); 1466 entp = readdir (dirp);
1467 done = !entp;
1468#endif
985 1469
1470 if (done)
1471 {
1472#ifndef _WIN32
1473 int old_errno = errno;
1474 closedir (dirp);
1475 errno = old_errno;
1476
986 if (!entp) 1477 if (errno)
1478 break;
1479#endif
1480
1481 /* sort etc. */
1482 req->int1 = flags;
1483 req->result = dentoffs;
1484
1485 if (flags & EIO_READDIR_STAT_ORDER)
1486 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1487 else if (flags & EIO_READDIR_DIRS_FIRST)
1488 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1489 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1490 else
1491 {
1492 /* in this case, all is known, and we just put dirs first and sort them */
1493 eio_dirent *oth = dents + dentoffs;
1494 eio_dirent *dir = dents;
1495
1496 /* now partition dirs to the front, and non-dirs to the back */
1497 /* by walking from both sides and swapping if necessary */
1498 while (oth > dir)
1499 {
1500 if (dir->type == EIO_DT_DIR)
1501 ++dir;
1502 else if ((--oth)->type == EIO_DT_DIR)
1503 {
1504 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1505
1506 ++dir;
1507 }
1508 }
1509
1510 /* now sort the dirs only (dirs all have the same score) */
1511 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1512 }
1513
987 break; 1514 break;
1515 }
988 1516
989 name = entp->d_name; 1517 /* now add the entry to our list(s) */
1518 name = D_NAME (entp);
990 1519
1520 /* skip . and .. entries */
991 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1521 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
992 { 1522 {
993 int len = strlen (name) + 1; 1523 int len = D_NAMLEN (entp) + 1;
994 1524
995 res++; 1525 while (ecb_expect_false (namesoffs + len > namesalloc))
996
997 while (memofs + len > memlen)
998 { 1526 {
999 memlen *= 2; 1527 namesalloc *= 2;
1000 X_LOCK (wrklock);
1001 req->ptr2 = names = realloc (names, memlen); 1528 req->ptr2 = names = realloc (names, namesalloc);
1002 X_UNLOCK (wrklock);
1003 1529
1004 if (!names) 1530 if (!names)
1531 break;
1532 }
1533
1534 memcpy (names + namesoffs, name, len);
1535
1536 if (dents)
1537 {
1538 struct eio_dirent *ent;
1539
1540 if (ecb_expect_false (dentoffs == dentalloc))
1541 {
1542 dentalloc *= 2;
1543 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1544
1545 if (!dents)
1005 break; 1546 break;
1006 } 1547 }
1007 1548
1008 memcpy (names + memofs, name, len); 1549 ent = dents + dentoffs;
1009 memofs += len; 1550
1551 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1552 ent->namelen = len - 1;
1553 ent->inode = D_INO (entp);
1554
1555 inode_bits |= ent->inode;
1556
1557 switch (D_TYPE (entp))
1558 {
1559 default:
1560 ent->type = EIO_DT_UNKNOWN;
1561 flags |= EIO_READDIR_FOUND_UNKNOWN;
1562 break;
1563
1564 #ifdef DT_FIFO
1565 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1566 #endif
1567 #ifdef DT_CHR
1568 case DT_CHR: ent->type = EIO_DT_CHR; break;
1569 #endif
1570 #ifdef DT_MPC
1571 case DT_MPC: ent->type = EIO_DT_MPC; break;
1572 #endif
1573 #ifdef DT_DIR
1574 case DT_DIR: ent->type = EIO_DT_DIR; break;
1575 #endif
1576 #ifdef DT_NAM
1577 case DT_NAM: ent->type = EIO_DT_NAM; break;
1578 #endif
1579 #ifdef DT_BLK
1580 case DT_BLK: ent->type = EIO_DT_BLK; break;
1581 #endif
1582 #ifdef DT_MPB
1583 case DT_MPB: ent->type = EIO_DT_MPB; break;
1584 #endif
1585 #ifdef DT_REG
1586 case DT_REG: ent->type = EIO_DT_REG; break;
1587 #endif
1588 #ifdef DT_NWK
1589 case DT_NWK: ent->type = EIO_DT_NWK; break;
1590 #endif
1591 #ifdef DT_CMP
1592 case DT_CMP: ent->type = EIO_DT_CMP; break;
1593 #endif
1594 #ifdef DT_LNK
1595 case DT_LNK: ent->type = EIO_DT_LNK; break;
1596 #endif
1597 #ifdef DT_SOCK
1598 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1599 #endif
1600 #ifdef DT_DOOR
1601 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1602 #endif
1603 #ifdef DT_WHT
1604 case DT_WHT: ent->type = EIO_DT_WHT; break;
1605 #endif
1606 }
1607
1608 ent->score = 7;
1609
1610 if (flags & EIO_READDIR_DIRS_FIRST)
1611 {
1612 if (ent->type == EIO_DT_UNKNOWN)
1613 {
1614 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1615 ent->score = 1;
1616 else if (!strchr (name, '.')) /* absence of dots indicate likely dirs */
1617 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1618 }
1619 else if (ent->type == EIO_DT_DIR)
1620 ent->score = 0;
1621 }
1010 } 1622 }
1623
1624 namesoffs += len;
1625 ++dentoffs;
1011 } 1626 }
1012 1627
1013 if (errno) 1628 if (EIO_CANCELLED (req))
1014 res = -1; 1629 {
1630 errno = ECANCELED;
1631 break;
1632 }
1633
1634#ifdef _WIN32
1635 if (!FindNextFile (dirp, &entp))
1636 {
1637 FindClose (dirp);
1638 dirp = 0;
1639 }
1640#endif
1641 }
1642}
1643
1644/*****************************************************************************/
1645/* working directory stuff */
1646/* various deficiencies in the posix 2008 api force us to */
1647/* keep the absolute path in string form at all times */
1648/* fuck yeah. */
1649
1650#if !HAVE_AT
1651
1652/* a bit like realpath, but usually faster because it doesn't have to return */
1653/* an absolute or canonical path */
1654static const char *
1655wd_expand (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
1656{
1657 if (!wd || *path == '/')
1658 return path;
1659
1660 if (path [0] == '.' && !path [1])
1661 return wd->str;
1662
1015 1663 {
1016 req->result = res; 1664 int l1 = wd->len;
1017} 1665 int l2 = strlen (path);
1018 1666
1019#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) 1667 char *res = etp_tmpbuf_get (tmpbuf, l1 + l2 + 2);
1020# undef msync 1668
1021# define msync(a,b,c) ENOSYS 1669 memcpy (res, wd->str, l1);
1670 res [l1] = '/';
1671 memcpy (res + l1 + 1, path, l2 + 1);
1672
1673 return res;
1674 }
1675}
1676
1677#endif
1678
1679static eio_wd
1680eio__wd_open_sync (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
1681{
1682 int fd;
1683 eio_wd res;
1684 int len = eio__realpath (tmpbuf, wd, path);
1685
1686 if (len < 0)
1687 return EIO_INVALID_WD;
1688
1689#if HAVE_AT
1690 fd = openat (WD2FD (wd), path, EIO_O_PATH | O_DIRECTORY | O_NONBLOCK | O_CLOEXEC);
1691
1692 /* 0 is a valid fd, but we use it for EIO_CWD, so in the very unlikely */
1693 /* case of fd 0 being available (almost certainly an a pplication bug) */
1694 /* make sure we use another fd value */
1695 #if EIO_CWD
1696 error EIO_CWD must be 0
1022#endif 1697 #endif
1698 if (ecb_expect_false (fd == 0))
1699 {
1700 int fd2 = fcntl (fd, F_DUPFD_CLOEXEC ? F_DUPFD_CLOEXEC : F_DUPFD);
1701 fcntl (fd2, F_SETFD, FD_CLOEXEC);
1702 eio__close (fd);
1703 fd = fd2;
1704 }
1023 1705
1024int 1706 if (fd < 0)
1025eio__mtouch (void *mem, size_t len, int flags) 1707 return EIO_INVALID_WD;
1708#endif
1709
1710 res = malloc (sizeof (*res) + len); /* one extra 0-byte */
1711
1712#if HAVE_AT
1713 res->fd = fd;
1714#endif
1715
1716 res->len = len;
1717 memcpy (res->str, tmpbuf->ptr, len);
1718 res->str [len] = 0;
1719
1720 return res;
1721}
1722
1723eio_wd
1724eio_wd_open_sync (eio_wd wd, const char *path)
1026{ 1725{
1027 intptr_t addr = (intptr_t)mem; 1726 struct etp_tmpbuf tmpbuf = { 0 };
1028 intptr_t end = addr + len; 1727 wd = eio__wd_open_sync (&tmpbuf, wd, path);
1029#ifdef PAGESIZE 1728 free (tmpbuf.ptr);
1030 const intptr_t page = PAGESIZE; 1729
1730 return wd;
1731}
1732
1733void
1734eio_wd_close_sync (eio_wd wd)
1735{
1736 if (wd != EIO_INVALID_WD && wd != EIO_CWD)
1737 {
1738 #if HAVE_AT
1739 eio__close (wd->fd);
1740 #endif
1741 free (wd);
1742 }
1743}
1744
1745#if HAVE_AT
1746
1747static int
1748eio__renameat2 (int olddirfd, const char *oldpath, int newdirfd, const char *newpath, unsigned int flags)
1749{
