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

Comparing libeio/eio.c (file contents):
Revision 1.36 by root, Fri Jun 12 00:43:16 2009 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) 2007,2008,2009 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
42#ifdef EIO_STACKSIZE
43# define XTHREAD_STACKSIZE EIO_STACKSIZE
44#endif
45#include "xthread.h" 45#include "ecb.h"
46 46
47#include <errno.h> 47#include <errno.h>
48#include <stddef.h> 48#include <stddef.h>
49#include <stdlib.h> 49#include <stdlib.h>
50#include <string.h> 50#include <string.h>
53#include <sys/stat.h> 53#include <sys/stat.h>
54#include <limits.h> 54#include <limits.h>
55#include <fcntl.h> 55#include <fcntl.h>
56#include <assert.h> 56#include <assert.h>
57 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
58#ifndef EIO_FINISH 77#ifndef EIO_FINISH
59# 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
60#endif 79#endif
61 80
62#ifndef EIO_DESTROY 81#ifndef EIO_DESTROY
65 84
66#ifndef EIO_FEED 85#ifndef EIO_FEED
67# 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)
68#endif 87#endif
69 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
70#ifdef _WIN32 100#ifdef _WIN32
71 101
72 /*doh*/ 102 #undef PAGESIZE
103 #define PAGESIZE 4096 /* GetSystemInfo? */
104
105 /* TODO: look at how perl does stat (non-sloppy), unlink (ro-files), utime, link */
106
107 #ifdef EIO_STRUCT_STATI64
108 /* look at perl's non-sloppy stat */
109 #define stat(path,buf) _stati64 (path,buf)
110 #define fstat(fd,buf) _fstati64 (fd,buf)
111 #endif
112 #define lstat(path,buf) stat (path,buf)
113 #define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
114 #define mkdir(path,mode) _mkdir (path)
115 #define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
116
117 #define chmod(path,mode) _chmod (path, mode)
118 #define dup(fd) _dup (fd)
119 #define dup2(fd1,fd2) _dup2 (fd1, fd2)
120 #define pipe(fds) _pipe (fds, 4096, O_BINARY)
121
122 #define fcntl(fd,cmd,arg) EIO_ENOSYS ()
123 #define ioctl(fd,cmd,arg) EIO_ENOSYS ()
124 #define fchmod(fd,mode) EIO_ENOSYS ()
125 #define chown(path,uid,gid) EIO_ENOSYS ()
126 #define fchown(fd,uid,gid) EIO_ENOSYS ()
127 #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
128 #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
129 #define mknod(path,mode,dev) EIO_ENOSYS ()
130 #define sync() EIO_ENOSYS ()
131 #define readlink(path,buf,s) EIO_ENOSYS ()
132 #define statvfs(path,buf) EIO_ENOSYS ()
133 #define fstatvfs(fd,buf) EIO_ENOSYS ()
134
135 #define pread(fd,buf,count,offset) eio__pread (fd, buf, count, offset)
136 #define pwrite(fd,buf,count,offset) eio__pwrite (fd, buf, count, offset)
137
138 #if __GNUC__
139 typedef long long eio_off_t; /* signed for compatibility to msvc */
140 #else
141 typedef __int64 eio_off_t; /* unsigned not supported by msvc */
142 #endif
143
144 static eio_ssize_t
145 eio__pread (int fd, void *buf, eio_ssize_t count, eio_off_t offset)
146 {
147 OVERLAPPED o = { 0 };
148 DWORD got;
149
150 o.Offset = offset;
151 o.OffsetHigh = offset >> 32;
152
153 return ReadFile ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd), buf, count, &got, &o)
154 ? got : -1;
155 }
156
157 static eio_ssize_t
158 eio__pwrite (int fd, void *buf, eio_ssize_t count, eio_off_t offset)
159 {
160 OVERLAPPED o = { 0 };
161 DWORD got;
162
163 o.Offset = offset;
164 o.OffsetHigh = offset >> 32;
165
166 return WriteFile ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd), buf, count, &got, &o)
167 ? got : -1;
168 }
169
170 /* rename() uses MoveFile, which fails to overwrite */
171 #define rename(old,neu) eio__rename (old, neu)
172
173 static int
174 eio__rename (const char *old, const char *neu)
175 {
176 if (MoveFileEx (old, neu, MOVEFILE_REPLACE_EXISTING))
177 return 0;
178
179 /* should steal _dosmaperr */
180 switch (GetLastError ())
181 {
182 case ERROR_FILE_NOT_FOUND:
183 case ERROR_PATH_NOT_FOUND:
184 case ERROR_INVALID_DRIVE:
185 case ERROR_NO_MORE_FILES:
186 case ERROR_BAD_NETPATH:
187 case ERROR_BAD_NET_NAME:
188 case ERROR_BAD_PATHNAME:
189 case ERROR_FILENAME_EXCED_RANGE:
190 errno = ENOENT;
191 break;
192
193 default:
194 errno = EACCES;
195 break;
196 }
197
198 return -1;
199 }
200
201 /* we could even stat and see if it exists */
202 static int
203 symlink (const char *old, const char *neu)
204 {
205 #if WINVER >= 0x0600
206 int flags;
207
208 /* This tries out all combinations of SYMBOLIC_LINK_FLAG_DIRECTORY
209 * and SYMBOLIC_LINK_FLAG_ALLOW_UNPRIVILEGED_CREATE,
210 * with directory first.
211 */
212 for (flags = 3; flags >= 0; --flags)
213 if (CreateSymbolicLink (neu, old, flags))
214 return 0;
215 #endif
216
217 return EIO_ERRNO (ENOENT, -1);
218 }
219
220 /* POSIX API only, causing trouble for win32 apps */
221 #define CreateHardLink(neu,old,flags) 0 /* not really creating hardlink, still using relative paths? */
222 #define CreateSymbolicLink(neu,old,flags) 0 /* vista+ only */
223
224 struct statvfs
225 {
226 int dummy;
227 };
228
229 #define DT_DIR EIO_DT_DIR
230 #define DT_REG EIO_DT_REG
231 #define D_NAME(entp) entp.cFileName
232 #define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
233
73#else 234#else
74 235
75# include "config.h" 236 #include <sys/ioctl.h>
76# include <sys/time.h> 237 #include <sys/time.h>
77# include <sys/select.h> 238 #include <sys/select.h>
78# 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
79# include <utime.h> 283# include <utime.h>
80# include <signal.h>
81# include <dirent.h>
82
83/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
84# if defined(__FreeBSD__) || defined(__NetBSD__) || defined(__OpenBSD__)
85# define _DIRENT_HAVE_D_TYPE /* sigh */
86# define D_INO(de) (de)->d_fileno
87# define D_NAMLEN(de) (de)->d_namlen
88# elif defined(__linux) || defined(d_ino) || _XOPEN_SOURCE >= 600
89# define D_INO(de) (de)->d_ino
90# endif 284#endif
91 285
92#ifdef _D_EXACT_NAMLEN 286#if HAVE_SYS_SYSCALL_H
93# undef D_NAMLEN 287# include <sys/syscall.h>
94# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
95#endif
96
97# ifdef _DIRENT_HAVE_D_TYPE
98# define D_TYPE(de) (de)->d_type
99# endif
100
101# ifndef EIO_STRUCT_DIRENT
102# define EIO_STRUCT_DIRENT struct dirent
103# endif
104
105#endif 288#endif
106 289
107#if HAVE_SENDFILE 290#if HAVE_SENDFILE
108# if __linux 291# if __linux
109# include <sys/sendfile.h> 292# include <sys/sendfile.h>
110# elif __freebsd 293# elif __FreeBSD__ || defined __APPLE__
111# include <sys/socket.h> 294# include <sys/socket.h>
112# include <sys/uio.h> 295# include <sys/uio.h>
113# elif __hpux 296# elif __hpux
114# include <sys/socket.h> 297# include <sys/socket.h>
115# elif __solaris /* not yet */ 298# elif __solaris
116# include <sys/sendfile.h> 299# include <sys/sendfile.h>
117# else 300# else
118# error sendfile support requested but not available 301# error sendfile support requested but not available
119# endif 302# endif
120#endif 303#endif
121 304
