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

Comparing libeio/eio.c (file contents):
Revision 1.85 by root, Wed Jul 13 00:53:03 2011 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) 2007,2008,2009,2010,2011 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,
… …
42#endif 42#endif
43 43
44#include "eio.h" 44#include "eio.h"
45#include "ecb.h" 45#include "ecb.h"
46 46
47#ifdef EIO_STACKSIZE
48# define XTHREAD_STACKSIZE EIO_STACKSIZE
49#endif
50#include "xthread.h"
51
52#include <errno.h> 47#include <errno.h>
53#include <stddef.h> 48#include <stddef.h>
54#include <stdlib.h> 49#include <stdlib.h>
55#include <string.h> 50#include <string.h>
56#include <errno.h> 51#include <errno.h>
57#include <sys/types.h> 52#include <sys/types.h>
58#include <sys/stat.h> 53#include <sys/stat.h>
59#include <sys/statvfs.h>
60#include <limits.h> 54#include <limits.h>
61#include <fcntl.h> 55#include <fcntl.h>
62#include <assert.h> 56#include <assert.h>
63 57
64/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */ 58/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
… …
68#endif 62#endif
69 63
70#ifndef ECANCELED 64#ifndef ECANCELED
71# define ECANCELED EDOM 65# define ECANCELED EDOM
72#endif 66#endif
67#ifndef ELOOP
68# define ELOOP EDOM
69#endif
70
71#if !defined(ENOTSOCK) && defined(WSAENOTSOCK)
72# define ENOTSOCK WSAENOTSOCK
73#endif
73 74
74static void eio_destroy (eio_req *req); 75static void eio_destroy (eio_req *req);
75 76
76#ifndef EIO_FINISH 77#ifndef EIO_FINISH
77# 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
… …
83 84
84#ifndef EIO_FEED 85#ifndef EIO_FEED
85# 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)
86#endif 87#endif
87 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
88#ifdef _WIN32 100#ifdef _WIN32
89 101
90 /*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
91#else 234#else
92 235
236 #include <sys/ioctl.h>
93# include <sys/time.h> 237 #include <sys/time.h>
94# include <sys/select.h> 238 #include <sys/select.h>
95# 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
96# include <utime.h> 283# include <utime.h>
97# include <signal.h>
98# include <dirent.h>
99
100#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
101# include <sys/mman.h>
102#endif
103
104/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
105# if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
106# define _DIRENT_HAVE_D_TYPE /* sigh */
107# define D_INO(de) (de)->d_fileno
108# define D_NAMLEN(de) (de)->d_namlen
109# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
110# define D_INO(de) (de)->d_ino
111# endif 284#endif
112 285
113#ifdef _D_EXACT_NAMLEN 286#if HAVE_SYS_SYSCALL_H
114# undef D_NAMLEN 287# include <sys/syscall.h>
115# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
116#endif
117
118# ifdef _DIRENT_HAVE_D_TYPE
119# define D_TYPE(de) (de)->d_type
120# endif
121
122# ifndef EIO_STRUCT_DIRENT
123# define EIO_STRUCT_DIRENT struct dirent
124# endif
125
126#endif 288#endif
127 289
128#if HAVE_SENDFILE 290#if HAVE_SENDFILE
129# if __linux 291# if __linux
130# include <sys/sendfile.h> 292# include <sys/sendfile.h>
… …
138# else 300# else
139# error sendfile support requested but not available 301# error sendfile support requested but not available
140# endif 302# endif
141#endif 303#endif
142 304
305#if HAVE_RENAMEAT2
306# include <sys/syscall.h>
307# include <linux/fs.h>
308#endif
309
143#ifndef D_TYPE 310#ifndef D_TYPE
144# define D_TYPE(de) 0 311# define D_TYPE(de) 0
145#endif 312#endif
146#ifndef D_INO 313#ifndef D_INO
147# define D_INO(de) 0 314# define D_INO(de) 0
148#endif 315#endif
149#ifndef D_NAMLEN 316#ifndef D_NAMLEN
150# define D_NAMLEN(de) strlen ((de)->d_name) 317# define D_NAMLEN(entp) strlen (D_NAME (entp))
151#endif 318#endif
152 319
153/* used for struct dirent, AIX doesn't provide it */ 320/* used for struct dirent, AIX doesn't provide it */
154#ifndef NAME_MAX 321#ifndef NAME_MAX
155# define NAME_MAX 4096 322# define NAME_MAX 4096
156#endif 323#endif
157 324
158/* used for readlink etc. */ 325/* used for readlink etc. */
159#ifndef PATH_MAX 326#ifndef PATH_MAX
160# define PATH_MAX 4096 327# define PATH_MAX 0
161#endif 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)
162 339
163/* buffer size for various temporary buffers */ 340/* buffer size for various temporary buffers */
164#define EIO_BUFSIZE 65536 341#define EIO_BUFSIZE 65536
165 342
166#define dBUF \ 343#define dBUF \
167 char *eio_buf; \
168 ETP_WORKER_LOCK (self); \
169 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 344 char *eio_buf = malloc (EIO_BUFSIZE); \
170 ETP_WORKER_UNLOCK (self); \
171 errno = ENOMEM; \ 345 errno = ENOMEM; \
172 if (!eio_buf) \ 346 if (!eio_buf) \
173 return -1; 347 return -1
174 348
175#define EIO_TICKS ((1000000 + 1023) >> 10) 349#define FUBd \
350 free (eio_buf)
351
352/*****************************************************************************/
353
354struct etp_tmpbuf;
355
356#if _POSIX_VERSION >= 200809L
357 #define HAVE_AT 1
358 #define WD2FD(wd) ((wd) ? (wd)->fd : AT_FDCWD)
359 #ifndef O_SEARCH
360 #define O_SEARCH O_RDONLY
361 #endif
362#else
363 #define HAVE_AT 0
364 static const char *wd_expand (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path);
365#endif
366
367struct eio_pwd
368{
369#if HAVE_AT
370 int fd;
371#endif
372 int len;
373 char str[1]; /* actually, a 0-terminated canonical path */
374};
375
376/*****************************************************************************/
176 377
177#define ETP_PRI_MIN EIO_PRI_MIN 378#define ETP_PRI_MIN EIO_PRI_MIN
178#define ETP_PRI_MAX EIO_PRI_MAX 379#define ETP_PRI_MAX EIO_PRI_MAX
179 380
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
180struct etp_worker; 391struct etp_worker;
181
182#define ETP_REQ eio_req 392#define ETP_REQ eio_req
183#define ETP_DESTROY(req) eio_destroy (req) 393#define ETP_DESTROY(req) eio_destroy (req)
184static int eio_finish (eio_req *req); 394static int eio_finish (eio_req *req);
185#define ETP_FINISH(req) eio_finish (req) 395#define ETP_FINISH(req) eio_finish (req)
186static void eio_execute (struct etp_worker *self, eio_req *req); 396static void eio_execute (struct etp_worker *self, eio_req *req);
187#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) 397#define ETP_EXECUTE(wrk,req) eio_execute (wrk, req)
188 398
189#define ETP_WORKER_CLEAR(req) \ 399#include "etp.c"
190 if (wrk->dbuf) \
191 { \
192 free (wrk->dbuf); \
193 wrk->dbuf = 0; \
194 } \
195 \
196 if (wrk->dirp) \
197 { \
198 closedir (wrk->dirp); \
199 wrk->dirp = 0; \
200 }
201 400
202#define ETP_WORKER_COMMON \ 401static struct etp_pool eio_pool;
203 void *dbuf; \ 402#define EIO_POOL (&eio_pool)
204 DIR *dirp;
205
206/*****************************************************************************/
207
208#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
209
210/* calculate time difference in ~1/EIO_TICKS of a second */
211ecb_inline int
212tvdiff (struct timeval *tv1, struct timeval *tv2)
213{
214 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
215 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
216}
217
218static unsigned int started, idle, wanted = 4;
219
220static void (*want_poll_cb) (void);
221static void (*done_poll_cb) (void);
222
223static unsigned int max_poll_time; /* reslock */
224static unsigned int max_poll_reqs; /* reslock */
225
226static volatile unsigned int nreqs; /* reqlock */
227static volatile unsigned int nready; /* reqlock */
228static volatile unsigned int npending; /* reqlock */
229static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
230static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
231
232static xmutex_t wrklock;
233static xmutex_t reslock;
234static xmutex_t reqlock;
235static xcond_t reqwait;
236
237#if !HAVE_PREADWRITE
238/*
239 * make our pread/pwrite emulation safe against themselves, but not against
240 * normal read/write by using a mutex. slows down execution a lot,
241 * but that's your problem, not mine.
242 */
243static xmutex_t preadwritelock = X_MUTEX_INIT;
244#endif
245
246typedef struct etp_worker
247{
248 /* locked by wrklock */
249 struct etp_worker *prev, *next;
250
251 xthread_t tid;
252
253 /* locked by reslock, reqlock or wrklock */
254 ETP_REQ *req; /* currently processed request */
255
256 ETP_WORKER_COMMON
257} etp_worker;
258
259static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */
260
261#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
262#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
263
264/* worker threads management */
265
266static void ecb_cold
267etp_worker_clear (etp_worker *wrk)
268{
269 ETP_WORKER_CLEAR (wrk);
270}
271
272static void ecb_cold
273etp_worker_free (etp_worker *wrk)
274{
275 wrk->next->prev = wrk->prev;
276 wrk->prev->next = wrk->next;
277
278 free (wrk);
279}
280
281static unsigned int
282etp_nreqs (void)
283{
284 int retval;
285 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
286 retval = nreqs;
287 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
288 return retval;
289}
290
291static unsigned int
292etp_nready (void)
293{
294 unsigned int retval;
295
296 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
297 retval = nready;
298 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
299
300 return retval;
301}
302
303static unsigned int
304etp_npending (void)
305{
306 unsigned int retval;
307
308 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
309 retval = npending;
310 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
311
312 return retval;
313}
314
315static unsigned int
316etp_nthreads (void)
317{
318 unsigned int retval;
319
320 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
321 retval = started;
322 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
323
324 return retval;
325}
326
327/*
328 * a somewhat faster data structure might be nice, but
329 * with 8 priorities this actually needs <20 insns
330 * per shift, the most expensive operation.
331 */
332typedef struct {
333 ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */
334 int size;
335} etp_reqq;
336
337static etp_reqq req_queue;
338static etp_reqq res_queue;
339
340static int ecb_noinline
341reqq_push (etp_reqq *q, ETP_REQ *req)
342{
343 int pri = req->pri;
344 req->next = 0;
345
346 if (q->qe[pri])
347 {
348 q->qe[pri]->next = req;
349 q->qe[pri] = req;
350 }
351 else
352 q->qe[pri] = q->qs[pri] = req;
353
354 return q->size++;
355}
356
357static ETP_REQ * ecb_noinline
358reqq_shift (etp_reqq *q)
359{
360 int pri;
361
362 if (!q->size)
363 return 0;
364
365 --q->size;
366
367 for (pri = ETP_NUM_PRI; pri--; )
368 {
369 eio_req *req = q->qs[pri];
370
371 if (req)
372 {
373 if (!(q->qs[pri] = (eio_req *)req->next))
374 q->qe[pri] = 0;
375
376 return req;
377 }
378 }
379
380 abort ();
381}
382
383static void ecb_cold
384etp_thread_init (void)
385{
386#if !HAVE_PREADWRITE
387 X_MUTEX_CREATE (preadwritelock);
388#endif
389 X_MUTEX_CREATE (wrklock);
390 X_MUTEX_CREATE (reslock);
391 X_MUTEX_CREATE (reqlock);
392 X_COND_CREATE (reqwait);
393}
394
395static void ecb_cold
396etp_atfork_prepare (void)
397{
398}
399
400static void ecb_cold
401etp_atfork_parent (void)
402{
403}
404
405static void ecb_cold
406etp_atfork_child (void)
407{
408 ETP_REQ *prv;
409
410 while ((prv = reqq_shift (&req_queue)))
411 ETP_DESTROY (prv);
412
413 while ((prv = reqq_shift (&res_queue)))
414 ETP_DESTROY (prv);
415
416 while (wrk_first.next != &wrk_first)
417 {
418 etp_worker *wrk = wrk_first.next;
419
420 if (wrk->req)
421 ETP_DESTROY (wrk->req);
422
423 etp_worker_clear (wrk);
424 etp_worker_free (wrk);
425 }
426
427 started = 0;
428 idle = 0;
429 nreqs = 0;
430 nready = 0;
431 npending = 0;
432
433 etp_thread_init ();
434}
435
436static void ecb_cold
437etp_once_init (void)
438{
439 etp_thread_init ();
440 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
441}
442
443static int ecb_cold
444etp_init (void (*want_poll)(void), void (*done_poll)(void))
445{
446 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
447
448 pthread_once (&doinit, etp_once_init);
449
450 want_poll_cb = want_poll;
451 done_poll_cb = done_poll;
452
453 return 0;
454}
455
456X_THREAD_PROC (etp_proc);
457
458static void ecb_cold
459etp_start_thread (void)
460{
461 etp_worker *wrk = calloc (1, sizeof (etp_worker));
462
463 /*TODO*/
464 assert (("unable to allocate worker thread data", wrk));
465
466 X_LOCK (wrklock);
467
468 if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
469 {
470 wrk->prev = &wrk_first;
471 wrk->next = wrk_first.next;
472 wrk_first.next->prev = wrk;
473 wrk_first.next = wrk;
474 ++started;
475 }
476 else
477 free (wrk);
478
479 X_UNLOCK (wrklock);
480}
481
482static void
483etp_maybe_start_thread (void)
484{
485 if (ecb_expect_true (etp_nthreads () >= wanted))
486 return;
487
488 /* todo: maybe use idle here, but might be less exact */
489 if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
490 return;
491
492 etp_start_thread ();
493}
494
495static void ecb_cold
496etp_end_thread (void)
497{
498 eio_req *req = calloc (1, sizeof (eio_req));
499
500 req->type = -1;
501 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
502
503 X_LOCK (reqlock);
504 reqq_push (&req_queue, req);
505 X_COND_SIGNAL (reqwait);
506 X_UNLOCK (reqlock);
507
508 X_LOCK (wrklock);
509 --started;
510 X_UNLOCK (wrklock);
511}
512
513static int
514etp_poll (void)
515{
516 unsigned int maxreqs;
517 unsigned int maxtime;
518 struct timeval tv_start, tv_now;
519
520 X_LOCK (reslock);
521 maxreqs = max_poll_reqs;
522 maxtime = max_poll_time;
523 X_UNLOCK (reslock);
524
525 if (maxtime)
526 gettimeofday (&tv_start, 0);
527
528 for (;;)
529 {
530 ETP_REQ *req;
531
532 etp_maybe_start_thread ();
533
534 X_LOCK (reslock);
535 req = reqq_shift (&res_queue);
536
537 if (req)
538 {
539 --npending;
540
541 if (!res_queue.size && done_poll_cb)
542 done_poll_cb ();
543 }
544
545 X_UNLOCK (reslock);
546
547 if (!req)
548 return 0;
549
550 X_LOCK (reqlock);
551 --nreqs;
552 X_UNLOCK (reqlock);
553
554 if (ecb_expect_false (req->type == EIO_GROUP && req->size))
555 {
556 req->int1 = 1; /* mark request as delayed */
557 continue;
558 }
559 else
560 {
561 int res = ETP_FINISH (req);
562 if (ecb_expect_false (res))
563 return res;
564 }
565
566 if (ecb_expect_false (maxreqs && !--maxreqs))
567 break;
568
569 if (maxtime)
570 {
571 gettimeofday (&tv_now, 0);
572
573 if (tvdiff (&tv_start, &tv_now) >= maxtime)
574 break;
575 }
576 }
577
578 errno = EAGAIN;
579 return -1;
580}
581
582static void
583etp_cancel (ETP_REQ *req)
584{
585 req->cancelled = 1;
586
587 eio_grp_cancel (req);
588}
589
590static void
591etp_submit (ETP_REQ *req)
592{
593 req->pri -= ETP_PRI_MIN;
594
595 if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
596 if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
597
598 if (ecb_expect_false (req->type == EIO_GROUP))
599 {
600 /* I hope this is worth it :/ */
601 X_LOCK (reqlock);
602 ++nreqs;
603 X_UNLOCK (reqlock);
604
605 X_LOCK (reslock);
606
607 ++npending;
608
609 if (!reqq_push (&res_queue, req) && want_poll_cb)
610 want_poll_cb ();
611
612 X_UNLOCK (reslock);
613 }
614 else
615 {
616 X_LOCK (reqlock);
617 ++nreqs;
618 ++nready;
619 reqq_push (&req_queue, req);
620 X_COND_SIGNAL (reqwait);
621 X_UNLOCK (reqlock);
622
623 etp_maybe_start_thread ();
624 }
625}
626
627static void ecb_cold
628etp_set_max_poll_time (double nseconds)
629{
630 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
631 max_poll_time = nseconds * EIO_TICKS;
632 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
633}
634
635static void ecb_cold
636etp_set_max_poll_reqs (unsigned int maxreqs)
637{
638 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
639 max_poll_reqs = maxreqs;
640 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
641}
642
643static void ecb_cold
644etp_set_max_idle (unsigned int nthreads)
645{
646 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
647 max_idle = nthreads;
648 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
649}
650
651static void ecb_cold
652etp_set_idle_timeout (unsigned int seconds)
653{
654 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
655 idle_timeout = seconds;
656 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
657}
658
659static void ecb_cold
660etp_set_min_parallel (unsigned int nthreads)
661{
662 if (wanted < nthreads)
663 wanted = nthreads;
664}
665
666static void ecb_cold
667etp_set_max_parallel (unsigned int nthreads)
668{
669 if (wanted > nthreads)
670 wanted = nthreads;
671
672 while (started > wanted)
673 etp_end_thread ();
674}
675 403