1750#if HAVE_RENAMEAT2
1751 return syscall (SYS_renameat2, olddirfd, oldpath, newdirfd, newpath, flags);
1031#else 1752#else
1032 static intptr_t page;
1033
1034 if (!page) 1753 if (flags)
1035 page = sysconf (_SC_PAGESIZE); 1754 return EIO_ENOSYS ();
1036#endif
1037 1755
1038 addr &= ~(page - 1); /* assume page size is always a power of two */ 1756 return renameat (olddirfd, oldpath, newdirfd, newpath);
1757#endif
1758}
1039 1759
1040 if (addr < end) 1760/* they forgot these */
1041 if (flags) /* modify */
1042 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
1043 else
1044 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
1045 1761
1762static int
1763eio__truncateat (int dirfd, const char *path, off_t length)
1764{
1765 int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC | O_NONBLOCK);
1766 int res;
1767
1768 if (fd < 0)
1769 return fd;
1770
1771 res = ftruncate (fd, length);
1772 silent_close (fd);
1046 return 0; 1773 return res;
1047} 1774}
1775
1776static int
1777eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
1778{
1779 int fd = openat (dirfd, path, EIO_O_PATH | O_CLOEXEC | O_NONBLOCK);
1780 int res;
1781
1782 if (fd < 0)
1783 return fd;
1784
1785 res = fstatvfs (fd, buf);
1786 silent_close (fd);
1787 return res;
1788}
1789
1790#endif
1048 1791
1049/*****************************************************************************/ 1792/*****************************************************************************/
1050 1793
1051#define ALLOC(len) \ 1794#define ALLOC(len) \
1052 if (!req->ptr2) \ 1795 if (!req->ptr2) \
1053 { \ 1796 { \
1054 X_LOCK (wrklock); \ 1797 X_LOCK (EIO_POOL->wrklock); \
1055 req->flags |= EIO_FLAG_PTR2_FREE; \ 1798 req->flags |= EIO_FLAG_PTR2_FREE; \
1056 X_UNLOCK (wrklock); \ 1799 X_UNLOCK (EIO_POOL->wrklock); \
1057 req->ptr2 = malloc (len); \ 1800 req->ptr2 = malloc (len); \
1058 if (!req->ptr2) \ 1801 if (!req->ptr2) \
1059 { \ 1802 { \
1060 errno = ENOMEM; \ 1803 errno = ENOMEM; \
1061 req->result = -1; \ 1804 req->result = -1; \
1062 break; \ 1805 goto alloc_fail; \
1063 } \ 1806 } \
1064 } 1807 }
1065 1808
1066X_THREAD_PROC (etp_proc) 1809/*****************************************************************************/
1067{
1068 ETP_REQ *req;
1069 struct timespec ts;
1070 etp_worker *self = (etp_worker *)thr_arg;
1071 1810
1072 /* try to distribute timeouts somewhat randomly */ 1811static void
1073 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 1812eio__slurp (int fd, eio_req *req)
1813{
1814 req->result = fd;
1074 1815
1075 for (;;) 1816 if (fd < 0)
1817 return;
1818
1819 if (req->offs < 0 || !req->size) /* do we need the size? */
1076 { 1820 {
1077 X_LOCK (reqlock); 1821 off_t size = lseek (fd, 0, SEEK_END);
1078 1822
1079 for (;;)
1080 {
1081 self->req = req = reqq_shift (&req_queue);
1082
1083 if (req)
1084 break;
1085
1086 ++idle;
1087
1088 ts.tv_sec = time (0) + IDLE_TIMEOUT;
1089 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
1090 {
1091 if (idle > max_idle)
1092 {
1093 --idle;
1094 X_UNLOCK (reqlock);
1095 X_LOCK (wrklock);
1096 --started;
1097 X_UNLOCK (wrklock);
1098 goto quit;
1099 }
1100
1101 /* we are allowed to idle, so do so without any timeout */
1102 X_COND_WAIT (reqwait, reqlock);
1103 }
1104
1105 --idle;
1106 }
1107
1108 --nready;
1109
1110 X_UNLOCK (reqlock);
1111
1112 if (req->type < 0) 1823 if (req->offs < 0)
1113 goto quit; 1824 req->offs += size;
1114 1825
1115 if (!EIO_CANCELLED (req)) 1826 if (!req->size)
1116 ETP_EXECUTE (self, req); 1827 req->size = size - req->offs;
1117
1118 X_LOCK (reslock);
1119
1120 ++npending;
1121
1122 if (!reqq_push (&res_queue, req) && want_poll_cb)
1123 want_poll_cb ();
1124
1125 self->req = 0;
1126 etp_worker_clear (self);
1127
1128 X_UNLOCK (reslock);
1129 } 1828 }
1130 1829