305#if HAVE_RENAMEAT2
306# include <sys/syscall.h>
307# include <linux/fs.h>
308#endif
309
122#ifndef D_TYPE 310#ifndef D_TYPE
123# define D_TYPE(de) 0 311# define D_TYPE(de) 0
124#endif 312#endif
125#ifndef D_INO 313#ifndef D_INO
126# define D_INO(de) 0 314# define D_INO(de) 0
127#endif 315#endif
128#ifndef D_NAMLEN 316#ifndef D_NAMLEN
129# define D_NAMLEN(de) strlen ((de)->d_name) 317# define D_NAMLEN(entp) strlen (D_NAME (entp))
130#endif 318#endif
131
132/* number of seconds after which an idle threads exit */
133#define IDLE_TIMEOUT 10
134 319
135/* used for struct dirent, AIX doesn't provide it */ 320/* used for struct dirent, AIX doesn't provide it */
136#ifndef NAME_MAX 321#ifndef NAME_MAX
137# define NAME_MAX 4096 322# define NAME_MAX 4096
138#endif 323#endif
139 324
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
140/* buffer size for various temporary buffers */ 340/* buffer size for various temporary buffers */
141#define EIO_BUFSIZE 65536 341#define EIO_BUFSIZE 65536
142 342
143#define dBUF \ 343#define dBUF \
144 char *eio_buf; \
145 ETP_WORKER_LOCK (self); \
146 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 344 char *eio_buf = malloc (EIO_BUFSIZE); \
147 ETP_WORKER_UNLOCK (self); \
148 errno = ENOMEM; \ 345 errno = ENOMEM; \
149 if (!eio_buf) \ 346 if (!eio_buf) \
150 return -1; 347 return -1
151 348
152#define EIO_TICKS ((1000000 + 1023) >> 10) 349#define FUBd \
350 free (eio_buf)
153 351
154/*****************************************************************************/ 352/*****************************************************************************/
155 353
156#if __GNUC__ >= 3 354struct etp_tmpbuf;
157# define expect(expr,value) __builtin_expect ((expr),(value)) 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
158#else 362#else
159# define expect(expr,value) (expr) 363 #define HAVE_AT 0
364 static const char *wd_expand (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path);
160#endif 365#endif
161 366
162#define expect_false(expr) expect ((expr) != 0, 0) 367struct eio_pwd
163#define expect_true(expr) expect ((expr) != 0, 1) 368{
369#if HAVE_AT
370 int fd;
371#endif
372 int len;
373 char str[1]; /* actually, a 0-terminated canonical path */
374};
164 375
165/*****************************************************************************/ 376/*****************************************************************************/
166 377
167#define ETP_PRI_MIN EIO_PRI_MIN 378#define ETP_PRI_MIN EIO_PRI_MIN
168#define ETP_PRI_MAX EIO_PRI_MAX 379#define ETP_PRI_MAX EIO_PRI_MAX
169 380
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
170struct etp_worker; 391struct etp_worker;
171
172#define ETP_REQ eio_req 392#define ETP_REQ eio_req
173#define ETP_DESTROY(req) eio_destroy (req) 393#define ETP_DESTROY(req) eio_destroy (req)
174static int eio_finish (eio_req *req); 394static int eio_finish (eio_req *req);
175#define ETP_FINISH(req) eio_finish (req) 395#define ETP_FINISH(req) eio_finish (req)
176static void eio_execute (struct etp_worker *self, eio_req *req); 396static void eio_execute (struct etp_worker *self, eio_req *req);
177#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) 397#define ETP_EXECUTE(wrk,req) eio_execute (wrk, req)
178 398
179#define ETP_WORKER_CLEAR(req) \ 399#include "etp.c"
180 if (wrk->dbuf) \
181 { \
182 free (wrk->dbuf); \
183 wrk->dbuf = 0; \
184 } \
185 \
186 if (wrk->dirp) \
187 { \
188 closedir (wrk->dirp); \
189 wrk->dirp = 0; \
190 }
191 400
192#define ETP_WORKER_COMMON \ 401static struct etp_pool eio_pool;
193 void *dbuf; \ 402#define EIO_POOL (&eio_pool)
194 DIR *dirp;
195 403
196/*****************************************************************************/ 404/*****************************************************************************/
197 405
198#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
199
200/* calculcate time difference in ~1/EIO_TICKS of a second */
201static int tvdiff (struct timeval *tv1, struct timeval *tv2)
202{
203 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
204 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
205}
206
207static unsigned int started, idle, wanted = 4;
208
209static void (*want_poll_cb) (void);
210static void (*done_poll_cb) (void);
211
212static unsigned int max_poll_time; /* reslock */
213static unsigned int max_poll_reqs; /* reslock */
214
215static volatile unsigned int nreqs; /* reqlock */
216static volatile unsigned int nready; /* reqlock */
217static volatile unsigned int npending; /* reqlock */
218static volatile unsigned int max_idle = 4;
219
220static mutex_t wrklock = X_MUTEX_INIT;
221static mutex_t reslock = X_MUTEX_INIT;
222static mutex_t reqlock = X_MUTEX_INIT;
223static cond_t reqwait = X_COND_INIT;
224
225#if !HAVE_PREADWRITE
226/*
227 * make our pread/pwrite emulation safe against themselves, but not against
228 * normal read/write by using a mutex. slows down execution a lot,
229 * but that's your problem, not mine.
230 */
231static mutex_t preadwritelock = X_MUTEX_INIT;
232#endif
233
234typedef struct etp_worker
235{
236 /* locked by wrklock */
237 struct etp_worker *prev, *next;
238
239 thread_t tid;
240
241 /* locked by reslock, reqlock or wrklock */
242 ETP_REQ *req; /* currently processed request */
243
244 ETP_WORKER_COMMON
245} etp_worker;
246
247static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */
248
249#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
250#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
251
252/* worker threads management */
253
254static void etp_worker_clear (etp_worker *wrk)
255{
256 ETP_WORKER_CLEAR (wrk);
257}
258
259static void etp_worker_free (etp_worker *wrk)
260{
261 wrk->next->prev = wrk->prev;
262 wrk->prev->next = wrk->next;
263
264 free (wrk);
265}
266
267static unsigned int etp_nreqs (void)
268{
269 int retval;
270 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
271 retval = nreqs;
272 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
273 return retval;
274}
275
276static unsigned int etp_nready (void)
277{
278 unsigned int retval;
279
280 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
281 retval = nready;
282 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
283
284 return retval;
285}
286
287static unsigned int etp_npending (void)
288{
289 unsigned int retval;
290
291 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
292 retval = npending;
293 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
294
295 return retval;
296}
297
298static unsigned int etp_nthreads (void)
299{
300 unsigned int retval;
301
302 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
303 retval = started;
304 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
305
306 return retval;
307}
308
309/*
310 * a somewhat faster data structure might be nice, but
311 * with 8 priorities this actually needs <20 insns
312 * per shift, the most expensive operation.
313 */
314typedef struct {
315 ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */
316 int size;
317} etp_reqq;
318
319static etp_reqq req_queue;
320static etp_reqq res_queue;
321
322static int reqq_push (etp_reqq *q, ETP_REQ *req)
323{
324 int pri = req->pri;
325 req->next = 0;
326
327 if (q->qe[pri])
328 {
329 q->qe[pri]->next = req;
330 q->qe[pri] = req;
331 }
332 else
333 q->qe[pri] = q->qs[pri] = req;
334
335 return q->size++;
336}
337
338static ETP_REQ *reqq_shift (etp_reqq *q)
339{
340 int pri;
341
342 if (!q->size)
343 return 0;
344
345 --q->size;
346
347 for (pri = ETP_NUM_PRI; pri--; )
348 {
349 eio_req *req = q->qs[pri];
350
351 if (req)
352 {
353 if (!(q->qs[pri] = (eio_req *)req->next))
354 q->qe[pri] = 0;
355
356 return req;
357 }
358 }
359
360 abort ();
361}
362
363static void etp_atfork_prepare (void)
364{
365 X_LOCK (wrklock);
366 X_LOCK (reqlock);
367 X_LOCK (reslock);
368#if !HAVE_PREADWRITE
369 X_LOCK (preadwritelock);
370#endif
371}
372
373static void etp_atfork_parent (void)
374{
375#if !HAVE_PREADWRITE
376 X_UNLOCK (preadwritelock);
377#endif
378 X_UNLOCK (reslock);
379 X_UNLOCK (reqlock);
380 X_UNLOCK (wrklock);
381}
382
383static void etp_atfork_child (void)
384{
385 ETP_REQ *prv;
386
387 while ((prv = reqq_shift (&req_queue)))
388 ETP_DESTROY (prv);
389
390 while ((prv = reqq_shift (&res_queue)))
391 ETP_DESTROY (prv);
392
393 while (wrk_first.next != &wrk_first)
394 {
395 etp_worker *wrk = wrk_first.next;
396
397 if (wrk->req)
398 ETP_DESTROY (wrk->req);
399
400 etp_worker_clear (wrk);
401 etp_worker_free (wrk);
402 }
403
404 started = 0;
405 idle = 0;
406 nreqs = 0;
407 nready = 0;
408 npending = 0;
409
410 etp_atfork_parent ();
411}
412
413static void 406static void
414etp_once_init (void)
415{
416 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
417}
418
419static int
420etp_init (void (*want_poll)(void), void (*done_poll)(void))
421{
422 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
423
424 pthread_once (&doinit, etp_once_init);
425
426 want_poll_cb = want_poll;
427 done_poll_cb = done_poll;
428
429 return 0;
430}
431
432X_THREAD_PROC (etp_proc);
433
434static void etp_start_thread (void)
435{
436 etp_worker *wrk = calloc (1, sizeof (etp_worker));
437
438 /*TODO*/
439 assert (("unable to allocate worker thread data", wrk));
440
441 X_LOCK (wrklock);
442
443 if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
444 {
445 wrk->prev = &wrk_first;
446 wrk->next = wrk_first.next;
447 wrk_first.next->prev = wrk;
448 wrk_first.next = wrk;
449 ++started;
450 }
451 else
452 free (wrk);
453
454 X_UNLOCK (wrklock);
455}
456
457static void etp_maybe_start_thread (void)
458{
459 if (expect_true (etp_nthreads () >= wanted))
460 return;
461
462 /* todo: maybe use idle here, but might be less exact */
463 if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
464 return;
465
466 etp_start_thread ();
467}
468
469static void etp_end_thread (void)
470{
471 eio_req *req = calloc (1, sizeof (eio_req));
472
473 req->type = -1;
474 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
475
476 X_LOCK (reqlock);
477 reqq_push (&req_queue, req);
478 X_COND_SIGNAL (reqwait);
479 X_UNLOCK (reqlock);
480
481 X_LOCK (wrklock);
482 --started;
483 X_UNLOCK (wrklock);
484}
485
486static int etp_poll (void)
487{
488 unsigned int maxreqs;
489 unsigned int maxtime;
490 struct timeval tv_start, tv_now;
491
492 X_LOCK (reslock);
493 maxreqs = max_poll_reqs;
494 maxtime = max_poll_time;
495 X_UNLOCK (reslock);
496
497 if (maxtime)
498 gettimeofday (&tv_start, 0);
499
500 for (;;)
501 {
502 ETP_REQ *req;
503
504 etp_maybe_start_thread ();
505
506 X_LOCK (reslock);
507 req = reqq_shift (&res_queue);
508
509 if (req)
510 {
511 --npending;
512
513 if (!res_queue.size && done_poll_cb)
514 done_poll_cb ();
515 }
516
517 X_UNLOCK (reslock);
518
519 if (!req)
520 return 0;
521
522 X_LOCK (reqlock);
523 --nreqs;
524 X_UNLOCK (reqlock);
525
526 if (expect_false (req->type == EIO_GROUP && req->size))
527 {
528 req->int1 = 1; /* mark request as delayed */
529 continue;
530 }
531 else
532 {
533 int res = ETP_FINISH (req);
534 if (expect_false (res))
535 return res;
536 }
537
538 if (expect_false (maxreqs && !--maxreqs))
539 break;
540
541 if (maxtime)
542 {
543 gettimeofday (&tv_now, 0);
544
545 if (tvdiff (&tv_start, &tv_now) >= maxtime)
546 break;
547 }
548 }
549
550 errno = EAGAIN;
551 return -1;
552}
553
554static void etp_cancel (ETP_REQ *req)
555{
556 X_LOCK (wrklock);
557 req->flags |= EIO_FLAG_CANCELLED;
558 X_UNLOCK (wrklock);
559
560 eio_grp_cancel (req);
561}
562
563static void etp_submit (ETP_REQ *req)
564{
565 req->pri -= ETP_PRI_MIN;
566
567 if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
568 if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
569
570 if (expect_false (req->type == EIO_GROUP))
571 {
572 /* I hope this is worth it :/ */
573 X_LOCK (reqlock);
574 ++nreqs;
575 X_UNLOCK (reqlock);
576
577 X_LOCK (reslock);
578
579 ++npending;
580
581 if (!reqq_push (&res_queue, req) && want_poll_cb)
582 want_poll_cb ();
583
584 X_UNLOCK (reslock);
585 }
586 else
587 {
588 X_LOCK (reqlock);
589 ++nreqs;
590 ++nready;
591 reqq_push (&req_queue, req);
592 X_COND_SIGNAL (reqwait);
593 X_UNLOCK (reqlock);
594
595 etp_maybe_start_thread ();
596 }
597}
598
599static void etp_set_max_poll_time (double nseconds)
600{
601 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
602 max_poll_time = nseconds;
603 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
604}
605
606static void etp_set_max_poll_reqs (unsigned int maxreqs)
607{
608 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
609 max_poll_reqs = maxreqs;
610 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
611}
612
613static void etp_set_max_idle (unsigned int nthreads)
614{
615 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
616 max_idle = nthreads <= 0 ? 1 : nthreads;
617 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
618}
619
620static void etp_set_min_parallel (unsigned int nthreads)
621{
622 if (wanted < nthreads)
623 wanted = nthreads;
624}
625
626static void etp_set_max_parallel (unsigned int nthreads)
627{
628 if (wanted > nthreads)
629 wanted = nthreads;
630
631 while (started > wanted)
632 etp_end_thread ();
633}
634
635/*****************************************************************************/
636
637static void grp_try_feed (eio_req *grp) 407grp_try_feed (eio_req *grp)
638{ 408{
639 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 409 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
640 { 410 {
641 grp->flags &= ~EIO_FLAG_GROUPADD; 411 grp->flags &= ~ETP_FLAG_GROUPADD;
642 412
643 EIO_FEED (grp); 413 EIO_FEED (grp);
644 414
645 /* stop if no progress has been made */ 415 /* stop if no progress has been made */
646 if (!(grp->flags & EIO_FLAG_GROUPADD)) 416 if (!(grp->flags & ETP_FLAG_GROUPADD))
647 { 417 {
648 grp->feed = 0; 418 grp->feed = 0;
649 break; 419 break;
650 } 420 }
651 } 421 }
652} 422}
653 423
424static int
654static int grp_dec (eio_req *grp) 425grp_dec (eio_req *grp)
655{ 426{
656 --grp->size; 427 --grp->size;
657 428
658 /* call feeder, if applicable */ 429 /* call feeder, if applicable */
659 grp_try_feed (grp); 430 grp_try_feed (grp);
660 431
661 /* finish, if done */ 432 /* finish, if done */
662 if (!grp->size && grp->int1) 433 if (!grp->size && grp->flags & ETP_FLAG_DELAYED)
663 return eio_finish (grp); 434 return eio_finish (grp);
664 else 435 else
665 return 0; 436 return 0;
666} 437}
667 438
439static void
668void eio_destroy (eio_req *req) 440eio_destroy (eio_req *req)
669{ 441{
670 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); 442 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
671 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); 443 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
672 444
673 EIO_DESTROY (req); 445 EIO_DESTROY (req);
674} 446}
675 447
448static int
676static int eio_finish (eio_req *req) 449eio_finish (eio_req *req)
677{ 450{
678 int res = EIO_FINISH (req); 451 int res = EIO_FINISH (req);
679 452
680 if (req->grp) 453 if (req->grp)
681 { 454 {
689 if (grp->grp_first == req) 462 if (grp->grp_first == req)
690 grp->grp_first = req->grp_next; 463 grp->grp_first = req->grp_next;
691 464
692 res2 = grp_dec (grp); 465 res2 = grp_dec (grp);
693 466
694 if (!res && res2) 467 if (!res)
695 res = res2; 468 res = res2;
696 } 469 }
697 470
698 eio_destroy (req); 471 eio_destroy (req);
699 472
700 return res; 473 return res;
701} 474}
702 475
476void
703void eio_grp_cancel (eio_req *grp) 477eio_grp_cancel (eio_req *grp)
704{ 478{
705 for (grp = grp->grp_first; grp; grp = grp->grp_next) 479 etp_grp_cancel (EIO_POOL, grp);
706 eio_cancel (grp);
707} 480}
708 481
482void
709void eio_cancel (eio_req *req) 483eio_cancel (eio_req *req)
710{ 484{
711 etp_cancel (req); 485 etp_cancel (EIO_POOL, req);
712} 486}
713 487
488void
714void eio_submit (eio_req *req) 489eio_submit (eio_req *req)
715{ 490{
716 etp_submit (req); 491 etp_submit (EIO_POOL, req);
717} 492}
718 493
719unsigned int eio_nreqs (void) 494unsigned int
495eio_nreqs (void)
720{ 496{
721 return etp_nreqs (); 497 return etp_nreqs (EIO_POOL);
722} 498}
723 499
724unsigned int eio_nready (void) 500unsigned int
501eio_nready (void)
725{ 502{
726 return etp_nready (); 503 return etp_nready (EIO_POOL);
727} 504}
728 505
729unsigned int eio_npending (void) 506unsigned int
507eio_npending (void)
730{ 508{
731 return etp_npending (); 509 return etp_npending (EIO_POOL);
732} 510}
733 511
734unsigned int eio_nthreads (void) 512unsigned int ecb_cold
513eio_nthreads (void)
735{ 514{
736 return etp_nthreads (); 515 return etp_nthreads (EIO_POOL);
737} 516}
738 517
518void ecb_cold
739void eio_set_max_poll_time (double nseconds) 519eio_set_max_poll_time (double nseconds)
740{ 520{
741 etp_set_max_poll_time (nseconds); 521 etp_set_max_poll_time (EIO_POOL, nseconds);
742} 522}
743 523
524void ecb_cold
744void eio_set_max_poll_reqs (unsigned int maxreqs) 525eio_set_max_poll_reqs (unsigned int maxreqs)
745{ 526{
746 etp_set_max_poll_reqs (maxreqs); 527 etp_set_max_poll_reqs (EIO_POOL, maxreqs);
747} 528}
748 529
530void ecb_cold
749void eio_set_max_idle (unsigned int nthreads) 531eio_set_max_idle (unsigned int nthreads)
750{ 532{
751 etp_set_max_idle (nthreads); 533 etp_set_max_idle (EIO_POOL, nthreads);
752} 534}
753 535
536void ecb_cold
537eio_set_idle_timeout (unsigned int seconds)
538{
539 etp_set_idle_timeout (EIO_POOL, seconds);
540}
541
542void ecb_cold
754void eio_set_min_parallel (unsigned int nthreads) 543eio_set_min_parallel (unsigned int nthreads)
755{ 544{
756 etp_set_min_parallel (nthreads); 545 etp_set_min_parallel (EIO_POOL, nthreads);
757} 546}
758 547
548void ecb_cold
759void eio_set_max_parallel (unsigned int nthreads) 549eio_set_max_parallel (unsigned int nthreads)
760{ 550{
761 etp_set_max_parallel (nthreads); 551 etp_set_max_parallel (EIO_POOL, nthreads);
762} 552}
763 553
764int eio_poll (void) 554int eio_poll (void)
765{ 555{
766 return etp_poll (); 556 return etp_poll (EIO_POOL);
767} 557}
768 558
769/*****************************************************************************/ 559/*****************************************************************************/
770/* work around various missing functions */ 560/* work around various missing functions */
771 561
772#if !HAVE_PREADWRITE 562#if HAVE_POSIX_CLOSE && !__linux
773# undef pread 563# define eio__close(fd) posix_close (fd, 0)
774# undef pwrite 564#else
775# define pread eio__pread 565# define eio__close(fd) close (fd)
776# define pwrite eio__pwrite
777
778static ssize_t
779eio__pread (int fd, void *buf, size_t count, off_t offset)
780{
781 ssize_t res;
782 off_t ooffset;
783
784 X_LOCK (preadwritelock);
785 ooffset = lseek (fd, 0, SEEK_CUR);
786 lseek (fd, offset, SEEK_SET);
787 res = read (fd, buf, count);
788 lseek (fd, ooffset, SEEK_SET);
789 X_UNLOCK (preadwritelock);
790
791 return res;
792}
793
794static ssize_t
795eio__pwrite (int fd, void *buf, size_t count, off_t offset)
796{
797 ssize_t res;
798 off_t ooffset;
799
800 X_LOCK (preadwritelock);
801 ooffset = lseek (fd, 0, SEEK_CUR);
802 lseek (fd, offset, SEEK_SET);
803 res = write (fd, buf, count);
804 lseek (fd, ooffset, SEEK_SET);
805 X_UNLOCK (preadwritelock);
806
807 return res;
808}
809#endif 566#endif
810 567
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
811#ifndef HAVE_FUTIMES 577#ifndef HAVE_UTIMES
812 578
813# undef utimes 579# undef utimes
814# undef futimes
815# define utimes(path,times) eio__utimes (path, times) 580# define utimes(path,times) eio__utimes (path, times)
816# define futimes(fd,times) eio__futimes (fd, times)
817 581
818static int 582static int
819eio__utimes (const char *filename, const struct timeval times[2]) 583eio__utimes (const char *filename, const struct timeval times[2])
820{ 584{
821 if (times) 585 if (times)
829 } 593 }
830 else 594 else
831 return utime (filename, 0); 595 return utime (filename, 0);
832} 596}
833 597
598#endif
599
600#ifndef HAVE_FUTIMES
601
602# undef futimes
603# define futimes(fd,times) eio__futimes (fd, times)
604
605static int
834static int eio__futimes (int fd, const struct timeval tv[2]) 606eio__futimes (int fd, const struct timeval tv[2])
835{ 607{
836 errno = ENOSYS; 608 errno = ENOSYS;
837 return -1; 609 return -1;
838} 610}
839 611
842#if !HAVE_FDATASYNC 614#if !HAVE_FDATASYNC
843# undef fdatasync 615# undef fdatasync
844# define fdatasync(fd) fsync (fd) 616# define fdatasync(fd) fsync (fd)
845#endif 617#endif
846 618
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
847/* sync_file_range always needs emulation */ 636/* sync_file_range always needs emulation */
848int 637static int
849eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags) 638eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
850{ 639{
851#if HAVE_SYNC_FILE_RANGE 640#if HAVE_SYNC_FILE_RANGE
852 int res; 641 int res;
853 642
866 if (!res || errno != ENOSYS) 655 if (!res || errno != ENOSYS)
867 return res; 656 return res;
868#endif 657#endif
869 658
870 /* even though we could play tricks with the flags, it's better to always 659 /* even though we could play tricks with the flags, it's better to always
871 * call fdatasync, as thta matches the expectation of it's users best */ 660 * call fdatasync, as that matches the expectation of its users best */
872 return fdatasync (fd); 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
873} 672}
874 673
875#if !HAVE_READAHEAD 674#if !HAVE_READAHEAD
876# undef readahead 675# undef readahead
877# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 676# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
878 677
879static ssize_t 678static eio_ssize_t
880eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) 679eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
881{ 680{
882 size_t todo = count; 681 size_t todo = count;
883 dBUF; 682 dBUF;
884 683
889 pread (fd, eio_buf, len, offset); 688 pread (fd, eio_buf, len, offset);
890 offset += len; 689 offset += len;
891 todo -= len; 690 todo -= len;
892 } 691 }
893 692
894 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 */
895 return count; 697 return 0;
896} 698}
897 699
898#endif 700#endif
899 701
900/* sendfile always needs emulation */ 702/* sendfile always needs emulation */
901static ssize_t 703static eio_ssize_t
902eio__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)
903{ 705{
706 eio_ssize_t written = 0;
904 ssize_t res; 707 eio_ssize_t res;
905 708
906 if (!count) 709 if (!count)
907 return 0; 710 return 0;
908 711
712 for (;;)
713 {
714#ifdef __APPLE__
715# undef HAVE_SENDFILE /* broken, as everything on os x */
716#endif
909#if HAVE_SENDFILE 717#if HAVE_SENDFILE
910# if __linux 718# if __linux
719 off_t soffset = offset;
911 res = sendfile (ofd, ifd, &offset, count); 720 res = sendfile (ofd, ifd, &soffset, count);
912 721
913# elif __freebsd 722# elif __FreeBSD__
914 /* 723 /*
915 * 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
916 * wasted on making it similar to other I/O functions. 725 * wasted on making it similar to other I/O functions.
917 */ 726 */
918 {
919 off_t sbytes; 727 off_t sbytes;
920 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 728 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
921 729
922 if (res < 0 && sbytes) 730 #if 0 /* according to the manpage, this is correct, but broken behaviour */
923 /* 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)
924 res = sbytes; 735 res = sbytes;
925 } 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;
926 749
927# elif __hpux 750# elif __hpux
928 res = sendfile (ofd, ifd, offset, count, 0, 0); 751 res = sendfile (ofd, ifd, offset, count, 0, 0);
929 752
930# elif __solaris 753# elif __solaris
931 {
932 struct sendfilevec vec; 754 struct sendfilevec vec;
933 size_t sbytes; 755 size_t sbytes;
934 756
935 vec.sfv_fd = ifd; 757 vec.sfv_fd = ifd;
936 vec.sfv_flag = 0; 758 vec.sfv_flag = 0;
937 vec.sfv_off = offset; 759 vec.sfv_off = offset;
938 vec.sfv_len = count; 760 vec.sfv_len = count;
939 761
940 res = sendfilev (ofd, &vec, 1, &sbytes); 762 res = sendfilev (ofd, &vec, 1, &sbytes);
941 763
942 if (res < 0 && sbytes) 764 if (res < 0 && sbytes)
943 res = sbytes; 765 res = sbytes;
944 }
945 766
946# 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
947#else 778#else
948 res = -1; 779 res = EIO_ENOSYS ();
949 errno = ENOSYS;
950#endif 780#endif
951 781
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
952 if (res < 0 805 if (res < 0
953 && (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
954#if __solaris 814#if __solaris
955 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 815 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
956#endif 816#endif
957 ) 817 )
958 ) 818 )
962 822
963 res = 0; 823 res = 0;
964 824
965 while (count) 825 while (count)
966 { 826 {
967 ssize_t cnt; 827 eio_ssize_t cnt;
968 828
969 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);
970 830
971 if (cnt <= 0) 831 if (cnt <= 0)
972 { 832 {
984 844
985 offset += cnt; 845 offset += cnt;
986 res += cnt; 846 res += cnt;
987 count -= cnt; 847 count -= cnt;
988 } 848 }
849
850 FUBd;
989 } 851 }
990 852
991 return res; 853 return res;
992} 854}
993 855
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
994static int 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
995eio_dent_cmp (const void *a_, const void *b_) 1164eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
996{ 1165{
997 const eio_dirent *a = (const eio_dirent *)a_; 1166 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
998 const eio_dirent *b = (const eio_dirent *)b_; 1167 : a->inode < b->inode ? -1
1168 : a->inode > b->inode ? 1
1169 : 0;
1170}
999 1171
1000 return (int)b->score - (int)a->score ? (int)b->score - (int)a->score 1172#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1001 : a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0; /* int might be < ino_t */ 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);
1002} 1326}
1003 1327
1004/* read a full directory */ 1328/* read a full directory */
1005static void 1329static void
1006eio__scandir (eio_req *req, etp_worker *self) 1330eio__scandir (eio_req *req, etp_worker *self)
1007{ 1331{
1008 DIR *dirp;
1009 EIO_STRUCT_DIRENT *entp;
1010 unsigned char *name, *names; 1332 char *name, *names;
1011 int namesalloc = 4096; 1333 int namesalloc = 4096 - sizeof (void *) * 4;
1012 int namesoffs = 0; 1334 int namesoffs = 0;
1013 int flags = req->int1; 1335 int flags = req->int1;
1014 eio_dirent *dents = 0; 1336 eio_dirent *dents = 0;
1015 int dentalloc = 128; 1337 int dentalloc = 128;
1016 int dentoffs = 0; 1338 int dentoffs = 0;
1339 eio_ino_t inode_bits = 0;
1340#ifdef _WIN32
1341 HANDLE dirp;
1342 WIN32_FIND_DATA entp;
1343#else
1344 DIR *dirp;
1345 EIO_STRUCT_DIRENT *entp;
1346#endif
1017 1347
1018 req->result = -1; 1348 req->result = -1;
1019 1349
1020 if (!(flags & EIO_READDIR_DENTS)) 1350 if (!(flags & EIO_READDIR_DENTS))
1021 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); 1351 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1022 1352
1023 X_LOCK (wrklock); 1353#ifdef _WIN32
1024 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 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)
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
1025 self->dirp = dirp = opendir (req->ptr1); 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
1026 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1425 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1027 req->ptr1 = names = malloc (namesalloc);
1028 req->ptr2 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; 1426 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1029 X_UNLOCK (wrklock); 1427 req->ptr2 = names = malloc (namesalloc);
1030 1428
1031 if (dirp && names && (!flags || dents)) 1429 if (!names || (flags && !dents))
1430 return;
1431
1032 for (;;) 1432 for (;;)
1033 { 1433 {
1434 int done;
1435
1436#ifdef _WIN32
1437 done = !dirp;
1438#else
1034 errno = 0; 1439 errno = 0;
1035 entp = readdir (dirp); 1440 entp = readdir (dirp);
1441 done = !entp;
1442#endif
1036 1443
1037 if (!entp) 1444 if (done)
1038 { 1445 {
1446#ifndef _WIN32
1447 int old_errno = errno;
1448 closedir (dirp);
1449 errno = old_errno;
1450
1039 if (errno) 1451 if (errno)
1040 break; 1452 break;
1453#endif
1041 1454
1042 /* sort etc. */ 1455 /* sort etc. */
1043 req->int1 = flags; 1456 req->int1 = flags;
1044 req->result = dentoffs; 1457 req->result = dentoffs;
1045 1458
1046 if (dents)
1047 {
1048 eio_dirent *ent = dents + dentoffs;
1049
1050 while (ent > dents)
1051 (--ent)->name = names + (size_t)ent->name;
1052 }
1053
1054 if (flags & EIO_READDIR_STAT_ORDER 1459 if (flags & EIO_READDIR_STAT_ORDER)
1055 || !(~flags & (EIO_READDIR_DIRS_FIRST | EIO_READDIR_FOUND_UNKNOWN))) 1460 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1056 {
1057 /* pray your qsort doesn't use quicksort */
1058 qsort (dents, dentoffs, sizeof (*dents), eio_dent_cmp); /* score depends of DIRS_FIRST */
1059 }
1060 else if (flags & EIO_READDIR_DIRS_FIRST) 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
1061 { 1465 {
1062 /* in this case, all is known, and we just put dirs first and sort them */ 1466 /* in this case, all is known, and we just put dirs first and sort them */
1063 eio_dirent *ent = dents + dentoffs; 1467 eio_dirent *oth = dents + dentoffs;
1064 eio_dirent *dir = dents; 1468 eio_dirent *dir = dents;
1065 1469
1066 /* now move dirs to the front, and non-dirs to the back */ 1470 /* now partition dirs to the front, and non-dirs to the back */
1067 /* by walkign from both sides and swapping if necessary */ 1471 /* by walking from both sides and swapping if necessary */
1068 while (ent > dir) 1472 while (oth > dir)
1069 { 1473 {
1070 if (dir->type == DT_DIR) 1474 if (dir->type == EIO_DT_DIR)
1071 ++dir; 1475 ++dir;
1072 else 1476 else if ((--oth)->type == EIO_DT_DIR)
1073 { 1477 {
1074 --ent;
1075
1076 if (ent->type == DT_DIR)
1077 {
1078 eio_dirent tmp = *dir; 1478 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1079 *dir = *ent;
1080 *ent = tmp;
1081 1479
1082 ++dir; 1480 ++dir;
1083 }
1084 } 1481 }
1085 } 1482 }
1086 1483
1087 /* now sort the dirs only */ 1484 /* now sort the dirs only (dirs all have the same score) */
1088 qsort (dents, dir - dents, sizeof (*dents), eio_dent_cmp); 1485 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1089 } 1486 }
1090 1487
1091 /* only provide the names array unless DENTS is specified */
1092 if (!(flags & EIO_READDIR_DENTS))
1093 {
1094 X_LOCK (wrklock);
1095 assert (!dents);
1096 req->ptr1 = 0;
1097 req->ptr2 = names;
1098 X_UNLOCK (wrklock);
1099 }
1100
1101 break; 1488 break;
1102 } 1489 }
1103 1490
1104 /* now add the entry to our list(s) */ 1491 /* now add the entry to our list(s) */
1105 name = entp->d_name; 1492 name = D_NAME (entp);
1106 1493
1107 /* skip . and .. entries */ 1494 /* skip . and .. entries */
1108 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1495 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1109 { 1496 {
1110 int len = D_NAMLEN (entp) + 1; 1497 int len = D_NAMLEN (entp) + 1;
1111 1498
1112 while (expect_false (namesoffs + len > namesalloc)) 1499 while (ecb_expect_false (namesoffs + len > namesalloc))
1113 { 1500 {
1114 namesalloc *= 2; 1501 namesalloc *= 2;
1115 X_LOCK (wrklock);
1116 req->ptr1 = names = realloc (names, namesalloc); 1502 req->ptr2 = names = realloc (names, namesalloc);
1117 X_UNLOCK (wrklock);
1118 1503
1119 if (!names) 1504 if (!names)
1120 break; 1505 break;
1121 } 1506 }
1122 1507
1123 memcpy (names + namesoffs, name, len); 1508 memcpy (names + namesoffs, name, len);
1124 1509
1125 if (dents) 1510 if (dents)
1126 { 1511 {
1127 struct eio_dirent *ent; 1512 struct eio_dirent *ent;
1128 1513
1129 if (expect_false (dentoffs == dentalloc)) 1514 if (ecb_expect_false (dentoffs == dentalloc))
1130 { 1515 {
1131 dentalloc *= 2; 1516 dentalloc *= 2;
1132 X_LOCK (wrklock);
1133 req->ptr2 = dents = realloc (dents, dentalloc * sizeof (eio_dirent)); 1517 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1134 X_UNLOCK (wrklock);
1135 1518
1136 if (!dents) 1519 if (!dents)
1137 break; 1520 break;
1138 } 1521 }
1139 1522
1140 ent = dents + dentoffs; 1523 ent = dents + dentoffs;
1141 1524
1142 ent->name = (char *)(size_t)namesoffs; /* rather dirtily we store the offset in the pointer */ 1525 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1143 ent->namelen = len - 1; 1526 ent->namelen = len - 1;
1144 ent->inode = D_INO (entp); 1527 ent->inode = D_INO (entp);
1145 1528
1529 inode_bits |= ent->inode;
1530
1146 switch (D_TYPE (entp)) 1531 switch (D_TYPE (entp))
1147 { 1532 {
1148 default: 1533 default:
1149 ent->type = EIO_DT_UNKNOWN; 1534 ent->type = EIO_DT_UNKNOWN;
1150 flags |= EIO_READDIR_FOUND_UNKNOWN; 1535 flags |= EIO_READDIR_FOUND_UNKNOWN;
1151 break; 1536 break;
1152 1537
1153 #ifdef DT_FIFO 1538 #ifdef DT_FIFO
1154 case DT_FIFO: ent->type = EIO_DT_FIFO; break; 1539 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1155 #endif 1540 #endif
1156 #ifdef DT_CHR 1541 #ifdef DT_CHR
1157 case DT_CHR: ent->type = EIO_DT_CHR; break; 1542 case DT_CHR: ent->type = EIO_DT_CHR; break;
1158 #endif 1543 #endif
1159 #ifdef DT_MPC 1544 #ifdef DT_MPC
1160 case DT_MPC: ent->type = EIO_DT_MPC; break; 1545 case DT_MPC: ent->type = EIO_DT_MPC; break;
1161 #endif 1546 #endif
1162 #ifdef DT_DIR 1547 #ifdef DT_DIR
1163 case DT_DIR: ent->type = EIO_DT_DIR; break; 1548 case DT_DIR: ent->type = EIO_DT_DIR; break;
1164 #endif 1549 #endif
1165 #ifdef DT_NAM 1550 #ifdef DT_NAM
1166 case DT_NAM: ent->type = EIO_DT_NAM; break; 1551 case DT_NAM: ent->type = EIO_DT_NAM; break;
1167 #endif 1552 #endif
1168 #ifdef DT_BLK 1553 #ifdef DT_BLK
1169 case DT_BLK: ent->type = EIO_DT_BLK; break; 1554 case DT_BLK: ent->type = EIO_DT_BLK; break;
1170 #endif 1555 #endif
1171 #ifdef DT_MPB 1556 #ifdef DT_MPB
1172 case DT_MPB: ent->type = EIO_DT_MPB; break; 1557 case DT_MPB: ent->type = EIO_DT_MPB; break;
1173 #endif 1558 #endif
1174 #ifdef DT_REG 1559 #ifdef DT_REG
1175 case DT_REG: ent->type = EIO_DT_REG; break; 1560 case DT_REG: ent->type = EIO_DT_REG; break;
1176 #endif 1561 #endif
1177 #ifdef DT_NWK 1562 #ifdef DT_NWK
1178 case DT_NWK: ent->type = EIO_DT_NWK; break; 1563 case DT_NWK: ent->type = EIO_DT_NWK; break;
1179 #endif 1564 #endif
1180 #ifdef DT_CMP 1565 #ifdef DT_CMP
1181 case DT_CMP: ent->type = EIO_DT_CMP; break; 1566 case DT_CMP: ent->type = EIO_DT_CMP; break;
1182 #endif 1567 #endif
1183 #ifdef DT_LNK 1568 #ifdef DT_LNK
1184 case DT_LNK: ent->type = EIO_DT_LNK; break; 1569 case DT_LNK: ent->type = EIO_DT_LNK; break;
1185 #endif 1570 #endif
1186 #ifdef DT_SOCK 1571 #ifdef DT_SOCK
1187 case DT_SOCK: ent->type = EIO_DT_SOCK; break; 1572 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1188 #endif 1573 #endif
1189 #ifdef DT_DOOR 1574 #ifdef DT_DOOR
1190 case DT_DOOR: ent->type = EIO_DT_DOOR; break; 1575 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1191 #endif 1576 #endif
1192 #ifdef DT_WHT 1577 #ifdef DT_WHT
1193 case DT_WHT: ent->type = EIO_DT_WHT; break; 1578 case DT_WHT: ent->type = EIO_DT_WHT; break;
1194 #endif 1579 #endif
1195 } 1580 }
1196 1581
1197 ent->score = 0; 1582 ent->score = 7;
1198 1583
1199 if (flags & EIO_READDIR_DIRS_FIRST) 1584 if (flags & EIO_READDIR_DIRS_FIRST)
1200 { 1585 {
1201 if (ent->type == EIO_DT_UNKNOWN) 1586 if (ent->type == EIO_DT_UNKNOWN)
1202 { 1587 {
1203 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */ 1588 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1204 ent->score = 98; 1589 ent->score = 1;
1205 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */ 1590 else if (!strchr (name, '.')) /* absence of dots indicate likely dirs */
1206 ent->score = len <= 2 ? len + 6 : len <= 4 ? 5 : len <= 7 ? 4 : 1; /* shorter == more likely dir, but avoid too many classes */ 1591 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1207 } 1592 }
1208 else if (ent->type == EIO_DT_DIR) 1593 else if (ent->type == EIO_DT_DIR)
1209 ent->score = 100; 1594 ent->score = 0;
1210 } 1595 }
1211 } 1596 }
1212 1597
1213 namesoffs += len; 1598 namesoffs += len;
1214 ++dentoffs; 1599 ++dentoffs;
1215 } 1600 }
1216 1601
1217 if (EIO_CANCELLED (req)) 1602 if (EIO_CANCELLED (req))
1218 { 1603 {
1219 errno = ECANCELED; 1604 errno = ECANCELED;
1220 break; 1605 break;
1221 } 1606 }
1607
1608#ifdef _WIN32
1609 if (!FindNextFile (dirp, &entp))
1610 {
1611 FindClose (dirp);
1612 dirp = 0;
1222 } 1613 }
1223}
1224
1225#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1226# undef msync
1227# define msync(a,b,c) ((errno = ENOSYS), -1)
1228#endif 1614#endif
1615 }
1616}
1229 1617
1230int 1618/*****************************************************************************/
1231eio__mtouch (void *mem, size_t len, int flags) 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)
1232{ 1630{
1233 intptr_t addr = (intptr_t)mem; 1631 if (!wd || *path == '/')
1234 intptr_t end = addr + len; 1632 return path;
1235#ifdef PAGESIZE 1633
1236 const intptr_t page = PAGESIZE; 1634 if (path [0] == '.' && !path [1])
1635 return wd->str;
1636
1637 {
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 }
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);
1237#else 1712#else
1238 static intptr_t page;
1239
1240 if (!page) 1713 if (flags)
1241 page = sysconf (_SC_PAGESIZE); 1714 return EIO_ENOSYS ();
1242#endif
1243 1715
1244 addr &= ~(page - 1); /* assume page size is always a power of two */ 1716 return renameat (olddirfd, oldpath, newdirfd, newpath);
1717#endif
1718}
1245 1719
1246 if (addr < end) 1720/* they forgot these */
1247 if (flags) /* modify */
1248 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
1249 else
1250 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
1251 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);
1252 return 0; 1733 return res;
1253} 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
1254 1751
1255/*****************************************************************************/ 1752/*****************************************************************************/
1256 1753
1257#define ALLOC(len) \ 1754#define ALLOC(len) \
1258 if (!req->ptr2) \ 1755 if (!req->ptr2) \
1259 { \ 1756 { \
1260 X_LOCK (wrklock); \ 1757 X_LOCK (EIO_POOL->wrklock); \
1261 req->flags |= EIO_FLAG_PTR2_FREE; \ 1758 req->flags |= EIO_FLAG_PTR2_FREE; \
1262 X_UNLOCK (wrklock); \ 1759 X_UNLOCK (EIO_POOL->wrklock); \
1263 req->ptr2 = malloc (len); \ 1760 req->ptr2 = malloc (len); \
1264 if (!req->ptr2) \ 1761 if (!req->ptr2) \
1265 { \ 1762 { \
1266 errno = ENOMEM; \ 1763 errno = ENOMEM; \
1267 req->result = -1; \ 1764 req->result = -1; \
1268 break; \ 1765 goto alloc_fail; \
1269 } \ 1766 } \
1270 } 1767 }
1271 1768
1272X_THREAD_PROC (etp_proc) 1769/*****************************************************************************/
1273{
1274 ETP_REQ *req;
1275 struct timespec ts;
1276 etp_worker *self = (etp_worker *)thr_arg;
1277 1770
1278 /* try to distribute timeouts somewhat randomly */ 1771static void
1279 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 1772eio__slurp (int fd, eio_req *req)
1773{
1774 req->result = fd;
1280 1775
1281 for (;;) 1776 if (fd < 0)
1777 return;
1778
1779 if (req->offs < 0 || !req->size) /* do we need the size? */
1282 { 1780 {
1283 X_LOCK (reqlock); 1781 off_t size = lseek (fd, 0, SEEK_END);
1284 1782
1285 for (;;)
1286 {
1287 self->req = req = reqq_shift (&req_queue);
1288
1289 if (req)
1290 break;
1291
1292 ++idle;
1293
1294 ts.tv_sec = time (0) + IDLE_TIMEOUT;
1295 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
1296 {
1297 if (idle > max_idle)
1298 {
1299 --idle;
1300 X_UNLOCK (reqlock);
1301 X_LOCK (wrklock);
1302 --started;
1303 X_UNLOCK (wrklock);
1304 goto quit;
1305 }
1306
1307 /* we are allowed to idle, so do so without any timeout */
1308 X_COND_WAIT (reqwait, reqlock);
1309 }
1310
1311 --idle;
1312 }
1313
1314 --nready;
1315
1316 X_UNLOCK (reqlock);
1317
1318 if (req->type < 0) 1783 if (req->offs < 0)
1319 goto quit; 1784 req->offs += size;
1320 1785
1321 if (!EIO_CANCELLED (req)) 1786 if (!req->size)
1322 ETP_EXECUTE (self, req); 1787 req->size = size - req->offs;
1323
1324 X_LOCK (reslock);
1325
1326 ++npending;
1327
1328 if (!reqq_push (&res_queue, req) && want_poll_cb)
1329 want_poll_cb ();
1330
1331 self->req = 0;
1332 etp_worker_clear (self);
1333
1334 X_UNLOCK (reslock);
1335 } 1788 }
1336 1789
1337quit: 1790 ALLOC (req->size);
1338 X_LOCK (wrklock); 1791 req->result = pread (fd, req->ptr2, req->size, req->offs);
1339 etp_worker_free (self);
1340 X_UNLOCK (wrklock);
1341 1792
1342 return 0; 1793 silent_close (fd);
1343}
1344 1794
1345/*****************************************************************************/ 1795alloc_fail:
1796 ;
1797}
1346 1798
1799int ecb_cold
1347int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1800eio_init (void (*want_poll)(void), void (*done_poll)(void))
1348{ 1801{
1349 return etp_init (want_poll, done_poll); 1802 eio_want_poll_cb = want_poll;
1350} 1803 eio_done_poll_cb = done_poll;
1351 1804
1805 return etp_init (EIO_POOL, 0, 0, 0);
1806}
1807
1808ecb_inline void
1352static void eio_api_destroy (eio_req *req) 1809eio_api_destroy (eio_req *req)
1353{ 1810{
1354 free (req); 1811 free (req);
1355} 1812}
1356 1813
1357#define REQ(rtype) \ 1814#define REQ(rtype) \
1358 eio_req *req; \ 1815 eio_req *req; \
1359 \ 1816 \
1360 req = (eio_req *)calloc (1, sizeof *req); \ 1817 req = (eio_req *)calloc (1, sizeof *req); \
1361 if (!req) \ 1818 if (!req) \
1362 return 0; \ 1819 return 0; \
1376 { \ 1833 { \
1377 eio_api_destroy (req); \ 1834 eio_api_destroy (req); \
1378 return 0; \ 1835 return 0; \
1379 } 1836 }
1380 1837
1838#define SINGLEDOT(ptr) (0[(char *)(ptr)] == '.' && !1[(char *)(ptr)])
1839
1840static void
1381static void eio_execute (etp_worker *self, eio_req *req) 1841eio_execute (etp_worker *self, eio_req *req)
1382{ 1842{
1383 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 }
1384 1871
1385 switch (req->type) 1872 switch (req->type)
1386 { 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;
1387 case EIO_READ: ALLOC (req->size); 1881 case EIO_READ: ALLOC (req->size);
1388 req->result = req->offs >= 0 1882 req->result = req->offs >= 0
1389 ? pread (req->int1, req->ptr2, req->size, req->offs) 1883 ? pread (req->int1, req->ptr2, req->size, req->offs)
1390 : read (req->int1, req->ptr2, req->size); break; 1884 : read (req->int1, req->ptr2, req->size); break;
1391 case EIO_WRITE: req->result = req->offs >= 0 1885 case EIO_WRITE: req->result = req->offs >= 0
1392 ? pwrite (req->int1, req->ptr2, req->size, req->offs) 1886 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1393 : write (req->int1, req->ptr2, req->size); break; 1887 : write (req->int1, req->ptr2, req->size); break;
1394 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
1395 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;
1396 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
1397 1896
1398 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1897 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1399 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;
1400 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1899 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1401 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;
1402 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1403 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1404
1405 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;
1406 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1407 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;
1408 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1409 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;
1410 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1411
1412 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;
1413 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;
1414 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break; 1906
1415 case EIO_UNLINK: req->result = unlink (req->ptr1); break; 1907 case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
1416 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;
1417 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;
1418 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;
1419 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;
1420 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; 1923 case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
1421 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break; 1924 case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1422 1925 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1926 req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1423 case EIO_READLINK: ALLOC (NAME_MAX); 1927 case EIO_READLINK: ALLOC (EIO_PATH_MAX);
1424 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; 1928 req->result = readlinkat (dirfd, req->ptr1, req->ptr2, EIO_PATH_MAX);
1425 1929 if (req->result == EIO_PATH_MAX)
1426 case EIO_SYNC: req->result = 0; sync (); break; 1930 {
1427 case EIO_FSYNC: req->result = fsync (req->int1); break; 1931 req->result = -1;
1428 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 1932 errno = ENAMETOOLONG;
1429 case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break; 1933 }
1430 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break; 1934 break;
1431 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1432
1433 case EIO_READDIR: eio__scandir (req, self); break;
1434
1435 case EIO_BUSY: 1935 case EIO_UTIME:
1436#ifdef _WIN32 1936 case EIO_FUTIME:
1437 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
1438#else 1959#else
1439 {
1440 struct timeval tv;
1441 1960
1442 tv.tv_sec = req->nv1; 1961 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1443 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;
1444 1970
1445 req->result = select (0, 0, 0, 0, &tv); 1971 case EIO_UNLINK: req->result = unlink (path ); break;
1446 } 1972 case EIO_RMDIR: req->result = rmdir (path ); break;
1447#endif 1973 case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
1448 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;
1449 1988
1450 case EIO_UTIME: 1989 case EIO_UTIME:
1451 case EIO_FUTIME: 1990 case EIO_FUTIME:
1452 { 1991 {
1453 struct timeval tv[2]; 1992 struct timeval tv[2];
1454 struct timeval *times; 1993 struct timeval *times;
1455 1994
1456 if (req->nv1 != -1. || req->nv2 != -1.) 1995 if (req->nv1 != -1. || req->nv2 != -1.)
1457 { 1996 {
1458 tv[0].tv_sec = req->nv1; 1997 tv[0].tv_sec = req->nv1;
1459 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.; 1998 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
1460 tv[1].tv_sec = req->nv2; 1999 tv[1].tv_sec = req->nv2;
1461 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.; 2000 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
1462 2001
1463 times = tv; 2002 times = tv;
1464 } 2003 }
1465 else 2004 else
1466 times = 0; 2005 times = 0;
1467
1468 2006
1469 req->result = req->type == EIO_FUTIME 2007 req->result = req->type == EIO_FUTIME
1470 ? futimes (req->int1, times) 2008 ? futimes (req->int1, times)
1471 : utimes (req->ptr1, times); 2009 : utimes (req->ptr1, times);
1472 } 2010 }
1473 break; 2011 break;
1474 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
1475 case EIO_GROUP: 2063 case EIO_GROUP:
1476 abort (); /* handled in eio_request */ 2064 abort (); /* handled in eio_request */
2065#endif
1477 2066
1478 case EIO_NOP: 2067 case EIO_NOP:
1479 req->result = 0; 2068 req->result = 0;
1480 break; 2069 break;
1481 2070
1482 case EIO_CUSTOM: 2071 case EIO_CUSTOM:
1483 ((void (*)(eio_req *))req->feed) (req); 2072 req->feed (req);
1484 break; 2073 break;
1485 2074
1486 default: 2075 default:
1487 req->result = -1; 2076 req->result = EIO_ENOSYS ();
1488 break; 2077 break;
1489 } 2078 }
1490 2079
2080alloc_fail:
1491 req->errorno = errno; 2081 req->errorno = errno;
1492} 2082}
1493 2083
1494#ifndef EIO_NO_WRAPPERS 2084#ifndef EIO_NO_WRAPPERS
1495 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
1496eio_req *eio_nop (int pri, eio_cb cb, void *data) 2096eio_req *eio_nop (int pri, eio_cb cb, void *data)
1497{ 2097{
1498 REQ (EIO_NOP); SEND; 2098 REQ (EIO_NOP); SEND;
1499} 2099}
1500 2100
1516eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2116eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1517{ 2117{
1518 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2118 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1519} 2119}
1520 2120
2121eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2122{
2123 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2124}
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
1521eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2136eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1522{ 2137{
1523 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2138 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1524} 2139}
1525 2140
1526eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) 2141eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
1527{ 2142{
1528 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; 2143 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
1529} 2144}
1530 2145
1531eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) 2146eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
1532{ 2147{
1533 REQ (EIO_FDATASYNC); req->int1 = fd; SEND; 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;
1534} 2154}
1535 2155
1536eio_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)
1537{ 2157{
1538 REQ (EIO_CLOSE); req->int1 = fd; SEND; 2158 REQ (EIO_CLOSE); req->int1 = fd; SEND;
1541eio_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)
1542{ 2162{
1543 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;
1544} 2164}
1545 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
1546eio_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)
1547{ 2172{
1548 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;
1549} 2174}
1550 2175
1551eio_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)
1552{ 2177{
1553 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;
1554} 2179}
1555 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
1556eio_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)
1557{ 2192{
1558 REQ (EIO_FSTAT); req->int1 = fd; SEND; 2193 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1559} 2194}
1560 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
1561eio_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)
1562{ 2202{
1563 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;
1564} 2204}
1565 2205
1571eio_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)
1572{ 2212{
1573 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2213 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1574} 2214}
1575 2215
1576eio_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)
1577{ 2217{
1578 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;
1579} 2219}
1580 2220
1581eio_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)
1601eio_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)
1602{ 2242{
1603 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2243 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1604} 2244}
1605 2245
1606eio_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)
1607{ 2247{
1608 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;
1609} 2249}
1610 2250
1611eio_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)
1627eio_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)
1628{ 2268{
1629 return eio__1path (EIO_READLINK, path, pri, cb, data); 2269 return eio__1path (EIO_READLINK, path, pri, cb, data);
1630} 2270}
1631 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
1632eio_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)
1633{ 2278{
1634 return eio__1path (EIO_STAT, path, pri, cb, data); 2279 return eio__1path (EIO_STAT, path, pri, cb, data);
1635} 2280}
1636 2281
1637eio_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)
1638{ 2283{
1639 return eio__1path (EIO_LSTAT, path, pri, cb, data); 2284 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1640} 2285}
1641 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
1642eio_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)
1643{ 2293{
1644 return eio__1path (EIO_UNLINK, path, pri, cb, data); 2294 return eio__1path (EIO_UNLINK, path, pri, cb, data);
1645} 2295}
1646 2296
1654 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND; 2304 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1655} 2305}
1656 2306
1657eio_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)
1658{ 2308{
1659 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; 2309 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1660} 2310}
1661 2311
1662static eio_req * 2312static eio_req *
1663eio__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)
1664{ 2314{
1688eio_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)
1689{ 2339{
1690 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2340 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1691} 2341}
1692 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
1693eio_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)
1694{ 2349{
1695 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND; 2350 REQ (EIO_CUSTOM); req->feed = execute; SEND;
1696} 2351}
1697 2352
1698#endif 2353#endif
1699 2354
1700eio_req *eio_grp (eio_cb cb, void *data) 2355eio_req *eio_grp (eio_cb cb, void *data)
1709#undef SEND 2364#undef SEND
1710 2365
1711/*****************************************************************************/ 2366/*****************************************************************************/
1712/* grp functions */ 2367/* grp functions */
1713 2368
2369void
1714void 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)
1715{ 2371{
1716 grp->int2 = limit; 2372 grp->int2 = limit;
1717 grp->feed = feed; 2373 grp->feed = feed;
1718 2374
1719 grp_try_feed (grp); 2375 grp_try_feed (grp);
1720} 2376}
1721 2377
2378void
1722void eio_grp_limit (eio_req *grp, int limit) 2379eio_grp_limit (eio_req *grp, int limit)
1723{ 2380{
1724 grp->int2 = limit; 2381 grp->int2 = limit;
1725 2382
1726 grp_try_feed (grp); 2383 grp_try_feed (grp);
1727} 2384}
1728 2385
2386void
1729void eio_grp_add (eio_req *grp, eio_req *req) 2387eio_grp_add (eio_req *grp, eio_req *req)
1730{ 2388{
1731 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));
1732 2390
1733 grp->flags |= EIO_FLAG_GROUPADD; 2391 grp->flags |= ETP_FLAG_GROUPADD;
1734 2392
1735 ++grp->size; 2393 ++grp->size;
1736 req->grp = grp; 2394 req->grp = grp;
1737 2395
1738 req->grp_prev = 0; 2396 req->grp_prev = 0;
1745} 2403}
1746 2404
1747/*****************************************************************************/ 2405/*****************************************************************************/
1748/* misc garbage */ 2406/* misc garbage */
1749 2407
2408eio_ssize_t
1750ssize_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)
1751{ 2410{
1752 etp_worker wrk;
1753
1754 wrk.dbuf = 0;
1755
1756 eio__sendfile (ofd, ifd, offset, count, &wrk); 2411 return eio__sendfile (ofd, ifd, offset, count);
1757
1758 if (wrk.dbuf)
1759 free (wrk.dbuf);
1760} 2412}
1761 2413

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines