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

Comparing libeio/eio.c (file contents):
Revision 1.88 by root, Thu Jul 14 19:31:43 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>
58#include <sys/stat.h> 53#include <sys/stat.h>
59#include <limits.h> 54#include <limits.h>
60#include <fcntl.h> 55#include <fcntl.h>
61#include <assert.h> 56#include <assert.h>
62 57
63#include <sys/statvfs.h>
64/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */ 58/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
65/* intptr_t only comes from stdint.h, says idiot openbsd coder */ 59/* intptr_t only comes from stdint.h, says idiot openbsd coder */
66#if HAVE_STDINT_H 60#if HAVE_STDINT_H
67# include <stdint.h> 61# include <stdint.h>
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
96 97
97#define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1) 98#define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1)
98 99
99#ifdef _WIN32 100#ifdef _WIN32
100 101
102 #undef PAGESIZE
101 #define PAGESIZE 4096 /* GetSystemInfo? */ 103 #define PAGESIZE 4096 /* GetSystemInfo? */
102 104
105 /* TODO: look at how perl does stat (non-sloppy), unlink (ro-files), utime, link */
106
103 #ifdef EIO_STRUCT_STATI64 107 #ifdef EIO_STRUCT_STATI64
108 /* look at perl's non-sloppy stat */
104 #define stat(path,buf) _stati64 (path,buf) 109 #define stat(path,buf) _stati64 (path,buf)
105 #define fstat(fd,buf) _fstati64 (path,buf) 110 #define fstat(fd,buf) _fstati64 (fd,buf)
106 #endif 111 #endif
107 #define lstat(path,buf) stat (path,buf) 112 #define lstat(path,buf) stat (path,buf)
108 #define fsync(fd) (FlushFileBuffers (EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1)) 113 #define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
109 #define mkdir(path,mode) _mkdir (path) 114 #define mkdir(path,mode) _mkdir (path)
110 #define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1)) 115 #define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
111 116
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 ()
112 #define chown(path,uid,gid) EIO_ENOSYS () 125 #define chown(path,uid,gid) EIO_ENOSYS ()
113 #define fchown(fd,uid,gid) EIO_ENOSYS () 126 #define fchown(fd,uid,gid) EIO_ENOSYS ()
114 #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */ 127 #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
115 #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */ 128 #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
116 #define mknod(path,mode,dev) EIO_ENOSYS () 129 #define mknod(path,mode,dev) EIO_ENOSYS ()
117 #define sync() 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 }
118 200
119 /* we could even stat and see if it exists */ 201 /* we could even stat and see if it exists */
120 static int 202 static int
121 symlink (const char *old, const char *neu) 203 symlink (const char *old, const char *neu)
122 { 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)
123 if (CreateSymbolicLink (neu, old, 1)) 213 if (CreateSymbolicLink (neu, old, flags))
124 return 0; 214 return 0;
125 215 #endif
126 if (CreateSymbolicLink (neu, old, 0))
127 return 0;
128 216
129 return EIO_ERRNO (ENOENT, -1); 217 return EIO_ERRNO (ENOENT, -1);
130 } 218 }
131 219
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
132#else 234#else
133 235
236 #include <sys/ioctl.h>
134 #include <sys/time.h> 237 #include <sys/time.h>
135 #include <sys/select.h> 238 #include <sys/select.h>
136 #include <sys/statvfs.h>
137 #include <unistd.h> 239 #include <unistd.h>
138 #include <utime.h>
139 #include <signal.h> 240 #include <signal.h>
140 #include <dirent.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
141 251
142 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES 252 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
143 #include <sys/mman.h> 253 #include <sys/mman.h>
144 #endif 254 #endif
145 255
256 #define D_NAME(entp) entp->d_name
257
146 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */ 258 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
147 #if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__ 259 #if __FreeBSD__ || __NetBSD__ || __OpenBSD__
148 #define _DIRENT_HAVE_D_TYPE /* sigh */ 260 #define _DIRENT_HAVE_D_TYPE /* sigh */
149 #define D_INO(de) (de)->d_fileno 261 #define D_INO(de) (de)->d_fileno
150 #define D_NAMLEN(de) (de)->d_namlen 262 #define D_NAMLEN(de) (de)->d_namlen
151 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600 263 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
152 #define D_INO(de) (de)->d_ino 264 #define D_INO(de) (de)->d_ino
163 275
164 #ifndef EIO_STRUCT_DIRENT 276 #ifndef EIO_STRUCT_DIRENT
165 #define EIO_STRUCT_DIRENT struct dirent 277 #define EIO_STRUCT_DIRENT struct dirent
166 #endif 278 #endif
167 279
280#endif
281
282#if HAVE_UTIMES
283# include <utime.h>
284#endif
285
286#if HAVE_SYS_SYSCALL_H
287# include <sys/syscall.h>
168#endif 288#endif
169 289
170#if HAVE_SENDFILE 290#if HAVE_SENDFILE
171# if __linux 291# if __linux
172# include <sys/sendfile.h> 292# include <sys/sendfile.h>
180# else 300# else
181# error sendfile support requested but not available 301# error sendfile support requested but not available
182# endif 302# endif
183#endif 303#endif
184 304
305#if HAVE_RENAMEAT2
306# include <sys/syscall.h>
307# include <linux/fs.h>
308#endif
309
185#ifndef D_TYPE 310#ifndef D_TYPE
186# define D_TYPE(de) 0 311# define D_TYPE(de) 0
187#endif 312#endif
188#ifndef D_INO 313#ifndef D_INO
189# define D_INO(de) 0 314# define D_INO(de) 0
190#endif 315#endif
191#ifndef D_NAMLEN 316#ifndef D_NAMLEN
192# define D_NAMLEN(de) strlen ((de)->d_name) 317# define D_NAMLEN(entp) strlen (D_NAME (entp))
193#endif 318#endif
194 319
195/* used for struct dirent, AIX doesn't provide it */ 320/* used for struct dirent, AIX doesn't provide it */
196#ifndef NAME_MAX 321#ifndef NAME_MAX
197# define NAME_MAX 4096 322# define NAME_MAX 4096
198#endif 323#endif
199 324
200/* used for readlink etc. */ 325/* used for readlink etc. */
201#ifndef PATH_MAX 326#ifndef PATH_MAX
202# define PATH_MAX 4096 327# define PATH_MAX 0
203#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)
204 339
205/* buffer size for various temporary buffers */ 340/* buffer size for various temporary buffers */
206#define EIO_BUFSIZE 65536 341#define EIO_BUFSIZE 65536
207 342
208#define dBUF \ 343#define dBUF \
209 char *eio_buf; \
210 ETP_WORKER_LOCK (self); \
211 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 344 char *eio_buf = malloc (EIO_BUFSIZE); \
212 ETP_WORKER_UNLOCK (self); \
213 errno = ENOMEM; \ 345 errno = ENOMEM; \
214 if (!eio_buf) \ 346 if (!eio_buf) \
215 return -1; 347 return -1
216 348
217#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/*****************************************************************************/
218 377
219#define ETP_PRI_MIN EIO_PRI_MIN 378#define ETP_PRI_MIN EIO_PRI_MIN
220#define ETP_PRI_MAX EIO_PRI_MAX 379#define ETP_PRI_MAX EIO_PRI_MAX
221 380
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
222struct etp_worker; 391struct etp_worker;
223
224#define ETP_REQ eio_req 392#define ETP_REQ eio_req
225#define ETP_DESTROY(req) eio_destroy (req) 393#define ETP_DESTROY(req) eio_destroy (req)
226static int eio_finish (eio_req *req); 394static int eio_finish (eio_req *req);
227#define ETP_FINISH(req) eio_finish (req) 395#define ETP_FINISH(req) eio_finish (req)
228static void eio_execute (struct etp_worker *self, eio_req *req); 396static void eio_execute (struct etp_worker *self, eio_req *req);
229#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) 397#define ETP_EXECUTE(wrk,req) eio_execute (wrk, req)
230 398
231#define ETP_WORKER_CLEAR(req) \ 399#include "etp.c"
232 if (wrk->dbuf) \
233 { \
234 free (wrk->dbuf); \
235 wrk->dbuf = 0; \
236 } \
237 \
238 if (wrk->dirp) \
239 { \
240 closedir (wrk->dirp); \
241 wrk->dirp = 0; \
242 }
243 400
244#define ETP_WORKER_COMMON \ 401static struct etp_pool eio_pool;
245 void *dbuf; \ 402#define EIO_POOL (&eio_pool)
246 DIR *dirp;
247
248/*****************************************************************************/
249
250#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
251
252/* calculate time difference in ~1/EIO_TICKS of a second */
253ecb_inline int
254tvdiff (struct timeval *tv1, struct timeval *tv2)
255{
256 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
257 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
258}
259
260static unsigned int started, idle, wanted = 4;
261
262static void (*want_poll_cb) (void);
263static void (*done_poll_cb) (void);
264
265static unsigned int max_poll_time; /* reslock */
266static unsigned int max_poll_reqs; /* reslock */
267
268static volatile unsigned int nreqs; /* reqlock */
269static volatile unsigned int nready; /* reqlock */
270static volatile unsigned int npending; /* reqlock */
271static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
272static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
273
274static xmutex_t wrklock;
275static xmutex_t reslock;
276static xmutex_t reqlock;
277static xcond_t reqwait;
278
279#if !HAVE_PREADWRITE
280/*
281 * make our pread/pwrite emulation safe against themselves, but not against
282 * normal read/write by using a mutex. slows down execution a lot,
283 * but that's your problem, not mine.
284 */
285static xmutex_t preadwritelock = X_MUTEX_INIT;
286#endif
287
288typedef struct etp_worker
289{
290 /* locked by wrklock */
291 struct etp_worker *prev, *next;
292
293 xthread_t tid;
294
295 /* locked by reslock, reqlock or wrklock */
296 ETP_REQ *req; /* currently processed request */
297
298 ETP_WORKER_COMMON
299} etp_worker;
300
301static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */
302
303#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
304#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
305
306/* worker threads management */
307
308static void ecb_cold
309etp_worker_clear (etp_worker *wrk)
310{
311 ETP_WORKER_CLEAR (wrk);
312}
313
314static void ecb_cold
315etp_worker_free (etp_worker *wrk)
316{
317 wrk->next->prev = wrk->prev;
318 wrk->prev->next = wrk->next;
319
320 free (wrk);
321}
322
323static unsigned int
324etp_nreqs (void)
325{
326 int retval;
327 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
328 retval = nreqs;
329 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
330 return retval;
331}
332
333static unsigned int
334etp_nready (void)
335{
336 unsigned int retval;
337
338 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
339 retval = nready;
340 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
341
342 return retval;
343}
344
345static unsigned int
346etp_npending (void)
347{
348 unsigned int retval;
349
350 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
351 retval = npending;
352 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
353
354 return retval;
355}
356
357static unsigned int
358etp_nthreads (void)
359{
360 unsigned int retval;
361
362 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
363 retval = started;
364 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
365
366 return retval;
367}
368
369/*
370 * a somewhat faster data structure might be nice, but
371 * with 8 priorities this actually needs <20 insns
372 * per shift, the most expensive operation.
373 */
374typedef struct {
375 ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */
376 int size;
377} etp_reqq;
378
379static etp_reqq req_queue;
380static etp_reqq res_queue;
381
382static int ecb_noinline
383reqq_push (etp_reqq *q, ETP_REQ *req)
384{
385 int pri = req->pri;
386 req->next = 0;
387
388 if (q->qe[pri])
389 {
390 q->qe[pri]->next = req;
391 q->qe[pri] = req;
392 }
393 else
394 q->qe[pri] = q->qs[pri] = req;
395
396 return q->size++;
397}
398
399static ETP_REQ * ecb_noinline
400reqq_shift (etp_reqq *q)
401{
402 int pri;
403
404 if (!q->size)
405 return 0;
406
407 --q->size;
408
409 for (pri = ETP_NUM_PRI; pri--; )
410 {
411 eio_req *req = q->qs[pri];
412
413 if (req)
414 {
415 if (!(q->qs[pri] = (eio_req *)req->next))
416 q->qe[pri] = 0;
417
418 return req;
419 }
420 }
421
422 abort ();
423}
424
425static void ecb_cold
426etp_thread_init (void)
427{
428#if !HAVE_PREADWRITE
429 X_MUTEX_CREATE (preadwritelock);
430#endif
431 X_MUTEX_CREATE (wrklock);
432 X_MUTEX_CREATE (reslock);
433 X_MUTEX_CREATE (reqlock);
434 X_COND_CREATE (reqwait);
435}
436
437static void ecb_cold
438etp_atfork_prepare (void)
439{
440}
441
442static void ecb_cold
443etp_atfork_parent (void)
444{
445}
446
447static void ecb_cold
448etp_atfork_child (void)
449{
450 ETP_REQ *prv;
451
452 while ((prv = reqq_shift (&req_queue)))
453 ETP_DESTROY (prv);
454
455 while ((prv = reqq_shift (&res_queue)))
456 ETP_DESTROY (prv);
457
458 while (wrk_first.next != &wrk_first)
459 {
460 etp_worker *wrk = wrk_first.next;
461
462 if (wrk->req)
463 ETP_DESTROY (wrk->req);
464
465 etp_worker_clear (wrk);
466 etp_worker_free (wrk);
467 }
468
469 started = 0;
470 idle = 0;
471 nreqs = 0;
472 nready = 0;
473 npending = 0;
474
475 etp_thread_init ();
476}
477
478static void ecb_cold
479etp_once_init (void)
480{
481 etp_thread_init ();
482 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
483}
484
485static int ecb_cold
486etp_init (void (*want_poll)(void), void (*done_poll)(void))
487{
488 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
489
490 pthread_once (&doinit, etp_once_init);
491
492 want_poll_cb = want_poll;
493 done_poll_cb = done_poll;
494
495 return 0;
496}
497
498X_THREAD_PROC (etp_proc);
499
500static void ecb_cold
501etp_start_thread (void)
502{
503 etp_worker *wrk = calloc (1, sizeof (etp_worker));
504
505 /*TODO*/
506 assert (("unable to allocate worker thread data", wrk));
507
508 X_LOCK (wrklock);
509
510 if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
511 {
512 wrk->prev = &wrk_first;
513 wrk->next = wrk_first.next;
514 wrk_first.next->prev = wrk;
515 wrk_first.next = wrk;
516 ++started;
517 }
518 else
519 free (wrk);
520
521 X_UNLOCK (wrklock);
522}
523
524static void
525etp_maybe_start_thread (void)
526{
527 if (ecb_expect_true (etp_nthreads () >= wanted))
528 return;
529
530 /* todo: maybe use idle here, but might be less exact */
531 if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
532 return;
533
534 etp_start_thread ();
535}
536
537static void ecb_cold
538etp_end_thread (void)
539{
540 eio_req *req = calloc (1, sizeof (eio_req));
541
542 req->type = -1;
543 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
544
545 X_LOCK (reqlock);
546 reqq_push (&req_queue, req);
547 X_COND_SIGNAL (reqwait);
548 X_UNLOCK (reqlock);
549
550 X_LOCK (wrklock);
551 --started;
552 X_UNLOCK (wrklock);
553}
554
555static int
556etp_poll (void)
557{
558 unsigned int maxreqs;
559 unsigned int maxtime;
560 struct timeval tv_start, tv_now;
561
562 X_LOCK (reslock);
563 maxreqs = max_poll_reqs;
564 maxtime = max_poll_time;
565 X_UNLOCK (reslock);
566
567 if (maxtime)
568 gettimeofday (&tv_start, 0);
569
570 for (;;)
571 {
572 ETP_REQ *req;
573
574 etp_maybe_start_thread ();
575
576 X_LOCK (reslock);
577 req = reqq_shift (&res_queue);
578
579 if (req)
580 {
581 --npending;
582
583 if (!res_queue.size && done_poll_cb)
584 done_poll_cb ();
585 }
586
587 X_UNLOCK (reslock);
588
589 if (!req)
590 return 0;
591
592 X_LOCK (reqlock);
593 --nreqs;
594 X_UNLOCK (reqlock);
595
596 if (ecb_expect_false (req->type == EIO_GROUP && req->size))
597 {
598 req->int1 = 1; /* mark request as delayed */
599 continue;
600 }
601 else
602 {
603 int res = ETP_FINISH (req);
604 if (ecb_expect_false (res))
605 return res;
606 }
607
608 if (ecb_expect_false (maxreqs && !--maxreqs))
609 break;
610
611 if (maxtime)
612 {
613 gettimeofday (&tv_now, 0);
614
615 if (tvdiff (&tv_start, &tv_now) >= maxtime)
616 break;
617 }
618 }
619
620 errno = EAGAIN;
621 return -1;
622}
623
624static void
625etp_cancel (ETP_REQ *req)
626{
627 req->cancelled = 1;
628
629 eio_grp_cancel (req);
630}
631
632static void
633etp_submit (ETP_REQ *req)
634{
635 req->pri -= ETP_PRI_MIN;
636
637 if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
638 if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
639
640 if (ecb_expect_false (req->type == EIO_GROUP))
641 {
642 /* I hope this is worth it :/ */
643 X_LOCK (reqlock);
644 ++nreqs;
645 X_UNLOCK (reqlock);
646
647 X_LOCK (reslock);
648
649 ++npending;
650
651 if (!reqq_push (&res_queue, req) && want_poll_cb)
652 want_poll_cb ();
653
654 X_UNLOCK (reslock);
655 }
656 else
657 {
658 X_LOCK (reqlock);
659 ++nreqs;
660 ++nready;
661 reqq_push (&req_queue, req);
662 X_COND_SIGNAL (reqwait);
663 X_UNLOCK (reqlock);
664
665 etp_maybe_start_thread ();
666 }
667}
668
669static void ecb_cold
670etp_set_max_poll_time (double nseconds)
671{
672 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
673 max_poll_time = nseconds * EIO_TICKS;
674 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
675}
676
677static void ecb_cold
678etp_set_max_poll_reqs (unsigned int maxreqs)
679{
680 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
681 max_poll_reqs = maxreqs;
682 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
683}
684
685static void ecb_cold
686etp_set_max_idle (unsigned int nthreads)
687{
688 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
689 max_idle = nthreads;
690 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
691}
692
693static void ecb_cold
694etp_set_idle_timeout (unsigned int seconds)
695{
696 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
697 idle_timeout = seconds;
698 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
699}
700
701static void ecb_cold
702etp_set_min_parallel (unsigned int nthreads)
703{
704 if (wanted < nthreads)
705 wanted = nthreads;
706}
707
708static void ecb_cold
709etp_set_max_parallel (unsigned int nthreads)
710{
711 if (wanted > nthreads)
712 wanted = nthreads;
713
714 while (started > wanted)
715 etp_end_thread ();
716}
717 403
718/*****************************************************************************/ 404/*****************************************************************************/
719 405
720static void 406static void
721grp_try_feed (eio_req *grp) 407grp_try_feed (eio_req *grp)
722{ 408{
723 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 409 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
724 { 410 {
725 grp->flags &= ~EIO_FLAG_GROUPADD; 411 grp->flags &= ~ETP_FLAG_GROUPADD;
726 412
727 EIO_FEED (grp); 413 EIO_FEED (grp);
728 414
729 /* stop if no progress has been made */ 415 /* stop if no progress has been made */
730 if (!(grp->flags & EIO_FLAG_GROUPADD)) 416 if (!(grp->flags & ETP_FLAG_GROUPADD))
731 { 417 {
732 grp->feed = 0; 418 grp->feed = 0;
733 break; 419 break;
734 } 420 }
735 } 421 }
742 428
743 /* call feeder, if applicable */ 429 /* call feeder, if applicable */
744 grp_try_feed (grp); 430 grp_try_feed (grp);
745 431
746 /* finish, if done */ 432 /* finish, if done */
747 if (!grp->size && grp->int1) 433 if (!grp->size && grp->flags & ETP_FLAG_DELAYED)
748 return eio_finish (grp); 434 return eio_finish (grp);
749 else 435 else
750 return 0; 436 return 0;
751} 437}
752 438
788} 474}
789 475
790void 476void
791eio_grp_cancel (eio_req *grp) 477eio_grp_cancel (eio_req *grp)
792{ 478{
793 for (grp = grp->grp_first; grp; grp = grp->grp_next) 479 etp_grp_cancel (EIO_POOL, grp);
794 eio_cancel (grp);
795} 480}
796 481
797void 482void
798eio_cancel (eio_req *req) 483eio_cancel (eio_req *req)
799{ 484{
800 etp_cancel (req); 485 etp_cancel (EIO_POOL, req);
801} 486}
802 487
803void 488void
804eio_submit (eio_req *req) 489eio_submit (eio_req *req)
805{ 490{
806 etp_submit (req); 491 etp_submit (EIO_POOL, req);
807} 492}
808 493
809unsigned int 494unsigned int
810eio_nreqs (void) 495eio_nreqs (void)
811{ 496{
812 return etp_nreqs (); 497 return etp_nreqs (EIO_POOL);
813} 498}
814 499
815unsigned int 500unsigned int
816eio_nready (void) 501eio_nready (void)
817{ 502{
818 return etp_nready (); 503 return etp_nready (EIO_POOL);
819} 504}
820 505
821unsigned int 506unsigned int
822eio_npending (void) 507eio_npending (void)
823{ 508{
824 return etp_npending (); 509 return etp_npending (EIO_POOL);
825} 510}
826 511
827unsigned int ecb_cold 512unsigned int ecb_cold
828eio_nthreads (void) 513eio_nthreads (void)
829{ 514{
830 return etp_nthreads (); 515 return etp_nthreads (EIO_POOL);
831} 516}
832 517
833void ecb_cold 518void ecb_cold
834eio_set_max_poll_time (double nseconds) 519eio_set_max_poll_time (double nseconds)
835{ 520{
836 etp_set_max_poll_time (nseconds); 521 etp_set_max_poll_time (EIO_POOL, nseconds);
837} 522}
838 523
839void ecb_cold 524void ecb_cold
840eio_set_max_poll_reqs (unsigned int maxreqs) 525eio_set_max_poll_reqs (unsigned int maxreqs)
841{ 526{
842 etp_set_max_poll_reqs (maxreqs); 527 etp_set_max_poll_reqs (EIO_POOL, maxreqs);
843} 528}
844 529
845void ecb_cold 530void ecb_cold
846eio_set_max_idle (unsigned int nthreads) 531eio_set_max_idle (unsigned int nthreads)
847{ 532{
848 etp_set_max_idle (nthreads); 533 etp_set_max_idle (EIO_POOL, nthreads);
849} 534}
850 535
851void ecb_cold 536void ecb_cold
852eio_set_idle_timeout (unsigned int seconds) 537eio_set_idle_timeout (unsigned int seconds)
853{ 538{
854 etp_set_idle_timeout (seconds); 539 etp_set_idle_timeout (EIO_POOL, seconds);
855} 540}
856 541
857void ecb_cold 542void ecb_cold
858eio_set_min_parallel (unsigned int nthreads) 543eio_set_min_parallel (unsigned int nthreads)
859{ 544{
860 etp_set_min_parallel (nthreads); 545 etp_set_min_parallel (EIO_POOL, nthreads);
861} 546}
862 547
863void ecb_cold 548void ecb_cold
864eio_set_max_parallel (unsigned int nthreads) 549eio_set_max_parallel (unsigned int nthreads)
865{ 550{
866 etp_set_max_parallel (nthreads); 551 etp_set_max_parallel (EIO_POOL, nthreads);
867} 552}
868 553
869int eio_poll (void) 554int eio_poll (void)
870{ 555{
871 return etp_poll (); 556 return etp_poll (EIO_POOL);
872} 557}
873 558
874/*****************************************************************************/ 559/*****************************************************************************/
875/* work around various missing functions */ 560/* work around various missing functions */
876 561
877#if !HAVE_PREADWRITE 562#if HAVE_POSIX_CLOSE && !__linux
878# undef pread 563# define eio__close(fd) posix_close (fd, 0)
879# undef pwrite 564#else
880# define pread eio__pread 565# define eio__close(fd) close (fd)
881# define pwrite eio__pwrite
882
883static eio_ssize_t
884eio__pread (int fd, void *buf, size_t count, off_t offset)
885{
886 eio_ssize_t res;
887 off_t ooffset;
888
889 X_LOCK (preadwritelock);
890 ooffset = lseek (fd, 0, SEEK_CUR);
891 lseek (fd, offset, SEEK_SET);
892 res = read (fd, buf, count);
893 lseek (fd, ooffset, SEEK_SET);
894 X_UNLOCK (preadwritelock);
895
896 return res;
897}
898
899static eio_ssize_t
900eio__pwrite (int fd, void *buf, size_t count, off_t offset)
901{
902 eio_ssize_t res;
903 off_t ooffset;
904
905 X_LOCK (preadwritelock);
906 ooffset = lseek (fd, 0, SEEK_CUR);
907 lseek (fd, offset, SEEK_SET);
908 res = write (fd, buf, count);
909 lseek (fd, ooffset, SEEK_SET);
910 X_UNLOCK (preadwritelock);
911
912 return res;
913}
914#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}
915 576
916#ifndef HAVE_UTIMES 577#ifndef HAVE_UTIMES
917 578
918# undef utimes 579# undef utimes
919# define utimes(path,times) eio__utimes (path, times) 580# define utimes(path,times) eio__utimes (path, times)
952 613
953#if !HAVE_FDATASYNC 614#if !HAVE_FDATASYNC
954# undef fdatasync 615# undef fdatasync
955# define fdatasync(fd) fsync (fd) 616# define fdatasync(fd) fsync (fd)
956#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}
957 635
958/* sync_file_range always needs emulation */ 636/* sync_file_range always needs emulation */
959static int 637static int
960eio__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)
961{ 639{
984} 662}
985 663
986static int 664static int
987eio__fallocate (int fd, int mode, off_t offset, size_t len) 665eio__fallocate (int fd, int mode, off_t offset, size_t len)
988{ 666{
989#if HAVE_FALLOCATE 667#if HAVE_LINUX_FALLOCATE
990 return fallocate (fd, mode, offset, len); 668 return fallocate (fd, mode, offset, len);
991#else 669#else
992 errno = ENOSYS; 670 return EIO_ENOSYS ();
993 return -1;
994#endif 671#endif
995} 672}
996 673
997#if !HAVE_READAHEAD 674#if !HAVE_READAHEAD
998# undef readahead 675# undef readahead
1011 pread (fd, eio_buf, len, offset); 688 pread (fd, eio_buf, len, offset);
1012 offset += len; 689 offset += len;
1013 todo -= len; 690 todo -= len;
1014 } 691 }
1015 692
1016 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 */
1017 return count; 697 return 0;
1018} 698}
1019 699
1020#endif 700#endif
1021 701
1022/* sendfile always needs emulation */ 702/* sendfile always needs emulation */
1023static eio_ssize_t 703static eio_ssize_t
1024eio__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)
1025{ 705{
1026 eio_ssize_t written = 0; 706 eio_ssize_t written = 0;
1027 eio_ssize_t res; 707 eio_ssize_t res;
1028 708
1029 if (!count) 709 if (!count)
1057 737
1058 /* according to source inspection, this is correct, and useful behaviour */ 738 /* according to source inspection, this is correct, and useful behaviour */
1059 if (sbytes) 739 if (sbytes)
1060 res = sbytes; 740 res = sbytes;
1061 741
1062# elif defined (__APPLE__) 742# elif defined __APPLE__
1063 off_t sbytes = count; 743 off_t sbytes = count;
1064 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0); 744 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
1065 745
1066 /* according to the manpage, sbytes is always valid */ 746 /* according to the manpage, sbytes is always valid */
1067 if (sbytes) 747 if (sbytes)
1094 HANDLE h = TO_SOCKET (ifd); 774 HANDLE h = TO_SOCKET (ifd);
1095 SetFilePointer (h, offset, 0, FILE_BEGIN); 775 SetFilePointer (h, offset, 0, FILE_BEGIN);
1096 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0); 776 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1097 777
1098#else 778#else
1099 res = -1; 779 res = EIO_ENOSYS ();
1100 errno = ENOSYS;
1101#endif 780#endif
1102 781
1103 /* we assume sendfile can copy at least 128mb in one go */ 782 /* we assume sendfile can copy at least 128mb in one go */
1104 if (res <= 128 * 1024 * 1024) 783 if (res <= 128 * 1024 * 1024)
1105 { 784 {
1127 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 806 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1128 /* BSDs */ 807 /* BSDs */
1129#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... */
1130 || errno == ENOTSUP 809 || errno == ENOTSUP
1131#endif 810#endif
811#ifdef EOPNOTSUPP /* windows */
1132 || errno == EOPNOTSUPP /* BSDs */ 812 || errno == EOPNOTSUPP /* BSDs */
813#endif
1133#if __solaris 814#if __solaris
1134 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 815 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
1135#endif 816#endif
1136 ) 817 )
1137 ) 818 )
1163 844
1164 offset += cnt; 845 offset += cnt;
1165 res += cnt; 846 res += cnt;
1166 count -= cnt; 847 count -= cnt;
1167 } 848 }
849
850 FUBd;
1168 } 851 }
1169 852
1170 return res; 853 return res;
1171} 854}
1172 855
1199 /* round up length */ 882 /* round up length */
1200 *length = (*length + mask) & ~mask; 883 *length = (*length + mask) & ~mask;
1201} 884}
1202 885
1203#if !_POSIX_MEMLOCK 886#if !_POSIX_MEMLOCK
1204# define eio__mlockall(a) eio_nosyscall() 887# define eio__mlockall(a) EIO_ENOSYS ()
1205#else 888#else
1206 889
1207static int 890static int
1208eio__mlockall (int flags) 891eio__mlockall (int flags)
1209{ 892{
1276 intptr_t end = addr + len; 959 intptr_t end = addr + len;
1277 intptr_t page = eio_pagesize (); 960 intptr_t page = eio_pagesize ();
1278 961
1279 if (addr < end) 962 if (addr < end)
1280 if (flags & EIO_MT_MODIFY) /* modify */ 963 if (flags & EIO_MT_MODIFY) /* modify */
1281 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));
1282 else 965 else
1283 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));
1284 } 967 }
1285 968
1286 return 0; 969 return 0;
1287} 970}
1288 971
1289/*****************************************************************************/ 972/*****************************************************************************/
1290/* requests implemented outside eio_execute, because they are so large */ 973/* requests implemented outside eio_execute, because they are so large */
1291 974
1292static void 975static void
1293eio__realpath (eio_req *req, etp_worker *self) 976eio__lseek (eio_req *req)
1294{ 977{
1295 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{
1296 char *res; 993 char *res;
994 const char *rel = path;
1297 char *tmp1, *tmp2; 995 char *tmp1, *tmp2;
1298#if SYMLOOP_MAX > 32 996#if SYMLOOP_MAX > 32
1299 int symlinks = SYMLOOP_MAX; 997 int symlinks = SYMLOOP_MAX;
1300#else 998#else
1301 int symlinks = 32; 999 int symlinks = 32;
1302#endif 1000#endif
1303 1001
1304 req->result = -1;
1305
1306 errno = EINVAL; 1002 errno = EINVAL;
1307 if (!rel) 1003 if (!rel)
1308 return; 1004 return -1;
1309 1005
1310 errno = ENOENT; 1006 errno = ENOENT;
1311 if (!*rel) 1007 if (!*rel)
1312 return; 1008 return -1;
1313 1009
1314 if (!req->ptr2) 1010 res = etp_tmpbuf_get (tmpbuf, EIO_PATH_MAX * 3);
1315 { 1011#ifdef _WIN32
1316 X_LOCK (wrklock); 1012 if (_access (rel, 4) != 0)
1317 req->flags |= EIO_FLAG_PTR2_FREE;
1318 X_UNLOCK (wrklock);
1319 req->ptr2 = malloc (PATH_MAX * 3);
1320
1321 errno = ENOMEM;
1322 if (!req->ptr2)
1323 return; 1013 return -1;
1324 }
1325 1014
1326 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
1327 tmp1 = res + PATH_MAX; 1028 tmp1 = res + EIO_PATH_MAX;
1328 tmp2 = tmp1 + PATH_MAX; 1029 tmp2 = tmp1 + EIO_PATH_MAX;
1329 1030
1330#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 */
1331#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME) 1032#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1332 /* on linux we may be able to ask the kernel */ 1033 /* on linux we may be able to ask the kernel */
1333 { 1034 {
1334 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);
1335 1036
1336 if (fd >= 0) 1037 if (fd >= 0)
1337 { 1038 {
1338 sprintf (tmp1, "/proc/self/fd/%d", fd); 1039 sprintf (tmp1, "/proc/self/fd/%d", fd);
1339 req->result = readlink (tmp1, res, PATH_MAX); 1040 req->result = readlink (tmp1, res, EIO_PATH_MAX);
1340 close (fd);
1341
1342 /* 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);
1343 1043
1344 if (req->result > 0) 1044 if (req->result > 0)
1345 goto done; 1045 goto done;
1346 } 1046 }