1131quit: 1830 ALLOC (req->size);
1132 X_LOCK (wrklock); 1831 req->result = pread (fd, req->ptr2, req->size, req->offs);
1133 etp_worker_free (self);
1134 X_UNLOCK (wrklock);
1135 1832
1136 return 0; 1833 silent_close (fd);
1137}
1138 1834
1139/*****************************************************************************/ 1835alloc_fail:
1836 ;
1837}
1140 1838
1839int ecb_cold
1141int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1840eio_init (void (*want_poll)(void), void (*done_poll)(void))
1142{ 1841{
1143 return etp_init (want_poll, done_poll); 1842 eio_want_poll_cb = want_poll;
1144} 1843 eio_done_poll_cb = done_poll;
1145 1844
1845 return etp_init (EIO_POOL, 0, 0, 0);
1846}
1847
1848ecb_inline void
1146static void eio_api_destroy (eio_req *req) 1849eio_api_destroy (eio_req *req)
1147{ 1850{
1148 free (req); 1851 free (req);
1149} 1852}
1150 1853
1151#define REQ(rtype) \ 1854#define REQ(rtype) \
1152 eio_req *req; \ 1855 eio_req *req; \
1153 \ 1856 \
1154 req = (eio_req *)calloc (1, sizeof *req); \ 1857 req = (eio_req *)calloc (1, sizeof *req); \
1155 if (!req) \ 1858 if (!req) \
1156 return 0; \ 1859 return 0; \
… …
1170 { \ 1873 { \
1171 eio_api_destroy (req); \ 1874 eio_api_destroy (req); \
1172 return 0; \ 1875 return 0; \
1173 } 1876 }
1174 1877
1878#define SINGLEDOT(ptr) (0[(char *)(ptr)] == '.' && !1[(char *)(ptr)])
1879
1880static void
1175static void eio_execute (etp_worker *self, eio_req *req) 1881eio_execute (etp_worker *self, eio_req *req)
1176{ 1882{
1177 errno = 0; 1883#if HAVE_AT
1884 int dirfd;
1885#else
1886 const char *path;
1887#endif
1888
1889 if (ecb_expect_false (EIO_CANCELLED (req)))
1890 {
1891 req->result = -1;
1892 req->errorno = ECANCELED;
1893 return;
1894 }
1895
1896 if (ecb_expect_false (req->wd == EIO_INVALID_WD))
1897 {
1898 req->result = -1;
1899 req->errorno = ENOENT;
1900 return;
1901 }
1902
1903 if (req->type >= EIO_OPEN)
1904 {
1905 #if HAVE_AT
1906 dirfd = WD2FD (req->wd);
1907 #else
1908 path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1909 #endif
1910 }
1178 1911
1179 switch (req->type) 1912 switch (req->type)
1180 { 1913 {
1914 case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
1915 req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
1916 break;
1917 case EIO_WD_CLOSE: req->result = 0;
1918 eio_wd_close_sync (req->wd); break;
1919
1920 case EIO_SEEK: eio__lseek (req); break;
1181 case EIO_READ: ALLOC (req->size); 1921 case EIO_READ: ALLOC (req->size);
1182 req->result = req->offs >= 0 1922 req->result = req->offs >= 0
1183 ? pread (req->int1, req->ptr2, req->size, req->offs) 1923 ? pread (req->int1, req->ptr2, req->size, req->offs)
1184 : read (req->int1, req->ptr2, req->size); break; 1924 : read (req->int1, req->ptr2, req->size); break;
1185 case EIO_WRITE: req->result = req->offs >= 0 1925 case EIO_WRITE: req->result = req->offs >= 0
1186 ? pwrite (req->int1, req->ptr2, req->size, req->offs) 1926 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1187 : write (req->int1, req->ptr2, req->size); break; 1927 : write (req->int1, req->ptr2, req->size); break;
1188 1928
1929 case EIO_FCNTL: req->result = fcntl (req->int1, (int) req->int2, req->ptr2); break;
1930 case EIO_IOCTL: req->result = ioctl (req->int1, (unsigned long)req->int2, req->ptr2); break;
1931
1189 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break; 1932 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1190 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break; 1933 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
1934
1935#if HAVE_AT
1191 1936
1192 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1937 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1193 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 1938 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break;
1194 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1939 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1195 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 1940 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, AT_SYMLINK_NOFOLLOW); break;
1196 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1197 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1198
1199 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; 1941 case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break;
1200 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1201 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break; 1942 case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break;
1202 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1203 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break; 1943 case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break;
1204 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1205
1206 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break; 1944 case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break;
1207 case EIO_CLOSE: req->result = close (req->int1); break; 1945 case EIO_SLURP: eio__slurp ( openat (dirfd, req->ptr1, O_RDONLY | O_CLOEXEC), req); break;
1208 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break; 1946
1209 case EIO_UNLINK: req->result = unlink (req->ptr1); break; 1947 case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
1210 case EIO_RMDIR: req->result = rmdir (req->ptr1); break; 1948 case EIO_RMDIR: /* complications arise because "." cannot be removed, so we might have to expand */
1949 req->result = req->wd && SINGLEDOT (req->ptr1)
1950 ? rmdir (req->wd->str)
1951 : unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break;
1211 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; 1952 case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); break;
1212 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; 1953 case EIO_RENAME: req->result = eio__renameat2 (
1954 dirfd,
1955 /* complications arise because "." cannot be renamed, so we might have to expand */
1956 req->wd && SINGLEDOT (req->ptr1) ? req->wd->str : req->ptr1,
1957 WD2FD ((eio_wd)req->int3),
1958 req->ptr2,
1959 req->int2
1960 );
1961 break;
1213 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; 1962 case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2, 0); break;
1214 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; 1963 case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
1215 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break; 1964 case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1216 1965 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1966 req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1217 case EIO_READLINK: ALLOC (NAME_MAX); 1967 case EIO_READLINK: ALLOC (EIO_PATH_MAX);
1218 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; 1968 req->result = readlinkat (dirfd, req->ptr1, req->ptr2, EIO_PATH_MAX);
1219 1969 if (req->result == EIO_PATH_MAX)
1220 case EIO_SYNC: req->result = 0; sync (); break; 1970 {
1221 case EIO_FSYNC: req->result = fsync (req->int1); break; 1971 req->result = -1;
1222 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 1972 errno = ENAMETOOLONG;
1223 case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break; 1973 }
1224 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break; 1974 break;
1225 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1226
1227 case EIO_READDIR: eio__scandir (req, self); break;
1228
1229 case EIO_BUSY: 1975 case EIO_UTIME:
1230#ifdef _WIN32 1976 case EIO_FUTIME:
1231 Sleep (req->nv1 * 1000.); 1977 {
1978 struct timespec ts[2];
1979 struct timespec *times;
1980
1981 if (req->nv1 != -1. || req->nv2 != -1.)
