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

Comparing libeio/eio.c (file contents):
Revision 1.89 by root, Thu Jul 14 22:36:17 2011 UTC vs.
Revision 1.128 by root, Mon Feb 18 03:16:30 2013 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 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,
43 43
44#include "eio.h" 44#include "eio.h"
45#include "ecb.h" 45#include "ecb.h"
46 46
47#ifdef EIO_STACKSIZE 47#ifdef EIO_STACKSIZE
48# define XTHREAD_STACKSIZE EIO_STACKSIZE 48# define X_STACKSIZE EIO_STACKSIZE
49#endif 49#endif
50#include "xthread.h" 50#include "xthread.h"
51 51
52#include <errno.h> 52#include <errno.h>
53#include <stddef.h> 53#include <stddef.h>
67#endif 67#endif
68 68
69#ifndef ECANCELED 69#ifndef ECANCELED
70# define ECANCELED EDOM 70# define ECANCELED EDOM
71#endif 71#endif
72#ifndef ELOOP
73# define ELOOP EDOM
74#endif
75
76#if !defined(ENOTSOCK) && defined(WSAENOTSOCK)
77# define ENOTSOCK WSAENOTSOCK
78#endif
72 79
73static void eio_destroy (eio_req *req); 80static void eio_destroy (eio_req *req);
74 81
75#ifndef EIO_FINISH 82#ifndef EIO_FINISH
76# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0 83# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
95 102
96#define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1) 103#define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1)
97 104
98#ifdef _WIN32 105#ifdef _WIN32
99 106
107 #undef PAGESIZE
100 #define PAGESIZE 4096 /* GetSystemInfo? */ 108 #define PAGESIZE 4096 /* GetSystemInfo? */
101 109
110 /* TODO: look at how perl does stat (non-sloppy), unlink (ro-files), utime, link */
111
102 #ifdef EIO_STRUCT_STATI64 112 #ifdef EIO_STRUCT_STATI64
113 /* look at perl's non-sloppy stat */
103 #define stat(path,buf) _stati64 (path,buf) 114 #define stat(path,buf) _stati64 (path,buf)
104 #define fstat(fd,buf) _fstati64 (path,buf) 115 #define fstat(fd,buf) _fstati64 (fd,buf)
105 #endif 116 #endif
106 #define lstat(path,buf) stat (path,buf) 117 #define lstat(path,buf) stat (path,buf)
107 #define fsync(fd) (FlushFileBuffers (EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1)) 118 #define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
108 #define mkdir(path,mode) _mkdir (path) 119 #define mkdir(path,mode) _mkdir (path)
109 #define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1)) 120 #define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
110 121
122 #define chmod(path,mode) _chmod (path, mode)
123 #define dup(fd) _dup (fd)
124 #define dup2(fd1,fd2) _dup2 (fd1, fd2)
125
126 #define fchmod(fd,mode) EIO_ENOSYS ()
111 #define chown(path,uid,gid) EIO_ENOSYS () 127 #define chown(path,uid,gid) EIO_ENOSYS ()
112 #define fchown(fd,uid,gid) EIO_ENOSYS () 128 #define fchown(fd,uid,gid) EIO_ENOSYS ()
113 #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */ 129 #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
114 #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */ 130 #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
115 #define mknod(path,mode,dev) EIO_ENOSYS () 131 #define mknod(path,mode,dev) EIO_ENOSYS ()
116 #define sync() EIO_ENOSYS () 132 #define sync() EIO_ENOSYS ()
133 #define readlink(path,buf,s) EIO_ENOSYS ()
134 #define statvfs(path,buf) EIO_ENOSYS ()
135 #define fstatvfs(fd,buf) EIO_ENOSYS ()
136
137 #define pread(fd,buf,count,offset) eio__pread (fd, buf, count, offset)
138 #define pwrite(fd,buf,count,offset) eio__pwrite (fd, buf, count, offset)
139
140 #if __GNUC__
141 typedef long long eio_off_t; /* signed for compatibility to msvc */
142 #else
143 typedef __int64 eio_off_t; /* unsigned not supported by msvc */
144 #endif
145
146 static eio_ssize_t
147 eio__pread (int fd, void *buf, eio_ssize_t count, eio_off_t offset)
148 {
149 OVERLAPPED o = { 0 };
150 DWORD got;
151
152 o.Offset = offset;
153 o.OffsetHigh = offset >> 32;
154
155 return ReadFile ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd), buf, count, &got, &o)
156 ? got : -1;
157 }
158
159 static eio_ssize_t
160 eio__pwrite (int fd, void *buf, eio_ssize_t count, eio_off_t offset)
161 {
162 OVERLAPPED o = { 0 };
163 DWORD got;
164
165 o.Offset = offset;
166 o.OffsetHigh = offset >> 32;
167
168 return WriteFile ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd), buf, count, &got, &o)
169 ? got : -1;
170 }
171
172 /* rename() uses MoveFile, which fails to overwrite */
173 #define rename(old,neu) eio__rename (old, neu)
174
175 static int
176 eio__rename (const char *old, const char *neu)
177 {
178 if (MoveFileEx (old, neu, MOVEFILE_REPLACE_EXISTING))
179 return 0;
180
181 /* should steal _dosmaperr */
182 switch (GetLastError ())
183 {
184 case ERROR_FILE_NOT_FOUND:
185 case ERROR_PATH_NOT_FOUND:
186 case ERROR_INVALID_DRIVE:
187 case ERROR_NO_MORE_FILES:
188 case ERROR_BAD_NETPATH:
189 case ERROR_BAD_NET_NAME:
190 case ERROR_BAD_PATHNAME:
191 case ERROR_FILENAME_EXCED_RANGE:
192 errno = ENOENT;
193 break;
194
195 default:
196 errno = EACCES;
197 break;
198 }
199
200 return -1;
201 }
117 202
118 /* we could even stat and see if it exists */ 203 /* we could even stat and see if it exists */
119 static int 204 static int
120 symlink (const char *old, const char *neu) 205 symlink (const char *old, const char *neu)
121 { 206 {
207 #if WINVER >= 0x0600
122 if (CreateSymbolicLink (neu, old, 1)) 208 if (CreateSymbolicLink (neu, old, 1))
123 return 0; 209 return 0;
124 210
125 if (CreateSymbolicLink (neu, old, 0)) 211 if (CreateSymbolicLink (neu, old, 0))
126 return 0; 212 return 0;
213 #endif
127 214
128 return EIO_ERRNO (ENOENT, -1); 215 return EIO_ERRNO (ENOENT, -1);
129 } 216 }
130 217
218 /* POSIX API only */
219 #define CreateHardLink(neu,old,flags) 0
220 #define CreateSymbolicLink(neu,old,flags) 0
221
222 struct statvfs
223 {
224 int dummy;
225 };
226
227 #define DT_DIR EIO_DT_DIR
228 #define DT_REG EIO_DT_REG
229 #define D_NAME(entp) entp.cFileName
230 #define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
231
131#else 232#else
132 233
133 #include <sys/time.h> 234 #include <sys/time.h>
134 #include <sys/select.h> 235 #include <sys/select.h>
135 #include <sys/statvfs.h>
136 #include <unistd.h> 236 #include <unistd.h>
137 #include <utime.h>
138 #include <signal.h> 237 #include <signal.h>
139 #include <dirent.h> 238 #include <dirent.h>
239
240 #ifdef ANDROID
241 #include <sys/vfs.h>
242 #define statvfs statfs
243 #define fstatvfs fstatfs
244 #include <asm/page.h> /* supposedly limits.h does #define PAGESIZE PAGESIZE */
245 #else
246 #include <sys/statvfs.h>
247 #endif
140 248
141 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES 249 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
142 #include <sys/mman.h> 250 #include <sys/mman.h>
143 #endif 251 #endif
144 252
253 #define D_NAME(entp) entp->d_name
254
145 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */ 255 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
146 #if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__ 256 #if __FreeBSD__ || __NetBSD__ || __OpenBSD__
147 #define _DIRENT_HAVE_D_TYPE /* sigh */ 257 #define _DIRENT_HAVE_D_TYPE /* sigh */
148 #define D_INO(de) (de)->d_fileno 258 #define D_INO(de) (de)->d_fileno
149 #define D_NAMLEN(de) (de)->d_namlen 259 #define D_NAMLEN(de) (de)->d_namlen
150 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600 260 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
151 #define D_INO(de) (de)->d_ino 261 #define D_INO(de) (de)->d_ino
162 272
163 #ifndef EIO_STRUCT_DIRENT 273 #ifndef EIO_STRUCT_DIRENT
164 #define EIO_STRUCT_DIRENT struct dirent 274 #define EIO_STRUCT_DIRENT struct dirent
165 #endif 275 #endif
166 276
277#endif
278
279#if HAVE_UTIMES
280# include <utime.h>
281#endif
282
283#if HAVE_SYS_SYSCALL_H
284# include <sys/syscall.h>
285#endif
286
287#if HAVE_SYS_PRCTL_H
288# include <sys/prctl.h>
167#endif 289#endif
168 290
169#if HAVE_SENDFILE 291#if HAVE_SENDFILE
170# if __linux 292# if __linux
171# include <sys/sendfile.h> 293# include <sys/sendfile.h>
186#endif 308#endif
187#ifndef D_INO 309#ifndef D_INO
188# define D_INO(de) 0 310# define D_INO(de) 0
189#endif 311#endif
190#ifndef D_NAMLEN 312#ifndef D_NAMLEN
191# define D_NAMLEN(de) strlen ((de)->d_name) 313# define D_NAMLEN(entp) strlen (D_NAME (entp))
192#endif 314#endif
193 315
194/* used for struct dirent, AIX doesn't provide it */ 316/* used for struct dirent, AIX doesn't provide it */
195#ifndef NAME_MAX 317#ifndef NAME_MAX
196# define NAME_MAX 4096 318# define NAME_MAX 4096
202#endif 324#endif
203 325
204/* buffer size for various temporary buffers */ 326/* buffer size for various temporary buffers */
205#define EIO_BUFSIZE 65536 327#define EIO_BUFSIZE 65536
206 328
207#define dBUF \ 329#define dBUF \
208 char *eio_buf; \
209 ETP_WORKER_LOCK (self); \
210 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 330 char *eio_buf = malloc (EIO_BUFSIZE); \
211 ETP_WORKER_UNLOCK (self); \
212 errno = ENOMEM; \ 331 errno = ENOMEM; \
213 if (!eio_buf) \ 332 if (!eio_buf) \
214 return -1; 333 return -1
334
335#define FUBd \
336 free (eio_buf)
215 337
216#define EIO_TICKS ((1000000 + 1023) >> 10) 338#define EIO_TICKS ((1000000 + 1023) >> 10)
339
340/*****************************************************************************/
341
342struct tmpbuf
343{
344 void *ptr;
345 int len;
346};
347
348static void *
349tmpbuf_get (struct tmpbuf *buf, int len)
350{
351 if (buf->len < len)
352 {
353 free (buf->ptr);
354 buf->ptr = malloc (buf->len = len);
355 }
356
357 return buf->ptr;
358}
359
360struct tmpbuf;
361
362#if _POSIX_VERSION >= 200809L
363 #define HAVE_AT 1
364 #define WD2FD(wd) ((wd) ? (wd)->fd : AT_FDCWD)
365 #ifndef O_SEARCH
366 #define O_SEARCH O_RDONLY
367 #endif
368#else
369 #define HAVE_AT 0
370 static const char *wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path);
371#endif
372
373struct eio_pwd
374{
375#if HAVE_AT
376 int fd;
377#endif
378 int len;
379 char str[1]; /* actually, a 0-terminated canonical path */
380};
381
382/*****************************************************************************/
217 383
218#define ETP_PRI_MIN EIO_PRI_MIN 384#define ETP_PRI_MIN EIO_PRI_MIN
219#define ETP_PRI_MAX EIO_PRI_MAX 385#define ETP_PRI_MAX EIO_PRI_MAX
220 386
221struct etp_worker; 387struct etp_worker;
225static int eio_finish (eio_req *req); 391static int eio_finish (eio_req *req);
226#define ETP_FINISH(req) eio_finish (req) 392#define ETP_FINISH(req) eio_finish (req)
227static void eio_execute (struct etp_worker *self, eio_req *req); 393static void eio_execute (struct etp_worker *self, eio_req *req);
228#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) 394#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
229 395
230#define ETP_WORKER_CLEAR(req) \
231 if (wrk->dbuf) \
232 { \
233 free (wrk->dbuf); \
234 wrk->dbuf = 0; \
235 } \
236 \
237 if (wrk->dirp) \
238 { \
239 closedir (wrk->dirp); \
240 wrk->dirp = 0; \
241 }
242
243#define ETP_WORKER_COMMON \
244 void *dbuf; \
245 DIR *dirp;
246
247/*****************************************************************************/ 396/*****************************************************************************/
248 397
249#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) 398#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
250 399
251/* calculate time difference in ~1/EIO_TICKS of a second */ 400/* calculate time difference in ~1/EIO_TICKS of a second */
262static void (*done_poll_cb) (void); 411static void (*done_poll_cb) (void);
263 412
264static unsigned int max_poll_time; /* reslock */ 413static unsigned int max_poll_time; /* reslock */
265static unsigned int max_poll_reqs; /* reslock */ 414static unsigned int max_poll_reqs; /* reslock */
266 415
267static volatile unsigned int nreqs; /* reqlock */ 416static unsigned int nreqs; /* reqlock */
268static volatile unsigned int nready; /* reqlock */ 417static unsigned int nready; /* reqlock */
269static volatile unsigned int npending; /* reqlock */ 418static unsigned int npending; /* reqlock */
270static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */ 419static unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
271static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */ 420static unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
272 421
273static xmutex_t wrklock; 422static xmutex_t wrklock;
274static xmutex_t reslock; 423static xmutex_t reslock;
275static xmutex_t reqlock; 424static xmutex_t reqlock;
276static xcond_t reqwait; 425static xcond_t reqwait;
277 426
278#if !HAVE_PREADWRITE
279/*
280 * make our pread/pwrite emulation safe against themselves, but not against
281 * normal read/write by using a mutex. slows down execution a lot,
282 * but that's your problem, not mine.
283 */
284static xmutex_t preadwritelock = X_MUTEX_INIT;
285#endif
286
287typedef struct etp_worker 427typedef struct etp_worker
288{ 428{
429 struct tmpbuf tmpbuf;
430
289 /* locked by wrklock */ 431 /* locked by wrklock */
290 struct etp_worker *prev, *next; 432 struct etp_worker *prev, *next;
291 433
292 xthread_t tid; 434 xthread_t tid;
293 435
294 /* locked by reslock, reqlock or wrklock */ 436#ifdef ETP_WORKER_COMMON
295 ETP_REQ *req; /* currently processed request */
296
297 ETP_WORKER_COMMON 437 ETP_WORKER_COMMON
438#endif
298} etp_worker; 439} etp_worker;
299 440
300static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */ 441static etp_worker wrk_first; /* NOT etp */
301 442
302#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock) 443#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
303#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock) 444#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
304 445
305/* worker threads management */ 446/* worker threads management */
306 447
307static void ecb_cold 448static void
308etp_worker_clear (etp_worker *wrk) 449etp_worker_clear (etp_worker *wrk)
309{ 450{
310 ETP_WORKER_CLEAR (wrk);
311} 451}
312 452
313static void ecb_cold 453static void ecb_cold
314etp_worker_free (etp_worker *wrk) 454etp_worker_free (etp_worker *wrk)
315{ 455{
456 free (wrk->tmpbuf.ptr);
457
316 wrk->next->prev = wrk->prev; 458 wrk->next->prev = wrk->prev;
317 wrk->prev->next = wrk->next; 459 wrk->prev->next = wrk->next;
318 460
319 free (wrk); 461 free (wrk);
320} 462}
376} etp_reqq; 518} etp_reqq;
377 519
378static etp_reqq req_queue; 520static etp_reqq req_queue;
379static etp_reqq res_queue; 521static etp_reqq res_queue;
380 522
523static void ecb_noinline ecb_cold
524reqq_init (etp_reqq *q)
525{
526 int pri;
527
528 for (pri = 0; pri < ETP_NUM_PRI; ++pri)
529 q->qs[pri] = q->qe[pri] = 0;
530
531 q->size = 0;
532}
533
381static int ecb_noinline 534static int ecb_noinline
382reqq_push (etp_reqq *q, ETP_REQ *req) 535reqq_push (etp_reqq *q, ETP_REQ *req)
383{ 536{
384 int pri = req->pri; 537 int pri = req->pri;
385 req->next = 0; 538 req->next = 0;
419 } 572 }
420 573
421 abort (); 574 abort ();
422} 575}
423 576
424static void ecb_cold 577static int ecb_cold
425etp_thread_init (void) 578etp_init (void (*want_poll)(void), void (*done_poll)(void))
426{ 579{
427#if !HAVE_PREADWRITE
428 X_MUTEX_CREATE (preadwritelock);
429#endif
430 X_MUTEX_CREATE (wrklock); 580 X_MUTEX_CREATE (wrklock);
431 X_MUTEX_CREATE (reslock); 581 X_MUTEX_CREATE (reslock);
432 X_MUTEX_CREATE (reqlock); 582 X_MUTEX_CREATE (reqlock);
433 X_COND_CREATE (reqwait); 583 X_COND_CREATE (reqwait);
434}
435 584
436static void ecb_cold 585 reqq_init (&req_queue);
437etp_atfork_prepare (void) 586 reqq_init (&res_queue);
438{
439}
440 587
441static void ecb_cold 588 wrk_first.next =
442etp_atfork_parent (void) 589 wrk_first.prev = &wrk_first;
443{
444}
445
446static void ecb_cold
447etp_atfork_child (void)
448{
449 ETP_REQ *prv;
450
451 while ((prv = reqq_shift (&req_queue)))
452 ETP_DESTROY (prv);
453
454 while ((prv = reqq_shift (&res_queue)))
455 ETP_DESTROY (prv);
456
457 while (wrk_first.next != &wrk_first)
458 {
459 etp_worker *wrk = wrk_first.next;
460
461 if (wrk->req)
462 ETP_DESTROY (wrk->req);
463
464 etp_worker_clear (wrk);
465 etp_worker_free (wrk);
466 }
467 590
468 started = 0; 591 started = 0;
469 idle = 0; 592 idle = 0;
470 nreqs = 0; 593 nreqs = 0;
471 nready = 0; 594 nready = 0;
472 npending = 0; 595 npending = 0;
473 596
474 etp_thread_init ();
475}
476
477static void ecb_cold
478etp_once_init (void)
479{
480 etp_thread_init ();
481 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
482}
483
484static int ecb_cold
485etp_init (void (*want_poll)(void), void (*done_poll)(void))
486{
487 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
488
489 pthread_once (&doinit, etp_once_init);
490
491 want_poll_cb = want_poll; 597 want_poll_cb = want_poll;
492 done_poll_cb = done_poll; 598 done_poll_cb = done_poll;
493 599
494 return 0; 600 return 0;
495} 601}
504 /*TODO*/ 610 /*TODO*/
505 assert (("unable to allocate worker thread data", wrk)); 611 assert (("unable to allocate worker thread data", wrk));
506 612
507 X_LOCK (wrklock); 613 X_LOCK (wrklock);
508 614
509 if (thread_create (&wrk->tid, etp_proc, (void *)wrk)) 615 if (xthread_create (&wrk->tid, etp_proc, (void *)wrk))
510 { 616 {
511 wrk->prev = &wrk_first; 617 wrk->prev = &wrk_first;
512 wrk->next = wrk_first.next; 618 wrk->next = wrk_first.next;
513 wrk_first.next->prev = wrk; 619 wrk_first.next->prev = wrk;
514 wrk_first.next = wrk; 620 wrk_first.next = wrk;
534} 640}
535 641
536static void ecb_cold 642static void ecb_cold
537etp_end_thread (void) 643etp_end_thread (void)
538{ 644{
539 eio_req *req = calloc (1, sizeof (eio_req)); 645 eio_req *req = calloc (1, sizeof (eio_req)); /* will be freed by worker */
540 646
541 req->type = -1; 647 req->type = -1;
542 req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 648 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
543 649
544 X_LOCK (reqlock); 650 X_LOCK (reqlock);
871} 977}
872 978
873/*****************************************************************************/ 979/*****************************************************************************/
874/* work around various missing functions */ 980/* work around various missing functions */
875 981
876#if !HAVE_PREADWRITE
877# undef pread
878# undef pwrite
879# define pread eio__pread
880# define pwrite eio__pwrite
881
882static eio_ssize_t
883eio__pread (int fd, void *buf, size_t count, off_t offset)
884{
885 eio_ssize_t res;
886 off_t ooffset;
887
888 X_LOCK (preadwritelock);
889 ooffset = lseek (fd, 0, SEEK_CUR);
890 lseek (fd, offset, SEEK_SET);
891 res = read (fd, buf, count);
892 lseek (fd, ooffset, SEEK_SET);
893 X_UNLOCK (preadwritelock);
894
895 return res;
896}
897
898static eio_ssize_t
899eio__pwrite (int fd, void *buf, size_t count, off_t offset)
900{
901 eio_ssize_t res;
902 off_t ooffset;
903
904 X_LOCK (preadwritelock);
905 ooffset = lseek (fd, 0, SEEK_CUR);
906 lseek (fd, offset, SEEK_SET);
907 res = write (fd, buf, count);
908 lseek (fd, ooffset, SEEK_SET);
909 X_UNLOCK (preadwritelock);
910
911 return res;
912}
913#endif
914
915#ifndef HAVE_UTIMES 982#ifndef HAVE_UTIMES
916 983
917# undef utimes 984# undef utimes
918# define utimes(path,times) eio__utimes (path, times) 985# define utimes(path,times) eio__utimes (path, times)
919 986
951 1018
952#if !HAVE_FDATASYNC 1019#if !HAVE_FDATASYNC
953# undef fdatasync 1020# undef fdatasync
954# define fdatasync(fd) fsync (fd) 1021# define fdatasync(fd) fsync (fd)
955#endif 1022#endif
1023
1024static int
1025eio__syncfs (int fd)
1026{
1027 int res;
1028
1029#if HAVE_SYS_SYNCFS
1030 res = (int)syscall (__NR_syncfs, (int)(fd));
1031#else
1032 res = EIO_ENOSYS ();
1033#endif
1034
1035 if (res < 0 && errno == ENOSYS && fd >= 0)
1036 sync ();
1037
1038 return res;
1039}
956 1040
957/* sync_file_range always needs emulation */ 1041/* sync_file_range always needs emulation */
958static int 1042static int
959eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags) 1043eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
960{ 1044{
983} 1067}
984 1068
985static int 1069static int
986eio__fallocate (int fd, int mode, off_t offset, size_t len) 1070eio__fallocate (int fd, int mode, off_t offset, size_t len)
987{ 1071{
988#if HAVE_FALLOCATE 1072#if HAVE_LINUX_FALLOCATE
989 return fallocate (fd, mode, offset, len); 1073 return fallocate (fd, mode, offset, len);
990#else 1074#else
991 errno = ENOSYS; 1075 return EIO_ENOSYS ();
992 return -1;
993#endif 1076#endif
994} 1077}
995 1078
996#if !HAVE_READAHEAD 1079#if !HAVE_READAHEAD
997# undef readahead 1080# undef readahead
1010 pread (fd, eio_buf, len, offset); 1093 pread (fd, eio_buf, len, offset);
1011 offset += len; 1094 offset += len;
1012 todo -= len; 1095 todo -= len;
1013 } 1096 }
1014 1097
1015 errno = 0; 1098 FUBd;
1099
1100 /* linux's readahead basically only fails for EBADF or EINVAL (not mmappable) */
1101 /* but not for e.g. EIO or eof, so we also never fail */
1016 return count; 1102 return 0;
1017} 1103}
1018 1104
1019#endif 1105#endif
1020 1106
1021/* sendfile always needs emulation */ 1107/* sendfile always needs emulation */
1022static eio_ssize_t 1108static eio_ssize_t
1023eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) 1109eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
1024{ 1110{
1025 eio_ssize_t written = 0; 1111 eio_ssize_t written = 0;
1026 eio_ssize_t res; 1112 eio_ssize_t res;
1027 1113
1028 if (!count) 1114 if (!count)
1056 1142
1057 /* according to source inspection, this is correct, and useful behaviour */ 1143 /* according to source inspection, this is correct, and useful behaviour */
1058 if (sbytes) 1144 if (sbytes)
1059 res = sbytes; 1145 res = sbytes;
1060 1146
1061# elif defined (__APPLE__) 1147# elif defined __APPLE__
1062 off_t sbytes = count; 1148 off_t sbytes = count;
1063 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0); 1149 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
1064 1150
1065 /* according to the manpage, sbytes is always valid */ 1151 /* according to the manpage, sbytes is always valid */
1066 if (sbytes) 1152 if (sbytes)
1093 HANDLE h = TO_SOCKET (ifd); 1179 HANDLE h = TO_SOCKET (ifd);
1094 SetFilePointer (h, offset, 0, FILE_BEGIN); 1180 SetFilePointer (h, offset, 0, FILE_BEGIN);
1095 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0); 1181 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1096 1182
1097#else 1183#else
1098 res = -1; 1184 res = EIO_ENOSYS ();
1099 errno = ENOSYS;
1100#endif 1185#endif
1101 1186
1102 /* we assume sendfile can copy at least 128mb in one go */ 1187 /* we assume sendfile can copy at least 128mb in one go */
1103 if (res <= 128 * 1024 * 1024) 1188 if (res <= 128 * 1024 * 1024)
1104 { 1189 {
1126 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1211 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1127 /* BSDs */ 1212 /* BSDs */
1128#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */ 1213#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1129 || errno == ENOTSUP 1214 || errno == ENOTSUP
1130#endif 1215#endif
1216#ifdef EOPNOTSUPP /* windows */
1131 || errno == EOPNOTSUPP /* BSDs */ 1217 || errno == EOPNOTSUPP /* BSDs */
1218#endif
1132#if __solaris 1219#if __solaris
1133 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1220 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
1134#endif 1221#endif
1135 ) 1222 )
1136 ) 1223 )
1162 1249
1163 offset += cnt; 1250 offset += cnt;
1164 res += cnt; 1251 res += cnt;
1165 count -= cnt; 1252 count -= cnt;
1166 } 1253 }
1254
1255 FUBd;
1167 } 1256 }
1168 1257
1169 return res; 1258 return res;
1170} 1259}
1171 1260
1287 1376
1288/*****************************************************************************/ 1377/*****************************************************************************/
1289/* requests implemented outside eio_execute, because they are so large */ 1378/* requests implemented outside eio_execute, because they are so large */
1290 1379
1291static void 1380static void
1292eio__realpath (eio_req *req, etp_worker *self) 1381eio__lseek (eio_req *req)
1293{ 1382{
1294 char *rel = req->ptr1; 1383 /* this usually gets optimised away completely, or your compiler sucks, */
1384 /* or the whence constants really are not 0, 1, 2 */
1385 int whence = req->int2 == EIO_SEEK_SET ? SEEK_SET
1386 : req->int2 == EIO_SEEK_CUR ? SEEK_CUR
1387 : req->int2 == EIO_SEEK_END ? SEEK_END
1388 : req->int2;
1389
1390 req->offs = lseek (req->int1, req->offs, whence);
1391 req->result = req->offs == (off_t)-1 ? -1 : 0;
1392}
1393
1394/* result will always end up in tmpbuf, there is always space for adding a 0-byte */
1395static int
1396eio__realpath (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
1397{
1398 const char *rel = path;
1295 char *res; 1399 char *res;
1296 char *tmp1, *tmp2; 1400 char *tmp1, *tmp2;
1297#if SYMLOOP_MAX > 32 1401#if SYMLOOP_MAX > 32
1298 int symlinks = SYMLOOP_MAX; 1402 int symlinks = SYMLOOP_MAX;
1299#else 1403#else
1300 int symlinks = 32; 1404 int symlinks = 32;
1301#endif 1405#endif
1302 1406
1303 req->result = -1;
1304
1305 errno = EINVAL; 1407 errno = EINVAL;
1306 if (!rel) 1408 if (!rel)
1307 return; 1409 return -1;
1308 1410
1309 errno = ENOENT; 1411 errno = ENOENT;
1310 if (!*rel) 1412 if (!*rel)
1311 return; 1413 return -1;
1312 1414
1313 if (!req->ptr2) 1415 res = tmpbuf_get (tmpbuf, PATH_MAX * 3);
1314 {
1315 X_LOCK (wrklock);
1316 req->flags |= EIO_FLAG_PTR2_FREE;
1317 X_UNLOCK (wrklock);
1318 req->ptr2 = malloc (PATH_MAX * 3);
1319
1320 errno = ENOMEM;
1321 if (!req->ptr2)
1322 return;
1323 }
1324
1325 res = req->ptr2;
1326 tmp1 = res + PATH_MAX; 1416 tmp1 = res + PATH_MAX;
1327 tmp2 = tmp1 + PATH_MAX; 1417 tmp2 = tmp1 + PATH_MAX;
1328 1418
1329#if 0 /* disabled, the musl way to do things is just too racy */ 1419#if 0 /* disabled, the musl way to do things is just too racy */
1330#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME) 1420#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1334 1424
1335 if (fd >= 0) 1425 if (fd >= 0)
1336 { 1426 {
1337 sprintf (tmp1, "/proc/self/fd/%d", fd); 1427 sprintf (tmp1, "/proc/self/fd/%d", fd);
1338 req->result = readlink (tmp1, res, PATH_MAX); 1428 req->result = readlink (tmp1, res, PATH_MAX);
1429 /* here we should probably stat the open file and the disk file, to make sure they still match */
1339 close (fd); 1430 close (fd);
1340
1341 /* here we should probably stat the open file and the disk file, to make sure they still match */
1342 1431
1343 if (req->result > 0) 1432 if (req->result > 0)
1344 goto done; 1433 goto done;
1345 } 1434 }
1346 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO) 1435 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1347 return; 1436 return -1;
1348 } 1437 }
1349#endif 1438#endif
1350#endif 1439#endif
1351 1440
1352 if (*rel != '/') 1441 if (*rel != '/')
1353 { 1442 {
1443 int len;
1444
1445 errno = ENOENT;
1446 if (wd == EIO_INVALID_WD)
1447 return -1;
1448
1449 if (wd == EIO_CWD)
1450 {
1354 if (!getcwd (res, PATH_MAX)) 1451 if (!getcwd (res, PATH_MAX))
1355 return; 1452 return -1;
1453
1454 len = strlen (res);
1455 }
1456 else
1457 memcpy (res, wd->str, len = wd->len);
1356 1458
1357 if (res [1]) /* only use if not / */ 1459 if (res [1]) /* only use if not / */
1358 res += strlen (res); 1460 res += len;
1359 } 1461 }
1360 1462
1361 while (*rel) 1463 while (*rel)
1362 { 1464 {
1363 eio_ssize_t len, linklen; 1465 eio_ssize_t len, linklen;
1364 char *beg = rel; 1466 const char *beg = rel;
1365 1467
1366 while (*rel && *rel != '/') 1468 while (*rel && *rel != '/')
1367 ++rel; 1469 ++rel;
1368 1470
1369 len = rel - beg; 1471 len = rel - beg;
1381 1483
1382 if (beg [1] == '.' && len == 2) 1484 if (beg [1] == '.' && len == 2)
1383 { 1485 {
1384 /* .. - back up one component, if possible */ 1486 /* .. - back up one component, if possible */
1385 1487
1386 while (res != req->ptr2) 1488 while (res != tmpbuf->ptr)
1387 if (*--res == '/') 1489 if (*--res == '/')
1388 break; 1490 break;
1389 1491
1390 continue; 1492 continue;
1391 } 1493 }
1392 } 1494 }
1393 1495
1394 errno = ENAMETOOLONG; 1496 errno = ENAMETOOLONG;
1395 if (res + 1 + len + 1 >= tmp1) 1497 if (res + 1 + len + 1 >= tmp1)
1396 return; 1498 return -1;
1397 1499
1398 /* copy one component */ 1500 /* copy one component */
1399 *res = '/'; 1501 *res = '/';
1400 memcpy (res + 1, beg, len); 1502 memcpy (res + 1, beg, len);
1401 1503
1402 /* zero-terminate, for readlink */ 1504 /* zero-terminate, for readlink */
1403 res [len + 1] = 0; 1505 res [len + 1] = 0;
1404 1506
1405 /* now check if it's a symlink */ 1507 /* now check if it's a symlink */
1406 linklen = readlink (req->ptr2, tmp1, PATH_MAX); 1508 linklen = readlink (tmpbuf->ptr, tmp1, PATH_MAX);
1407 1509
1408 if (linklen < 0) 1510 if (linklen < 0)
1409 { 1511 {
1410 if (errno != EINVAL) 1512 if (errno != EINVAL)
1411 return; 1513 return -1;
1412 1514
1413 /* it's a normal directory. hopefully */ 1515 /* it's a normal directory. hopefully */
1414 res += len + 1; 1516 res += len + 1;
1415 } 1517 }
1416 else 1518 else
1418 /* yay, it was a symlink - build new path in tmp2 */ 1520 /* yay, it was a symlink - build new path in tmp2 */
1419 int rellen = strlen (rel); 1521 int rellen = strlen (rel);
1420 1522
1421 errno = ENAMETOOLONG; 1523 errno = ENAMETOOLONG;
1422 if (linklen + 1 + rellen >= PATH_MAX) 1524 if (linklen + 1 + rellen >= PATH_MAX)
1423 return; 1525 return -1;
1424 1526
1425 errno = ELOOP; 1527 errno = ELOOP;
1426 if (!--symlinks) 1528 if (!--symlinks)
1427 return; 1529 return -1;
1428 1530
1429 if (*tmp1 == '/') 1531 if (*tmp1 == '/')
1430 res = req->ptr2; /* symlink resolves to an absolute path */ 1532 res = tmpbuf->ptr; /* symlink resolves to an absolute path */
1431 1533
1432 /* we need to be careful, as rel might point into tmp2 already */ 1534 /* we need to be careful, as rel might point into tmp2 already */
1433 memmove (tmp2 + linklen + 1, rel, rellen + 1); 1535 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1434 tmp2 [linklen] = '/'; 1536 tmp2 [linklen] = '/';
1435 memcpy (tmp2, tmp1, linklen); 1537 memcpy (tmp2, tmp1, linklen);
1437 rel = tmp2; 1539 rel = tmp2;
1438 } 1540 }
1439 } 1541 }
1440 1542
1441 /* special case for the lone root path */ 1543 /* special case for the lone root path */
1442 if (res == req->ptr2) 1544 if (res == tmpbuf->ptr)
1443 *res++ = '/'; 1545 *res++ = '/';
1444 1546
1445 req->result = res - (char *)req->ptr2; 1547 return res - (char *)tmpbuf->ptr;
1446
1447done:
1448 req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
1449} 1548}
1450 1549
1451static signed char 1550static signed char
1452eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) 1551eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1453{ 1552{
1461 1560
1462#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */ 1561#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1463#define EIO_SORT_FAST 60 /* when to only use insertion sort */ 1562#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1464 1563
1465static void 1564static void
1466eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1565eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1467{ 1566{
1468 unsigned char bits [9 + sizeof (ino_t) * 8]; 1567 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1469 unsigned char *bit = bits; 1568 unsigned char *bit = bits;
1470 1569
1471 assert (CHAR_BIT == 8); 1570 assert (CHAR_BIT == 8);
1472 assert (sizeof (eio_dirent) * 8 < 256); 1571 assert (sizeof (eio_dirent) * 8 < 256);
1473 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */ 1572 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1475 1574
1476 if (size <= EIO_SORT_FAST) 1575 if (size <= EIO_SORT_FAST)
1477 return; 1576 return;
1478 1577
1479 /* first prepare an array of bits to test in our radix sort */ 1578 /* first prepare an array of bits to test in our radix sort */
1480 /* try to take endianness into account, as well as differences in ino_t sizes */ 1579 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1481 /* inode_bits must contain all inodes ORed together */ 1580 /* inode_bits must contain all inodes ORed together */
1482 /* which is used to skip bits that are 0 everywhere, which is very common */ 1581 /* which is used to skip bits that are 0 everywhere, which is very common */
1483 { 1582 {
1484 ino_t endianness; 1583 eio_ino_t endianness;
1485 int i, j; 1584 int i, j;
1486 1585
1487 /* we store the byte offset of byte n into byte n of "endianness" */ 1586 /* we store the byte offset of byte n into byte n of "endianness" */
1488 for (i = 0; i < sizeof (ino_t); ++i) 1587 for (i = 0; i < sizeof (eio_ino_t); ++i)
1489 ((unsigned char *)&endianness)[i] = i; 1588 ((unsigned char *)&endianness)[i] = i;
1490 1589
1491 *bit++ = 0; 1590 *bit++ = 0;
1492 1591
1493 for (i = 0; i < sizeof (ino_t); ++i) 1592 for (i = 0; i < sizeof (eio_ino_t); ++i)
1494 { 1593 {
1495 /* shifting off the byte offsets out of "endianness" */ 1594 /* shifting off the byte offsets out of "endianness" */
1496 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8; 1595 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1497 endianness >>= 8; 1596 endianness >>= 8;
1498 1597
1499 for (j = 0; j < 8; ++j) 1598 for (j = 0; j < 8; ++j)
1500 if (inode_bits & (((ino_t)1) << (i * 8 + j))) 1599 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1501 *bit++ = offs + j; 1600 *bit++ = offs + j;
1502 } 1601 }
1503 1602
1504 for (j = 0; j < 8; ++j) 1603 for (j = 0; j < 8; ++j)
1505 if (score_bits & (1 << j)) 1604 if (score_bits & (1 << j))
1506 *bit++ = offsetof (eio_dirent, score) * 8 + j; 1605 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1507 } 1606 }
1508 1607
1509 /* now actually do the sorting (a variant of MSD radix sort) */ 1608 /* now actually do the sorting (a variant of MSD radix sort) */
1510 { 1609 {
1511 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base; 1610 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1512 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end; 1611 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1513 unsigned char *bit_stk [9 + sizeof (ino_t) * 8]; 1612 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1514 int stk_idx = 0; 1613 int stk_idx = 0;
1515 1614
1516 base_stk [stk_idx] = dents; 1615 base_stk [stk_idx] = dents;
1517 end_stk [stk_idx] = dents + size; 1616 end_stk [stk_idx] = dents + size;
1518 bit_stk [stk_idx] = bit - 1; 1617 bit_stk [stk_idx] = bit - 1;
1597 } 1696 }
1598 } 1697 }
1599} 1698}
1600 1699
1601static void 1700static void
1602eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1701eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1603{ 1702{
1604 if (size <= 1) 1703 if (size <= 1)
1605 return; /* our insertion sort relies on size > 0 */ 1704 return; /* our insertion sort relies on size > 0 */
1606 1705
1607 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */ 1706 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1615 1714
1616/* read a full directory */ 1715/* read a full directory */
1617static void 1716static void
1618eio__scandir (eio_req *req, etp_worker *self) 1717eio__scandir (eio_req *req, etp_worker *self)
1619{ 1718{
1620 DIR *dirp;
1621 EIO_STRUCT_DIRENT *entp;
1622 char *name, *names; 1719 char *name, *names;
1623 int namesalloc = 4096; 1720 int namesalloc = 4096 - sizeof (void *) * 4;
1624 int namesoffs = 0; 1721 int namesoffs = 0;
1625 int flags = req->int1; 1722 int flags = req->int1;
1626 eio_dirent *dents = 0; 1723 eio_dirent *dents = 0;
1627 int dentalloc = 128; 1724 int dentalloc = 128;
1628 int dentoffs = 0; 1725 int dentoffs = 0;
1629 ino_t inode_bits = 0; 1726 eio_ino_t inode_bits = 0;
1727#ifdef _WIN32
1728 HANDLE dirp;
1729 WIN32_FIND_DATA entp;
1730#else
1731 DIR *dirp;
1732 EIO_STRUCT_DIRENT *entp;
1733#endif
1630 1734
1631 req->result = -1; 1735 req->result = -1;
1632 1736
1633 if (!(flags & EIO_READDIR_DENTS)) 1737 if (!(flags & EIO_READDIR_DENTS))
1634 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); 1738 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1635 1739
1636 X_LOCK (wrklock); 1740#ifdef _WIN32
1637 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 1741 {
1742 int len = strlen ((const char *)req->ptr1);
1743 char *path = malloc (MAX_PATH);
1744 const char *fmt;
1745 const char *reqpath = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1746
1747 if (!len)
1748 fmt = "./*";
1749 else if (reqpath[len - 1] == '/' || reqpath[len - 1] == '\\')
1750 fmt = "%s*";
1751 else
1752 fmt = "%s/*";
1753
1754 _snprintf (path, MAX_PATH, fmt, reqpath);
1755 dirp = FindFirstFile (path, &entp);
1756 free (path);
1757
1758 if (dirp == INVALID_HANDLE_VALUE)
1759 {
1760 /* should steal _dosmaperr */
1761 switch (GetLastError ())
1762 {
1763 case ERROR_FILE_NOT_FOUND:
1764 req->result = 0;
1765 break;
1766
1767 case ERROR_INVALID_NAME:
1768 case ERROR_PATH_NOT_FOUND:
1769 case ERROR_NO_MORE_FILES:
1770 errno = ENOENT;
1771 break;
1772
1773 case ERROR_NOT_ENOUGH_MEMORY:
1774 errno = ENOMEM;
1775 break;
1776
1777 default:
1778 errno = EINVAL;
1779 break;
1780 }
1781
1782 return;
1783 }
1784 }
1785#else
1786 #if HAVE_AT
1787 if (req->wd)
1788 {
1789 int fd = openat (WD2FD (req->wd), req->ptr1, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1790
1791 if (fd < 0)
1792 return;
1793
1794 dirp = fdopendir (fd);
1795
1796 if (!dirp)
1797 close (fd);
1798 }
1799 else
1638 self->dirp = dirp = opendir (req->ptr1); 1800 dirp = opendir (req->ptr1);
1801 #else
1802 dirp = opendir (wd_expand (&self->tmpbuf, req->wd, req->ptr1));
1803 #endif
1804
1805 if (!dirp)
1806 return;
1807#endif
1639 1808
1640 if (req->flags & EIO_FLAG_PTR1_FREE) 1809 if (req->flags & EIO_FLAG_PTR1_FREE)
1641 free (req->ptr1); 1810 free (req->ptr1);
1642 1811
1643 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1812 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1644 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; 1813 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1645 req->ptr2 = names = malloc (namesalloc); 1814 req->ptr2 = names = malloc (namesalloc);
1646 X_UNLOCK (wrklock);
1647 1815
1648 if (dirp && names && (!flags || dents)) 1816 if (!names || (flags && !dents))
1817 return;
1818
1649 for (;;) 1819 for (;;)
1650 { 1820 {
1821 int done;
1822
1823#ifdef _WIN32
1824 done = !dirp;
1825#else
1651 errno = 0; 1826 errno = 0;
1652 entp = readdir (dirp); 1827 entp = readdir (dirp);
1828 done = !entp;
1829#endif
1653 1830
1654 if (!entp) 1831 if (done)
1655 { 1832 {
1833#ifndef _WIN32
1834 int old_errno = errno;
1835 closedir (dirp);
1836 errno = old_errno;
1837
1656 if (errno) 1838 if (errno)
1657 break; 1839 break;
1840#endif
1658 1841
1659 /* sort etc. */ 1842 /* sort etc. */
1660 req->int1 = flags; 1843 req->int1 = flags;
1661 req->result = dentoffs; 1844 req->result = dentoffs;
1662 1845
1663 if (flags & EIO_READDIR_STAT_ORDER) 1846 if (flags & EIO_READDIR_STAT_ORDER)
1664 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits); 1847 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1665 else if (flags & EIO_READDIR_DIRS_FIRST) 1848 else if (flags & EIO_READDIR_DIRS_FIRST)
1666 if (flags & EIO_READDIR_FOUND_UNKNOWN) 1849 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1667 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */ 1850 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1668 else 1851 else
1852 {
1853 /* in this case, all is known, and we just put dirs first and sort them */
1854 eio_dirent *oth = dents + dentoffs;
1855 eio_dirent *dir = dents;
1856
1857 /* now partition dirs to the front, and non-dirs to the back */
1858 /* by walking from both sides and swapping if necessary */
1859 while (oth > dir)
1860 {
1861 if (dir->type == EIO_DT_DIR)
1862 ++dir;
1863 else if ((--oth)->type == EIO_DT_DIR)
1864 {
1865 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1866
1867 ++dir;
1868 }
1869 }
1870
1871 /* now sort the dirs only (dirs all have the same score) */
1872 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1873 }
1874
1875 break;
1876 }
1877
1878 /* now add the entry to our list(s) */
1879 name = D_NAME (entp);
1880
1881 /* skip . and .. entries */
1882 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1883 {
1884 int len = D_NAMLEN (entp) + 1;
1885
1886 while (ecb_expect_false (namesoffs + len > namesalloc))
1887 {
1888 namesalloc *= 2;
1889 req->ptr2 = names = realloc (names, namesalloc);
1890
1891 if (!names)
1892 break;
1893 }
1894
1895 memcpy (names + namesoffs, name, len);
1896
1897 if (dents)
1898 {
1899 struct eio_dirent *ent;
1900
1901 if (ecb_expect_false (dentoffs == dentalloc))
1669 { 1902 {
1670 /* in this case, all is known, and we just put dirs first and sort them */ 1903 dentalloc *= 2;
1904 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1905
1906 if (!dents)
1907 break;
1908 }
1909
1671 eio_dirent *oth = dents + dentoffs; 1910 ent = dents + dentoffs;
1672 eio_dirent *dir = dents;
1673 1911
1674 /* now partition dirs to the front, and non-dirs to the back */ 1912 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1675 /* by walking from both sides and swapping if necessary */ 1913 ent->namelen = len - 1;
1676 while (oth > dir) 1914 ent->inode = D_INO (entp);
1915
1916 inode_bits |= ent->inode;
1917
1918 switch (D_TYPE (entp))
1919 {
1920 default:
1921 ent->type = EIO_DT_UNKNOWN;
1922 flags |= EIO_READDIR_FOUND_UNKNOWN;
1923 break;
1924
1925 #ifdef DT_FIFO
1926 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1927 #endif
1928 #ifdef DT_CHR
1929 case DT_CHR: ent->type = EIO_DT_CHR; break;
1930 #endif
1931 #ifdef DT_MPC
1932 case DT_MPC: ent->type = EIO_DT_MPC; break;
1933 #endif
1934 #ifdef DT_DIR
1935 case DT_DIR: ent->type = EIO_DT_DIR; break;
1936 #endif
1937 #ifdef DT_NAM
1938 case DT_NAM: ent->type = EIO_DT_NAM; break;
1939 #endif
1940 #ifdef DT_BLK
1941 case DT_BLK: ent->type = EIO_DT_BLK; break;
1942 #endif
1943 #ifdef DT_MPB
1944 case DT_MPB: ent->type = EIO_DT_MPB; break;
1945 #endif
1946 #ifdef DT_REG
1947 case DT_REG: ent->type = EIO_DT_REG; break;
1948 #endif
1949 #ifdef DT_NWK
1950 case DT_NWK: ent->type = EIO_DT_NWK; break;
1951 #endif
1952 #ifdef DT_CMP
1953 case DT_CMP: ent->type = EIO_DT_CMP; break;
1954 #endif
1955 #ifdef DT_LNK
1956 case DT_LNK: ent->type = EIO_DT_LNK; break;
1957 #endif
1958 #ifdef DT_SOCK
1959 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1960 #endif
1961 #ifdef DT_DOOR
1962 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1963 #endif
1964 #ifdef DT_WHT
1965 case DT_WHT: ent->type = EIO_DT_WHT; break;
1966 #endif
1967 }
1968
1969 ent->score = 7;
1970
1971 if (flags & EIO_READDIR_DIRS_FIRST)
1972 {
1973 if (ent->type == EIO_DT_UNKNOWN)
1677 { 1974 {
1678 if (dir->type == EIO_DT_DIR) 1975 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1679 ++dir; 1976 ent->score = 1;
1680 else if ((--oth)->type == EIO_DT_DIR) 1977 else if (!strchr (name, '.')) /* absence of dots indicate likely dirs */
1681 { 1978 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1682 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1683
1684 ++dir;
1685 }
1686 } 1979 }
1687 1980 else if (ent->type == EIO_DT_DIR)
1688 /* now sort the dirs only (dirs all have the same score) */ 1981 ent->score = 0;
1689 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1690 } 1982 }
1691
1692 break;
1693 }
1694
1695 /* now add the entry to our list(s) */
1696 name = entp->d_name;
1697
1698 /* skip . and .. entries */
1699 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1700 {
1701 int len = D_NAMLEN (entp) + 1;
1702
1703 while (ecb_expect_false (namesoffs + len > namesalloc))
1704 {
1705 namesalloc *= 2;
1706 X_LOCK (wrklock);
1707 req->ptr2 = names = realloc (names, namesalloc);
1708 X_UNLOCK (wrklock);
1709
1710 if (!names)
1711 break;
1712 } 1983 }
1713 1984
1714 memcpy (names + namesoffs, name, len);
1715
1716 if (dents)
1717 {
1718 struct eio_dirent *ent;
1719
1720 if (ecb_expect_false (dentoffs == dentalloc))
1721 {
1722 dentalloc *= 2;
1723 X_LOCK (wrklock);
1724 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1725 X_UNLOCK (wrklock);
1726
1727 if (!dents)
1728 break;
1729 }
1730
1731 ent = dents + dentoffs;
1732
1733 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1734 ent->namelen = len - 1;
1735 ent->inode = D_INO (entp);
1736
1737 inode_bits |= ent->inode;
1738
1739 switch (D_TYPE (entp))
1740 {
1741 default:
1742 ent->type = EIO_DT_UNKNOWN;
1743 flags |= EIO_READDIR_FOUND_UNKNOWN;
1744 break;
1745
1746 #ifdef DT_FIFO
1747 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1748 #endif
1749 #ifdef DT_CHR
1750 case DT_CHR: ent->type = EIO_DT_CHR; break;
1751 #endif
1752 #ifdef DT_MPC
1753 case DT_MPC: ent->type = EIO_DT_MPC; break;
1754 #endif
1755 #ifdef DT_DIR
1756 case DT_DIR: ent->type = EIO_DT_DIR; break;
1757 #endif
1758 #ifdef DT_NAM
1759 case DT_NAM: ent->type = EIO_DT_NAM; break;
1760 #endif
1761 #ifdef DT_BLK
1762 case DT_BLK: ent->type = EIO_DT_BLK; break;
1763 #endif
1764 #ifdef DT_MPB
1765 case DT_MPB: ent->type = EIO_DT_MPB; break;
1766 #endif
1767 #ifdef DT_REG
1768 case DT_REG: ent->type = EIO_DT_REG; break;
1769 #endif
1770 #ifdef DT_NWK
1771 case DT_NWK: ent->type = EIO_DT_NWK; break;
1772 #endif
1773 #ifdef DT_CMP
1774 case DT_CMP: ent->type = EIO_DT_CMP; break;
1775 #endif
1776 #ifdef DT_LNK
1777 case DT_LNK: ent->type = EIO_DT_LNK; break;
1778 #endif
1779 #ifdef DT_SOCK
1780 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1781 #endif
1782 #ifdef DT_DOOR
1783 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1784 #endif
1785 #ifdef DT_WHT
1786 case DT_WHT: ent->type = EIO_DT_WHT; break;
1787 #endif
1788 }
1789
1790 ent->score = 7;
1791
1792 if (flags & EIO_READDIR_DIRS_FIRST)
1793 {
1794 if (ent->type == EIO_DT_UNKNOWN)
1795 {
1796 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1797 ent->score = 1;
1798 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1799 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1800 }
1801 else if (ent->type == EIO_DT_DIR)
1802 ent->score = 0;
1803 }
1804 }
1805
1806 namesoffs += len; 1985 namesoffs += len;
1807 ++dentoffs; 1986 ++dentoffs;
1808 } 1987 }
1809 1988
1810 if (EIO_CANCELLED (req)) 1989 if (EIO_CANCELLED (req))
1811 { 1990 {
1812 errno = ECANCELED; 1991 errno = ECANCELED;
1813 break; 1992 break;
1814 } 1993 }
1994
1995#ifdef _WIN32
1996 if (!FindNextFile (dirp, &entp))
1997 {
1998 FindClose (dirp);
1999 dirp = 0;
1815 } 2000 }
2001#endif
2002 }
1816} 2003}
2004
2005/*****************************************************************************/
2006/* working directory stuff */
2007/* various deficiencies in the posix 2008 api force us to */
2008/* keep the absolute path in string form at all times */
2009/* fuck yeah. */
2010
2011#if !HAVE_AT
2012
2013/* a bit like realpath, but usually faster because it doesn'T have to return */
2014/* an absolute or canonical path */
2015static const char *
2016wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
2017{
2018 if (!wd || *path == '/')
2019 return path;
2020
2021 if (path [0] == '.' && !path [1])
2022 return wd->str;
2023
2024 {
2025 int l1 = wd->len;
2026 int l2 = strlen (path);
2027
2028 char *res = tmpbuf_get (tmpbuf, l1 + l2 + 2);
2029
2030 memcpy (res, wd->str, l1);
2031 res [l1] = '/';
2032 memcpy (res + l1 + 1, path, l2 + 1);
2033
2034 return res;
2035 }
2036}
2037
2038#endif
2039
2040static eio_wd
2041eio__wd_open_sync (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
2042{
2043 int fd;
2044 eio_wd res;
2045 int len = eio__realpath (tmpbuf, wd, path);
2046
2047 if (len < 0)
2048 return EIO_INVALID_WD;
2049
2050#if HAVE_AT
2051 fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
2052
2053 if (fd < 0)
2054 return EIO_INVALID_WD;
2055#endif
2056
2057 res = malloc (sizeof (*res) + len); /* one extra 0-byte */
2058
2059#if HAVE_AT
2060 res->fd = fd;
2061#endif
2062
2063 res->len = len;
2064 memcpy (res->str, tmpbuf->ptr, len);
2065 res->str [len] = 0;
2066
2067 return res;
2068}
2069
2070eio_wd
2071eio_wd_open_sync (eio_wd wd, const char *path)
2072{
2073 struct tmpbuf tmpbuf = { 0 };
2074 wd = eio__wd_open_sync (&tmpbuf, wd, path);
2075 free (tmpbuf.ptr);
2076
2077 return wd;
2078}
2079
2080void
2081eio_wd_close_sync (eio_wd wd)
2082{
2083 if (wd != EIO_INVALID_WD && wd != EIO_CWD)
2084 {
2085 #if HAVE_AT
2086 close (wd->fd);
2087 #endif
2088 free (wd);
2089 }
2090}
2091
2092#if HAVE_AT
2093
2094/* they forgot these */
2095
2096static int
2097eio__truncateat (int dirfd, const char *path, off_t length)
2098{
2099 int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
2100 int res;
2101
2102 if (fd < 0)
2103 return fd;
2104
2105 res = ftruncate (fd, length);
2106 close (fd);
2107 return res;
2108}
2109
2110static int
2111eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
2112{
2113 int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
2114 int res;
2115
2116 if (fd < 0)
2117 return fd;
2118
2119 res = fstatvfs (fd, buf);
2120 close (fd);
2121 return res;
2122
2123}
2124
2125#endif
1817 2126
1818/*****************************************************************************/ 2127/*****************************************************************************/
1819 2128
1820#define ALLOC(len) \ 2129#define ALLOC(len) \
1821 if (!req->ptr2) \ 2130 if (!req->ptr2) \
1830 req->result = -1; \ 2139 req->result = -1; \
1831 break; \ 2140 break; \
1832 } \ 2141 } \
1833 } 2142 }
1834 2143
2144static void ecb_noinline ecb_cold
2145etp_proc_init (void)
2146{
2147#if HAVE_PRCTL_SET_NAME
2148 /* provide a more sensible "thread name" */
2149 char name[16 + 1];
2150 const int namelen = sizeof (name) - 1;
2151 int len;
2152
2153 prctl (PR_GET_NAME, (unsigned long)name, 0, 0, 0);
2154 name [namelen] = 0;
2155 len = strlen (name);
2156 strcpy (name + (len <= namelen - 4 ? len : namelen - 4), "/eio");
2157 prctl (PR_SET_NAME, (unsigned long)name, 0, 0, 0);
2158#endif
2159}
2160
1835X_THREAD_PROC (etp_proc) 2161X_THREAD_PROC (etp_proc)
1836{ 2162{
1837 ETP_REQ *req; 2163 ETP_REQ *req;
1838 struct timespec ts; 2164 struct timespec ts;
1839 etp_worker *self = (etp_worker *)thr_arg; 2165 etp_worker *self = (etp_worker *)thr_arg;
1840 int timeout; 2166
2167 etp_proc_init ();
1841 2168
1842 /* try to distribute timeouts somewhat evenly */ 2169 /* try to distribute timeouts somewhat evenly */
1843 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 2170 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1844 2171
1845 for (;;) 2172 for (;;)
1848 2175
1849 X_LOCK (reqlock); 2176 X_LOCK (reqlock);
1850 2177
1851 for (;;) 2178 for (;;)
1852 { 2179 {
1853 self->req = req = reqq_shift (&req_queue); 2180 req = reqq_shift (&req_queue);
1854 2181
1855 if (req) 2182 if (req)
1856 break; 2183 break;
1857 2184
1858 if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */ 2185 if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
1896 ++npending; 2223 ++npending;
1897 2224
1898 if (!reqq_push (&res_queue, req) && want_poll_cb) 2225 if (!reqq_push (&res_queue, req) && want_poll_cb)
1899 want_poll_cb (); 2226 want_poll_cb ();
1900 2227
1901 self->req = 0;
1902 etp_worker_clear (self); 2228 etp_worker_clear (self);
1903 2229
1904 X_UNLOCK (reslock); 2230 X_UNLOCK (reslock);
1905 } 2231 }
1906 2232
1907quit: 2233quit:
2234 free (req);
2235
1908 X_LOCK (wrklock); 2236 X_LOCK (wrklock);
1909 etp_worker_free (self); 2237 etp_worker_free (self);
1910 X_UNLOCK (wrklock); 2238 X_UNLOCK (wrklock);
1911 2239
1912 return 0; 2240 return 0;
1924eio_api_destroy (eio_req *req) 2252eio_api_destroy (eio_req *req)
1925{ 2253{
1926 free (req); 2254 free (req);
1927} 2255}
1928 2256
1929#define REQ(rtype) \ 2257#define REQ(rtype) \
1930 eio_req *req; \ 2258 eio_req *req; \
1931 \ 2259 \
1932 req = (eio_req *)calloc (1, sizeof *req); \ 2260 req = (eio_req *)calloc (1, sizeof *req); \
1933 if (!req) \ 2261 if (!req) \
1934 return 0; \ 2262 return 0; \
1948 { \ 2276 { \
1949 eio_api_destroy (req); \ 2277 eio_api_destroy (req); \
1950 return 0; \ 2278 return 0; \
1951 } 2279 }
1952 2280
2281#define SINGLEDOT(ptr) (0[(char *)(ptr)] == '.' && !1[(char *)(ptr)])
2282
1953static void 2283static void
1954eio_execute (etp_worker *self, eio_req *req) 2284eio_execute (etp_worker *self, eio_req *req)
1955{ 2285{
2286#if HAVE_AT
2287 int dirfd;
2288#else
2289 const char *path;
2290#endif
2291
1956 if (ecb_expect_false (EIO_CANCELLED (req))) 2292 if (ecb_expect_false (EIO_CANCELLED (req)))
1957 { 2293 {
1958 req->result = -1; 2294 req->result = -1;
1959 req->errorno = ECANCELED; 2295 req->errorno = ECANCELED;
1960 return; 2296 return;
1961 } 2297 }
1962 2298
2299 if (ecb_expect_false (req->wd == EIO_INVALID_WD))
2300 {
2301 req->result = -1;
2302 req->errorno = ENOENT;
2303 return;
2304 }
2305
2306 if (req->type >= EIO_OPEN)
2307 {
2308 #if HAVE_AT
2309 dirfd = WD2FD (req->wd);
2310 #else
2311 path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
2312 #endif
2313 }
2314
1963 switch (req->type) 2315 switch (req->type)
1964 { 2316 {
2317 case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
2318 req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
2319 break;
2320 case EIO_WD_CLOSE: req->result = 0;
2321 eio_wd_close_sync (req->wd); break;
2322
2323 case EIO_SEEK: eio__lseek (req); break;
1965 case EIO_READ: ALLOC (req->size); 2324 case EIO_READ: ALLOC (req->size);
1966 req->result = req->offs >= 0 2325 req->result = req->offs >= 0
1967 ? pread (req->int1, req->ptr2, req->size, req->offs) 2326 ? pread (req->int1, req->ptr2, req->size, req->offs)
1968 : read (req->int1, req->ptr2, req->size); break; 2327 : read (req->int1, req->ptr2, req->size); break;
1969 case EIO_WRITE: req->result = req->offs >= 0 2328 case EIO_WRITE: req->result = req->offs >= 0
1970 ? pwrite (req->int1, req->ptr2, req->size, req->offs) 2329 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1971 : write (req->int1, req->ptr2, req->size); break; 2330 : write (req->int1, req->ptr2, req->size); break;
1972 2331
1973 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break; 2332 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1974 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break; 2333 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
2334
2335#if HAVE_AT
1975 2336
1976 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2337 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1977 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2338 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break;
1978 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2339 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2340 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, AT_SYMLINK_NOFOLLOW); break;
2341 case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break;
2342 case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break;
2343 case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break;
2344 case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break;
2345
2346 case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
2347 case EIO_RMDIR: /* complications arise because "." cannot be removed, so we might have to expand */
2348 req->result = req->wd && SINGLEDOT (req->ptr1)
2349 ? rmdir (req->wd->str)
2350 : unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break;
2351 case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); break;
2352 case EIO_RENAME: /* complications arise because "." cannot be renamed, so we might have to expand */
2353 req->result = req->wd && SINGLEDOT (req->ptr1)
2354 ? rename (req->wd->str, req->ptr2)
2355 : renameat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2); break;
2356 case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2, 0); break;
2357 case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
2358 case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
2359 case EIO_READLINK: ALLOC (PATH_MAX);
2360 req->result = readlinkat (dirfd, req->ptr1, req->ptr2, PATH_MAX); break;
2361 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2362 req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2363 case EIO_UTIME:
2364 case EIO_FUTIME:
2365 {
2366 struct timespec ts[2];
2367 struct timespec *times;
2368
2369 if (req->nv1 != -1. || req->nv2 != -1.)
2370 {
2371 ts[0].tv_sec = req->nv1;
2372 ts[0].tv_nsec = (req->nv1 - ts[0].tv_sec) * 1e9;
2373 ts[1].tv_sec = req->nv2;
2374 ts[1].tv_nsec = (req->nv2 - ts[1].tv_sec) * 1e9;
2375
2376 times = ts;
2377 }
2378 else
2379 times = 0;
2380
2381 req->result = req->type == EIO_FUTIME
2382 ? futimens (req->int1, times)
2383 : utimensat (dirfd, req->ptr1, times, 0);
2384 }
2385 break;
2386
2387#else
2388
2389 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2390 req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
2391 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1979 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2392 req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
2393 case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
2394 case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
2395 case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
2396 case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
2397
2398 case EIO_UNLINK: req->result = unlink (path ); break;
2399 case EIO_RMDIR: req->result = rmdir (path ); break;
2400 case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
2401 case EIO_RENAME: req->result = rename (path , req->ptr2); break;
2402 case EIO_LINK: req->result = link (path , req->ptr2); break;
2403 case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
2404 case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
2405 case EIO_READLINK: ALLOC (PATH_MAX);
2406 req->result = readlink (path, req->ptr2, PATH_MAX); break;
2407 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2408 req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
2409
2410 case EIO_UTIME:
2411 case EIO_FUTIME:
2412 {
2413 struct timeval tv[2];
2414 struct timeval *times;
2415
2416 if (req->nv1 != -1. || req->nv2 != -1.)
2417 {
2418 tv[0].tv_sec = req->nv1;
2419 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
2420 tv[1].tv_sec = req->nv2;
2421 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
2422
2423 times = tv;
2424 }
2425 else
2426 times = 0;
2427
2428 req->result = req->type == EIO_FUTIME
2429 ? futimes (req->int1, times)
2430 : utimes (req->ptr1, times);
2431 }
2432 break;
2433
2434#endif
2435
2436 case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
2437 {
2438 ALLOC (req->result);
2439 memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
2440 }
2441 break;
2442
1980 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2443 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1981 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; 2444 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1982 2445
1983 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1984 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1985 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS)); 2446 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1986 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break; 2447 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1987 2448
1988 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1989 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; 2449 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1990 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1991 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break; 2450 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1992 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1993 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break; 2451 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1994 2452
1995 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
1996 case EIO_CLOSE: req->result = close (req->int1); break; 2453 case EIO_CLOSE: req->result = close (req->int1); break;
1997 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break; 2454 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1998 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
1999 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
2000 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
2001 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
2002 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
2003 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
2004 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
2005
2006 case EIO_REALPATH: eio__realpath (req, self); break;
2007
2008 case EIO_READLINK: ALLOC (PATH_MAX);
2009 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
2010
2011 case EIO_SYNC: req->result = 0; sync (); break; 2455 case EIO_SYNC: req->result = 0; sync (); break;
2012 case EIO_FSYNC: req->result = fsync (req->int1); break; 2456 case EIO_FSYNC: req->result = fsync (req->int1); break;
2013 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 2457 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
2458 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
2459 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
2014 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break; 2460 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
2015 case EIO_MTOUCH: req->result = eio__mtouch (req); break; 2461 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
2016 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break; 2462 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
2017 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break; 2463 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
2018 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
2019 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break; 2464 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
2020 2465
2021 case EIO_READDIR: eio__scandir (req, self); break; 2466 case EIO_READDIR: eio__scandir (req, self); break;
2022 2467
2023 case EIO_BUSY: 2468 case EIO_BUSY:
2033 req->result = select (0, 0, 0, 0, &tv); 2478 req->result = select (0, 0, 0, 0, &tv);
2034 } 2479 }
2035#endif 2480#endif
2036 break; 2481 break;
2037 2482
2038 case EIO_UTIME:
2039 case EIO_FUTIME:
2040 {
2041 struct timeval tv[2];
2042 struct timeval *times;
2043
2044 if (req->nv1 != -1. || req->nv2 != -1.)
2045 {
2046 tv[0].tv_sec = req->nv1;
2047 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
2048 tv[1].tv_sec = req->nv2;
2049 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
2050
2051 times = tv;
2052 }
2053 else
2054 times = 0;
2055
2056 req->result = req->type == EIO_FUTIME
2057 ? futimes (req->int1, times)
2058 : utimes (req->ptr1, times);
2059 }
2060 break;
2061
2062 case EIO_GROUP: 2483 case EIO_GROUP:
2063 abort (); /* handled in eio_request */ 2484 abort (); /* handled in eio_request */
2064 2485
2065 case EIO_NOP: 2486 case EIO_NOP:
2066 req->result = 0; 2487 req->result = 0;
2069 case EIO_CUSTOM: 2490 case EIO_CUSTOM:
2070 req->feed (req); 2491 req->feed (req);
2071 break; 2492 break;
2072 2493
2073 default: 2494 default:
2074 errno = ENOSYS;
2075 req->result = -1; 2495 req->result = EIO_ENOSYS ();
2076 break; 2496 break;
2077 } 2497 }
2078 2498
2079 req->errorno = errno; 2499 req->errorno = errno;
2080} 2500}
2081 2501
2082#ifndef EIO_NO_WRAPPERS 2502#ifndef EIO_NO_WRAPPERS
2083 2503
2504eio_req *eio_wd_open (const char *path, int pri, eio_cb cb, void *data)
2505{
2506 REQ (EIO_WD_OPEN); PATH; SEND;
2507}
2508
2509eio_req *eio_wd_close (eio_wd wd, int pri, eio_cb cb, void *data)
2510{
2511 REQ (EIO_WD_CLOSE); req->wd = wd; SEND;
2512}
2513
2084eio_req *eio_nop (int pri, eio_cb cb, void *data) 2514eio_req *eio_nop (int pri, eio_cb cb, void *data)
2085{ 2515{
2086 REQ (EIO_NOP); SEND; 2516 REQ (EIO_NOP); SEND;
2087} 2517}
2088 2518
2104eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2534eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2105{ 2535{
2106 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2536 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2107} 2537}
2108 2538
2539eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2540{
2541 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2542}
2543
2544eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
2545{
2546 REQ (EIO_SYNCFS); req->int1 = fd; SEND;
2547}
2548
2549eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2550{
2551 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2552}
2553
2109eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2554eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2110{ 2555{
2111 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2556 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2112} 2557}
2113 2558
2119eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data) 2564eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
2120{ 2565{
2121 REQ (EIO_MLOCKALL); req->int1 = flags; SEND; 2566 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2122} 2567}
2123 2568
2124eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2125{
2126 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2127}
2128
2129eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data) 2569eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2130{ 2570{
2131 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND; 2571 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
2132} 2572}
2133 2573
2134eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2135{
2136 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2137}
2138
2139eio_req *eio_close (int fd, int pri, eio_cb cb, void *data) 2574eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
2140{ 2575{
2141 REQ (EIO_CLOSE); req->int1 = fd; SEND; 2576 REQ (EIO_CLOSE); req->int1 = fd; SEND;
2142} 2577}
2143 2578
2144eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data) 2579eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
2145{ 2580{
2146 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND; 2581 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
2582}
2583
2584eio_req *eio_seek (int fd, off_t offset, int whence, int pri, eio_cb cb, void *data)
2585{
2586 REQ (EIO_SEEK); req->int1 = fd; req->offs = offset; req->int2 = whence; SEND;
2147} 2587}
2148 2588
2149eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data) 2589eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
2150{ 2590{
2151 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND; 2591 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
2369/* misc garbage */ 2809/* misc garbage */
2370 2810
2371eio_ssize_t 2811eio_ssize_t
2372eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2812eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2373{ 2813{
2374 etp_worker wrk;
2375 eio_ssize_t ret;
2376
2377 wrk.dbuf = 0;
2378
2379 ret = eio__sendfile (ofd, ifd, offset, count, &wrk); 2814 return eio__sendfile (ofd, ifd, offset, count);
2380
2381 if (wrk.dbuf)
2382 free (wrk.dbuf);
2383
2384 return ret;
2385} 2815}
2386 2816

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines