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

Comparing libeio/eio.c (file contents):
Revision 1.16 by root, Fri May 30 04:50:21 2008 UTC vs.
Revision 1.150 by root, Fri Jun 15 02:57:36 2018 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 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 EIO_PATH_MIN
335# define EIO_PATH_MIN 8160
336#endif
337
338#define EIO_PATH_MAX (PATH_MAX <= EIO_PATH_MIN ? EIO_PATH_MIN : PATH_MAX)
339
109/* buffer size for various temporary buffers */ 340/* buffer size for various temporary buffers */
110#define EIO_BUFSIZE 65536 341#define EIO_BUFSIZE 65536
111 342
112#define dBUF \ 343#define dBUF \
113 char *eio_buf; \
114 ETP_WORKER_LOCK (self); \
115 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 344 char *eio_buf = malloc (EIO_BUFSIZE); \
116 ETP_WORKER_UNLOCK (self); \
117 errno = ENOMEM; \ 345 errno = ENOMEM; \
118 if (!eio_buf) \ 346 if (!eio_buf) \
119 return -1; 347 return -1
120 348
121#define EIO_TICKS ((1000000 + 1023) >> 10) 349#define FUBd \
350 free (eio_buf)
351
352/*****************************************************************************/
353
354struct etp_tmpbuf;
355
356#if _POSIX_VERSION >= 200809L
357 #define HAVE_AT 1
358 #define WD2FD(wd) ((wd) ? (wd)->fd : AT_FDCWD)
359 #ifndef O_SEARCH
360 #define O_SEARCH O_RDONLY
361 #endif
362#else
363 #define HAVE_AT 0
364 static const char *wd_expand (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path);
365#endif
366
367struct eio_pwd
368{
369#if HAVE_AT
370 int fd;
371#endif
372 int len;
373 char str[1]; /* actually, a 0-terminated canonical path */
374};
122 375
123/*****************************************************************************/ 376/*****************************************************************************/
124 377
125#define ETP_PRI_MIN EIO_PRI_MIN 378#define ETP_PRI_MIN EIO_PRI_MIN
126#define ETP_PRI_MAX EIO_PRI_MAX 379#define ETP_PRI_MAX EIO_PRI_MAX
127 380
381#define ETP_TYPE_QUIT -1
382#define ETP_TYPE_GROUP EIO_GROUP
383
384static void eio_nop_callback (void) { }
385static void (*eio_want_poll_cb)(void) = eio_nop_callback;
386static void (*eio_done_poll_cb)(void) = eio_nop_callback;
387
388#define ETP_WANT_POLL(pool) eio_want_poll_cb ()
389#define ETP_DONE_POLL(pool) eio_done_poll_cb ()
390
128struct etp_worker; 391struct etp_worker;
129
130#define ETP_REQ eio_req 392#define ETP_REQ eio_req
131#define ETP_DESTROY(req) eio_destroy (req) 393#define ETP_DESTROY(req) eio_destroy (req)
132static int eio_finish (eio_req *req); 394static int eio_finish (eio_req *req);
133#define ETP_FINISH(req) eio_finish (req) 395#define ETP_FINISH(req) eio_finish (req)
134static void eio_execute (struct etp_worker *self, eio_req *req); 396static void eio_execute (struct etp_worker *self, eio_req *req);
135#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) 397#define ETP_EXECUTE(wrk,req) eio_execute (wrk, req)
136 398
137#define ETP_WORKER_CLEAR(req) \ 399#include "etp.c"
138 if (wrk->dbuf) \ 400
139 { \ 401static struct etp_pool eio_pool;
140 free (wrk->dbuf); \ 402#define EIO_POOL (&eio_pool)
141 wrk->dbuf = 0; \
142 } \
143 \
144 if (wrk->dirp) \
145 { \
146 closedir (wrk->dirp); \
147 wrk->dirp = 0; \
148 }
149#define ETP_WORKER_COMMON \
150 void *dbuf; \
151 DIR *dirp;
152 403
153/*****************************************************************************/ 404/*****************************************************************************/
154 405
155#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
156
157/* calculcate time difference in ~1/EIO_TICKS of a second */
158static int tvdiff (struct timeval *tv1, struct timeval *tv2)
159{
160 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
161 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
162}
163
164static unsigned int started, idle, wanted = 4;
165
166static void (*want_poll_cb) (void);
167static void (*done_poll_cb) (void);
168
169static unsigned int max_poll_time; /* reslock */
170static unsigned int max_poll_reqs; /* reslock */
171
172static volatile unsigned int nreqs; /* reqlock */
173static volatile unsigned int nready; /* reqlock */
174static volatile unsigned int npending; /* reqlock */
175static volatile unsigned int max_idle = 4;
176
177static mutex_t wrklock = X_MUTEX_INIT;
178static mutex_t reslock = X_MUTEX_INIT;
179static mutex_t reqlock = X_MUTEX_INIT;
180static cond_t reqwait = X_COND_INIT;
181
182typedef struct etp_worker
183{
184 /* locked by wrklock */
185 struct etp_worker *prev, *next;
186
187 thread_t tid;
188
189 /* locked by reslock, reqlock or wrklock */
190 ETP_REQ *req; /* currently processed request */
191
192 ETP_WORKER_COMMON
193} etp_worker;
194
195static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */
196
197#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
198#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
199
200/* worker threads management */
201
202static void etp_worker_clear (etp_worker *wrk)
203{
204 ETP_WORKER_CLEAR (wrk);
205}
206
207static void etp_worker_free (etp_worker *wrk)
208{
209 wrk->next->prev = wrk->prev;
210 wrk->prev->next = wrk->next;
211
212 free (wrk);
213}
214
215static unsigned int etp_nreqs (void)
216{
217 int retval;
218 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
219 retval = nreqs;
220 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
221 return retval;
222}
223
224static unsigned int etp_nready (void)
225{
226 unsigned int retval;
227
228 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
229 retval = nready;
230 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
231
232 return retval;
233}
234
235static unsigned int etp_npending (void)
236{
237 unsigned int retval;
238
239 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
240 retval = npending;
241 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
242
243 return retval;
244}
245
246static unsigned int etp_nthreads (void)
247{
248 unsigned int retval;
249
250 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
251 retval = started;
252 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
253
254 return retval;
255}
256
257/*
258 * a somewhat faster data structure might be nice, but
259 * with 8 priorities this actually needs <20 insns
260 * per shift, the most expensive operation.
261 */
262typedef struct {
263 ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */
264 int size;
265} etp_reqq;
266
267static etp_reqq req_queue;
268static etp_reqq res_queue;
269
270static int reqq_push (etp_reqq *q, ETP_REQ *req)
271{
272 int pri = req->pri;
273 req->next = 0;
274
275 if (q->qe[pri])
276 {
277 q->qe[pri]->next = req;
278 q->qe[pri] = req;
279 }
280 else
281 q->qe[pri] = q->qs[pri] = req;
282
283 return q->size++;
284}
285
286static ETP_REQ *reqq_shift (etp_reqq *q)
287{
288 int pri;
289
290 if (!q->size)
291 return 0;
292
293 --q->size;
294
295 for (pri = ETP_NUM_PRI; pri--; )
296 {
297 eio_req *req = q->qs[pri];
298
299 if (req)
300 {
301 if (!(q->qs[pri] = (eio_req *)req->next))
302 q->qe[pri] = 0;
303
304 return req;
305 }
306 }
307
308 abort ();
309}
310
311static void etp_atfork_prepare (void)
312{
313 X_LOCK (wrklock);
314 X_LOCK (reqlock);
315 X_LOCK (reslock);
316#if !HAVE_PREADWRITE
317 X_LOCK (preadwritelock);
318#endif
319}
320
321static void etp_atfork_parent (void)
322{
323#if !HAVE_PREADWRITE
324 X_UNLOCK (preadwritelock);
325#endif
326 X_UNLOCK (reslock);
327 X_UNLOCK (reqlock);
328 X_UNLOCK (wrklock);
329}
330
331static void etp_atfork_child (void)
332{
333 ETP_REQ *prv;
334
335 while (prv = reqq_shift (&req_queue))
336 ETP_DESTROY (prv);
337
338 while (prv = reqq_shift (&res_queue))
339 ETP_DESTROY (prv);
340
341 while (wrk_first.next != &wrk_first)
342 {
343 etp_worker *wrk = wrk_first.next;
344
345 if (wrk->req)
346 ETP_DESTROY (wrk->req);
347
348 etp_worker_clear (wrk);
349 etp_worker_free (wrk);
350 }
351
352 started = 0;
353 idle = 0;
354 nreqs = 0;
355 nready = 0;
356 npending = 0;
357
358 etp_atfork_parent ();
359}
360
361static void 406static void
362etp_once_init (void)
363{
364 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
365}
366
367static int
368etp_init (void (*want_poll)(void), void (*done_poll)(void))
369{
370 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
371
372 pthread_once (&doinit, etp_once_init);
373
374 want_poll_cb = want_poll;
375 done_poll_cb = done_poll;
376}
377
378X_THREAD_PROC (etp_proc);
379
380static void etp_start_thread (void)
381{
382 etp_worker *wrk = calloc (1, sizeof (etp_worker));
383
384 /*TODO*/
385 assert (("unable to allocate worker thread data", wrk));
386
387 X_LOCK (wrklock);
388
389 if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
390 {
391 wrk->prev = &wrk_first;
392 wrk->next = wrk_first.next;
393 wrk_first.next->prev = wrk;
394 wrk_first.next = wrk;
395 ++started;
396 }
397 else
398 free (wrk);
399
400 X_UNLOCK (wrklock);
401}
402
403static void etp_maybe_start_thread (void)
404{
405 if (etp_nthreads () >= wanted)
406 return;
407
408 /* todo: maybe use idle here, but might be less exact */
409 if (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ())
410 return;
411
412 etp_start_thread ();
413}
414
415static void etp_end_thread (void)
416{
417 eio_req *req = calloc (1, sizeof (eio_req));
418
419 req->type = -1;
420 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
421
422 X_LOCK (reqlock);
423 reqq_push (&req_queue, req);
424 X_COND_SIGNAL (reqwait);
425 X_UNLOCK (reqlock);
426
427 X_LOCK (wrklock);
428 --started;
429 X_UNLOCK (wrklock);
430}
431
432static int etp_poll (void)
433{
434 unsigned int maxreqs;
435 unsigned int maxtime;
436 struct timeval tv_start, tv_now;
437
438 X_LOCK (reslock);
439 maxreqs = max_poll_reqs;
440 maxtime = max_poll_time;
441 X_UNLOCK (reslock);
442
443 if (maxtime)
444 gettimeofday (&tv_start, 0);
445
446 for (;;)
447 {
448 ETP_REQ *req;
449
450 etp_maybe_start_thread ();
451
452 X_LOCK (reslock);
453 req = reqq_shift (&res_queue);
454
455 if (req)
456 {
457 --npending;
458
459 if (!res_queue.size && done_poll_cb)
460 done_poll_cb ();
461 }
462
463 X_UNLOCK (reslock);
464
465 if (!req)
466 return 0;
467
468 X_LOCK (reqlock);
469 --nreqs;
470 X_UNLOCK (reqlock);
471
472 if (req->type == EIO_GROUP && req->size)
473 {
474 req->int1 = 1; /* mark request as delayed */
475 continue;
476 }
477 else
478 {
479 int res = ETP_FINISH (req);
480 if (res)
481 return res;
482 }
483
484 if (maxreqs && !--maxreqs)
485 break;
486
487 if (maxtime)
488 {
489 gettimeofday (&tv_now, 0);
490
491 if (tvdiff (&tv_start, &tv_now) >= maxtime)
492 break;
493 }
494 }
495
496 errno = EAGAIN;
497 return -1;
498}
499
500static void etp_cancel (ETP_REQ *req)
501{
502 X_LOCK (wrklock);
503 req->flags |= EIO_FLAG_CANCELLED;
504 X_UNLOCK (wrklock);
505
506 eio_grp_cancel (req);
507}
508
509static void etp_submit (ETP_REQ *req)
510{
511 req->pri -= ETP_PRI_MIN;
512
513 if (req->pri < ETP_PRI_MIN - ETP_PRI_MIN) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
514 if (req->pri > ETP_PRI_MAX - ETP_PRI_MIN) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
515
516 X_LOCK (reqlock);
517 ++nreqs;
518 ++nready;
519 reqq_push (&req_queue, req);
520 X_COND_SIGNAL (reqwait);
521 X_UNLOCK (reqlock);
522
523 etp_maybe_start_thread ();
524}
525
526static void etp_set_max_poll_time (double nseconds)
527{
528 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
529 max_poll_time = nseconds;
530 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
531}
532
533static void etp_set_max_poll_reqs (unsigned int maxreqs)
534{
535 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
536 max_poll_reqs = maxreqs;
537 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
538}
539
540static void etp_set_max_idle (unsigned int nthreads)
541{
542 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
543 max_idle = nthreads <= 0 ? 1 : nthreads;
544 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
545}
546
547static void etp_set_min_parallel (unsigned int nthreads)
548{
549 if (wanted < nthreads)
550 wanted = nthreads;
551}
552
553static void etp_set_max_parallel (unsigned int nthreads)
554{
555 if (wanted > nthreads)
556 wanted = nthreads;
557
558 while (started > wanted)
559 etp_end_thread ();
560}
561
562/*****************************************************************************/
563
564static void grp_try_feed (eio_req *grp) 407grp_try_feed (eio_req *grp)
565{ 408{
566 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 409 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
567 { 410 {
568 int old_len = grp->size; 411 grp->flags &= ~ETP_FLAG_GROUPADD;
569 412
570 EIO_FEED (grp); 413 EIO_FEED (grp);
571 414
572 /* stop if no progress has been made */ 415 /* stop if no progress has been made */
573 if (old_len == grp->size) 416 if (!(grp->flags & ETP_FLAG_GROUPADD))
574 { 417 {
575 grp->feed = 0; 418 grp->feed = 0;
576 break; 419 break;
577 } 420 }
578 } 421 }
579} 422}
580 423
424static int
581static int grp_dec (eio_req *grp) 425grp_dec (eio_req *grp)
582{ 426{
583 --grp->size; 427 --grp->size;
584 428
585 /* call feeder, if applicable */ 429 /* call feeder, if applicable */
586 grp_try_feed (grp); 430 grp_try_feed (grp);
587 431
588 /* finish, if done */ 432 /* finish, if done */
589 if (!grp->size && grp->int1) 433 if (!grp->size && grp->flags & ETP_FLAG_DELAYED)
590 return eio_finish (grp); 434 return eio_finish (grp);
591 else 435 else
592 return 0; 436 return 0;
593} 437}
594 438
439static void
595void eio_destroy (eio_req *req) 440eio_destroy (eio_req *req)
596{ 441{
597 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); 442 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
598 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); 443 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
599 444
600 EIO_DESTROY (req); 445 EIO_DESTROY (req);
601} 446}
602 447
448static int
603static int eio_finish (eio_req *req) 449eio_finish (eio_req *req)
604{ 450{
605 int res = EIO_FINISH (req); 451 int res = EIO_FINISH (req);
606 452
607 if (req->grp) 453 if (req->grp)
608 { 454 {
… …
616 if (grp->grp_first == req) 462 if (grp->grp_first == req)
617 grp->grp_first = req->grp_next; 463 grp->grp_first = req->grp_next;
618 464
619 res2 = grp_dec (grp); 465 res2 = grp_dec (grp);
620 466
621 if (!res && res2) 467 if (!res)
622 res = res2; 468 res = res2;
623 } 469 }
624 470
625 eio_destroy (req); 471 eio_destroy (req);
626 472
627 return res; 473 return res;
628} 474}
629 475
476void
630void eio_grp_cancel (eio_req *grp) 477eio_grp_cancel (eio_req *grp)
631{ 478{
632 for (grp = grp->grp_first; grp; grp = grp->grp_next) 479 etp_grp_cancel (EIO_POOL, grp);
633 eio_cancel (grp);
634} 480}
635 481
482void
636void eio_cancel (eio_req *req) 483eio_cancel (eio_req *req)
637{ 484{
638 etp_cancel (req); 485 etp_cancel (EIO_POOL, req);
639} 486}
640 487
488void
641void eio_submit (eio_req *req) 489eio_submit (eio_req *req)
642{ 490{
643 etp_submit (req); 491 etp_submit (EIO_POOL, req);
644} 492}
645 493
646unsigned int eio_nreqs (void) 494unsigned int
495eio_nreqs (void)
647{ 496{
648 return etp_nreqs (); 497 return etp_nreqs (EIO_POOL);
649} 498}
650 499
651unsigned int eio_nready (void) 500unsigned int
501eio_nready (void)
652{ 502{
653 return etp_nready (); 503 return etp_nready (EIO_POOL);
654} 504}
655 505
656unsigned int eio_npending (void) 506unsigned int
507eio_npending (void)
657{ 508{
658 return etp_npending (); 509 return etp_npending (EIO_POOL);
659} 510}
660 511
661unsigned int eio_nthreads (void) 512unsigned int ecb_cold
513eio_nthreads (void)
662{ 514{
663 return etp_nthreads (); 515 return etp_nthreads (EIO_POOL);
664} 516}
665 517
518void ecb_cold
666void eio_set_max_poll_time (double nseconds) 519eio_set_max_poll_time (double nseconds)
667{ 520{
668 etp_set_max_poll_time (nseconds); 521 etp_set_max_poll_time (EIO_POOL, nseconds);
669} 522}
670 523
524void ecb_cold
671void eio_set_max_poll_reqs (unsigned int maxreqs) 525eio_set_max_poll_reqs (unsigned int maxreqs)
672{ 526{
673 etp_set_max_poll_reqs (maxreqs); 527 etp_set_max_poll_reqs (EIO_POOL, maxreqs);
674} 528}
675 529
530void ecb_cold
676void eio_set_max_idle (unsigned int nthreads) 531eio_set_max_idle (unsigned int nthreads)
677{ 532{
678 etp_set_max_idle (nthreads); 533 etp_set_max_idle (EIO_POOL, nthreads);
679} 534}
680 535
536void ecb_cold
537eio_set_idle_timeout (unsigned int seconds)
538{
539 etp_set_idle_timeout (EIO_POOL, seconds);
540}
541
542void ecb_cold
681void eio_set_min_parallel (unsigned int nthreads) 543eio_set_min_parallel (unsigned int nthreads)
682{ 544{
683 etp_set_min_parallel (nthreads); 545 etp_set_min_parallel (EIO_POOL, nthreads);
684} 546}
685 547
548void ecb_cold
686void eio_set_max_parallel (unsigned int nthreads) 549eio_set_max_parallel (unsigned int nthreads)
687{ 550{
688 etp_set_max_parallel (nthreads); 551 etp_set_max_parallel (EIO_POOL, nthreads);
689} 552}
690 553
691int eio_poll (void) 554int eio_poll (void)
692{ 555{
693 return etp_poll (); 556 return etp_poll (EIO_POOL);
694} 557}
695 558
696/*****************************************************************************/ 559/*****************************************************************************/
697/* work around various missing functions */ 560/* work around various missing functions */
698 561
699#if !HAVE_PREADWRITE 562#if HAVE_POSIX_CLOSE && !__linux
700# define pread eio__pread 563# define eio__close(fd) posix_close (fd, 0)
701# define pwrite eio__pwrite 564#else
702 565# define eio__close(fd) close (fd)
703/*
704 * make our pread/pwrite safe against themselves, but not against
705 * normal read/write by using a mutex. slows down execution a lot,
706 * but that's your problem, not mine.
707 */
708static mutex_t preadwritelock = X_MUTEX_INIT;
709
710static ssize_t
711eio__pread (int fd, void *buf, size_t count, off_t offset)
712{
713 ssize_t res;
714 off_t ooffset;
715
716 X_LOCK (preadwritelock);
717 ooffset = lseek (fd, 0, SEEK_CUR);
718 lseek (fd, offset, SEEK_SET);
719 res = read (fd, buf, count);
720 lseek (fd, ooffset, SEEK_SET);
721 X_UNLOCK (preadwritelock);
722
723 return res;
724}
725
726static ssize_t
727eio__pwrite (int fd, void *buf, size_t count, off_t offset)
728{
729 ssize_t res;
730 off_t ooffset;
731
732 X_LOCK (preadwritelock);
733 ooffset = lseek (fd, 0, SEEK_CUR);
734 lseek (fd, offset, SEEK_SET);
735 res = write (fd, buf, count);
736 lseek (fd, offset, SEEK_SET);
737 X_UNLOCK (preadwritelock);
738
739 return res;
740}
741#endif 566#endif
742 567
568/* close() without disturbing errno */
569static void
570silent_close (int fd)
571{
572 int saved_errno = errno;
573 eio__close (fd);
574 errno = saved_errno;
575}
576
743#ifndef HAVE_FUTIMES 577#ifndef HAVE_UTIMES
744 578
579# undef utimes
745# define utimes(path,times) eio__utimes (path, times) 580# define utimes(path,times) eio__utimes (path, times)
746# define futimes(fd,times) eio__futimes (fd, times)
747 581
748static int 582static int
749eio__utimes (const char *filename, const struct timeval times[2]) 583eio__utimes (const char *filename, const struct timeval times[2])
750{ 584{
751 if (times) 585 if (times)
… …
759 } 593 }
760 else 594 else
761 return utime (filename, 0); 595 return utime (filename, 0);
762} 596}
763 597
598#endif
599
600#ifndef HAVE_FUTIMES
601
602# undef futimes
603# define futimes(fd,times) eio__futimes (fd, times)
604
605static int
764static int eio__futimes (int fd, const struct timeval tv[2]) 606eio__futimes (int fd, const struct timeval tv[2])
765{ 607{
766 errno = ENOSYS; 608 errno = ENOSYS;
767 return -1; 609 return -1;
768} 610}
769 611
770#endif 612#endif
771 613
772#if !HAVE_FDATASYNC 614#if !HAVE_FDATASYNC
615# undef fdatasync
773# define fdatasync fsync 616# define fdatasync(fd) fsync (fd)
774#endif 617#endif
618
619static int
620eio__syncfs (int fd)
621{
622 int res;
623
624#if HAVE_SYS_SYNCFS
625 res = (int)syscall (__NR_syncfs, (int)(fd));
626#else
627 res = EIO_ENOSYS ();
628#endif
629
630 if (res < 0 && errno == ENOSYS && fd >= 0)
631 sync ();
632
633 return res;
634}
635
636/* sync_file_range always needs emulation */
637static int
638eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
639{
640#if HAVE_SYNC_FILE_RANGE
641 int res;
642
643 if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
644 || EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
645 || EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
646 {
647 flags = 0
648 | (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
649 | (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
650 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
651 }
652
653 res = sync_file_range (fd, offset, nbytes, flags);
654
655 if (!res || errno != ENOSYS)
656 return res;
657#endif
658
659 /* even though we could play tricks with the flags, it's better to always
660 * call fdatasync, as that matches the expectation of its users best */
661 return fdatasync (fd);
662}
663
664static int
665eio__fallocate (int fd, int mode, off_t offset, size_t len)
666{
667#if HAVE_LINUX_FALLOCATE
668 return fallocate (fd, mode, offset, len);
669#else
670 return EIO_ENOSYS ();
671#endif
672}
775 673
776#if !HAVE_READAHEAD 674#if !HAVE_READAHEAD
675# undef readahead
777# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 676# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
778 677
779static ssize_t 678static eio_ssize_t
780eio__readahead (int fd, off_t offset, size_t count, worker *self) 679eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
781{ 680{
782 size_t todo = count; 681 size_t todo = count;
783 dBUF; 682 dBUF;
784 683
785 while (todo > 0) 684 while (todo > 0)
… …
789 pread (fd, eio_buf, len, offset); 688 pread (fd, eio_buf, len, offset);
790 offset += len; 689 offset += len;
791 todo -= len; 690 todo -= len;
792 } 691 }
793 692
794 errno = 0; 693 FUBd;
694
695 /* linux's readahead basically only fails for EBADF or EINVAL (not mmappable) */
696 /* but not for e.g. EIO or eof, so we also never fail */
795 return count; 697 return 0;
796} 698}
797 699
798#endif 700#endif
799 701
800/* sendfile always needs emulation */ 702/* sendfile always needs emulation */
801static ssize_t 703static eio_ssize_t
802eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) 704eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
803{ 705{
706 eio_ssize_t written = 0;
804 ssize_t res; 707 eio_ssize_t res;
805 708
806 if (!count) 709 if (!count)
807 return 0; 710 return 0;
808 711
712 for (;;)
713 {
714#ifdef __APPLE__
715# undef HAVE_SENDFILE /* broken, as everything on os x */
716#endif
809#if HAVE_SENDFILE 717#if HAVE_SENDFILE
810# if __linux 718# if __linux
719 off_t soffset = offset;
811 res = sendfile (ofd, ifd, &offset, count); 720 res = sendfile (ofd, ifd, &soffset, count);
812 721
813# elif __freebsd 722# elif __FreeBSD__
814 /* 723 /*
815 * Of course, the freebsd sendfile is a dire hack with no thoughts 724 * Of course, the freebsd sendfile is a dire hack with no thoughts
816 * wasted on making it similar to other I/O functions. 725 * wasted on making it similar to other I/O functions.
817 */ 726 */
818 {
819 off_t sbytes; 727 off_t sbytes;
820 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 728 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
821 729
822 if (res < 0 && sbytes) 730 #if 0 /* according to the manpage, this is correct, but broken behaviour */
823 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 731 /* freebsd' sendfile will return 0 on success */
732 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
733 /* not on e.g. EIO or EPIPE - sounds broken */
734 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
824 res = sbytes; 735 res = sbytes;
825 } 736 #endif
737
738 /* according to source inspection, this is correct, and useful behaviour */
739 if (sbytes)
740 res = sbytes;
741
742# elif defined __APPLE__
743 off_t sbytes = count;
744 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
745
746 /* according to the manpage, sbytes is always valid */
747 if (sbytes)
748 res = sbytes;
826 749
827# elif __hpux 750# elif __hpux
828 res = sendfile (ofd, ifd, offset, count, 0, 0); 751 res = sendfile (ofd, ifd, offset, count, 0, 0);
829 752
830# elif __solaris 753# elif __solaris
831 {
832 struct sendfilevec vec; 754 struct sendfilevec vec;
833 size_t sbytes; 755 size_t sbytes;
834 756
835 vec.sfv_fd = ifd; 757 vec.sfv_fd = ifd;
836 vec.sfv_flag = 0; 758 vec.sfv_flag = 0;
837 vec.sfv_off = offset; 759 vec.sfv_off = offset;
838 vec.sfv_len = count; 760 vec.sfv_len = count;
839 761
840 res = sendfilev (ofd, &vec, 1, &sbytes); 762 res = sendfilev (ofd, &vec, 1, &sbytes);
841 763
842 if (res < 0 && sbytes) 764 if (res < 0 && sbytes)
843 res = sbytes; 765 res = sbytes;
844 }
845 766
846# endif 767# endif
768
769#elif defined (_WIN32) && 0
770 /* does not work, just for documentation of what would need to be done */
771 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
772 /* libeio guarantees that the file offset does not change, and windows */
773 /* has no way to get an independent handle to the same file description */
774 HANDLE h = TO_SOCKET (ifd);
775 SetFilePointer (h, offset, 0, FILE_BEGIN);
776 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
777
847#else 778#else
848 res = -1; 779 res = EIO_ENOSYS ();
849 errno = ENOSYS;
850#endif 780#endif
851 781
782 /* we assume sendfile can copy at least 128mb in one go */
783 if (res <= 128 * 1024 * 1024)
784 {
785 if (res > 0)
786 written += res;
787
788 if (written)
789 return written;
790
791 break;
792 }
793 else
794 {
795 /* if we requested more, then probably the kernel was lazy */
796 written += res;
797 offset += res;
798 count -= res;
799
800 if (!count)
801 return written;
802 }
803 }
804
852 if (res < 0 805 if (res < 0
853 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 806 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
807 /* BSDs */
808#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
809 || errno == ENOTSUP
810#endif
811#ifdef EOPNOTSUPP /* windows */
812 || errno == EOPNOTSUPP /* BSDs */
813#endif
854#if __solaris 814#if __solaris
855 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 815 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
856#endif 816#endif
857 ) 817 )
858 ) 818 )
… …
862 822
863 res = 0; 823 res = 0;
864 824
865 while (count) 825 while (count)
866 { 826 {
867 ssize_t cnt; 827 eio_ssize_t cnt;
868 828
869 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 829 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
870 830
871 if (cnt <= 0) 831 if (cnt <= 0)
872 { 832 {
… …
884 844
885 offset += cnt; 845 offset += cnt;
886 res += cnt; 846 res += cnt;
887 count -= cnt; 847 count -= cnt;
888 } 848 }
849
850 FUBd;
889 } 851 }
890 852
891 return res; 853 return res;
854}
855
856#ifdef PAGESIZE
857# define eio_pagesize() PAGESIZE
858#else
859static intptr_t
860eio_pagesize (void)
861{
862 static intptr_t page;
863
864 if (!page)
865 page = sysconf (_SC_PAGESIZE);
866
867 return page;
868}
869#endif
870
871static void
872eio_page_align (void **addr, size_t *length)
873{
874 intptr_t mask = eio_pagesize () - 1;
875
876 /* round down addr */
877 intptr_t adj = mask & (intptr_t)*addr;
878
879 *addr = (void *)((intptr_t)*addr - adj);
880 *length += adj;
881
882 /* round up length */
883 *length = (*length + mask) & ~mask;
884}
885
886#if !_POSIX_MEMLOCK
887# define eio__mlockall(a) EIO_ENOSYS ()
888#else
889
890static int
891eio__mlockall (int flags)
892{
893 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
894 extern int mallopt (int, int);
895 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
896 #endif
897
898 if (EIO_MCL_CURRENT != MCL_CURRENT
899 || EIO_MCL_FUTURE != MCL_FUTURE)
900 {
901 flags = 0
902 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
903 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
904 }
905
906 return mlockall (flags);
907}
908#endif
909
910#if !_POSIX_MEMLOCK_RANGE
911# define eio__mlock(a,b) EIO_ENOSYS ()
912#else
913
914static int
915eio__mlock (void *addr, size_t length)
916{
917 eio_page_align (&addr, &length);
918
919 return mlock (addr, length);
920}
921
922#endif
923
924#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
925# define eio__msync(a,b,c) EIO_ENOSYS ()
926#else
927
928static int
929eio__msync (void *mem, size_t len, int flags)
930{
931 eio_page_align (&mem, &len);
932
933 if (EIO_MS_ASYNC != MS_SYNC
934 || EIO_MS_INVALIDATE != MS_INVALIDATE
935 || EIO_MS_SYNC != MS_SYNC)
936 {
937 flags = 0
938 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
939 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
940 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
941 }
942
943 return msync (mem, len, flags);
944}
945
946#endif
947
948static int
949eio__mtouch (eio_req *req)
950{
951 void *mem = req->ptr2;
952 size_t len = req->size;
953 int flags = req->int1;
954
955 eio_page_align (&mem, &len);
956
957 {
958 intptr_t addr = (intptr_t)mem;
959 intptr_t end = addr + len;
960 intptr_t page = eio_pagesize ();
961
962 if (addr < end)
963 if (flags & EIO_MT_MODIFY) /* modify */
964 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < end && !EIO_CANCELLED (req));
965 else
966 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < end && !EIO_CANCELLED (req));
967 }
968
969 return 0;
970}
971
972/*****************************************************************************/
973/* requests implemented outside eio_execute, because they are so large */
974
975static void
976eio__lseek (eio_req *req)
977{
978 /* this usually gets optimised away completely, or your compiler sucks, */
979 /* or the whence constants really are not 0, 1, 2 */
980 int whence = req->int2 == EIO_SEEK_SET ? SEEK_SET
981 : req->int2 == EIO_SEEK_CUR ? SEEK_CUR
982 : req->int2 == EIO_SEEK_END ? SEEK_END
983 : req->int2;
984
985 req->offs = lseek (req->int1, req->offs, whence);
986 req->result = req->offs == (off_t)-1 ? -1 : 0;
987}
988
989/* result will always end up in tmpbuf, there is always space for adding a 0-byte */
990static int
991eio__realpath (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
992{
993 char *res;
994 const char *rel = path;
995 char *tmp1, *tmp2;
996#if SYMLOOP_MAX > 32
997 int symlinks = SYMLOOP_MAX;
998#else
999 int symlinks = 32;
1000#endif
1001
1002 errno = EINVAL;
1003 if (!rel)
1004 return -1;
1005
1006 errno = ENOENT;
1007 if (!*rel)
1008 return -1;
1009
1010 res = etp_tmpbuf_get (tmpbuf, EIO_PATH_MAX * 3);
1011#ifdef _WIN32
1012 if (_access (rel, 4) != 0)
1013 return -1;
1014
1015 symlinks = GetFullPathName (rel, EIO_PATH_MAX * 3, res, 0);
1016
1017 errno = ENAMETOOLONG;
1018 if (symlinks >= EIO_PATH_MAX * 3)
1019 return -1;
1020
1021 errno = EIO;
1022 if (symlinks <= 0)
1023 return -1;
1024
1025 return symlinks;
1026
1027#else
1028 tmp1 = res + EIO_PATH_MAX;
1029 tmp2 = tmp1 + EIO_PATH_MAX;
1030
1031#if 0 /* disabled, the musl way to do things is just too racy */
1032#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1033 /* on linux we may be able to ask the kernel */
1034 {
1035 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1036
1037 if (fd >= 0)
1038 {
1039 sprintf (tmp1, "/proc/self/fd/%d", fd);
1040 req->result = readlink (tmp1, res, EIO_PATH_MAX);
1041 /* here we should probably stat the open file and the disk file, to make sure they still match */
1042 eio__close (fd);
1043
1044 if (req->result > 0)
1045 goto done;
1046 }
1047 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1048 return -1;
1049 }
1050#endif
1051#endif
1052
1053 if (*rel != '/')
1054 {
1055 int len;
1056
1057 errno = ENOENT;
1058 if (wd == EIO_INVALID_WD)
1059 return -1;
1060
1061 if (wd == EIO_CWD)
1062 {
1063 if (!getcwd (res, EIO_PATH_MAX))
1064 return -1;
1065
1066 len = strlen (res);
1067 }
1068 else
1069 memcpy (res, wd->str, len = wd->len);
1070
1071 if (res [1]) /* only use if not / */
1072 res += len;
1073 }
1074
1075 while (*rel)
1076 {
1077 eio_ssize_t len, linklen;
1078 const char *beg = rel;
1079
1080 while (*rel && *rel != '/')
1081 ++rel;
1082
1083 len = rel - beg;
1084
1085 if (!len) /* skip slashes */
1086 {
1087 ++rel;
1088 continue;
1089 }
1090
1091 if (beg [0] == '.')
1092 {
1093 if (len == 1)
1094 continue; /* . - nop */
1095
1096 if (beg [1] == '.' && len == 2)
1097 {
1098 /* .. - back up one component, if possible */
1099
1100 while (res != tmpbuf->ptr)
1101 if (*--res == '/')
1102 break;
1103
1104 continue;
1105 }
1106 }
1107
1108 errno = ENAMETOOLONG;
1109 if (res + 1 + len + 1 >= tmp1)
1110 return -1;
1111
1112 /* copy one component */
1113 *res = '/';
1114 memcpy (res + 1, beg, len);
1115
1116 /* zero-terminate, for readlink */
1117 res [len + 1] = 0;
1118
1119 /* now check if it's a symlink */
1120 linklen = readlink (tmpbuf->ptr, tmp1, EIO_PATH_MAX);
1121
1122 if (linklen < 0)
1123 {
1124 if (errno != EINVAL)
1125 return -1;
1126
1127 /* it's a normal directory. hopefully */
1128 res += len + 1;
1129 }
1130 else
1131 {
1132 /* yay, it was a symlink - build new path in tmp2 */
1133 int rellen = strlen (rel);
1134
1135 errno = ENAMETOOLONG;
1136 if (linklen + 1 + rellen >= EIO_PATH_MAX) /* also catch linklen >= EIO_PATH_MAX */
1137 return -1;
1138
1139 errno = ELOOP;
1140 if (!--symlinks)
1141 return -1;
1142
1143 if (*tmp1 == '/')
1144 res = tmpbuf->ptr; /* symlink resolves to an absolute path */
1145
1146 /* we need to be careful, as rel might point into tmp2 already */
1147 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1148 tmp2 [linklen] = '/';
1149 memcpy (tmp2, tmp1, linklen);
1150
1151 rel = tmp2;
1152 }
1153 }
1154
1155 /* special case for the lone root path */
1156 if (res == tmpbuf->ptr)
1157 *res++ = '/';
1158
1159 return res - (char *)tmpbuf->ptr;
1160#endif
1161}
1162
1163static signed char
1164eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1165{
1166 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1167 : a->inode < b->inode ? -1
1168 : a->inode > b->inode ? 1
1169 : 0;
1170}
1171
1172#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1173
1174#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1175#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1176
1177static void
1178eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1179{
1180 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1181 unsigned char *bit = bits;
1182
1183 assert (CHAR_BIT == 8);
1184 assert (sizeof (eio_dirent) * 8 < 256);
1185 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1186 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1187
1188 if (size <= EIO_SORT_FAST)
1189 return;
1190
1191 /* first prepare an array of bits to test in our radix sort */
1192 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1193 /* inode_bits must contain all inodes ORed together */
1194 /* which is used to skip bits that are 0 everywhere, which is very common */
1195 {
1196 eio_ino_t endianness;
1197 int i, j;
1198
1199 /* we store the byte offset of byte n into byte n of "endianness" */
1200 for (i = 0; i < sizeof (eio_ino_t); ++i)
1201 ((unsigned char *)&endianness)[i] = i;
1202
1203 *bit++ = 0;
1204
1205 for (i = 0; i < sizeof (eio_ino_t); ++i)
1206 {
1207 /* shifting off the byte offsets out of "endianness" */
1208 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1209 endianness >>= 8;
1210
1211 for (j = 0; j < 8; ++j)
1212 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1213 *bit++ = offs + j;
1214 }
1215
1216 for (j = 0; j < 8; ++j)
1217 if (score_bits & (1 << j))
1218 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1219 }
1220
1221 /* now actually do the sorting (a variant of MSD radix sort) */
1222 {
1223 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1224 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1225 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1226 int stk_idx = 0;
1227
1228 base_stk [stk_idx] = dents;
1229 end_stk [stk_idx] = dents + size;
1230 bit_stk [stk_idx] = bit - 1;
1231
1232 do
1233 {
1234 base = base_stk [stk_idx];
1235 end = end_stk [stk_idx];
1236 bit = bit_stk [stk_idx];
1237
1238 for (;;)
1239 {
1240 unsigned char O = *bit >> 3;
1241 unsigned char M = 1 << (*bit & 7);
1242
1243 eio_dirent *a = base;
1244 eio_dirent *b = end;
1245
1246 if (b - a < EIO_SORT_CUTOFF)
1247 break;
1248
1249 /* now bit-partition the array on the bit */
1250 /* this ugly asymmetric loop seems to perform much better than typical */
1251 /* partition algos found in the literature */
1252 do
1253 if (!(((unsigned char *)a)[O] & M))
1254 ++a;
1255 else if (!(((unsigned char *)--b)[O] & M))
1256 {
1257 eio_dirent tmp = *a; *a = *b; *b = tmp;
1258 ++a;
1259 }
1260 while (b > a);
1261
1262 /* next bit, or stop, if no bits left in this path */
1263 if (!*--bit)
1264 break;
1265
1266 base_stk [stk_idx] = a;
1267 end_stk [stk_idx] = end;
1268 bit_stk [stk_idx] = bit;
1269 ++stk_idx;
1270
1271 end = a;
1272 }
1273 }
1274 while (stk_idx--);
1275 }
1276}
1277
1278static void
1279eio_dent_insertion_sort (eio_dirent *dents, int size)
1280{
1281 /* first move the smallest element to the front, to act as a sentinel */
1282 {
1283 int i;
1284 eio_dirent *min = dents;
1285
1286 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1287 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1288 if (EIO_DENT_CMP (dents [i], <, *min))
1289 min = &dents [i];
1290
1291 /* swap elements 0 and j (minimum) */
1292 {
1293 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1294 }
1295 }
1296
1297 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1298 {
1299 eio_dirent *i, *j;
1300
1301 for (i = dents + 1; i < dents + size; ++i)
1302 {
1303 eio_dirent value = *i;
1304
1305 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1306 j [1] = j [0];
1307
1308 j [1] = value;
1309 }
1310 }
1311}
1312
1313static void
1314eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1315{
1316 if (size <= 1)
1317 return; /* our insertion sort relies on size > 0 */
1318
1319 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1320 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1321 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1322
1323 /* use an insertion sort at the end, or for small arrays, */
1324 /* as insertion sort is more efficient for small partitions */
1325 eio_dent_insertion_sort (dents, size);
892} 1326}
893 1327
894/* read a full directory */ 1328/* read a full directory */
895static void 1329static void
896eio__scandir (eio_req *req, etp_worker *self) 1330eio__scandir (eio_req *req, etp_worker *self)
897{ 1331{
1332 char *name, *names;
1333 int namesalloc = 4096 - sizeof (void *) * 4;
1334 int namesoffs = 0;
1335 int flags = req->int1;
1336 eio_dirent *dents = 0;
1337 int dentalloc = 128;
1338 int dentoffs = 0;
1339 eio_ino_t inode_bits = 0;
1340#ifdef _WIN32
1341 HANDLE dirp;
1342 WIN32_FIND_DATA entp;
1343#else
898 DIR *dirp; 1344 DIR *dirp;
899 EIO_STRUCT_DIRENT *entp; 1345 EIO_STRUCT_DIRENT *entp;
900 char *name, *names; 1346#endif
901 int memlen = 4096;
902 int memofs = 0;
903 int res = 0;
904 1347
905 X_LOCK (wrklock); 1348 req->result = -1;
906 self->dirp = dirp = opendir (req->ptr1);
907 req->flags |= EIO_FLAG_PTR2_FREE;
908 req->ptr2 = names = malloc (memlen);
909 X_UNLOCK (wrklock);
910 1349
911 if (dirp && names) 1350 if (!(flags & EIO_READDIR_DENTS))
912 for (;;) 1351 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1352
1353#ifdef _WIN32
1354 {
1355 int len = strlen ((const char *)req->ptr1);
1356 char *path = malloc (MAX_PATH);
1357 const char *fmt;
1358 const char *reqpath = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1359
1360 if (!len)
1361 fmt = "./*";
1362 else if (reqpath[len - 1] == '/' || reqpath[len - 1] == '\\')
1363 fmt = "%s*";
1364 else
1365 fmt = "%s/*";
1366
1367 _snprintf (path, MAX_PATH, fmt, reqpath);
1368 dirp = FindFirstFile (path, &entp);
1369 free (path);
1370
1371 if (dirp == INVALID_HANDLE_VALUE)
1372 {
1373 /* should steal _dosmaperr */
1374 switch (GetLastError ())
1375 {
1376 case ERROR_FILE_NOT_FOUND:
1377 req->result = 0;
1378 break;
1379
1380 case ERROR_INVALID_NAME:
1381 case ERROR_PATH_NOT_FOUND:
1382 case ERROR_NO_MORE_FILES:
1383 errno = ENOENT;
1384 break;
1385
1386 case ERROR_NOT_ENOUGH_MEMORY:
1387 errno = ENOMEM;
1388 break;
1389
1390 default:
1391 errno = EINVAL;
1392 break;
1393 }
1394
1395 return;
1396 }
1397 }
1398#else
1399 #if HAVE_AT
1400 if (req->wd)
913 { 1401 {
1402 int fd = openat (WD2FD (req->wd), req->ptr1, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1403
1404 if (fd < 0)
1405 return;
1406
1407 dirp = fdopendir (fd);
1408
1409 if (!dirp)
1410 silent_close (fd);
1411 }
1412 else
1413 dirp = opendir (req->ptr1);
1414 #else
1415 dirp = opendir (wd_expand (&self->tmpbuf, req->wd, req->ptr1));
1416 #endif
1417
1418 if (!dirp)
1419 return;
1420#endif
1421
1422 if (req->flags & EIO_FLAG_PTR1_FREE)
1423 free (req->ptr1);
1424
1425 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1426 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1427 req->ptr2 = names = malloc (namesalloc);
1428
1429 if (!names || (flags && !dents))
1430 return;
1431
1432 for (;;)
1433 {
1434 int done;
1435
1436#ifdef _WIN32
1437 done = !dirp;
1438#else
914 errno = 0; 1439 errno = 0;
915 entp = readdir (dirp); 1440 entp = readdir (dirp);
1441 done = !entp;
1442#endif
916 1443
1444 if (done)
1445 {
1446#ifndef _WIN32
1447 int old_errno = errno;
1448 closedir (dirp);
1449 errno = old_errno;
1450
917 if (!entp) 1451 if (errno)
1452 break;
1453#endif
1454
1455 /* sort etc. */
1456 req->int1 = flags;
1457 req->result = dentoffs;
1458
1459 if (flags & EIO_READDIR_STAT_ORDER)
1460 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1461 else if (flags & EIO_READDIR_DIRS_FIRST)
1462 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1463 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1464 else
1465 {
1466 /* in this case, all is known, and we just put dirs first and sort them */
1467 eio_dirent *oth = dents + dentoffs;
1468 eio_dirent *dir = dents;
1469
1470 /* now partition dirs to the front, and non-dirs to the back */
1471 /* by walking from both sides and swapping if necessary */
1472 while (oth > dir)
1473 {
1474 if (dir->type == EIO_DT_DIR)
1475 ++dir;
1476 else if ((--oth)->type == EIO_DT_DIR)
1477 {
1478 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1479
1480 ++dir;
1481 }
1482 }
1483
1484 /* now sort the dirs only (dirs all have the same score) */
1485 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1486 }
1487
918 break; 1488 break;
1489 }
919 1490
920 name = entp->d_name; 1491 /* now add the entry to our list(s) */
1492 name = D_NAME (entp);
921 1493
1494 /* skip . and .. entries */
922 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1495 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
923 { 1496 {
924 int len = strlen (name) + 1; 1497 int len = D_NAMLEN (entp) + 1;
925 1498
926 res++; 1499 while (ecb_expect_false (namesoffs + len > namesalloc))
927
928 while (memofs + len > memlen)
929 { 1500 {
930 memlen *= 2; 1501 namesalloc *= 2;
931 X_LOCK (wrklock);
932 req->ptr2 = names = realloc (names, memlen); 1502 req->ptr2 = names = realloc (names, namesalloc);
933 X_UNLOCK (wrklock);
934 1503
935 if (!names) 1504 if (!names)
1505 break;
1506 }
1507
1508 memcpy (names + namesoffs, name, len);
1509
1510 if (dents)
1511 {
1512 struct eio_dirent *ent;
1513
1514 if (ecb_expect_false (dentoffs == dentalloc))
1515 {
1516 dentalloc *= 2;
1517 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1518
1519 if (!dents)
936 break; 1520 break;
937 } 1521 }
938 1522
939 memcpy (names + memofs, name, len); 1523 ent = dents + dentoffs;
940 memofs += len; 1524
1525 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1526 ent->namelen = len - 1;
1527 ent->inode = D_INO (entp);
1528
1529 inode_bits |= ent->inode;
1530
1531 switch (D_TYPE (entp))
1532 {
1533 default:
1534 ent->type = EIO_DT_UNKNOWN;
1535 flags |= EIO_READDIR_FOUND_UNKNOWN;
1536 break;
1537
1538 #ifdef DT_FIFO
1539 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1540 #endif
1541 #ifdef DT_CHR
1542 case DT_CHR: ent->type = EIO_DT_CHR; break;
1543 #endif
1544 #ifdef DT_MPC
1545 case DT_MPC: ent->type = EIO_DT_MPC; break;
1546 #endif
1547 #ifdef DT_DIR
1548 case DT_DIR: ent->type = EIO_DT_DIR; break;
1549 #endif
1550 #ifdef DT_NAM
1551 case DT_NAM: ent->type = EIO_DT_NAM; break;
1552 #endif
1553 #ifdef DT_BLK
1554 case DT_BLK: ent->type = EIO_DT_BLK; break;
1555 #endif
1556 #ifdef DT_MPB
1557 case DT_MPB: ent->type = EIO_DT_MPB; break;
1558 #endif
1559 #ifdef DT_REG
1560 case DT_REG: ent->type = EIO_DT_REG; break;
1561 #endif
1562 #ifdef DT_NWK
1563 case DT_NWK: ent->type = EIO_DT_NWK; break;
1564 #endif
1565 #ifdef DT_CMP
1566 case DT_CMP: ent->type = EIO_DT_CMP; break;
1567 #endif
1568 #ifdef DT_LNK
1569 case DT_LNK: ent->type = EIO_DT_LNK; break;
1570 #endif
1571 #ifdef DT_SOCK
1572 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1573 #endif
1574 #ifdef DT_DOOR
1575 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1576 #endif
1577 #ifdef DT_WHT
1578 case DT_WHT: ent->type = EIO_DT_WHT; break;
1579 #endif
1580 }
1581
1582 ent->score = 7;
1583
1584 if (flags & EIO_READDIR_DIRS_FIRST)
1585 {
1586 if (ent->type == EIO_DT_UNKNOWN)
1587 {
1588 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1589 ent->score = 1;
1590 else if (!strchr (name, '.')) /* absence of dots indicate likely dirs */
1591 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1592 }
1593 else if (ent->type == EIO_DT_DIR)
1594 ent->score = 0;
1595 }
941 } 1596 }
1597
1598 namesoffs += len;
1599 ++dentoffs;
942 } 1600 }
943 1601
944 if (errno) 1602 if (EIO_CANCELLED (req))
945 res = -1; 1603 {
1604 errno = ECANCELED;
1605 break;
1606 }
1607
1608#ifdef _WIN32
1609 if (!FindNextFile (dirp, &entp))
1610 {
1611 FindClose (dirp);
1612 dirp = 0;
1613 }
1614#endif
1615 }
1616}
1617
1618/*****************************************************************************/
1619/* working directory stuff */
1620/* various deficiencies in the posix 2008 api force us to */
1621/* keep the absolute path in string form at all times */
1622/* fuck yeah. */
1623
1624#if !HAVE_AT
1625
1626/* a bit like realpath, but usually faster because it doesn'T have to return */
1627/* an absolute or canonical path */
1628static const char *
1629wd_expand (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
1630{
1631 if (!wd || *path == '/')
1632 return path;
1633
1634 if (path [0] == '.' && !path [1])
1635 return wd->str;
1636
946 1637 {
947 req->result = res; 1638 int l1 = wd->len;
1639 int l2 = strlen (path);
1640
1641 char *res = etp_tmpbuf_get (tmpbuf, l1 + l2 + 2);
1642
1643 memcpy (res, wd->str, l1);
1644 res [l1] = '/';
1645 memcpy (res + l1 + 1, path, l2 + 1);
1646
1647 return res;
1648 }
948} 1649}
1650
1651#endif
1652
1653static eio_wd
1654eio__wd_open_sync (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
1655{
1656 int fd;
1657 eio_wd res;
1658 int len = eio__realpath (tmpbuf, wd, path);
1659
1660 if (len < 0)
1661 return EIO_INVALID_WD;
1662
1663#if HAVE_AT
1664 fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1665
1666 if (fd < 0)
1667 return EIO_INVALID_WD;
1668#endif
1669
1670 res = malloc (sizeof (*res) + len); /* one extra 0-byte */
1671
1672#if HAVE_AT
1673 res->fd = fd;
1674#endif
1675
1676 res->len = len;
1677 memcpy (res->str, tmpbuf->ptr, len);
1678 res->str [len] = 0;
1679
1680 return res;
1681}
1682
1683eio_wd
1684eio_wd_open_sync (eio_wd wd, const char *path)
1685{
1686 struct etp_tmpbuf tmpbuf = { };
1687 wd = eio__wd_open_sync (&tmpbuf, wd, path);
1688 free (tmpbuf.ptr);
1689
1690 return wd;
1691}
1692
1693void
1694eio_wd_close_sync (eio_wd wd)
1695{
1696 if (wd != EIO_INVALID_WD && wd != EIO_CWD)
1697 {
1698 #if HAVE_AT
1699 eio__close (wd->fd);
1700 #endif
1701 free (wd);
1702 }
1703}
1704
1705#if HAVE_AT
1706
1707static int
1708eio__renameat2 (int olddirfd, const char *oldpath, int newdirfd, const char *newpath, unsigned int flags)
1709{
1710#if HAVE_RENAMEAT2
1711 return syscall (SYS_renameat2, olddirfd, oldpath, newdirfd, newpath, flags);
1712#else
1713 if (flags)
1714 return EIO_ENOSYS ();
1715
1716 return renameat (olddirfd, oldpath, newdirfd, newpath);
1717#endif
1718}
1719
1720/* they forgot these */
1721
1722static int
1723eio__truncateat (int dirfd, const char *path, off_t length)
1724{
1725 int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
1726 int res;
1727
1728 if (fd < 0)
1729 return fd;
1730
1731 res = ftruncate (fd, length);
1732 silent_close (fd);
1733 return res;
1734}
1735
1736static int
1737eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
1738{
1739 int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
1740 int res;
1741
1742 if (fd < 0)
1743 return fd;
1744
1745 res = fstatvfs (fd, buf);
1746 silent_close (fd);
1747 return res;
1748}
1749
1750#endif
949 1751
950/*****************************************************************************/ 1752/*****************************************************************************/
951 1753
952#define ALLOC(len) \ 1754#define ALLOC(len) \
953 if (!req->ptr2) \ 1755 if (!req->ptr2) \
954 { \ 1756 { \
955 X_LOCK (wrklock); \ 1757 X_LOCK (EIO_POOL->wrklock); \
956 req->flags |= EIO_FLAG_PTR2_FREE; \ 1758 req->flags |= EIO_FLAG_PTR2_FREE; \
957 X_UNLOCK (wrklock); \ 1759 X_UNLOCK (EIO_POOL->wrklock); \
958 req->ptr2 = malloc (len); \ 1760 req->ptr2 = malloc (len); \
959 if (!req->ptr2) \ 1761 if (!req->ptr2) \
960 { \ 1762 { \
961 errno = ENOMEM; \ 1763 errno = ENOMEM; \
962 req->result = -1; \ 1764 req->result = -1; \
963 break; \ 1765 goto alloc_fail; \
964 } \ 1766 } \
965 } 1767 }
966 1768
967X_THREAD_PROC (etp_proc) 1769/*****************************************************************************/
968{
969 ETP_REQ *req;
970 struct timespec ts;
971 etp_worker *self = (etp_worker *)thr_arg;
972 1770
973 /* try to distribute timeouts somewhat randomly */ 1771static void
974 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 1772eio__slurp (int fd, eio_req *req)
1773{
1774 req->result = fd;
975 1775
976 for (;;) 1776 if (fd < 0)
1777 return;
1778
1779 if (req->offs < 0 || !req->size) /* do we need the size? */
977 { 1780 {
978 X_LOCK (reqlock); 1781 off_t size = lseek (fd, 0, SEEK_END);
979 1782
980 for (;;)
981 {
982 self->req = req = reqq_shift (&req_queue);
983
984 if (req)
985 break;
986
987 ++idle;
988
989 ts.tv_sec = time (0) + IDLE_TIMEOUT;
990 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
991 {
992 if (idle > max_idle)
993 {
994 --idle;
995 X_UNLOCK (reqlock);
996 X_LOCK (wrklock);
997 --started;
998 X_UNLOCK (wrklock);
999 goto quit;
1000 }
1001
1002 /* we are allowed to idle, so do so without any timeout */
1003 X_COND_WAIT (reqwait, reqlock);
1004 }
1005
1006 --idle;
1007 }
1008
1009 --nready;
1010
1011 X_UNLOCK (reqlock);
1012
1013 if (req->type < 0) 1783 if (req->offs < 0)
1014 goto quit; 1784 req->offs += size;
1015 1785
1016 if (!EIO_CANCELLED (req)) 1786 if (!req->size)
1017 ETP_EXECUTE (self, req); 1787 req->size = size - req->offs;
1018
1019 X_LOCK (reslock);
1020
1021 ++npending;
1022
1023 if (!reqq_push (&res_queue, req) && want_poll_cb)
1024 want_poll_cb ();
1025
1026 self->req = 0;
1027 etp_worker_clear (self);
1028
1029 X_UNLOCK (reslock);
1030 } 1788 }
1031 1789
1032quit: 1790 ALLOC (req->size);
1033 X_LOCK (wrklock); 1791 req->result = pread (fd, req->ptr2, req->size, req->offs);
1034 etp_worker_free (self);
1035 X_UNLOCK (wrklock);
1036 1792
1037 return 0; 1793 silent_close (fd);
1038}
1039 1794
1040/*****************************************************************************/ 1795alloc_fail:
1796 ;
1797}
1041 1798
1799int ecb_cold
1042int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1800eio_init (void (*want_poll)(void), void (*done_poll)(void))
1043{ 1801{
1044 etp_init (want_poll, done_poll); 1802 eio_want_poll_cb = want_poll;
1045} 1803 eio_done_poll_cb = done_poll;
1046 1804
1805 return etp_init (EIO_POOL, 0, 0, 0);
1806}
1807
1808ecb_inline void
1047static void eio_api_destroy (eio_req *req) 1809eio_api_destroy (eio_req *req)
1048{ 1810{
1049 free (req); 1811 free (req);
1050} 1812}
1051 1813
1052#define REQ(rtype) \ 1814#define REQ(rtype) \
1053 eio_req *req; \ 1815 eio_req *req; \
1054 \ 1816 \
1055 req = (eio_req *)calloc (1, sizeof *req); \ 1817 req = (eio_req *)calloc (1, sizeof *req); \
1056 if (!req) \ 1818 if (!req) \
1057 return 0; \ 1819 return 0; \
… …
1071 { \ 1833 { \
1072 eio_api_destroy (req); \ 1834 eio_api_destroy (req); \
1073 return 0; \ 1835 return 0; \
1074 } 1836 }
1075 1837
1838#define SINGLEDOT(ptr) (0[(char *)(ptr)] == '.' && !1[(char *)(ptr)])
1839
1840static void
1076static void eio_execute (etp_worker *self, eio_req *req) 1841eio_execute (etp_worker *self, eio_req *req)
1077{ 1842{
1078 errno = 0; 1843#if HAVE_AT
1844 int dirfd;
1845#else
1846 const char *path;
1847#endif
1848
1849 if (ecb_expect_false (EIO_CANCELLED (req)))
1850 {
1851 req->result = -1;
1852 req->errorno = ECANCELED;
1853 return;
1854 }
1855
1856 if (ecb_expect_false (req->wd == EIO_INVALID_WD))
1857 {
1858 req->result = -1;
1859 req->errorno = ENOENT;
1860 return;
1861 }
1862
1863 if (req->type >= EIO_OPEN)
1864 {
1865 #if HAVE_AT
1866 dirfd = WD2FD (req->wd);
1867 #else
1868 path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1869 #endif
1870 }
1079 1871
1080 switch (req->type) 1872 switch (req->type)
1081 { 1873 {
1874 case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
1875 req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
1876 break;
1877 case EIO_WD_CLOSE: req->result = 0;
1878 eio_wd_close_sync (req->wd); break;
1879
1880 case EIO_SEEK: eio__lseek (req); break;
1082 case EIO_READ: ALLOC (req->size); 1881 case EIO_READ: ALLOC (req->size);
1083 req->result = req->offs >= 0 1882 req->result = req->offs >= 0
1084 ? pread (req->int1, req->ptr2, req->size, req->offs) 1883 ? pread (req->int1, req->ptr2, req->size, req->offs)
1085 : read (req->int1, req->ptr2, req->size); break; 1884 : read (req->int1, req->ptr2, req->size); break;
1086 case EIO_WRITE: req->result = req->offs >= 0 1885 case EIO_WRITE: req->result = req->offs >= 0
1087 ? pwrite (req->int1, req->ptr2, req->size, req->offs) 1886 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1088 : write (req->int1, req->ptr2, req->size); break; 1887 : write (req->int1, req->ptr2, req->size); break;
1089 1888
1889 case EIO_FCNTL: req->result = fcntl (req->int1, (int) req->int2, req->ptr2); break;
1890 case EIO_IOCTL: req->result = ioctl (req->int1, (unsigned long)req->int2, req->ptr2); break;
1891
1090 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break; 1892 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1091 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break; 1893 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
1894
1895#if HAVE_AT
1092 1896
1093 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1897 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1094 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 1898 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break;
1095 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1899 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1096 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 1900 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, AT_SYMLINK_NOFOLLOW); break;
1097 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1098 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1099
1100 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; 1901 case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break;
1101 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1102 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break; 1902 case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break;
1103 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1104 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break; 1903 case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break;
1105 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1106
1107 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break; 1904 case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break;
1108 case EIO_CLOSE: req->result = close (req->int1); break; 1905 case EIO_SLURP: eio__slurp ( openat (dirfd, req->ptr1, O_RDONLY | O_CLOEXEC), req); break;
1109 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break; 1906
1110 case EIO_UNLINK: req->result = unlink (req->ptr1); break; 1907 case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
1111 case EIO_RMDIR: req->result = rmdir (req->ptr1); break; 1908 case EIO_RMDIR: /* complications arise because "." cannot be removed, so we might have to expand */
1909 req->result = req->wd && SINGLEDOT (req->ptr1)
1910 ? rmdir (req->wd->str)
1911 : unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break;
1112 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; 1912 case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); break;
1113 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; 1913 case EIO_RENAME: req->result = eio__renameat2 (
1914 dirfd,
1915 /* complications arise because "." cannot be renamed, so we might have to expand */
1916 req->wd && SINGLEDOT (req->ptr1) ? req->wd->str : req->ptr1,
1917 WD2FD ((eio_wd)req->int3),
1918 req->ptr2,
1919 req->int2
1920 );
1921 break;
1114 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; 1922 case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2, 0); break;
1115 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; 1923 case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
1116 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break; 1924 case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1117 1925 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1926 req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1118 case EIO_READLINK: ALLOC (NAME_MAX); 1927 case EIO_READLINK: ALLOC (EIO_PATH_MAX);
1119 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; 1928 req->result = readlinkat (dirfd, req->ptr1, req->ptr2, EIO_PATH_MAX);
1120 1929 if (req->result == EIO_PATH_MAX)
1121 case EIO_SYNC: req->result = 0; sync (); break; 1930 {
1122 case EIO_FSYNC: req->result = fsync (req->int1); break; 1931 req->result = -1;
1123 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 1932 errno = ENAMETOOLONG;
1124 1933 }
1125 case EIO_READDIR: eio__scandir (req, self); break; 1934 break;
1126
1127 case EIO_BUSY: 1935 case EIO_UTIME:
1128#ifdef _WIN32 1936 case EIO_FUTIME:
1129 Sleep (req->nv1 * 1000.); 1937 {
1938 struct timespec ts[2];
1939 struct timespec *times;
1940
1941 if (req->nv1 != -1. || req->nv2 != -1.)
1942 {
1943 ts[0].tv_sec = req->nv1;
1944 ts[0].tv_nsec = (req->nv1 - ts[0].tv_sec) * 1e9;
1945 ts[1].tv_sec = req->nv2;
1946 ts[1].tv_nsec = (req->nv2 - ts[1].tv_sec) * 1e9;
1947
1948 times = ts;
1949 }
1950 else
1951 times = 0;
1952
1953 req->result = req->type == EIO_FUTIME
1954 ? futimens (req->int1, times)
1955 : utimensat (dirfd, req->ptr1, times, 0);
1956 }
1957 break;
1958
1130#else 1959#else
1131 {
1132 struct timeval tv;
1133 1960
1134 tv.tv_sec = req->nv1; 1961 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1135 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; 1962 req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
1963 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1964 req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
1965 case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
1966 case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
1967 case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
1968 case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
1969 case EIO_SLURP: eio__slurp ( open (path , O_RDONLY | O_CLOEXEC), req); break;
1136 1970
1137 req->result = select (0, 0, 0, 0, &tv); 1971 case EIO_UNLINK: req->result = unlink (path ); break;
1138 } 1972 case EIO_RMDIR: req->result = rmdir (path ); break;
1139#endif 1973 case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
1140 break; 1974 case EIO_RENAME: req->result = req->int2 ? EIO_ENOSYS () : rename (path, req->ptr2); break;
1975 case EIO_LINK: req->result = link (path , req->ptr2); break;
1976 case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
1977 case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
1978 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1979 req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
1980 case EIO_READLINK: ALLOC (EIO_PATH_MAX);
1981 req->result = readlink (path, req->ptr2, EIO_PATH_MAX);
1982 if (req->result == EIO_PATH_MAX)
1983 {
1984 req->result = -1;
1985 errno = ENAMETOOLONG;
1986 }
1987 break;
1141 1988
1142 case EIO_UTIME: 1989 case EIO_UTIME:
1143 case EIO_FUTIME: 1990 case EIO_FUTIME:
1144 { 1991 {
1145 struct timeval tv[2]; 1992 struct timeval tv[2];
1146 struct timeval *times; 1993 struct timeval *times;
1147 1994
1148 if (req->nv1 != -1. || req->nv2 != -1.) 1995 if (req->nv1 != -1. || req->nv2 != -1.)
1149 { 1996 {
1150 tv[0].tv_sec = req->nv1; 1997 tv[0].tv_sec = req->nv1;
1151 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.; 1998 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
1152 tv[1].tv_sec = req->nv2; 1999 tv[1].tv_sec = req->nv2;
1153 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.; 2000 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
1154 2001
1155 times = tv; 2002 times = tv;
1156 } 2003 }
1157 else 2004 else
1158 times = 0; 2005 times = 0;
1159 2006
1160
1161 req->result = req->type == EIO_FUTIME 2007 req->result = req->type == EIO_FUTIME
1162 ? futimes (req->int1, times) 2008 ? futimes (req->int1, times)
1163 : utimes (req->ptr1, times); 2009 : utimes (req->ptr1, times);
1164 } 2010 }
2011 break;
1165 2012
2013#endif
2014
2015 case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
2016 {
2017 ALLOC (req->result);
2018 memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
2019 }
2020 break;
2021
2022 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2023 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
2024
2025 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2026 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2027
2028 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
2029 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
2030 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
2031
2032 case EIO_CLOSE: req->result = eio__close (req->int1); break;
2033 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
2034 case EIO_SYNC: req->result = 0; sync (); break;
2035 case EIO_FSYNC: req->result = fsync (req->int1); break;
2036 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
2037 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
2038 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
2039 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
2040 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
2041 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
2042 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
2043 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
2044
2045 case EIO_READDIR: eio__scandir (req, self); break;
2046
2047 case EIO_BUSY:
2048#ifdef _WIN32
2049 Sleep (req->nv1 * 1e3);
2050#else
2051 {
2052 struct timeval tv;
2053
2054 tv.tv_sec = req->nv1;
2055 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
2056
2057 req->result = select (0, 0, 0, 0, &tv);
2058 }
2059#endif
2060 break;
2061
2062#if 0
1166 case EIO_GROUP: 2063 case EIO_GROUP:
2064 abort (); /* handled in eio_request */
2065#endif
2066
1167 case EIO_NOP: 2067 case EIO_NOP:
1168 req->result = 0; 2068 req->result = 0;
1169 break; 2069 break;
1170 2070
1171 case EIO_CUSTOM: 2071 case EIO_CUSTOM:
1172 req->feed (req); 2072 req->feed (req);
1173 break; 2073 break;
1174 2074
1175 default: 2075 default:
1176 req->result = -1; 2076 req->result = EIO_ENOSYS ();
1177 break; 2077 break;
1178 } 2078 }
1179 2079
2080alloc_fail:
1180 req->errorno = errno; 2081 req->errorno = errno;
1181} 2082}
1182 2083
1183#ifndef EIO_NO_WRAPPERS 2084#ifndef EIO_NO_WRAPPERS
1184 2085
2086eio_req *eio_wd_open (const char *path, int pri, eio_cb cb, void *data)
2087{
2088 REQ (EIO_WD_OPEN); PATH; SEND;
2089}
2090
2091eio_req *eio_wd_close (eio_wd wd, int pri, eio_cb cb, void *data)
2092{
2093 REQ (EIO_WD_CLOSE); req->wd = wd; SEND;
2094}
2095
1185eio_req *eio_nop (int pri, eio_cb cb, void *data) 2096eio_req *eio_nop (int pri, eio_cb cb, void *data)
1186{ 2097{
1187 REQ (EIO_NOP); SEND; 2098 REQ (EIO_NOP); SEND;
1188} 2099}
1189 2100
… …
1200eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data) 2111eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
1201{ 2112{
1202 REQ (EIO_FSYNC); req->int1 = fd; SEND; 2113 REQ (EIO_FSYNC); req->int1 = fd; SEND;
1203} 2114}
1204 2115
2116eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2117{
2118 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2119}
2120
1205eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) 2121eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1206{ 2122{
1207 REQ (EIO_FDATASYNC); req->int1 = fd; SEND; 2123 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1208} 2124}
1209 2125
2126eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
2127{
2128 REQ (EIO_SYNCFS); req->int1 = fd; SEND;
2129}
2130
2131eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2132{
2133 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2134}
2135
2136eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2137{
2138 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2139}
2140
2141eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
2142{
2143 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
2144}
2145
2146eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
2147{
2148 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2149}
2150
2151eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2152{
2153 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
2154}
2155
1210eio_req *eio_close (int fd, int pri, eio_cb cb, void *data) 2156eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
1211{ 2157{
1212 REQ (EIO_CLOSE); req->int1 = fd; SEND; 2158 REQ (EIO_CLOSE); req->int1 = fd; SEND;
1213} 2159}
1214 2160
1215eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data) 2161eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
1216{ 2162{
1217 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND; 2163 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
1218} 2164}
1219 2165
2166eio_req *eio_seek (int fd, off_t offset, int whence, int pri, eio_cb cb, void *data)
2167{
2168 REQ (EIO_SEEK); req->int1 = fd; req->offs = offset; req->int2 = whence; SEND;
2169}
2170
1220eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data) 2171eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1221{ 2172{
1222 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND; 2173 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1223} 2174}
1224 2175
1225eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data) 2176eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1226{ 2177{
1227 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND; 2178 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1228} 2179}
1229 2180
2181eio_req *eio_fcntl (int fd, int cmd, void *arg, int pri, eio_cb cb, void *data)
2182{
2183 REQ (EIO_IOCTL); req->int1 = fd; req->int2 = cmd; req->ptr2 = arg; SEND;
2184}
2185
2186eio_req *eio_ioctl (int fd, unsigned long request, void *buf, int pri, eio_cb cb, void *data)
2187{
2188 REQ (EIO_IOCTL); req->int1 = fd; req->int2 = request; req->ptr2 = buf; SEND;
2189}
2190
1230eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) 2191eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1231{ 2192{
1232 REQ (EIO_FSTAT); req->int1 = fd; SEND; 2193 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1233} 2194}
1234 2195
2196eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
2197{
2198 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
2199}
2200
1235eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) 2201eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1236{ 2202{
1237 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; 2203 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1238} 2204}
1239 2205
… …
1245eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) 2211eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1246{ 2212{
1247 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2213 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1248} 2214}
1249 2215
1250eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2216eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1251{ 2217{
1252 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2218 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1253} 2219}
1254 2220
1255eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) 2221eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
… …
1275eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data) 2241eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1276{ 2242{
1277 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2243 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1278} 2244}
1279 2245
1280eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2246eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1281{ 2247{
1282 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2248 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1283} 2249}
1284 2250
1285eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) 2251eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
… …
1301eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data) 2267eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1302{ 2268{
1303 return eio__1path (EIO_READLINK, path, pri, cb, data); 2269 return eio__1path (EIO_READLINK, path, pri, cb, data);
1304} 2270}
1305 2271
2272eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2273{
2274 return eio__1path (EIO_REALPATH, path, pri, cb, data);
2275}
2276
1306eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) 2277eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1307{ 2278{
1308 return eio__1path (EIO_STAT, path, pri, cb, data); 2279 return eio__1path (EIO_STAT, path, pri, cb, data);
1309} 2280}
1310 2281
1311eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) 2282eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1312{ 2283{
1313 return eio__1path (EIO_LSTAT, path, pri, cb, data); 2284 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1314} 2285}
1315 2286
2287eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
2288{
2289 return eio__1path (EIO_STATVFS, path, pri, cb, data);
2290}
2291
1316eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) 2292eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1317{ 2293{
1318 return eio__1path (EIO_UNLINK, path, pri, cb, data); 2294 return eio__1path (EIO_UNLINK, path, pri, cb, data);
1319} 2295}
1320 2296
1321eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data) 2297eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1322{ 2298{
1323 return eio__1path (EIO_RMDIR, path, pri, cb, data); 2299 return eio__1path (EIO_RMDIR, path, pri, cb, data);
1324} 2300}
1325 2301
1326eio_req *eio_readdir (const char *path, int pri, eio_cb cb, void *data) 2302eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1327{ 2303{
1328 return eio__1path (EIO_READDIR, path, pri, cb, data); 2304 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1329} 2305}
1330 2306
1331eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) 2307eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1332{ 2308{
1333 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int2 = (long)dev; SEND; 2309 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1334} 2310}
1335 2311
1336static eio_req * 2312static eio_req *
1337eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2313eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1338{ 2314{
… …
1362eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2338eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1363{ 2339{
1364 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2340 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1365} 2341}
1366 2342
2343eio_req *eio_slurp (const char *path, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
2344{
2345 REQ (EIO_SLURP); PATH; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
2346}
2347
1367eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) 2348eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
1368{ 2349{
1369 REQ (EIO_CUSTOM); req->feed = execute; SEND; 2350 REQ (EIO_CUSTOM); req->feed = execute; SEND;
1370} 2351}
1371 2352
1372#endif 2353#endif
… …
1383#undef SEND 2364#undef SEND
1384 2365
1385/*****************************************************************************/ 2366/*****************************************************************************/
1386/* grp functions */ 2367/* grp functions */
1387 2368
2369void
1388void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2370eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1389{ 2371{
1390 grp->int2 = limit; 2372 grp->int2 = limit;
1391 grp->feed = feed; 2373 grp->feed = feed;
1392 2374
1393 grp_try_feed (grp); 2375 grp_try_feed (grp);
1394} 2376}
1395 2377
2378void
1396void eio_grp_limit (eio_req *grp, int limit) 2379eio_grp_limit (eio_req *grp, int limit)
1397{ 2380{
1398 grp->int2 = limit; 2381 grp->int2 = limit;
1399 2382
1400 grp_try_feed (grp); 2383 grp_try_feed (grp);
1401} 2384}
1402 2385
2386void
1403void eio_grp_add (eio_req *grp, eio_req *req) 2387eio_grp_add (eio_req *grp, eio_req *req)
1404{ 2388{
1405 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2389 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
2390
2391 grp->flags |= ETP_FLAG_GROUPADD;
1406 2392
1407 ++grp->size; 2393 ++grp->size;
1408 req->grp = grp; 2394 req->grp = grp;
1409 2395
1410 req->grp_prev = 0; 2396 req->grp_prev = 0;
… …
1417} 2403}
1418 2404
1419/*****************************************************************************/ 2405/*****************************************************************************/
1420/* misc garbage */ 2406/* misc garbage */
1421 2407
2408eio_ssize_t
1422ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2409eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1423{ 2410{
1424 etp_worker wrk;
1425
1426 wrk.dbuf = 0;
1427
1428 eio__sendfile (ofd, ifd, offset, count, &wrk); 2411 return eio__sendfile (ofd, ifd, offset, count);
1429
1430 if (wrk.dbuf)
1431 free (wrk.dbuf);
1432} 2412}
1433 2413

Diff Legend

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