1982 {
1983 ts[0].tv_sec = req->nv1;
1984 ts[0].tv_nsec = (req->nv1 - ts[0].tv_sec) * 1e9;
1985 ts[1].tv_sec = req->nv2;
1986 ts[1].tv_nsec = (req->nv2 - ts[1].tv_sec) * 1e9;
1987
1988 times = ts;
1989 }
1990 else
1991 times = 0;
1992
1993 req->result = req->type == EIO_FUTIME
1994 ? futimens (req->int1, times)
1995 : utimensat (dirfd, req->ptr1, times, 0);
1996 }
1997 break;
1998
1232#else 1999#else
1233 {
1234 struct timeval tv;
1235 2000
1236 tv.tv_sec = req->nv1; 2001 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1237 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; 2002 req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
2003 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2004 req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
2005 case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
2006 case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
2007 case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
2008 case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
2009 case EIO_SLURP: eio__slurp ( open (path , O_RDONLY | O_CLOEXEC), req); break;
1238 2010
1239 req->result = select (0, 0, 0, 0, &tv); 2011 case EIO_UNLINK: req->result = unlink (path ); break;
1240 } 2012 case EIO_RMDIR: req->result = rmdir (path ); break;
1241#endif 2013 case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
1242 break; 2014 case EIO_RENAME: req->result = req->int2 ? EIO_ENOSYS () : rename (path, req->ptr2); break;
2015 case EIO_LINK: req->result = link (path , req->ptr2); break;
2016 case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
2017 case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
2018 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2019 req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
2020 case EIO_READLINK: ALLOC (EIO_PATH_MAX);
2021 req->result = readlink (path, req->ptr2, EIO_PATH_MAX);
2022 if (req->result == EIO_PATH_MAX)
2023 {
2024 req->result = -1;
2025 errno = ENAMETOOLONG;
2026 }
2027 break;
1243 2028
1244 case EIO_UTIME: 2029 case EIO_UTIME:
1245 case EIO_FUTIME: 2030 case EIO_FUTIME:
1246 { 2031 {
1247 struct timeval tv[2]; 2032 struct timeval tv[2];
1248 struct timeval *times; 2033 struct timeval *times;
1249 2034
1250 if (req->nv1 != -1. || req->nv2 != -1.) 2035 if (req->nv1 != -1. || req->nv2 != -1.)
1251 { 2036 {
1252 tv[0].tv_sec = req->nv1; 2037 tv[0].tv_sec = req->nv1;
1253 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.; 2038 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
1254 tv[1].tv_sec = req->nv2; 2039 tv[1].tv_sec = req->nv2;
1255 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.; 2040 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
1256 2041
1257 times = tv; 2042 times = tv;
1258 } 2043 }
1259 else 2044 else
1260 times = 0; 2045 times = 0;
1261
1262 2046
1263 req->result = req->type == EIO_FUTIME 2047 req->result = req->type == EIO_FUTIME
1264 ? futimes (req->int1, times) 2048 ? futimes (req->int1, times)
1265 : utimes (req->ptr1, times); 2049 : utimes (req->ptr1, times);
1266 } 2050 }
1267 break; 2051 break;
1268 2052
2053#endif
2054
2055 case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
2056 {
2057 ALLOC (req->result);
2058 memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
2059 }
2060 break;
2061
2062 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2063 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
2064
2065 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2066 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2067
2068 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
2069 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
2070 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
2071
2072 case EIO_CLOSE: req->result = eio__close (req->int1); break;
2073 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
2074 case EIO_SYNC: req->result = 0; sync (); break;
2075 case EIO_FSYNC: req->result = fsync (req->int1); break;
2076 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
2077 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
2078 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
2079 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
2080 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
2081 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
2082 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
2083 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
2084
2085 case EIO_READDIR: eio__scandir (req, self); break;
2086
2087 case EIO_BUSY:
2088#ifdef _WIN32
2089 Sleep (req->nv1 * 1e3);
2090#else
2091 {
2092 struct timeval tv;
2093
2094 tv.tv_sec = req->nv1;
2095 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
2096
2097 req->result = select (0, 0, 0, 0, &tv);
2098 }
2099#endif
2100 break;
2101
2102#if 0
1269 case EIO_GROUP: 2103 case EIO_GROUP:
1270 abort (); /* handled in eio_request */ 2104 abort (); /* handled in eio_request */
2105#endif
1271 2106
1272 case EIO_NOP: 2107 case EIO_NOP:
1273 req->result = 0; 2108 req->result = 0;
1274 break; 2109 break;
1275 2110
1276 case EIO_CUSTOM: 2111 case EIO_CUSTOM:
1277 ((void (*)(eio_req *))req->feed) (req); 2112 req->feed (req);
1278 break; 2113 break;
1279 2114
1280 default: 2115 default:
1281 req->result = -1; 2116 req->result = EIO_ENOSYS ();
1282 break; 2117 break;
1283 } 2118 }
1284 2119
2120alloc_fail:
1285 req->errorno = errno; 2121 req->errorno = errno;
1286} 2122}
1287 2123
1288#ifndef EIO_NO_WRAPPERS 2124#ifndef EIO_NO_WRAPPERS
1289 2125
2126eio_req *eio_wd_open (const char *path, int pri, eio_cb cb, void *data)
2127{
2128 REQ (EIO_WD_OPEN); PATH; SEND;
2129}
2130
2131eio_req *eio_wd_close (eio_wd wd, int pri, eio_cb cb, void *data)
2132{
2133 REQ (EIO_WD_CLOSE); req->wd = wd; SEND;
2134}
2135
1290eio_req *eio_nop (int pri, eio_cb cb, void *data) 2136eio_req *eio_nop (int pri, eio_cb cb, void *data)
1291{ 2137{
1292 REQ (EIO_NOP); SEND; 2138 REQ (EIO_NOP); SEND;
1293} 2139}
1294 2140
… …
1310eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2156eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1311{ 2157{
1312 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2158 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1313} 2159}
1314 2160
2161eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2162{
2163 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2164}
2165
2166eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
2167{
2168 REQ (EIO_SYNCFS); req->int1 = fd; SEND;
2169}
2170
2171eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2172{
2173 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2174}
2175
1315eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2176eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1316{ 2177{
1317 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2178 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1318} 2179}
1319 2180
1320eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) 2181eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
1321{ 2182{
1322 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; 2183 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
1323} 2184}
1324 2185
1325eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) 2186eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
1326{ 2187{
1327 REQ (EIO_FDATASYNC); req->int1 = fd; SEND; 2188 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2189}
2190
2191eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2192{
2193 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
1328} 2194}
1329 2195
1330eio_req *eio_close (int fd, int pri, eio_cb cb, void *data) 2196eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
1331{ 2197{
1332 REQ (EIO_CLOSE); req->int1 = fd; SEND; 2198 REQ (EIO_CLOSE); req->int1 = fd; SEND;
… …
1335eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data) 2201eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
1336{ 2202{
1337 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND; 2203 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
1338} 2204}
1339 2205
2206eio_req *eio_seek (int fd, off_t offset, int whence, int pri, eio_cb cb, void *data)
2207{
2208 REQ (EIO_SEEK); req->int1 = fd; req->offs = offset; req->int2 = whence; SEND;
2209}
2210
1340eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data) 2211eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1341{ 2212{
1342 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND; 2213 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1343} 2214}
1344 2215
1345eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data) 2216eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1346{ 2217{
1347 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND; 2218 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1348} 2219}
1349 2220
2221eio_req *eio_fcntl (int fd, int cmd, void *arg, int pri, eio_cb cb, void *data)
2222{
2223 REQ (EIO_IOCTL); req->int1 = fd; req->int2 = cmd; req->ptr2 = arg; SEND;
2224}
2225
2226eio_req *eio_ioctl (int fd, unsigned long request, void *buf, int pri, eio_cb cb, void *data)
2227{
2228 REQ (EIO_IOCTL); req->int1 = fd; req->int2 = request; req->ptr2 = buf; SEND;
2229}
2230
1350eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) 2231eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1351{ 2232{
1352 REQ (EIO_FSTAT); req->int1 = fd; SEND; 2233 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1353} 2234}
1354 2235
2236eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
2237{
2238 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
2239}
2240
1355eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) 2241eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1356{ 2242{
1357 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; 2243 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1358} 2244}
1359 2245
… …
1365eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) 2251eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1366{ 2252{
1367 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2253 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1368} 2254}
1369 2255
1370eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2256eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1371{ 2257{
1372 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2258 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1373} 2259}
1374 2260
1375eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) 2261eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
… …
1395eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data) 2281eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1396{ 2282{
1397 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2283 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1398} 2284}
1399 2285
1400eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2286eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1401{ 2287{
1402 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2288 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1403} 2289}
1404 2290
1405eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) 2291eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
… …
1421eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data) 2307eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1422{ 2308{
1423 return eio__1path (EIO_READLINK, path, pri, cb, data); 2309 return eio__1path (EIO_READLINK, path, pri, cb, data);
1424} 2310}
1425 2311
2312eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2313{
2314 return eio__1path (EIO_REALPATH, path, pri, cb, data);
2315}
2316
1426eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) 2317eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1427{ 2318{
1428 return eio__1path (EIO_STAT, path, pri, cb, data); 2319 return eio__1path (EIO_STAT, path, pri, cb, data);
1429} 2320}
1430 2321
1431eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) 2322eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1432{ 2323{
1433 return eio__1path (EIO_LSTAT, path, pri, cb, data); 2324 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1434} 2325}
1435 2326
2327eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
2328{
2329 return eio__1path (EIO_STATVFS, path, pri, cb, data);
2330}
2331
1436eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) 2332eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1437{ 2333{
1438 return eio__1path (EIO_UNLINK, path, pri, cb, data); 2334 return eio__1path (EIO_UNLINK, path, pri, cb, data);
1439} 2335}
1440 2336
1441eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data) 2337eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1442{ 2338{
1443 return eio__1path (EIO_RMDIR, path, pri, cb, data); 2339 return eio__1path (EIO_RMDIR, path, pri, cb, data);
1444} 2340}
1445 2341
1446eio_req *eio_readdir (const char *path, int pri, eio_cb cb, void *data) 2342eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1447{ 2343{
1448 return eio__1path (EIO_READDIR, path, pri, cb, data); 2344 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1449} 2345}
1450 2346
1451eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) 2347eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1452{ 2348{
1453 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; 2349 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1454} 2350}
1455 2351
1456static eio_req * 2352static eio_req *
1457eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2353eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1458{ 2354{
… …
1482eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2378eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1483{ 2379{
1484 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2380 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1485} 2381}
1486 2382
2383eio_req *eio_slurp (const char *path, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
2384{
2385 REQ (EIO_SLURP); PATH; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
2386}
2387
1487eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) 2388eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
1488{ 2389{
1489 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND; 2390 REQ (EIO_CUSTOM); req->feed = execute; SEND;
1490} 2391}
1491 2392
1492#endif 2393#endif
1493 2394
1494eio_req *eio_grp (eio_cb cb, void *data) 2395eio_req *eio_grp (eio_cb cb, void *data)
… …
1503#undef SEND 2404#undef SEND
1504 2405
1505/*****************************************************************************/ 2406/*****************************************************************************/
1506/* grp functions */ 2407/* grp functions */
1507 2408
2409void
1508void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2410eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1509{ 2411{
1510 grp->int2 = limit; 2412 grp->int2 = limit;
1511 grp->feed = feed; 2413 grp->feed = feed;
1512 2414
1513 grp_try_feed (grp); 2415 grp_try_feed (grp);
1514} 2416}
1515 2417
2418void
1516void eio_grp_limit (eio_req *grp, int limit) 2419eio_grp_limit (eio_req *grp, int limit)
1517{ 2420{
1518 grp->int2 = limit; 2421 grp->int2 = limit;
1519 2422
1520 grp_try_feed (grp); 2423 grp_try_feed (grp);
1521} 2424}
1522 2425
2426void
1523void eio_grp_add (eio_req *grp, eio_req *req) 2427eio_grp_add (eio_req *grp, eio_req *req)
1524{ 2428{
1525 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2429 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1526 2430
1527 grp->flags |= EIO_FLAG_GROUPADD; 2431 grp->flags |= ETP_FLAG_GROUPADD;
1528 2432
1529 ++grp->size; 2433 ++grp->size;
1530 req->grp = grp; 2434 req->grp = grp;
1531 2435
1532 req->grp_prev = 0; 2436 req->grp_prev = 0;
… …
1539} 2443}
1540 2444
1541/*****************************************************************************/ 2445/*****************************************************************************/
1542/* misc garbage */ 2446/* misc garbage */
1543 2447
2448eio_ssize_t
1544ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2449eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1545{ 2450{
1546 etp_worker wrk;
1547
1548 wrk.dbuf = 0;
1549
1550 eio__sendfile (ofd, ifd, offset, count, &wrk); 2451 return eio__sendfile (ofd, ifd, offset, count);
1551
1552 if (wrk.dbuf)
1553 free (wrk.dbuf);
1554} 2452}
1555 2453
2454int eio_mlockall_sync (int flags)
2455{
2456 return eio__mlockall (flags);
2457}
2458

Diff Legend

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