676/*****************************************************************************/ 404/*****************************************************************************/
677 405
678static void 406static void
679grp_try_feed (eio_req *grp) 407grp_try_feed (eio_req *grp)
680{ 408{
681 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 409 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
682 { 410 {
683 grp->flags &= ~EIO_FLAG_GROUPADD; 411 grp->flags &= ~ETP_FLAG_GROUPADD;
684 412
685 EIO_FEED (grp); 413 EIO_FEED (grp);
686 414
687 /* stop if no progress has been made */ 415 /* stop if no progress has been made */
688 if (!(grp->flags & EIO_FLAG_GROUPADD)) 416 if (!(grp->flags & ETP_FLAG_GROUPADD))
689 { 417 {
690 grp->feed = 0; 418 grp->feed = 0;
691 break; 419 break;
692 } 420 }
693 } 421 }
… …
700 428
701 /* call feeder, if applicable */ 429 /* call feeder, if applicable */
702 grp_try_feed (grp); 430 grp_try_feed (grp);
703 431
704 /* finish, if done */ 432 /* finish, if done */
705 if (!grp->size && grp->int1) 433 if (!grp->size && grp->flags & ETP_FLAG_DELAYED)
706 return eio_finish (grp); 434 return eio_finish (grp);
707 else 435 else
708 return 0; 436 return 0;
709} 437}
710 438
… …
746} 474}
747 475
748void 476void
749eio_grp_cancel (eio_req *grp) 477eio_grp_cancel (eio_req *grp)
750{ 478{
751 for (grp = grp->grp_first; grp; grp = grp->grp_next) 479 etp_grp_cancel (EIO_POOL, grp);
752 eio_cancel (grp);
753} 480}
754 481
755void 482void
756eio_cancel (eio_req *req) 483eio_cancel (eio_req *req)
757{ 484{
758 etp_cancel (req); 485 etp_cancel (EIO_POOL, req);
759} 486}
760 487
761void 488void
762eio_submit (eio_req *req) 489eio_submit (eio_req *req)
763{ 490{
764 etp_submit (req); 491 etp_submit (EIO_POOL, req);
765} 492}
766 493
767unsigned int 494unsigned int
768eio_nreqs (void) 495eio_nreqs (void)
769{ 496{
770 return etp_nreqs (); 497 return etp_nreqs (EIO_POOL);
771} 498}
772 499
773unsigned int 500unsigned int
774eio_nready (void) 501eio_nready (void)
775{ 502{
776 return etp_nready (); 503 return etp_nready (EIO_POOL);
777} 504}
778 505
779unsigned int 506unsigned int
780eio_npending (void) 507eio_npending (void)
781{ 508{
782 return etp_npending (); 509 return etp_npending (EIO_POOL);
783} 510}
784 511
785unsigned int ecb_cold 512unsigned int ecb_cold
786eio_nthreads (void) 513eio_nthreads (void)
787{ 514{
788 return etp_nthreads (); 515 return etp_nthreads (EIO_POOL);
789} 516}
790 517
791void ecb_cold 518void ecb_cold
792eio_set_max_poll_time (double nseconds) 519eio_set_max_poll_time (double nseconds)
793{ 520{
794 etp_set_max_poll_time (nseconds); 521 etp_set_max_poll_time (EIO_POOL, nseconds);
795} 522}
796 523
797void ecb_cold 524void ecb_cold
798eio_set_max_poll_reqs (unsigned int maxreqs) 525eio_set_max_poll_reqs (unsigned int maxreqs)
799{ 526{
800 etp_set_max_poll_reqs (maxreqs); 527 etp_set_max_poll_reqs (EIO_POOL, maxreqs);
801} 528}
802 529
803void ecb_cold 530void ecb_cold
804eio_set_max_idle (unsigned int nthreads) 531eio_set_max_idle (unsigned int nthreads)
805{ 532{
806 etp_set_max_idle (nthreads); 533 etp_set_max_idle (EIO_POOL, nthreads);
807} 534}
808 535
809void ecb_cold 536void ecb_cold
810eio_set_idle_timeout (unsigned int seconds) 537eio_set_idle_timeout (unsigned int seconds)
811{ 538{
812 etp_set_idle_timeout (seconds); 539 etp_set_idle_timeout (EIO_POOL, seconds);
813} 540}
814 541
815void ecb_cold 542void ecb_cold
816eio_set_min_parallel (unsigned int nthreads) 543eio_set_min_parallel (unsigned int nthreads)
817{ 544{
818 etp_set_min_parallel (nthreads); 545 etp_set_min_parallel (EIO_POOL, nthreads);
819} 546}
820 547
821void ecb_cold 548void ecb_cold
822eio_set_max_parallel (unsigned int nthreads) 549eio_set_max_parallel (unsigned int nthreads)
823{ 550{
824 etp_set_max_parallel (nthreads); 551 etp_set_max_parallel (EIO_POOL, nthreads);
825} 552}
826 553
827int eio_poll (void) 554int eio_poll (void)
828{ 555{
829 return etp_poll (); 556 return etp_poll (EIO_POOL);
830} 557}
831 558
832/*****************************************************************************/ 559/*****************************************************************************/
833/* work around various missing functions */ 560/* work around various missing functions */
834 561
835#if !HAVE_PREADWRITE 562#if HAVE_POSIX_CLOSE && !__linux
836# undef pread 563# define eio__close(fd) posix_close (fd, 0)
837# undef pwrite 564#else
838# define pread eio__pread 565# define eio__close(fd) close (fd)
839# define pwrite eio__pwrite
840
841static ssize_t
842eio__pread (int fd, void *buf, size_t count, off_t offset)
843{
844 ssize_t res;
845 off_t ooffset;
846
847 X_LOCK (preadwritelock);
848 ooffset = lseek (fd, 0, SEEK_CUR);
849 lseek (fd, offset, SEEK_SET);
850 res = read (fd, buf, count);
851 lseek (fd, ooffset, SEEK_SET);
852 X_UNLOCK (preadwritelock);
853
854 return res;
855}
856
857static ssize_t
858eio__pwrite (int fd, void *buf, size_t count, off_t offset)
859{
860 ssize_t res;
861 off_t ooffset;
862
863 X_LOCK (preadwritelock);
864 ooffset = lseek (fd, 0, SEEK_CUR);
865 lseek (fd, offset, SEEK_SET);
866 res = write (fd, buf, count);
867 lseek (fd, ooffset, SEEK_SET);
868 X_UNLOCK (preadwritelock);
869
870 return res;
871}
872#endif 566#endif
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}
873 576
874#ifndef HAVE_UTIMES 577#ifndef HAVE_UTIMES
875 578
876# undef utimes 579# undef utimes
877# define utimes(path,times) eio__utimes (path, times) 580# define utimes(path,times) eio__utimes (path, times)
… …
910 613
911#if !HAVE_FDATASYNC 614#if !HAVE_FDATASYNC
912# undef fdatasync 615# undef fdatasync
913# define fdatasync(fd) fsync (fd) 616# define fdatasync(fd) fsync (fd)
914#endif 617#endif
618
619static int
620eio__syncfs (int fd)
621{
622 int res;
623
624#if HAVE_SYS_SYNCFS
625 res = (int)syscall (__NR_syncfs, (int)(fd));
626#else
627 res = EIO_ENOSYS ();
628#endif
629
630 if (res < 0 && errno == ENOSYS && fd >= 0)
631 sync ();
632
633 return res;
634}
915 635
916/* sync_file_range always needs emulation */ 636/* sync_file_range always needs emulation */
917static int 637static int
918eio__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)
919{ 639{
… …
942} 662}
943 663
944static int 664static int
945eio__fallocate (int fd, int mode, off_t offset, size_t len) 665eio__fallocate (int fd, int mode, off_t offset, size_t len)
946{ 666{
947#if HAVE_FALLOCATE 667#if HAVE_LINUX_FALLOCATE
948 return fallocate (fd, mode, offset, len); 668 return fallocate (fd, mode, offset, len);
949#else 669#else
950 errno = ENOSYS; 670 return EIO_ENOSYS ();
951 return -1;
952#endif 671#endif
953} 672}
954 673
955#if !HAVE_READAHEAD 674#if !HAVE_READAHEAD
956# undef readahead 675# undef readahead
957# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 676# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
958 677
959static ssize_t 678static eio_ssize_t
960eio__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)
961{ 680{
962 size_t todo = count; 681 size_t todo = count;
963 dBUF; 682 dBUF;
964 683
… …
969 pread (fd, eio_buf, len, offset); 688 pread (fd, eio_buf, len, offset);
970 offset += len; 689 offset += len;
971 todo -= len; 690 todo -= len;
972 } 691 }
973 692
974 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 */
975 return count; 697 return 0;
976} 698}
977 699
978#endif 700#endif
979 701
980/* sendfile always needs emulation */ 702/* sendfile always needs emulation */
981static ssize_t 703static eio_ssize_t
982eio__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)
983{ 705{
984 ssize_t written = 0; 706 eio_ssize_t written = 0;
985 ssize_t res; 707 eio_ssize_t res;
986 708
987 if (!count) 709 if (!count)
988 return 0; 710 return 0;
989 711
990 for (;;) 712 for (;;)
… …
1015 737
1016 /* according to source inspection, this is correct, and useful behaviour */ 738 /* according to source inspection, this is correct, and useful behaviour */
1017 if (sbytes) 739 if (sbytes)
1018 res = sbytes; 740 res = sbytes;
1019 741
1020# elif defined (__APPLE__) 742# elif defined __APPLE__
1021 off_t sbytes = count; 743 off_t sbytes = count;
1022 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0); 744 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
1023 745
1024 /* according to the manpage, sbytes is always valid */ 746 /* according to the manpage, sbytes is always valid */
1025 if (sbytes) 747 if (sbytes)
… …
1042 if (res < 0 && sbytes) 764 if (res < 0 && sbytes)
1043 res = sbytes; 765 res = sbytes;
1044 766
1045# endif 767# endif
1046 768
1047#elif defined (_WIN32) 769#elif defined (_WIN32) && 0
1048 /* does not work, just for documentation of what would need to be done */ 770 /* does not work, just for documentation of what would need to be done */
1049 /* actually, cannot be done like this, as TransmitFile changes the file offset, */ 771 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
1050 /* libeio guarantees that the file offset does not change, and windows */ 772 /* libeio guarantees that the file offset does not change, and windows */
1051 /* has no way to get an independent handle to the same file description */ 773 /* has no way to get an independent handle to the same file description */
1052 HANDLE h = TO_SOCKET (ifd); 774 HANDLE h = TO_SOCKET (ifd);
1053 SetFilePointer (h, offset, 0, FILE_BEGIN); 775 SetFilePointer (h, offset, 0, FILE_BEGIN);
1054 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0); 776 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1055 777
1056#else 778#else
1057 res = -1; 779 res = EIO_ENOSYS ();
1058 errno = ENOSYS;
1059#endif 780#endif
1060 781
1061 /* we assume sendfile can copy at least 128mb in one go */ 782 /* we assume sendfile can copy at least 128mb in one go */
1062 if (res <= 128 * 1024 * 1024) 783 if (res <= 128 * 1024 * 1024)
1063 { 784 {
… …
1085 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 806 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1086 /* BSDs */ 807 /* BSDs */
1087#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */ 808#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1088 || errno == ENOTSUP 809 || errno == ENOTSUP
1089#endif 810#endif
811#ifdef EOPNOTSUPP /* windows */
1090 || errno == EOPNOTSUPP /* BSDs */ 812 || errno == EOPNOTSUPP /* BSDs */
813#endif
1091#if __solaris 814#if __solaris
1092 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 815 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
1093#endif 816#endif
1094 ) 817 )
1095 ) 818 )
… …
1099 822
1100 res = 0; 823 res = 0;
1101 824
1102 while (count) 825 while (count)
1103 { 826 {
1104 ssize_t cnt; 827 eio_ssize_t cnt;
1105 828
1106 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);
1107 830
1108 if (cnt <= 0) 831 if (cnt <= 0)
1109 { 832 {
… …
1121 844
1122 offset += cnt; 845 offset += cnt;
1123 res += cnt; 846 res += cnt;
1124 count -= cnt; 847 count -= cnt;
1125 } 848 }
849
850 FUBd;
1126 } 851 }
1127 852
1128 return res; 853 return res;
1129} 854}
1130 855
… …
1157 /* round up length */ 882 /* round up length */
1158 *length = (*length + mask) & ~mask; 883 *length = (*length + mask) & ~mask;
1159} 884}
1160 885
1161#if !_POSIX_MEMLOCK 886#if !_POSIX_MEMLOCK
1162# define eio__mlockall(a) ((errno = ENOSYS), -1) 887# define eio__mlockall(a) EIO_ENOSYS ()
1163#else 888#else
1164 889
1165static int 890static int
1166eio__mlockall (int flags) 891eio__mlockall (int flags)
1167{ 892{
… …
1181 return mlockall (flags); 906 return mlockall (flags);
1182} 907}
1183#endif 908#endif
1184 909
1185#if !_POSIX_MEMLOCK_RANGE 910#if !_POSIX_MEMLOCK_RANGE
1186# define eio__mlock(a,b) ((errno = ENOSYS), -1) 911# define eio__mlock(a,b) EIO_ENOSYS ()
1187#else 912#else
1188 913
1189static int 914static int
1190eio__mlock (void *addr, size_t length) 915eio__mlock (void *addr, size_t length)
1191{ 916{
… …
1195} 920}
1196 921
1197#endif 922#endif
1198 923
1199#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) 924#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1200# define eio__msync(a,b,c) ((errno = ENOSYS), -1) 925# define eio__msync(a,b,c) EIO_ENOSYS ()
1201#else 926#else
1202 927
1203static int 928static int
1204eio__msync (void *mem, size_t len, int flags) 929eio__msync (void *mem, size_t len, int flags)
1205{ 930{
… …
1234 intptr_t end = addr + len; 959 intptr_t end = addr + len;
1235 intptr_t page = eio_pagesize (); 960 intptr_t page = eio_pagesize ();
1236 961
1237 if (addr < end) 962 if (addr < end)
1238 if (flags & EIO_MT_MODIFY) /* modify */ 963 if (flags & EIO_MT_MODIFY) /* modify */
1239 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req)); 964 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < end && !EIO_CANCELLED (req));
1240 else 965 else
1241 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req)); 966 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < end && !EIO_CANCELLED (req));
1242 } 967 }
1243 968
1244 return 0; 969 return 0;
1245} 970}
1246 971
1247/*****************************************************************************/ 972/*****************************************************************************/
1248/* requests implemented outside eio_execute, because they are so large */ 973/* requests implemented outside eio_execute, because they are so large */
1249 974
1250static void 975static void
1251eio__realpath (eio_req *req, etp_worker *self) 976eio__lseek (eio_req *req)
1252{ 977{
1253 char *rel = req->ptr1; 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{
1254 char *res; 993 char *res;
994 const char *rel = path;
1255 char *tmp1, *tmp2; 995 char *tmp1, *tmp2;
1256#if SYMLOOP_MAX > 32 996#if SYMLOOP_MAX > 32
1257 int symlinks = SYMLOOP_MAX; 997 int symlinks = SYMLOOP_MAX;
1258#else 998#else
1259 int symlinks = 32; 999 int symlinks = 32;
1260#endif 1000#endif
1261 1001
1262 req->result = -1;
1263
1264 errno = EINVAL; 1002 errno = EINVAL;
1265 if (!rel) 1003 if (!rel)
1266 return; 1004 return -1;
1267 1005
1268 errno = ENOENT; 1006 errno = ENOENT;
1269 if (!*rel) 1007 if (!*rel)
1270 return; 1008 return -1;
1271 1009
1272 if (!req->ptr2) 1010 res = etp_tmpbuf_get (tmpbuf, EIO_PATH_MAX * 3);
1273 { 1011#ifdef _WIN32
1274 X_LOCK (wrklock); 1012 if (_access (rel, 4) != 0)
1275 req->flags |= EIO_FLAG_PTR2_FREE;
1276 X_UNLOCK (wrklock);
1277 req->ptr2 = malloc (PATH_MAX * 3);
1278
1279 errno = ENOMEM;
1280 if (!req->ptr2)
1281 return; 1013 return -1;
1282 }
1283 1014
1284 res = req->ptr2; 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
1285 tmp1 = res + PATH_MAX; 1028 tmp1 = res + EIO_PATH_MAX;
1286 tmp2 = tmp1 + PATH_MAX; 1029 tmp2 = tmp1 + EIO_PATH_MAX;
1287 1030
1288#if 0 /* disabled, the musl way to do things is just too racy */ 1031#if 0 /* disabled, the musl way to do things is just too racy */
1289#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME) 1032#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1290 /* on linux we may be able to ask the kernel */ 1033 /* on linux we may be able to ask the kernel */
1291 { 1034 {
1292 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME); 1035 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1293 1036
1294 if (fd >= 0) 1037 if (fd >= 0)
1295 { 1038 {
1296 sprintf (tmp1, "/proc/self/fd/%d", fd); 1039 sprintf (tmp1, "/proc/self/fd/%d", fd);
1297 req->result = readlink (tmp1, res, PATH_MAX); 1040 req->result = readlink (tmp1, res, EIO_PATH_MAX);
1298 close (fd);
1299
1300 /* here we should probably stat the open file and the disk file, to make sure they still match */ 1041 /* here we should probably stat the open file and the disk file, to make sure they still match */
1042 eio__close (fd);
1301 1043
1302 if (req->result > 0) 1044 if (req->result > 0)
1303 goto done; 1045 goto done;
1304 } 1046 }
1305 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO) 1047 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1306 return; 1048 return -1;
1307 } 1049 }
1308#endif 1050#endif
1309#endif 1051#endif
1310 1052
1311 if (*rel != '/') 1053 if (*rel != '/')
1312 { 1054 {
1313 if (!getcwd (res, PATH_MAX)) 1055 int len;
1056
1057 errno = ENOENT;
1058 if (wd == EIO_INVALID_WD)
1314 return; 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);
1315 1070
1316 if (res [1]) /* only use if not / */ 1071 if (res [1]) /* only use if not / */
1317 res += strlen (res); 1072 res += len;
1318 } 1073 }
1319 1074
1320 while (*rel) 1075 while (*rel)
1321 { 1076 {
1322 ssize_t len, linklen; 1077 eio_ssize_t len, linklen;
1323 char *beg = rel; 1078 const char *beg = rel;
1324 1079
1325 while (*rel && *rel != '/') 1080 while (*rel && *rel != '/')
1326 ++rel; 1081 ++rel;
1327 1082
1328 len = rel - beg; 1083 len = rel - beg;
… …
1340 1095
1341 if (beg [1] == '.' && len == 2) 1096 if (beg [1] == '.' && len == 2)
1342 { 1097 {
1343 /* .. - back up one component, if possible */ 1098 /* .. - back up one component, if possible */
1344 1099
1345 while (res != req->ptr2) 1100 while (res != tmpbuf->ptr)
1346 if (*--res == '/') 1101 if (*--res == '/')
1347 break; 1102 break;
1348 1103
1349 continue; 1104 continue;
1350 } 1105 }
1351 } 1106 }
1352 1107
1353 errno = ENAMETOOLONG; 1108 errno = ENAMETOOLONG;
1354 if (res + 1 + len + 1 >= tmp1) 1109 if (res + 1 + len + 1 >= tmp1)
1355 return; 1110 return -1;
1356 1111
1357 /* copy one component */ 1112 /* copy one component */
1358 *res = '/'; 1113 *res = '/';
1359 memcpy (res + 1, beg, len); 1114 memcpy (res + 1, beg, len);
1360 1115
1361 /* zero-terminate, for readlink */ 1116 /* zero-terminate, for readlink */
1362 res [len + 1] = 0; 1117 res [len + 1] = 0;
1363 1118
1364 /* now check if it's a symlink */ 1119 /* now check if it's a symlink */
1365 linklen = readlink (req->ptr2, tmp1, PATH_MAX); 1120 linklen = readlink (tmpbuf->ptr, tmp1, EIO_PATH_MAX);
1366 1121
1367 if (linklen < 0) 1122 if (linklen < 0)
1368 { 1123 {
1369 if (errno != EINVAL) 1124 if (errno != EINVAL)
1370 return; 1125 return -1;
1371 1126
1372 /* it's a normal directory. hopefully */ 1127 /* it's a normal directory. hopefully */
1373 res += len + 1; 1128 res += len + 1;
1374 } 1129 }
1375 else 1130 else
1376 { 1131 {
1377 /* yay, it was a symlink - build new path in tmp2 */ 1132 /* yay, it was a symlink - build new path in tmp2 */
1378 int rellen = strlen (rel); 1133 int rellen = strlen (rel);
1379 1134
1380 errno = ENAMETOOLONG; 1135 errno = ENAMETOOLONG;
1381 if (linklen + 1 + rellen >= PATH_MAX) 1136 if (linklen + 1 + rellen >= EIO_PATH_MAX) /* also catch linklen >= EIO_PATH_MAX */
1382 return; 1137 return -1;
1383 1138
1384 errno = ELOOP; 1139 errno = ELOOP;
1385 if (!--symlinks) 1140 if (!--symlinks)
1386 return; 1141 return -1;
1387 1142
1388 if (*tmp1 == '/') 1143 if (*tmp1 == '/')
1389 res = req->ptr2; /* symlink resolves to an absolute path */ 1144 res = tmpbuf->ptr; /* symlink resolves to an absolute path */
1390 1145
1391 /* we need to be careful, as rel might point into tmp2 already */ 1146 /* we need to be careful, as rel might point into tmp2 already */
1392 memmove (tmp2 + linklen + 1, rel, rellen + 1); 1147 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1393 tmp2 [linklen] = '/'; 1148 tmp2 [linklen] = '/';
1394 memcpy (tmp2, tmp1, linklen); 1149 memcpy (tmp2, tmp1, linklen);
… …
1396 rel = tmp2; 1151 rel = tmp2;
1397 } 1152 }
1398 } 1153 }
1399 1154
1400 /* special case for the lone root path */ 1155 /* special case for the lone root path */
1401 if (res == req->ptr2) 1156 if (res == tmpbuf->ptr)
1402 *res++ = '/'; 1157 *res++ = '/';
1403 1158
1404 req->result = res - (char *)req->ptr2; 1159 return res - (char *)tmpbuf->ptr;
1405 1160#endif
1406done:
1407 req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
1408} 1161}
1409 1162
1410static signed char 1163static signed char
1411eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) 1164eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1412{ 1165{
… …
1420 1173
1421#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */ 1174#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1422#define EIO_SORT_FAST 60 /* when to only use insertion sort */ 1175#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1423 1176
1424static void 1177static void
1425eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1178eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1426{ 1179{
1427 unsigned char bits [9 + sizeof (ino_t) * 8]; 1180 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1428 unsigned char *bit = bits; 1181 unsigned char *bit = bits;
1429 1182
1430 assert (CHAR_BIT == 8); 1183 assert (CHAR_BIT == 8);
1431 assert (sizeof (eio_dirent) * 8 < 256); 1184 assert (sizeof (eio_dirent) * 8 < 256);
1432 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */ 1185 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
… …
1434 1187
1435 if (size <= EIO_SORT_FAST) 1188 if (size <= EIO_SORT_FAST)
1436 return; 1189 return;
1437 1190
1438 /* first prepare an array of bits to test in our radix sort */ 1191 /* first prepare an array of bits to test in our radix sort */
1439 /* try to take endianness into account, as well as differences in ino_t sizes */ 1192 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1440 /* inode_bits must contain all inodes ORed together */ 1193 /* inode_bits must contain all inodes ORed together */
1441 /* which is used to skip bits that are 0 everywhere, which is very common */ 1194 /* which is used to skip bits that are 0 everywhere, which is very common */
1442 { 1195 {
1443 ino_t endianness; 1196 eio_ino_t endianness;
1444 int i, j; 1197 int i, j;
1445 1198
1446 /* we store the byte offset of byte n into byte n of "endianness" */ 1199 /* we store the byte offset of byte n into byte n of "endianness" */
1447 for (i = 0; i < sizeof (ino_t); ++i) 1200 for (i = 0; i < sizeof (eio_ino_t); ++i)
1448 ((unsigned char *)&endianness)[i] = i; 1201 ((unsigned char *)&endianness)[i] = i;
1449 1202
1450 *bit++ = 0; 1203 *bit++ = 0;
1451 1204
1452 for (i = 0; i < sizeof (ino_t); ++i) 1205 for (i = 0; i < sizeof (eio_ino_t); ++i)
1453 { 1206 {
1454 /* shifting off the byte offsets out of "endianness" */ 1207 /* shifting off the byte offsets out of "endianness" */
1455 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8; 1208 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1456 endianness >>= 8; 1209 endianness >>= 8;
1457 1210
1458 for (j = 0; j < 8; ++j) 1211 for (j = 0; j < 8; ++j)
1459 if (inode_bits & (((ino_t)1) << (i * 8 + j))) 1212 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1460 *bit++ = offs + j; 1213 *bit++ = offs + j;
1461 } 1214 }
1462 1215
1463 for (j = 0; j < 8; ++j) 1216 for (j = 0; j < 8; ++j)
1464 if (score_bits & (1 << j)) 1217 if (score_bits & (1 << j))
1465 *bit++ = offsetof (eio_dirent, score) * 8 + j; 1218 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1466 } 1219 }
1467 1220
1468 /* now actually do the sorting (a variant of MSD radix sort) */ 1221 /* now actually do the sorting (a variant of MSD radix sort) */
1469 { 1222 {
1470 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base; 1223 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1471 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end; 1224 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1472 unsigned char *bit_stk [9 + sizeof (ino_t) * 8]; 1225 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1473 int stk_idx = 0; 1226 int stk_idx = 0;
1474 1227
1475 base_stk [stk_idx] = dents; 1228 base_stk [stk_idx] = dents;
1476 end_stk [stk_idx] = dents + size; 1229 end_stk [stk_idx] = dents + size;
1477 bit_stk [stk_idx] = bit - 1; 1230 bit_stk [stk_idx] = bit - 1;
… …
1556 } 1309 }
1557 } 1310 }
1558} 1311}
1559 1312
1560static void 1313static void
1561eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1314eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1562{ 1315{
1563 if (size <= 1) 1316 if (size <= 1)
1564 return; /* our insertion sort relies on size > 0 */ 1317 return; /* our insertion sort relies on size > 0 */
1565 1318
1566 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */ 1319 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
… …
1574 1327
1575/* read a full directory */ 1328/* read a full directory */
1576static void 1329static void
1577eio__scandir (eio_req *req, etp_worker *self) 1330eio__scandir (eio_req *req, etp_worker *self)
1578{ 1331{
1579 DIR *dirp;
1580 EIO_STRUCT_DIRENT *entp;
1581 char *name, *names; 1332 char *name, *names;
1582 int namesalloc = 4096; 1333 int namesalloc = 4096 - sizeof (void *) * 4;
1583 int namesoffs = 0; 1334 int namesoffs = 0;
1584 int flags = req->int1; 1335 int flags = req->int1;
1585 eio_dirent *dents = 0; 1336 eio_dirent *dents = 0;
1586 int dentalloc = 128; 1337 int dentalloc = 128;
1587 int dentoffs = 0; 1338 int dentoffs = 0;
1588 ino_t inode_bits = 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
1589 1347
1590 req->result = -1; 1348 req->result = -1;
1591 1349
1592 if (!(flags & EIO_READDIR_DENTS)) 1350 if (!(flags & EIO_READDIR_DENTS))
1593 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); 1351 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1594 1352
1595 X_LOCK (wrklock); 1353#ifdef _WIN32
1596 /* 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
1597 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
1598 1421
1599 if (req->flags & EIO_FLAG_PTR1_FREE) 1422 if (req->flags & EIO_FLAG_PTR1_FREE)
1600 free (req->ptr1); 1423 free (req->ptr1);
1601 1424
1602 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1425 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1603 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; 1426 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1604 req->ptr2 = names = malloc (namesalloc); 1427 req->ptr2 = names = malloc (namesalloc);
1605 X_UNLOCK (wrklock);
1606 1428
1607 if (dirp && names && (!flags || dents)) 1429 if (!names || (flags && !dents))
1430 return;
1431
1608 for (;;) 1432 for (;;)
1609 { 1433 {
1434 int done;
1435
1436#ifdef _WIN32
1437 done = !dirp;
1438#else
1610 errno = 0; 1439 errno = 0;
1611 entp = readdir (dirp); 1440 entp = readdir (dirp);
1441 done = !entp;
1442#endif
1612 1443
1613 if (!entp) 1444 if (done)
1614 { 1445 {
1446#ifndef _WIN32
1447 int old_errno = errno;
1448 closedir (dirp);
1449 errno = old_errno;
1450
1615 if (errno) 1451 if (errno)
1616 break; 1452 break;
1453#endif
1617 1454
1618 /* sort etc. */ 1455 /* sort etc. */
1619 req->int1 = flags; 1456 req->int1 = flags;
1620 req->result = dentoffs; 1457 req->result = dentoffs;
1621 1458
1622 if (flags & EIO_READDIR_STAT_ORDER) 1459 if (flags & EIO_READDIR_STAT_ORDER)
1623 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits); 1460 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1624 else if (flags & EIO_READDIR_DIRS_FIRST) 1461 else if (flags & EIO_READDIR_DIRS_FIRST)
1625 if (flags & EIO_READDIR_FOUND_UNKNOWN) 1462 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1626 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */ 1463 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1627 else 1464 else
1465 {
1466 /* in this case, all is known, and we just put dirs first and sort them */
1467 eio_dirent *oth = dents + dentoffs;
1468 eio_dirent *dir = dents;
1469
1470 /* now partition dirs to the front, and non-dirs to the back */
1471 /* by walking from both sides and swapping if necessary */
1472 while (oth > dir)
1473 {
1474 if (dir->type == EIO_DT_DIR)
1475 ++dir;
1476 else if ((--oth)->type == EIO_DT_DIR)
1477 {
1478 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1479
1480 ++dir;
1481 }
1482 }
1483
1484 /* now sort the dirs only (dirs all have the same score) */
1485 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1486 }
1487
1488 break;
1489 }
1490
1491 /* now add the entry to our list(s) */
1492 name = D_NAME (entp);
1493
1494 /* skip . and .. entries */
1495 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1496 {
1497 int len = D_NAMLEN (entp) + 1;
1498
1499 while (ecb_expect_false (namesoffs + len > namesalloc))
1500 {
1501 namesalloc *= 2;
1502 req->ptr2 = names = realloc (names, namesalloc);
1503
1504 if (!names)
1505 break;
1506 }
1507
1508 memcpy (names + namesoffs, name, len);
1509
1510 if (dents)
1511 {
1512 struct eio_dirent *ent;
1513
1514 if (ecb_expect_false (dentoffs == dentalloc))
1628 { 1515 {
1629 /* in this case, all is known, and we just put dirs first and sort them */ 1516 dentalloc *= 2;
1517 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1518
1519 if (!dents)
1520 break;
1521 }
1522
1630 eio_dirent *oth = dents + dentoffs; 1523 ent = dents + dentoffs;
1631 eio_dirent *dir = dents;
1632 1524
1633 /* now partition dirs to the front, and non-dirs to the back */ 1525 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1634 /* by walking from both sides and swapping if necessary */ 1526 ent->namelen = len - 1;
1635 while (oth > dir) 1527 ent->inode = D_INO (entp);
1528
1529 inode_bits |= ent->inode;
1530
1531 switch (D_TYPE (entp))
1532 {
1533 default:
1534 ent->type = EIO_DT_UNKNOWN;
1535 flags |= EIO_READDIR_FOUND_UNKNOWN;
1536 break;
1537
1538 #ifdef DT_FIFO
1539 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1540 #endif
1541 #ifdef DT_CHR
1542 case DT_CHR: ent->type = EIO_DT_CHR; break;
1543 #endif
1544 #ifdef DT_MPC
1545 case DT_MPC: ent->type = EIO_DT_MPC; break;
1546 #endif
1547 #ifdef DT_DIR
1548 case DT_DIR: ent->type = EIO_DT_DIR; break;
1549 #endif
1550 #ifdef DT_NAM
1551 case DT_NAM: ent->type = EIO_DT_NAM; break;
1552 #endif
1553 #ifdef DT_BLK
1554 case DT_BLK: ent->type = EIO_DT_BLK; break;
1555 #endif
1556 #ifdef DT_MPB
1557 case DT_MPB: ent->type = EIO_DT_MPB; break;
1558 #endif
1559 #ifdef DT_REG
1560 case DT_REG: ent->type = EIO_DT_REG; break;
1561 #endif
1562 #ifdef DT_NWK
1563 case DT_NWK: ent->type = EIO_DT_NWK; break;
1564 #endif
1565 #ifdef DT_CMP
1566 case DT_CMP: ent->type = EIO_DT_CMP; break;
1567 #endif
1568 #ifdef DT_LNK
1569 case DT_LNK: ent->type = EIO_DT_LNK; break;
1570 #endif
1571 #ifdef DT_SOCK
1572 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1573 #endif
1574 #ifdef DT_DOOR
1575 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1576 #endif
1577 #ifdef DT_WHT
1578 case DT_WHT: ent->type = EIO_DT_WHT; break;
1579 #endif
1580 }
1581
1582 ent->score = 7;
1583
1584 if (flags & EIO_READDIR_DIRS_FIRST)
1585 {
1586 if (ent->type == EIO_DT_UNKNOWN)
1636 { 1587 {
1637 if (dir->type == EIO_DT_DIR) 1588 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1638 ++dir; 1589 ent->score = 1;
1639 else if ((--oth)->type == EIO_DT_DIR) 1590 else if (!strchr (name, '.')) /* absence of dots indicate likely dirs */
1640 { 1591 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1641 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1642
1643 ++dir;
1644 }
1645 } 1592 }
1646 1593 else if (ent->type == EIO_DT_DIR)
1647 /* now sort the dirs only (dirs all have the same score) */ 1594 ent->score = 0;
1648 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1649 } 1595 }
1650
1651 break;
1652 }
1653
1654 /* now add the entry to our list(s) */
1655 name = entp->d_name;
1656
1657 /* skip . and .. entries */
1658 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1659 {
1660 int len = D_NAMLEN (entp) + 1;
1661
1662 while (ecb_expect_false (namesoffs + len > namesalloc))
1663 {
1664 namesalloc *= 2;
1665 X_LOCK (wrklock);
1666 req->ptr2 = names = realloc (names, namesalloc);
1667 X_UNLOCK (wrklock);
1668
1669 if (!names)
1670 break;
1671 } 1596 }
1672 1597
1673 memcpy (names + namesoffs, name, len);
1674
1675 if (dents)
1676 {
1677 struct eio_dirent *ent;
1678
1679 if (ecb_expect_false (dentoffs == dentalloc))
1680 {
1681 dentalloc *= 2;
1682 X_LOCK (wrklock);
1683 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1684 X_UNLOCK (wrklock);
1685
1686 if (!dents)
1687 break;
1688 }
1689
1690 ent = dents + dentoffs;
1691
1692 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1693 ent->namelen = len - 1;
1694 ent->inode = D_INO (entp);
1695
1696 inode_bits |= ent->inode;
1697
1698 switch (D_TYPE (entp))
1699 {
1700 default:
1701 ent->type = EIO_DT_UNKNOWN;
1702 flags |= EIO_READDIR_FOUND_UNKNOWN;
1703 break;
1704
1705 #ifdef DT_FIFO
1706 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1707 #endif
1708 #ifdef DT_CHR
1709 case DT_CHR: ent->type = EIO_DT_CHR; break;
1710 #endif
1711 #ifdef DT_MPC
1712 case DT_MPC: ent->type = EIO_DT_MPC; break;
1713 #endif
1714 #ifdef DT_DIR
1715 case DT_DIR: ent->type = EIO_DT_DIR; break;
1716 #endif
1717 #ifdef DT_NAM
1718 case DT_NAM: ent->type = EIO_DT_NAM; break;
1719 #endif
1720 #ifdef DT_BLK
1721 case DT_BLK: ent->type = EIO_DT_BLK; break;
1722 #endif
1723 #ifdef DT_MPB
1724 case DT_MPB: ent->type = EIO_DT_MPB; break;
1725 #endif
1726 #ifdef DT_REG
1727 case DT_REG: ent->type = EIO_DT_REG; break;
1728 #endif
1729 #ifdef DT_NWK
1730 case DT_NWK: ent->type = EIO_DT_NWK; break;
1731 #endif
1732 #ifdef DT_CMP
1733 case DT_CMP: ent->type = EIO_DT_CMP; break;
1734 #endif
1735 #ifdef DT_LNK
1736 case DT_LNK: ent->type = EIO_DT_LNK; break;
1737 #endif
1738 #ifdef DT_SOCK
1739 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1740 #endif
1741 #ifdef DT_DOOR
1742 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1743 #endif
1744 #ifdef DT_WHT
1745 case DT_WHT: ent->type = EIO_DT_WHT; break;
1746 #endif
1747 }
1748
1749 ent->score = 7;
1750
1751 if (flags & EIO_READDIR_DIRS_FIRST)
1752 {
1753 if (ent->type == EIO_DT_UNKNOWN)
1754 {
1755 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1756 ent->score = 1;
1757 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1758 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1759 }
1760 else if (ent->type == EIO_DT_DIR)
1761 ent->score = 0;
1762 }
1763 }
1764
1765 namesoffs += len; 1598 namesoffs += len;
1766 ++dentoffs; 1599 ++dentoffs;
1767 } 1600 }
1768 1601
1769 if (EIO_CANCELLED (req)) 1602 if (EIO_CANCELLED (req))
1770 { 1603 {
1771 errno = ECANCELED; 1604 errno = ECANCELED;
1772 break; 1605 break;
1773 } 1606 }
1607
1608#ifdef _WIN32
1609 if (!FindNextFile (dirp, &entp))
1610 {
1611 FindClose (dirp);
1612 dirp = 0;
1774 } 1613 }
1614#endif
1615 }
1775} 1616}
1617
1618/*****************************************************************************/
1619/* working directory stuff */
1620/* various deficiencies in the posix 2008 api force us to */
1621/* keep the absolute path in string form at all times */
1622/* fuck yeah. */
1623
1624#if !HAVE_AT
1625
1626/* a bit like realpath, but usually faster because it doesn'T have to return */
1627/* an absolute or canonical path */
1628static const char *
1629wd_expand (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
1630{
1631 if (!wd || *path == '/')
1632 return path;
1633
1634 if (path [0] == '.' && !path [1])
1635 return wd->str;
1636
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);
1712#else
1713 if (flags)
1714 return EIO_ENOSYS ();
1715
1716 return renameat (olddirfd, oldpath, newdirfd, newpath);
1717#endif
1718}
1719
1720/* they forgot these */
1721
1722static int
1723eio__truncateat (int dirfd, const char *path, off_t length)
1724{
1725 int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
1726 int res;
1727
1728 if (fd < 0)
1729 return fd;
1730
1731 res = ftruncate (fd, length);
1732 silent_close (fd);
1733 return res;
1734}
1735
1736static int
1737eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
1738{
1739 int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
1740 int res;
1741
1742 if (fd < 0)
1743 return fd;
1744
1745 res = fstatvfs (fd, buf);
1746 silent_close (fd);
1747 return res;
1748}
1749
1750#endif
1776 1751
1777/*****************************************************************************/ 1752/*****************************************************************************/
1778 1753
1779#define ALLOC(len) \ 1754#define ALLOC(len) \
1780 if (!req->ptr2) \ 1755 if (!req->ptr2) \
1781 { \ 1756 { \
1782 X_LOCK (wrklock); \ 1757 X_LOCK (EIO_POOL->wrklock); \
1783 req->flags |= EIO_FLAG_PTR2_FREE; \ 1758 req->flags |= EIO_FLAG_PTR2_FREE; \
1784 X_UNLOCK (wrklock); \ 1759 X_UNLOCK (EIO_POOL->wrklock); \
1785 req->ptr2 = malloc (len); \ 1760 req->ptr2 = malloc (len); \
1786 if (!req->ptr2) \ 1761 if (!req->ptr2) \
1787 { \ 1762 { \
1788 errno = ENOMEM; \ 1763 errno = ENOMEM; \
1789 req->result = -1; \ 1764 req->result = -1; \
1790 break; \ 1765 goto alloc_fail; \
1791 } \ 1766 } \
1792 } 1767 }
1793 1768
1794X_THREAD_PROC (etp_proc) 1769/*****************************************************************************/
1795{
1796 ETP_REQ *req;
1797 struct timespec ts;
1798 etp_worker *self = (etp_worker *)thr_arg;
1799 int timeout;
1800 1770
1801 /* try to distribute timeouts somewhat evenly */ 1771static void
1802 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 1772eio__slurp (int fd, eio_req *req)
1773{
1774 req->result = fd;
1803 1775
1804 for (;;) 1776 if (fd < 0)
1777 return;
1778
1779 if (req->offs < 0 || !req->size) /* do we need the size? */
1805 { 1780 {
1806 ts.tv_sec = 0; 1781 off_t size = lseek (fd, 0, SEEK_END);
1807 1782
1808 X_LOCK (reqlock);
1809
1810 for (;;)
1811 {
1812 self->req = req = reqq_shift (&req_queue);
1813
1814 if (req)
1815 break;
1816
1817 if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
1818 {
1819 X_UNLOCK (reqlock);
1820 X_LOCK (wrklock);
1821 --started;
1822 X_UNLOCK (wrklock);
1823 goto quit;
1824 }
1825
1826 ++idle;
1827
1828 if (idle <= max_idle)
1829 /* we are allowed to idle, so do so without any timeout */
1830 X_COND_WAIT (reqwait, reqlock);
1831 else
1832 {
1833 /* initialise timeout once */
1834 if (!ts.tv_sec)
1835 ts.tv_sec = time (0) + idle_timeout;
1836
1837 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
1838 ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1839 }
1840
1841 --idle;
1842 }
1843
1844 --nready;
1845
1846 X_UNLOCK (reqlock);
1847
1848 if (req->type < 0) 1783 if (req->offs < 0)
1849 goto quit; 1784 req->offs += size;
1850 1785
1851 ETP_EXECUTE (self, req); 1786 if (!req->size)
1852 1787 req->size = size - req->offs;
1853 X_LOCK (reslock);
1854
1855 ++npending;
1856
1857 if (!reqq_push (&res_queue, req) && want_poll_cb)
1858 want_poll_cb ();
1859
1860 self->req = 0;
1861 etp_worker_clear (self);
1862
1863 X_UNLOCK (reslock);
1864 } 1788 }
1865 1789
1866quit: 1790 ALLOC (req->size);
1867 X_LOCK (wrklock); 1791 req->result = pread (fd, req->ptr2, req->size, req->offs);
1868 etp_worker_free (self);
1869 X_UNLOCK (wrklock);
1870 1792
1871 return 0; 1793 silent_close (fd);
1872}
1873 1794
1874/*****************************************************************************/ 1795alloc_fail:
1796 ;
1797}
1875 1798
1876int ecb_cold 1799int ecb_cold
1877eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1800eio_init (void (*want_poll)(void), void (*done_poll)(void))
1878{ 1801{
1879 return etp_init (want_poll, done_poll); 1802 eio_want_poll_cb = want_poll;
1803 eio_done_poll_cb = done_poll;
1804
1805 return etp_init (EIO_POOL, 0, 0, 0);
1880} 1806}
1881 1807
1882ecb_inline void 1808ecb_inline void
1883eio_api_destroy (eio_req *req) 1809eio_api_destroy (eio_req *req)
1884{ 1810{
1885 free (req); 1811 free (req);
1886} 1812}
1887 1813
1888#define REQ(rtype) \ 1814#define REQ(rtype) \
1889 eio_req *req; \ 1815 eio_req *req; \
1890 \ 1816 \
1891 req = (eio_req *)calloc (1, sizeof *req); \ 1817 req = (eio_req *)calloc (1, sizeof *req); \
1892 if (!req) \ 1818 if (!req) \
1893 return 0; \ 1819 return 0; \
… …
1907 { \ 1833 { \
1908 eio_api_destroy (req); \ 1834 eio_api_destroy (req); \
1909 return 0; \ 1835 return 0; \
1910 } 1836 }
1911 1837
1838#define SINGLEDOT(ptr) (0[(char *)(ptr)] == '.' && !1[(char *)(ptr)])
1839
1912static void 1840static void
1913eio_execute (etp_worker *self, eio_req *req) 1841eio_execute (etp_worker *self, eio_req *req)
1914{ 1842{
1843#if HAVE_AT
1844 int dirfd;
1845#else
1846 const char *path;
1847#endif
1848
1915 if (ecb_expect_false (EIO_CANCELLED (req))) 1849 if (ecb_expect_false (EIO_CANCELLED (req)))
1916 { 1850 {
1917 req->result = -1; 1851 req->result = -1;
1918 req->errorno = ECANCELED; 1852 req->errorno = ECANCELED;
1919 return; 1853 return;
1920 } 1854 }
1921 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 }
1871
1922 switch (req->type) 1872 switch (req->type)
1923 { 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;
1924 case EIO_READ: ALLOC (req->size); 1881 case EIO_READ: ALLOC (req->size);
1925 req->result = req->offs >= 0 1882 req->result = req->offs >= 0
1926 ? pread (req->int1, req->ptr2, req->size, req->offs) 1883 ? pread (req->int1, req->ptr2, req->size, req->offs)
1927 : read (req->int1, req->ptr2, req->size); break; 1884 : read (req->int1, req->ptr2, req->size); break;
1928 case EIO_WRITE: req->result = req->offs >= 0 1885 case EIO_WRITE: req->result = req->offs >= 0
1929 ? pwrite (req->int1, req->ptr2, req->size, req->offs) 1886 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1930 : write (req->int1, req->ptr2, req->size); break; 1887 : write (req->int1, req->ptr2, req->size); break;
1931 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
1932 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;
1933 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
1934 1896
1935 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1897 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1936 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;
1937 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1899 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1900 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, AT_SYMLINK_NOFOLLOW); break;
1901 case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break;
1902 case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break;
1903 case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break;
1904 case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break;
1905 case EIO_SLURP: eio__slurp ( openat (dirfd, req->ptr1, O_RDONLY | O_CLOEXEC), req); break;
1906
1907 case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); 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;
1912 case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); 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;
1922 case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2, 0); break;
1923 case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
1924 case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1925 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1926 req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1927 case EIO_READLINK: ALLOC (EIO_PATH_MAX);
1928 req->result = readlinkat (dirfd, req->ptr1, req->ptr2, EIO_PATH_MAX);
1929 if (req->result == EIO_PATH_MAX)
1930 {
1931 req->result = -1;
1932 errno = ENAMETOOLONG;
1933 }
1934 break;
1935 case EIO_UTIME:
1936 case EIO_FUTIME:
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
1959#else
1960
1961 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1962 req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
1963 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1938 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 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;
1970
1971 case EIO_UNLINK: req->result = unlink (path ); break;
1972 case EIO_RMDIR: req->result = rmdir (path ); break;
1973 case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); 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;
1988
1989 case EIO_UTIME:
1990 case EIO_FUTIME:
1991 {
1992 struct timeval tv[2];
1993 struct timeval *times;
1994
1995 if (req->nv1 != -1. || req->nv2 != -1.)
1996 {
1997 tv[0].tv_sec = req->nv1;
1998 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
1999 tv[1].tv_sec = req->nv2;
2000 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
2001
2002 times = tv;
2003 }
2004 else
2005 times = 0;
2006
2007 req->result = req->type == EIO_FUTIME
2008 ? futimes (req->int1, times)
2009 : utimes (req->ptr1, times);
2010 }
2011 break;
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
1939 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2022 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1940 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; 2023 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1941 2024
1942 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1943 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1944 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS)); 2025 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1945 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break; 2026 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1946 2027
1947 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1948 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; 2028 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1949 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1950 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break; 2029 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1951 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1952 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break; 2030 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1953 2031
1954 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
1955 case EIO_CLOSE: req->result = close (req->int1); break; 2032 case EIO_CLOSE: req->result = eio__close (req->int1); break;
1956 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break; 2033 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1957 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
1958 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1959 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1960 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1961 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1962 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1963 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1964
1965 case EIO_REALPATH: eio__realpath (req, self); break;
1966
1967 case EIO_READLINK: ALLOC (PATH_MAX);
1968 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1969
1970 case EIO_SYNC: req->result = 0; sync (); break; 2034 case EIO_SYNC: req->result = 0; sync (); break;
1971 case EIO_FSYNC: req->result = fsync (req->int1); break; 2035 case EIO_FSYNC: req->result = fsync (req->int1); break;
1972 case EIO_FDATASYNC: req->result = fdatasync (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;
1973 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break; 2039 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1974 case EIO_MTOUCH: req->result = eio__mtouch (req); break; 2040 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1975 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break; 2041 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1976 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break; 2042 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1977 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1978 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break; 2043 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
1979 2044
1980 case EIO_READDIR: eio__scandir (req, self); break; 2045 case EIO_READDIR: eio__scandir (req, self); break;
1981 2046
1982 case EIO_BUSY: 2047 case EIO_BUSY:
… …
1992 req->result = select (0, 0, 0, 0, &tv); 2057 req->result = select (0, 0, 0, 0, &tv);
1993 } 2058 }
1994#endif 2059#endif
1995 break; 2060 break;
1996 2061
1997 case EIO_UTIME: 2062#if 0
1998 case EIO_FUTIME:
1999 {
2000 struct timeval tv[2];
2001 struct timeval *times;
2002
2003 if (req->nv1 != -1. || req->nv2 != -1.)
2004 {
2005 tv[0].tv_sec = req->nv1;
2006 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
2007 tv[1].tv_sec = req->nv2;
2008 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
2009
2010 times = tv;
2011 }
2012 else
2013 times = 0;
2014
2015 req->result = req->type == EIO_FUTIME
2016 ? futimes (req->int1, times)
2017 : utimes (req->ptr1, times);
2018 }
2019 break;
2020
2021 case EIO_GROUP: 2063 case EIO_GROUP:
2022 abort (); /* handled in eio_request */ 2064 abort (); /* handled in eio_request */
2065#endif
2023 2066
2024 case EIO_NOP: 2067 case EIO_NOP:
2025 req->result = 0; 2068 req->result = 0;
2026 break; 2069 break;
2027 2070
2028 case EIO_CUSTOM: 2071 case EIO_CUSTOM:
2029 req->feed (req); 2072 req->feed (req);
2030 break; 2073 break;
2031 2074
2032 default: 2075 default:
2033 errno = ENOSYS;
2034 req->result = -1; 2076 req->result = EIO_ENOSYS ();
2035 break; 2077 break;
2036 } 2078 }
2037 2079
2080alloc_fail:
2038 req->errorno = errno; 2081 req->errorno = errno;
2039} 2082}
2040 2083
2041#ifndef EIO_NO_WRAPPERS 2084#ifndef EIO_NO_WRAPPERS
2042 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
2043eio_req *eio_nop (int pri, eio_cb cb, void *data) 2096eio_req *eio_nop (int pri, eio_cb cb, void *data)
2044{ 2097{
2045 REQ (EIO_NOP); SEND; 2098 REQ (EIO_NOP); SEND;
2046} 2099}
2047 2100
… …
2063eio_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)
2064{ 2117{
2065 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;
2066} 2119}
2067 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
2068eio_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)
2069{ 2137{
2070 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;
2071} 2139}
2072 2140
… …
2078eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data) 2146eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
2079{ 2147{
2080 REQ (EIO_MLOCKALL); req->int1 = flags; SEND; 2148 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2081} 2149}
2082 2150
2083eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2084{
2085 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2086}
2087
2088eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data) 2151eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2089{ 2152{
2090 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND; 2153 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
2091} 2154}
2092 2155
2093eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2094{
2095 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2096}
2097
2098eio_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)
2099{ 2157{
2100 REQ (EIO_CLOSE); req->int1 = fd; SEND; 2158 REQ (EIO_CLOSE); req->int1 = fd; SEND;
2101} 2159}
2102 2160
2103eio_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)
2104{ 2162{
2105 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;
2106} 2164}
2107 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
2108eio_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)
2109{ 2172{
2110 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;
2111} 2174}
2112 2175
2113eio_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)
2114{ 2177{
2115 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;
2116} 2179}
2117 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
2118eio_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)
2119{ 2192{
2120 REQ (EIO_FSTAT); req->int1 = fd; SEND; 2193 REQ (EIO_FSTAT); req->int1 = fd; SEND;
2121} 2194}
2122 2195
… …
2138eio_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)
2139{ 2212{
2140 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2213 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
2141} 2214}
2142 2215
2143eio_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)
2144{ 2217{
2145 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;
2146} 2219}
2147 2220
2148eio_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)
… …
2168eio_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)
2169{ 2242{
2170 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2243 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
2171} 2244}
2172 2245
2173eio_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)
2174{ 2247{
2175 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;
2176} 2249}
2177 2250
2178eio_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)
… …
2265eio_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)
2266{ 2339{
2267 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2340 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
2268} 2341}
2269 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
2270eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data) 2348eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
2271{ 2349{
2272 REQ (EIO_CUSTOM); req->feed = execute; SEND; 2350 REQ (EIO_CUSTOM); req->feed = execute; SEND;
2273} 2351}
2274 2352
… …
2308void 2386void
2309eio_grp_add (eio_req *grp, eio_req *req) 2387eio_grp_add (eio_req *grp, eio_req *req)
2310{ 2388{
2311 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));
2312 2390
2313 grp->flags |= EIO_FLAG_GROUPADD; 2391 grp->flags |= ETP_FLAG_GROUPADD;
2314 2392
2315 ++grp->size; 2393 ++grp->size;
2316 req->grp = grp; 2394 req->grp = grp;
2317 2395
2318 req->grp_prev = 0; 2396 req->grp_prev = 0;
… …
2325} 2403}
2326 2404
2327/*****************************************************************************/ 2405/*****************************************************************************/
2328/* misc garbage */ 2406/* misc garbage */
2329 2407
2330ssize_t 2408eio_ssize_t
2331eio_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)
2332{ 2410{
2333 etp_worker wrk;
2334 ssize_t ret;
2335
2336 wrk.dbuf = 0;
2337
2338 ret = eio__sendfile (ofd, ifd, offset, count, &wrk); 2411 return eio__sendfile (ofd, ifd, offset, count);
2339
2340 if (wrk.dbuf)
2341 free (wrk.dbuf);
2342
2343 return ret;
2344} 2412}
2345 2413

Diff Legend

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