1347 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)
1348 return; 1048 return -1;
1349 } 1049 }
1350#endif 1050#endif
1351#endif 1051#endif
1352 1052
1353 if (*rel != '/') 1053 if (*rel != '/')
1354 { 1054 {
1355 if (!getcwd (res, PATH_MAX)) 1055 int len;
1056
1057 errno = ENOENT;
1058 if (wd == EIO_INVALID_WD)
1356 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);
1357 1070
1358 if (res [1]) /* only use if not / */ 1071 if (res [1]) /* only use if not / */
1359 res += strlen (res); 1072 res += len;
1360 } 1073 }
1361 1074
1362 while (*rel) 1075 while (*rel)
1363 { 1076 {
1364 eio_ssize_t len, linklen; 1077 eio_ssize_t len, linklen;
1365 char *beg = rel; 1078 const char *beg = rel;
1366 1079
1367 while (*rel && *rel != '/') 1080 while (*rel && *rel != '/')
1368 ++rel; 1081 ++rel;
1369 1082
1370 len = rel - beg; 1083 len = rel - beg;
1382 1095
1383 if (beg [1] == '.' && len == 2) 1096 if (beg [1] == '.' && len == 2)
1384 { 1097 {
1385 /* .. - back up one component, if possible */ 1098 /* .. - back up one component, if possible */
1386 1099
1387 while (res != req->ptr2) 1100 while (res != tmpbuf->ptr)
1388 if (*--res == '/') 1101 if (*--res == '/')
1389 break; 1102 break;
1390 1103
1391 continue; 1104 continue;
1392 } 1105 }
1393 } 1106 }
1394 1107
1395 errno = ENAMETOOLONG; 1108 errno = ENAMETOOLONG;
1396 if (res + 1 + len + 1 >= tmp1) 1109 if (res + 1 + len + 1 >= tmp1)
1397 return; 1110 return -1;
1398 1111
1399 /* copy one component */ 1112 /* copy one component */
1400 *res = '/'; 1113 *res = '/';
1401 memcpy (res + 1, beg, len); 1114 memcpy (res + 1, beg, len);
1402 1115
1403 /* zero-terminate, for readlink */ 1116 /* zero-terminate, for readlink */
1404 res [len + 1] = 0; 1117 res [len + 1] = 0;
1405 1118
1406 /* now check if it's a symlink */ 1119 /* now check if it's a symlink */
1407 linklen = readlink (req->ptr2, tmp1, PATH_MAX); 1120 linklen = readlink (tmpbuf->ptr, tmp1, EIO_PATH_MAX);
1408 1121
1409 if (linklen < 0) 1122 if (linklen < 0)
1410 { 1123 {
1411 if (errno != EINVAL) 1124 if (errno != EINVAL)
1412 return; 1125 return -1;
1413 1126
1414 /* it's a normal directory. hopefully */ 1127 /* it's a normal directory. hopefully */
1415 res += len + 1; 1128 res += len + 1;
1416 } 1129 }
1417 else 1130 else
1418 { 1131 {
1419 /* yay, it was a symlink - build new path in tmp2 */ 1132 /* yay, it was a symlink - build new path in tmp2 */
1420 int rellen = strlen (rel); 1133 int rellen = strlen (rel);
1421 1134
1422 errno = ENAMETOOLONG; 1135 errno = ENAMETOOLONG;
1423 if (linklen + 1 + rellen >= PATH_MAX) 1136 if (linklen + 1 + rellen >= EIO_PATH_MAX) /* also catch linklen >= EIO_PATH_MAX */
1424 return; 1137 return -1;
1425 1138
1426 errno = ELOOP; 1139 errno = ELOOP;
1427 if (!--symlinks) 1140 if (!--symlinks)
1428 return; 1141 return -1;
1429 1142
1430 if (*tmp1 == '/') 1143 if (*tmp1 == '/')
1431 res = req->ptr2; /* symlink resolves to an absolute path */ 1144 res = tmpbuf->ptr; /* symlink resolves to an absolute path */
1432 1145
1433 /* 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 */
1434 memmove (tmp2 + linklen + 1, rel, rellen + 1); 1147 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1435 tmp2 [linklen] = '/'; 1148 tmp2 [linklen] = '/';
1436 memcpy (tmp2, tmp1, linklen); 1149 memcpy (tmp2, tmp1, linklen);
1438 rel = tmp2; 1151 rel = tmp2;
1439 } 1152 }
1440 } 1153 }
1441 1154
1442 /* special case for the lone root path */ 1155 /* special case for the lone root path */
1443 if (res == req->ptr2) 1156 if (res == tmpbuf->ptr)
1444 *res++ = '/'; 1157 *res++ = '/';
1445 1158
1446 req->result = res - (char *)req->ptr2; 1159 return res - (char *)tmpbuf->ptr;
1447 1160#endif
1448done:
1449 req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
1450} 1161}
1451 1162
1452static signed char 1163static signed char
1453eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) 1164eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1454{ 1165{
1462 1173
1463#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 */
1464#define EIO_SORT_FAST 60 /* when to only use insertion sort */ 1175#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1465 1176
1466static void 1177static void
1467eio_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)
1468{ 1179{
1469 unsigned char bits [9 + sizeof (ino_t) * 8]; 1180 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1470 unsigned char *bit = bits; 1181 unsigned char *bit = bits;
1471 1182
1472 assert (CHAR_BIT == 8); 1183 assert (CHAR_BIT == 8);
1473 assert (sizeof (eio_dirent) * 8 < 256); 1184 assert (sizeof (eio_dirent) * 8 < 256);
1474 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */ 1185 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1476 1187
1477 if (size <= EIO_SORT_FAST) 1188 if (size <= EIO_SORT_FAST)
1478 return; 1189 return;
1479 1190
1480 /* 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 */
1481 /* 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 */
1482 /* inode_bits must contain all inodes ORed together */ 1193 /* inode_bits must contain all inodes ORed together */
1483 /* 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 */
1484 { 1195 {
1485 ino_t endianness; 1196 eio_ino_t endianness;
1486 int i, j; 1197 int i, j;
1487 1198
1488 /* 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" */
1489 for (i = 0; i < sizeof (ino_t); ++i) 1200 for (i = 0; i < sizeof (eio_ino_t); ++i)
1490 ((unsigned char *)&endianness)[i] = i; 1201 ((unsigned char *)&endianness)[i] = i;
1491 1202
1492 *bit++ = 0; 1203 *bit++ = 0;
1493 1204
1494 for (i = 0; i < sizeof (ino_t); ++i) 1205 for (i = 0; i < sizeof (eio_ino_t); ++i)
1495 { 1206 {
1496 /* shifting off the byte offsets out of "endianness" */ 1207 /* shifting off the byte offsets out of "endianness" */
1497 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8; 1208 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1498 endianness >>= 8; 1209 endianness >>= 8;
1499 1210
1500 for (j = 0; j < 8; ++j) 1211 for (j = 0; j < 8; ++j)
1501 if (inode_bits & (((ino_t)1) << (i * 8 + j))) 1212 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1502 *bit++ = offs + j; 1213 *bit++ = offs + j;
1503 } 1214 }
1504 1215
1505 for (j = 0; j < 8; ++j) 1216 for (j = 0; j < 8; ++j)
1506 if (score_bits & (1 << j)) 1217 if (score_bits & (1 << j))
1507 *bit++ = offsetof (eio_dirent, score) * 8 + j; 1218 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1508 } 1219 }
1509 1220
1510 /* now actually do the sorting (a variant of MSD radix sort) */ 1221 /* now actually do the sorting (a variant of MSD radix sort) */
1511 { 1222 {
1512 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base; 1223 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1513 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end; 1224 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1514 unsigned char *bit_stk [9 + sizeof (ino_t) * 8]; 1225 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1515 int stk_idx = 0; 1226 int stk_idx = 0;
1516 1227
1517 base_stk [stk_idx] = dents; 1228 base_stk [stk_idx] = dents;
1518 end_stk [stk_idx] = dents + size; 1229 end_stk [stk_idx] = dents + size;
1519 bit_stk [stk_idx] = bit - 1; 1230 bit_stk [stk_idx] = bit - 1;
1598 } 1309 }
1599 } 1310 }
1600} 1311}
1601 1312
1602static void 1313static void
1603eio_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)
1604{ 1315{
1605 if (size <= 1) 1316 if (size <= 1)
1606 return; /* our insertion sort relies on size > 0 */ 1317 return; /* our insertion sort relies on size > 0 */
1607 1318
1608 /* 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 */
1616 1327
1617/* read a full directory */ 1328/* read a full directory */
1618static void 1329static void
1619eio__scandir (eio_req *req, etp_worker *self) 1330eio__scandir (eio_req *req, etp_worker *self)
1620{ 1331{
1621 DIR *dirp;
1622 EIO_STRUCT_DIRENT *entp;
1623 char *name, *names; 1332 char *name, *names;
1624 int namesalloc = 4096; 1333 int namesalloc = 4096 - sizeof (void *) * 4;
1625 int namesoffs = 0; 1334 int namesoffs = 0;
1626 int flags = req->int1; 1335 int flags = req->int1;
1627 eio_dirent *dents = 0; 1336 eio_dirent *dents = 0;
1628 int dentalloc = 128; 1337 int dentalloc = 128;
1629 int dentoffs = 0; 1338 int dentoffs = 0;
1630 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
1631 1347
1632 req->result = -1; 1348 req->result = -1;
1633 1349
1634 if (!(flags & EIO_READDIR_DENTS)) 1350 if (!(flags & EIO_READDIR_DENTS))
1635 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); 1351 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1636 1352
1637 X_LOCK (wrklock); 1353#ifdef _WIN32
1638 /* 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
1639 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
1640 1421
1641 if (req->flags & EIO_FLAG_PTR1_FREE) 1422 if (req->flags & EIO_FLAG_PTR1_FREE)
1642 free (req->ptr1); 1423 free (req->ptr1);
1643 1424
1644 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1425 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1645 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; 1426 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1646 req->ptr2 = names = malloc (namesalloc); 1427 req->ptr2 = names = malloc (namesalloc);
1647 X_UNLOCK (wrklock);
1648 1428
1649 if (dirp && names && (!flags || dents)) 1429 if (!names || (flags && !dents))
1430 return;
1431
1650 for (;;) 1432 for (;;)
1651 { 1433 {
1434 int done;
1435
1436#ifdef _WIN32
1437 done = !dirp;
1438#else
1652 errno = 0; 1439 errno = 0;
1653 entp = readdir (dirp); 1440 entp = readdir (dirp);
1441 done = !entp;
1442#endif
1654 1443
1655 if (!entp) 1444 if (done)
1656 { 1445 {
1446#ifndef _WIN32
1447 int old_errno = errno;
1448 closedir (dirp);
1449 errno = old_errno;
1450
1657 if (errno) 1451 if (errno)
1658 break; 1452 break;
1453#endif
1659 1454
1660 /* sort etc. */ 1455 /* sort etc. */
1661 req->int1 = flags; 1456 req->int1 = flags;
1662 req->result = dentoffs; 1457 req->result = dentoffs;
1663 1458
1664 if (flags & EIO_READDIR_STAT_ORDER) 1459 if (flags & EIO_READDIR_STAT_ORDER)
1665 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);
1666 else if (flags & EIO_READDIR_DIRS_FIRST) 1461 else if (flags & EIO_READDIR_DIRS_FIRST)
1667 if (flags & EIO_READDIR_FOUND_UNKNOWN) 1462 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1668 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 */
1669 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))
1670 { 1515 {
1671 /* 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
1672 eio_dirent *oth = dents + dentoffs; 1523 ent = dents + dentoffs;
1673 eio_dirent *dir = dents;
1674 1524
1675 /* 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 */
1676 /* by walking from both sides and swapping if necessary */ 1526 ent->namelen = len - 1;
1677 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)
1678 { 1587 {
1679 if (dir->type == EIO_DT_DIR) 1588 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1680 ++dir; 1589 ent->score = 1;
1681 else if ((--oth)->type == EIO_DT_DIR) 1590 else if (!strchr (name, '.')) /* absence of dots indicate likely dirs */
1682 { 1591 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1683 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1684
1685 ++dir;
1686 }
1687 } 1592 }
1688 1593 else if (ent->type == EIO_DT_DIR)
1689 /* now sort the dirs only (dirs all have the same score) */ 1594 ent->score = 0;
1690 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1691 } 1595 }
1692
1693 break;
1694 }
1695
1696 /* now add the entry to our list(s) */
1697 name = entp->d_name;
1698
1699 /* skip . and .. entries */
1700 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1701 {
1702 int len = D_NAMLEN (entp) + 1;
1703
1704 while (ecb_expect_false (namesoffs + len > namesalloc))
1705 {
1706 namesalloc *= 2;
1707 X_LOCK (wrklock);
1708 req->ptr2 = names = realloc (names, namesalloc);
1709 X_UNLOCK (wrklock);
1710
1711 if (!names)
1712 break;
1713 } 1596 }
1714 1597
1715 memcpy (names + namesoffs, name, len);
1716
1717 if (dents)
1718 {
1719 struct eio_dirent *ent;
1720
1721 if (ecb_expect_false (dentoffs == dentalloc))
1722 {
1723 dentalloc *= 2;
1724 X_LOCK (wrklock);
1725 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1726 X_UNLOCK (wrklock);
1727
1728 if (!dents)
1729 break;
1730 }
1731
1732 ent = dents + dentoffs;
1733
1734 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1735 ent->namelen = len - 1;
1736 ent->inode = D_INO (entp);
1737
1738 inode_bits |= ent->inode;
1739
1740 switch (D_TYPE (entp))
1741 {
1742 default:
1743 ent->type = EIO_DT_UNKNOWN;
1744 flags |= EIO_READDIR_FOUND_UNKNOWN;
1745 break;
1746
1747 #ifdef DT_FIFO
1748 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1749 #endif
1750 #ifdef DT_CHR
1751 case DT_CHR: ent->type = EIO_DT_CHR; break;
1752 #endif
1753 #ifdef DT_MPC
1754 case DT_MPC: ent->type = EIO_DT_MPC; break;
1755 #endif
1756 #ifdef DT_DIR
1757 case DT_DIR: ent->type = EIO_DT_DIR; break;
1758 #endif
1759 #ifdef DT_NAM
1760 case DT_NAM: ent->type = EIO_DT_NAM; break;
1761 #endif
1762 #ifdef DT_BLK
1763 case DT_BLK: ent->type = EIO_DT_BLK; break;
1764 #endif
1765 #ifdef DT_MPB
1766 case DT_MPB: ent->type = EIO_DT_MPB; break;
1767 #endif
1768 #ifdef DT_REG
1769 case DT_REG: ent->type = EIO_DT_REG; break;
1770 #endif
1771 #ifdef DT_NWK
1772 case DT_NWK: ent->type = EIO_DT_NWK; break;
1773 #endif
1774 #ifdef DT_CMP
1775 case DT_CMP: ent->type = EIO_DT_CMP; break;
1776 #endif
1777 #ifdef DT_LNK
1778 case DT_LNK: ent->type = EIO_DT_LNK; break;
1779 #endif
1780 #ifdef DT_SOCK
1781 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1782 #endif
1783 #ifdef DT_DOOR
1784 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1785 #endif
1786 #ifdef DT_WHT
1787 case DT_WHT: ent->type = EIO_DT_WHT; break;
1788 #endif
1789 }
1790
1791 ent->score = 7;
1792
1793 if (flags & EIO_READDIR_DIRS_FIRST)
1794 {
1795 if (ent->type == EIO_DT_UNKNOWN)
1796 {
1797 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1798 ent->score = 1;
1799 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1800 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1801 }
1802 else if (ent->type == EIO_DT_DIR)
1803 ent->score = 0;
1804 }
1805 }
1806
1807 namesoffs += len; 1598 namesoffs += len;
1808 ++dentoffs; 1599 ++dentoffs;
1809 } 1600 }
1810 1601
1811 if (EIO_CANCELLED (req)) 1602 if (EIO_CANCELLED (req))
1812 { 1603 {
1813 errno = ECANCELED; 1604 errno = ECANCELED;
1814 break; 1605 break;
1815 } 1606 }
1607
1608#ifdef _WIN32
1609 if (!FindNextFile (dirp, &entp))
1610 {
1611 FindClose (dirp);
1612 dirp = 0;
1816 } 1613 }
1614#endif
1615 }
1817} 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
1818 1751
1819/*****************************************************************************/ 1752/*****************************************************************************/
1820 1753
1821#define ALLOC(len) \ 1754#define ALLOC(len) \
1822 if (!req->ptr2) \ 1755 if (!req->ptr2) \
1823 { \ 1756 { \
1824 X_LOCK (wrklock); \ 1757 X_LOCK (EIO_POOL->wrklock); \
1825 req->flags |= EIO_FLAG_PTR2_FREE; \ 1758 req->flags |= EIO_FLAG_PTR2_FREE; \
1826 X_UNLOCK (wrklock); \ 1759 X_UNLOCK (EIO_POOL->wrklock); \
1827 req->ptr2 = malloc (len); \ 1760 req->ptr2 = malloc (len); \
1828 if (!req->ptr2) \ 1761 if (!req->ptr2) \
1829 { \ 1762 { \
1830 errno = ENOMEM; \ 1763 errno = ENOMEM; \
1831 req->result = -1; \ 1764 req->result = -1; \
1832 break; \ 1765 goto alloc_fail; \
1833 } \ 1766 } \
1834 } 1767 }
1835 1768
1836X_THREAD_PROC (etp_proc) 1769/*****************************************************************************/
1837{
1838 ETP_REQ *req;
1839 struct timespec ts;
1840 etp_worker *self = (etp_worker *)thr_arg;
1841 int timeout;
1842 1770
1843 /* try to distribute timeouts somewhat evenly */ 1771static void
1844 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 1772eio__slurp (int fd, eio_req *req)
1773{
1774 req->result = fd;
1845 1775
1846 for (;;) 1776 if (fd < 0)
1777 return;
1778
1779 if (req->offs < 0 || !req->size) /* do we need the size? */
1847 { 1780 {
1848 ts.tv_sec = 0; 1781 off_t size = lseek (fd, 0, SEEK_END);
1849 1782
1850 X_LOCK (reqlock);
1851
1852 for (;;)
1853 {
1854 self->req = req = reqq_shift (&req_queue);
1855
1856 if (req)
1857 break;
1858
1859 if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
1860 {
1861 X_UNLOCK (reqlock);
1862 X_LOCK (wrklock);
1863 --started;
1864 X_UNLOCK (wrklock);
1865 goto quit;
1866 }
1867
1868 ++idle;
1869
1870 if (idle <= max_idle)
1871 /* we are allowed to idle, so do so without any timeout */
1872 X_COND_WAIT (reqwait, reqlock);
1873 else
1874 {
1875 /* initialise timeout once */
1876 if (!ts.tv_sec)
1877 ts.tv_sec = time (0) + idle_timeout;
1878
1879 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
1880 ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1881 }
1882
1883 --idle;
1884 }
1885
1886 --nready;
1887
1888 X_UNLOCK (reqlock);
1889
1890 if (req->type < 0) 1783 if (req->offs < 0)
1891 goto quit; 1784 req->offs += size;
1892 1785
1893 ETP_EXECUTE (self, req); 1786 if (!req->size)
1894 1787 req->size = size - req->offs;
1895 X_LOCK (reslock);
1896
1897 ++npending;
1898
1899 if (!reqq_push (&res_queue, req) && want_poll_cb)
1900 want_poll_cb ();
1901
1902 self->req = 0;
1903 etp_worker_clear (self);
1904
1905 X_UNLOCK (reslock);
1906 } 1788 }
1907 1789
1908quit: 1790 ALLOC (req->size);
1909 X_LOCK (wrklock); 1791 req->result = pread (fd, req->ptr2, req->size, req->offs);
1910 etp_worker_free (self);
1911 X_UNLOCK (wrklock);
1912 1792
1913 return 0; 1793 silent_close (fd);
1914}
1915 1794
1916/*****************************************************************************/ 1795alloc_fail:
1796 ;
1797}
1917 1798
1918int ecb_cold 1799int ecb_cold
1919eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1800eio_init (void (*want_poll)(void), void (*done_poll)(void))
1920{ 1801{
1921 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);
1922} 1806}
1923 1807
1924ecb_inline void 1808ecb_inline void
1925eio_api_destroy (eio_req *req) 1809eio_api_destroy (eio_req *req)
1926{ 1810{
1927 free (req); 1811 free (req);
1928} 1812}
1929 1813
1930#define REQ(rtype) \ 1814#define REQ(rtype) \
1931 eio_req *req; \ 1815 eio_req *req; \
1932 \ 1816 \
1933 req = (eio_req *)calloc (1, sizeof *req); \ 1817 req = (eio_req *)calloc (1, sizeof *req); \
1934 if (!req) \ 1818 if (!req) \
1935 return 0; \ 1819 return 0; \
1949 { \ 1833 { \
1950 eio_api_destroy (req); \ 1834 eio_api_destroy (req); \
1951 return 0; \ 1835 return 0; \
1952 } 1836 }
1953 1837
1838#define SINGLEDOT(ptr) (0[(char *)(ptr)] == '.' && !1[(char *)(ptr)])
1839
1954static void 1840static void
1955eio_execute (etp_worker *self, eio_req *req) 1841eio_execute (etp_worker *self, eio_req *req)
1956{ 1842{
1843#if HAVE_AT
1844 int dirfd;
1845#else
1846 const char *path;
1847#endif
1848
1957 if (ecb_expect_false (EIO_CANCELLED (req))) 1849 if (ecb_expect_false (EIO_CANCELLED (req)))
1958 { 1850 {
1959 req->result = -1; 1851 req->result = -1;
1960 req->errorno = ECANCELED; 1852 req->errorno = ECANCELED;
1961 return; 1853 return;
1962 } 1854 }
1963 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
1964 switch (req->type) 1872 switch (req->type)
1965 { 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;
1966 case EIO_READ: ALLOC (req->size); 1881 case EIO_READ: ALLOC (req->size);
1967 req->result = req->offs >= 0 1882 req->result = req->offs >= 0
1968 ? pread (req->int1, req->ptr2, req->size, req->offs) 1883 ? pread (req->int1, req->ptr2, req->size, req->offs)
1969 : read (req->int1, req->ptr2, req->size); break; 1884 : read (req->int1, req->ptr2, req->size); break;
1970 case EIO_WRITE: req->result = req->offs >= 0 1885 case EIO_WRITE: req->result = req->offs >= 0
1971 ? pwrite (req->int1, req->ptr2, req->size, req->offs) 1886 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1972 : write (req->int1, req->ptr2, req->size); break; 1887 : write (req->int1, req->ptr2, req->size); break;
1973 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
1974 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;
1975 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
1976 1896
1977 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1897 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1978 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;
1979 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));
1980 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
1981 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2022 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1982 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; 2023 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1983 2024
1984 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1985 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1986 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS)); 2025 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1987 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break; 2026 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1988 2027
1989 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1990 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;
1991 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1992 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;
1993 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1994 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break; 2030 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1995 2031
1996 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
1997 case EIO_CLOSE: req->result = close (req->int1); break; 2032 case EIO_CLOSE: req->result = eio__close (req->int1); break;
1998 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break; 2033 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1999 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
2000 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
2001 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
2002 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
2003 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
2004 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
2005 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
2006
2007 case EIO_REALPATH: eio__realpath (req, self); break;
2008
2009 case EIO_READLINK: ALLOC (PATH_MAX);
2010 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
2011
2012 case EIO_SYNC: req->result = 0; sync (); break; 2034 case EIO_SYNC: req->result = 0; sync (); break;
2013 case EIO_FSYNC: req->result = fsync (req->int1); break; 2035 case EIO_FSYNC: req->result = fsync (req->int1); break;
2014 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;
2015 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;
2016 case EIO_MTOUCH: req->result = eio__mtouch (req); break; 2040 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
2017 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;
2018 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break; 2042 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
2019 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
2020 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;
2021 2044
2022 case EIO_READDIR: eio__scandir (req, self); break; 2045 case EIO_READDIR: eio__scandir (req, self); break;
2023 2046
2024 case EIO_BUSY: 2047 case EIO_BUSY:
2034 req->result = select (0, 0, 0, 0, &tv); 2057 req->result = select (0, 0, 0, 0, &tv);
2035 } 2058 }
2036#endif 2059#endif
2037 break; 2060 break;
2038 2061
2039 case EIO_UTIME: 2062#if 0
2040 case EIO_FUTIME:
2041 {
2042 struct timeval tv[2];
2043 struct timeval *times;
2044
2045 if (req->nv1 != -1. || req->nv2 != -1.)
2046 {
2047 tv[0].tv_sec = req->nv1;
2048 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
2049 tv[1].tv_sec = req->nv2;
2050 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
2051
2052 times = tv;
2053 }
2054 else
2055 times = 0;
2056
2057 req->result = req->type == EIO_FUTIME
2058 ? futimes (req->int1, times)
2059 : utimes (req->ptr1, times);
2060 }
2061 break;
2062
2063 case EIO_GROUP: 2063 case EIO_GROUP:
2064 abort (); /* handled in eio_request */ 2064 abort (); /* handled in eio_request */
2065#endif
2065 2066
2066 case EIO_NOP: 2067 case EIO_NOP:
2067 req->result = 0; 2068 req->result = 0;
2068 break; 2069 break;
2069 2070
2070 case EIO_CUSTOM: 2071 case EIO_CUSTOM:
2071 req->feed (req); 2072 req->feed (req);
2072 break; 2073 break;
2073 2074
2074 default: 2075 default:
2075 errno = ENOSYS;
2076 req->result = -1; 2076 req->result = EIO_ENOSYS ();
2077 break; 2077 break;
2078 } 2078 }
2079 2079
2080alloc_fail:
2080 req->errorno = errno; 2081 req->errorno = errno;
2081} 2082}
2082 2083
2083#ifndef EIO_NO_WRAPPERS 2084#ifndef EIO_NO_WRAPPERS
2084 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
2085eio_req *eio_nop (int pri, eio_cb cb, void *data) 2096eio_req *eio_nop (int pri, eio_cb cb, void *data)
2086{ 2097{
2087 REQ (EIO_NOP); SEND; 2098 REQ (EIO_NOP); SEND;
2088} 2099}
2089 2100
2105eio_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)
2106{ 2117{
2107 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;
2108} 2119}
2109 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
2110eio_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)
2111{ 2137{
2112 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;
2113} 2139}
2114 2140
2120eio_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)
2121{ 2147{
2122 REQ (EIO_MLOCKALL); req->int1 = flags; SEND; 2148 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2123} 2149}
2124 2150
2125eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2126{
2127 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2128}
2129
2130eio_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)
2131{ 2152{
2132 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;
2133} 2154}
2134 2155
2135eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2136{
2137 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2138}
2139
2140eio_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)
2141{ 2157{
2142 REQ (EIO_CLOSE); req->int1 = fd; SEND; 2158 REQ (EIO_CLOSE); req->int1 = fd; SEND;
2143} 2159}
2144 2160
2145eio_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)
2146{ 2162{
2147 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;
2148} 2164}
2149 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
2150eio_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)
2151{ 2172{
2152 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;
2153} 2174}
2154 2175
2155eio_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)
2156{ 2177{
2157 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;
2179}
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;
2158} 2189}
2159 2190
2160eio_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)
2161{ 2192{
2162 REQ (EIO_FSTAT); req->int1 = fd; SEND; 2193 REQ (EIO_FSTAT); req->int1 = fd; SEND;
2307eio_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)
2308{ 2339{
2309 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2340 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
2310} 2341}
2311 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
2312eio_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)
2313{ 2349{
2314 REQ (EIO_CUSTOM); req->feed = execute; SEND; 2350 REQ (EIO_CUSTOM); req->feed = execute; SEND;
2315} 2351}
2316 2352
2350void 2386void
2351eio_grp_add (eio_req *grp, eio_req *req) 2387eio_grp_add (eio_req *grp, eio_req *req)
2352{ 2388{
2353 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));
2354 2390
2355 grp->flags |= EIO_FLAG_GROUPADD; 2391 grp->flags |= ETP_FLAG_GROUPADD;
2356 2392
2357 ++grp->size; 2393 ++grp->size;
2358 req->grp = grp; 2394 req->grp = grp;
2359 2395
2360 req->grp_prev = 0; 2396 req->grp_prev = 0;
2370/* misc garbage */ 2406/* misc garbage */
2371 2407
2372eio_ssize_t 2408eio_ssize_t
2373eio_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)
2374{ 2410{
2375 etp_worker wrk;
2376 eio_ssize_t ret;
2377
2378 wrk.dbuf = 0;
2379
2380 ret = eio__sendfile (ofd, ifd, offset, count, &wrk); 2411 return eio__sendfile (ofd, ifd, offset, count);
2381
2382 if (wrk.dbuf)
2383 free (wrk.dbuf);
2384
2385 return ret;
2386} 2412}
2387 2413

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines