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

Comparing libeio/eio.c (file contents):
Revision 1.27 by root, Wed Oct 22 18:15:36 2008 UTC vs.
Revision 1.121 by root, Sat Jun 2 20:13:26 2012 UTC

1/* 1/*
2 * libeio implementation 2 * libeio implementation
3 * 3 *
4 * Copyright (c) 2007,2008 Marc Alexander Lehmann <libeio@schmorp.de> 4 * Copyright (c) 2007,2008,2009,2010,2011,2012 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 *
… …
35 * and other provisions required by the GPL. If you do not delete the 35 * and other provisions required by the GPL. If you do not delete the
36 * provisions above, a recipient may use your version of this file under 36 * provisions above, a recipient may use your version of this file under
37 * either the BSD or the GPL. 37 * either the BSD or the GPL.
38 */ 38 */
39 39
40#ifndef _WIN32
41# include "config.h"
42#endif
43
40#include "eio.h" 44#include "eio.h"
45#include "ecb.h"
46
47#ifdef EIO_STACKSIZE
48# define X_STACKSIZE EIO_STACKSIZE
49#endif
41#include "xthread.h" 50#include "xthread.h"
42 51
43#include <errno.h> 52#include <errno.h>
44#include <stddef.h> 53#include <stddef.h>
45#include <stdlib.h> 54#include <stdlib.h>
… …
49#include <sys/stat.h> 58#include <sys/stat.h>
50#include <limits.h> 59#include <limits.h>
51#include <fcntl.h> 60#include <fcntl.h>
52#include <assert.h> 61#include <assert.h>
53 62
63/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
64/* intptr_t only comes from stdint.h, says idiot openbsd coder */
65#if HAVE_STDINT_H
66# include <stdint.h>
67#endif
68
69#ifndef ECANCELED
70# define ECANCELED EDOM
71#endif
72#ifndef ELOOP
73# define ELOOP EDOM
74#endif
75
76#if !defined(ENOTSOCK) && defined(WSAENOTSOCK)
77# define ENOTSOCK WSAENOTSOCK
78#endif
79
80static void eio_destroy (eio_req *req);
81
54#ifndef EIO_FINISH 82#ifndef EIO_FINISH
55# 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
56#endif 84#endif
57 85
58#ifndef EIO_DESTROY 86#ifndef EIO_DESTROY
… …
61 89
62#ifndef EIO_FEED 90#ifndef EIO_FEED
63# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0) 91# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)
64#endif 92#endif
65 93
94#ifndef EIO_FD_TO_WIN32_HANDLE
95# define EIO_FD_TO_WIN32_HANDLE(fd) _get_osfhandle (fd)
96#endif
97#ifndef EIO_WIN32_HANDLE_TO_FD
98# define EIO_WIN32_HANDLE_TO_FD(handle) _open_osfhandle (handle, 0)
99#endif
100
101#define EIO_ERRNO(errval,retval) ((errno = errval), retval)
102
103#define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1)
104
66#ifdef _WIN32 105#ifdef _WIN32
67 106
68 /*doh*/ 107 #undef PAGESIZE
108 #define PAGESIZE 4096 /* GetSystemInfo? */
109
110 /* TODO: look at how perl does stat (non-sloppy), unlink (ro-files), utime, link */
111
112 #ifdef EIO_STRUCT_STATI64
113 /* look at perl's non-sloppy stat */
114 #define stat(path,buf) _stati64 (path,buf)
115 #define fstat(fd,buf) _fstati64 (fd,buf)
116 #endif
117 #define lstat(path,buf) stat (path,buf)
118 #define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
119 #define mkdir(path,mode) _mkdir (path)
120 #define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
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 ()
127 #define chown(path,uid,gid) EIO_ENOSYS ()
128 #define fchown(fd,uid,gid) EIO_ENOSYS ()
129 #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
130 #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
131 #define mknod(path,mode,dev) 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 /* rename() uses MoveFile, which fails to overwrite */
138 #define rename(old,neu) eio__rename (old, neu)
139
140 static int
141 eio__rename (const char *old, const char *neu)
142 {
143 if (MoveFileEx (old, neu, MOVEFILE_REPLACE_EXISTING))
144 return 0;
145
146 /* should steal _dosmaperr */
147 switch (GetLastError ())
148 {
149 case ERROR_FILE_NOT_FOUND:
150 case ERROR_PATH_NOT_FOUND:
151 case ERROR_INVALID_DRIVE:
152 case ERROR_NO_MORE_FILES:
153 case ERROR_BAD_NETPATH:
154 case ERROR_BAD_NET_NAME:
155 case ERROR_BAD_PATHNAME:
156 case ERROR_FILENAME_EXCED_RANGE:
157 errno = ENOENT;
158 break;
159
160 default:
161 errno = EACCES;
162 break;
163 }
164
165 return -1;
166 }
167
168 /* we could even stat and see if it exists */
169 static int
170 symlink (const char *old, const char *neu)
171 {
172 #if WINVER >= 0x0600
173 if (CreateSymbolicLink (neu, old, 1))
174 return 0;
175
176 if (CreateSymbolicLink (neu, old, 0))
177 return 0;
178 #endif
179
180 return EIO_ERRNO (ENOENT, -1);
181 }
182
183 /* POSIX API only */
184 #define CreateHardLink(neu,old,flags) 0
185 #define CreateSymbolicLink(neu,old,flags) 0
186
187 struct statvfs
188 {
189 int dummy;
190 };
191
192 #define DT_DIR EIO_DT_DIR
193 #define DT_REG EIO_DT_REG
194 #define D_NAME(entp) entp.cFileName
195 #define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
69 196
70#else 197#else
71 198
72# include "config.h"
73# include <sys/time.h> 199 #include <sys/time.h>
74# include <sys/select.h> 200 #include <sys/select.h>
201 #include <sys/statvfs.h>
75# include <unistd.h> 202 #include <unistd.h>
203 #include <signal.h>
204 #include <dirent.h>
205
206 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
207 #include <sys/mman.h>
208 #endif
209
210 #define D_NAME(entp) entp->d_name
211
212 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
213 #if __FreeBSD__ || __NetBSD__ || __OpenBSD__
214 #define _DIRENT_HAVE_D_TYPE /* sigh */
215 #define D_INO(de) (de)->d_fileno
216 #define D_NAMLEN(de) (de)->d_namlen
217 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
218 #define D_INO(de) (de)->d_ino
219 #endif
220
221 #ifdef _D_EXACT_NAMLEN
222 #undef D_NAMLEN
223 #define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
224 #endif
225
226 #ifdef _DIRENT_HAVE_D_TYPE
227 #define D_TYPE(de) (de)->d_type
228 #endif
229
230 #ifndef EIO_STRUCT_DIRENT
231 #define EIO_STRUCT_DIRENT struct dirent
232 #endif
233
234#endif
235
236#if HAVE_UTIMES
76# include <utime.h> 237# include <utime.h>
77# include <signal.h>
78# include <dirent.h>
79
80# ifndef EIO_STRUCT_DIRENT
81# define EIO_STRUCT_DIRENT struct dirent
82# endif 238#endif
83 239
240#if HAVE_SYS_SYSCALL_H
241# include <sys/syscall.h>
242#endif
243
244#if HAVE_SYS_PRCTL_H
245# include <sys/prctl.h>
84#endif 246#endif
85 247
86#if HAVE_SENDFILE 248#if HAVE_SENDFILE
87# if __linux 249# if __linux
88# include <sys/sendfile.h> 250# include <sys/sendfile.h>
89# elif __freebsd 251# elif __FreeBSD__ || defined __APPLE__
90# include <sys/socket.h> 252# include <sys/socket.h>
91# include <sys/uio.h> 253# include <sys/uio.h>
92# elif __hpux 254# elif __hpux
93# include <sys/socket.h> 255# include <sys/socket.h>
94# elif __solaris /* not yet */ 256# elif __solaris
95# include <sys/sendfile.h> 257# include <sys/sendfile.h>
96# else 258# else
97# error sendfile support requested but not available 259# error sendfile support requested but not available
98# endif 260# endif
99#endif 261#endif
100 262
101/* number of seconds after which an idle threads exit */ 263#ifndef D_TYPE
102#define IDLE_TIMEOUT 10 264# define D_TYPE(de) 0
265#endif
266#ifndef D_INO
267# define D_INO(de) 0
268#endif
269#ifndef D_NAMLEN
270# define D_NAMLEN(entp) strlen (D_NAME (entp))
271#endif
103 272
104/* used for struct dirent, AIX doesn't provide it */ 273/* used for struct dirent, AIX doesn't provide it */
105#ifndef NAME_MAX 274#ifndef NAME_MAX
106# define NAME_MAX 4096 275# define NAME_MAX 4096
107#endif 276#endif
108 277
278/* used for readlink etc. */
279#ifndef PATH_MAX
280# define PATH_MAX 4096
281#endif
282
109/* buffer size for various temporary buffers */ 283/* buffer size for various temporary buffers */
110#define EIO_BUFSIZE 65536 284#define EIO_BUFSIZE 65536
111 285
112#define dBUF \ 286#define dBUF \
113 char *eio_buf; \
114 ETP_WORKER_LOCK (self); \
115 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 287 char *eio_buf = malloc (EIO_BUFSIZE); \
116 ETP_WORKER_UNLOCK (self); \
117 errno = ENOMEM; \ 288 errno = ENOMEM; \
118 if (!eio_buf) \ 289 if (!eio_buf) \
119 return -1; 290 return -1
291
292#define FUBd \
293 free (eio_buf)
120 294
121#define EIO_TICKS ((1000000 + 1023) >> 10) 295#define EIO_TICKS ((1000000 + 1023) >> 10)
122 296
123/*****************************************************************************/ 297/*****************************************************************************/
124 298
125#if __GNUC__ >= 3 299struct tmpbuf
126# define expect(expr,value) __builtin_expect ((expr),(value)) 300{
301 void *ptr;
302 int len;
303};
304
305static void *
306tmpbuf_get (struct tmpbuf *buf, int len)
307{
308 if (buf->len < len)
309 {
310 free (buf->ptr);
311 buf->ptr = malloc (buf->len = len);
312 }
313
314 return buf->ptr;
315}
316
317struct tmpbuf;
318
319#if _POSIX_VERSION >= 200809L
320 #define HAVE_AT 1
321 #define WD2FD(wd) ((wd) ? (wd)->fd : AT_FDCWD)
322 #ifndef O_SEARCH
323 #define O_SEARCH O_RDONLY
324 #endif
127#else 325#else
128# define expect(expr,value) (expr) 326 #define HAVE_AT 0
327 static const char *wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path);
129#endif 328#endif
130 329
131#define expect_false(expr) expect ((expr) != 0, 0) 330struct eio_pwd
132#define expect_true(expr) expect ((expr) != 0, 1) 331{
332#if HAVE_AT
333 int fd;
334#endif
335 int len;
336 char str[1]; /* actually, a 0-terminated canonical path */
337};
133 338
134/*****************************************************************************/ 339/*****************************************************************************/
135 340
136#define ETP_PRI_MIN EIO_PRI_MIN 341#define ETP_PRI_MIN EIO_PRI_MIN
137#define ETP_PRI_MAX EIO_PRI_MAX 342#define ETP_PRI_MAX EIO_PRI_MAX
… …
143static int eio_finish (eio_req *req); 348static int eio_finish (eio_req *req);
144#define ETP_FINISH(req) eio_finish (req) 349#define ETP_FINISH(req) eio_finish (req)
145static void eio_execute (struct etp_worker *self, eio_req *req); 350static void eio_execute (struct etp_worker *self, eio_req *req);
146#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) 351#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
147 352
148#define ETP_WORKER_CLEAR(req) \
149 if (wrk->dbuf) \
150 { \
151 free (wrk->dbuf); \
152 wrk->dbuf = 0; \
153 } \
154 \
155 if (wrk->dirp) \
156 { \
157 closedir (wrk->dirp); \
158 wrk->dirp = 0; \
159 }
160#define ETP_WORKER_COMMON \
161 void *dbuf; \
162 DIR *dirp;
163
164/*****************************************************************************/ 353/*****************************************************************************/
165 354
166#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) 355#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
167 356
168/* calculcate time difference in ~1/EIO_TICKS of a second */ 357/* calculate time difference in ~1/EIO_TICKS of a second */
358ecb_inline int
169static int tvdiff (struct timeval *tv1, struct timeval *tv2) 359tvdiff (struct timeval *tv1, struct timeval *tv2)
170{ 360{
171 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 361 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
172 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 362 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
173} 363}
174 364
… …
178static void (*done_poll_cb) (void); 368static void (*done_poll_cb) (void);
179 369
180static unsigned int max_poll_time; /* reslock */ 370static unsigned int max_poll_time; /* reslock */
181static unsigned int max_poll_reqs; /* reslock */ 371static unsigned int max_poll_reqs; /* reslock */
182 372
183static volatile unsigned int nreqs; /* reqlock */ 373static unsigned int nreqs; /* reqlock */
184static volatile unsigned int nready; /* reqlock */ 374static unsigned int nready; /* reqlock */
185static volatile unsigned int npending; /* reqlock */ 375static unsigned int npending; /* reqlock */
186static volatile unsigned int max_idle = 4; 376static unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
377static unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
187 378
188static mutex_t wrklock = X_MUTEX_INIT; 379static xmutex_t wrklock;
189static mutex_t reslock = X_MUTEX_INIT; 380static xmutex_t reslock;
190static mutex_t reqlock = X_MUTEX_INIT; 381static xmutex_t reqlock;
191static cond_t reqwait = X_COND_INIT; 382static xcond_t reqwait;
192
193#if !HAVE_PREADWRITE
194/*
195 * make our pread/pwrite emulation safe against themselves, but not against
196 * normal read/write by using a mutex. slows down execution a lot,
197 * but that's your problem, not mine.
198 */
199static mutex_t preadwritelock = X_MUTEX_INIT;
200#endif
201 383
202typedef struct etp_worker 384typedef struct etp_worker
203{ 385{
386 struct tmpbuf tmpbuf;
387
204 /* locked by wrklock */ 388 /* locked by wrklock */
205 struct etp_worker *prev, *next; 389 struct etp_worker *prev, *next;
206 390
207 thread_t tid; 391 xthread_t tid;
208 392
209 /* locked by reslock, reqlock or wrklock */ 393#ifdef ETP_WORKER_COMMON
210 ETP_REQ *req; /* currently processed request */
211
212 ETP_WORKER_COMMON 394 ETP_WORKER_COMMON
395#endif
213} etp_worker; 396} etp_worker;
214 397
215static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */ 398static etp_worker wrk_first; /* NOT etp */
216 399
217#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock) 400#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
218#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock) 401#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
219 402
220/* worker threads management */ 403/* worker threads management */
221 404
405static void
222static void etp_worker_clear (etp_worker *wrk) 406etp_worker_clear (etp_worker *wrk)
223{ 407{
224 ETP_WORKER_CLEAR (wrk);
225} 408}
226 409
410static void ecb_cold
227static void etp_worker_free (etp_worker *wrk) 411etp_worker_free (etp_worker *wrk)
228{ 412{
413 free (wrk->tmpbuf.ptr);
414
229 wrk->next->prev = wrk->prev; 415 wrk->next->prev = wrk->prev;
230 wrk->prev->next = wrk->next; 416 wrk->prev->next = wrk->next;
231 417
232 free (wrk); 418 free (wrk);
233} 419}
234 420
235static unsigned int etp_nreqs (void) 421static unsigned int
422etp_nreqs (void)
236{ 423{
237 int retval; 424 int retval;
238 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 425 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
239 retval = nreqs; 426 retval = nreqs;
240 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 427 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
241 return retval; 428 return retval;
242} 429}
243 430
244static unsigned int etp_nready (void) 431static unsigned int
432etp_nready (void)
245{ 433{
246 unsigned int retval; 434 unsigned int retval;
247 435
248 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 436 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
249 retval = nready; 437 retval = nready;
250 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 438 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
251 439
252 return retval; 440 return retval;
253} 441}
254 442
255static unsigned int etp_npending (void) 443static unsigned int
444etp_npending (void)
256{ 445{
257 unsigned int retval; 446 unsigned int retval;
258 447
259 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 448 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
260 retval = npending; 449 retval = npending;
261 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 450 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
262 451
263 return retval; 452 return retval;
264} 453}
265 454
266static unsigned int etp_nthreads (void) 455static unsigned int
456etp_nthreads (void)
267{ 457{
268 unsigned int retval; 458 unsigned int retval;
269 459
270 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 460 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
271 retval = started; 461 retval = started;
… …
285} etp_reqq; 475} etp_reqq;
286 476
287static etp_reqq req_queue; 477static etp_reqq req_queue;
288static etp_reqq res_queue; 478static etp_reqq res_queue;
289 479
480static void ecb_noinline ecb_cold
481reqq_init (etp_reqq *q)
482{
483 int pri;
484
485 for (pri = 0; pri < ETP_NUM_PRI; ++pri)
486 q->qs[pri] = q->qe[pri] = 0;
487
488 q->size = 0;
489}
490
491static int ecb_noinline
290static int reqq_push (etp_reqq *q, ETP_REQ *req) 492reqq_push (etp_reqq *q, ETP_REQ *req)
291{ 493{
292 int pri = req->pri; 494 int pri = req->pri;
293 req->next = 0; 495 req->next = 0;
294 496
295 if (q->qe[pri]) 497 if (q->qe[pri])
… …
301 q->qe[pri] = q->qs[pri] = req; 503 q->qe[pri] = q->qs[pri] = req;
302 504
303 return q->size++; 505 return q->size++;
304} 506}
305 507
508static ETP_REQ * ecb_noinline
306static ETP_REQ *reqq_shift (etp_reqq *q) 509reqq_shift (etp_reqq *q)
307{ 510{
308 int pri; 511 int pri;
309 512
310 if (!q->size) 513 if (!q->size)
311 return 0; 514 return 0;
… …
326 } 529 }
327 530
328 abort (); 531 abort ();
329} 532}
330 533
331static void etp_atfork_prepare (void) 534static int ecb_cold
535etp_init (void (*want_poll)(void), void (*done_poll)(void))
332{ 536{
333 X_LOCK (wrklock); 537 X_MUTEX_CREATE (wrklock);
334 X_LOCK (reqlock); 538 X_MUTEX_CREATE (reslock);
335 X_LOCK (reslock); 539 X_MUTEX_CREATE (reqlock);
336#if !HAVE_PREADWRITE 540 X_COND_CREATE (reqwait);
337 X_LOCK (preadwritelock);
338#endif
339}
340 541
341static void etp_atfork_parent (void) 542 reqq_init (&req_queue);
342{ 543 reqq_init (&res_queue);
343#if !HAVE_PREADWRITE
344 X_UNLOCK (preadwritelock);
345#endif
346 X_UNLOCK (reslock);
347 X_UNLOCK (reqlock);
348 X_UNLOCK (wrklock);
349}
350 544
351static void etp_atfork_child (void) 545 wrk_first.next =
352{ 546 wrk_first.prev = &wrk_first;
353 ETP_REQ *prv;
354
355 while ((prv = reqq_shift (&req_queue)))
356 ETP_DESTROY (prv);
357
358 while ((prv = reqq_shift (&res_queue)))
359 ETP_DESTROY (prv);
360
361 while (wrk_first.next != &wrk_first)
362 {
363 etp_worker *wrk = wrk_first.next;
364
365 if (wrk->req)
366 ETP_DESTROY (wrk->req);
367
368 etp_worker_clear (wrk);
369 etp_worker_free (wrk);
370 }
371 547
372 started = 0; 548 started = 0;
373 idle = 0; 549 idle = 0;
374 nreqs = 0; 550 nreqs = 0;
375 nready = 0; 551 nready = 0;
376 npending = 0; 552 npending = 0;
377 553
378 etp_atfork_parent ();
379}
380
381static void
382etp_once_init (void)
383{
384 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
385}
386
387static int
388etp_init (void (*want_poll)(void), void (*done_poll)(void))
389{
390 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
391
392 pthread_once (&doinit, etp_once_init);
393
394 want_poll_cb = want_poll; 554 want_poll_cb = want_poll;
395 done_poll_cb = done_poll; 555 done_poll_cb = done_poll;
396 556
397 return 0; 557 return 0;
398} 558}
399 559
400X_THREAD_PROC (etp_proc); 560X_THREAD_PROC (etp_proc);
401 561
562static void ecb_cold
402static void etp_start_thread (void) 563etp_start_thread (void)
403{ 564{
404 etp_worker *wrk = calloc (1, sizeof (etp_worker)); 565 etp_worker *wrk = calloc (1, sizeof (etp_worker));
405 566
406 /*TODO*/ 567 /*TODO*/
407 assert (("unable to allocate worker thread data", wrk)); 568 assert (("unable to allocate worker thread data", wrk));
… …
420 free (wrk); 581 free (wrk);
421 582
422 X_UNLOCK (wrklock); 583 X_UNLOCK (wrklock);
423} 584}
424 585
586static void
425static void etp_maybe_start_thread (void) 587etp_maybe_start_thread (void)
426{ 588{
427 if (expect_true (etp_nthreads () >= wanted)) 589 if (ecb_expect_true (etp_nthreads () >= wanted))
428 return; 590 return;
429 591
430 /* todo: maybe use idle here, but might be less exact */ 592 /* todo: maybe use idle here, but might be less exact */
431 if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ())) 593 if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
432 return; 594 return;
433 595
434 etp_start_thread (); 596 etp_start_thread ();
435} 597}
436 598
599static void ecb_cold
437static void etp_end_thread (void) 600etp_end_thread (void)
438{ 601{
439 eio_req *req = calloc (1, sizeof (eio_req)); 602 eio_req *req = calloc (1, sizeof (eio_req)); /* will be freed by worker */
440 603
441 req->type = -1; 604 req->type = -1;
442 req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 605 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
443 606
444 X_LOCK (reqlock); 607 X_LOCK (reqlock);
… …
449 X_LOCK (wrklock); 612 X_LOCK (wrklock);
450 --started; 613 --started;
451 X_UNLOCK (wrklock); 614 X_UNLOCK (wrklock);
452} 615}
453 616
454static int etp_poll (void) 617static int
618etp_poll (void)
455{ 619{
456 unsigned int maxreqs; 620 unsigned int maxreqs;
457 unsigned int maxtime; 621 unsigned int maxtime;
458 struct timeval tv_start, tv_now; 622 struct timeval tv_start, tv_now;
459 623
… …
489 653
490 X_LOCK (reqlock); 654 X_LOCK (reqlock);
491 --nreqs; 655 --nreqs;
492 X_UNLOCK (reqlock); 656 X_UNLOCK (reqlock);
493 657
494 if (expect_false (req->type == EIO_GROUP && req->size)) 658 if (ecb_expect_false (req->type == EIO_GROUP && req->size))
495 { 659 {
496 req->int1 = 1; /* mark request as delayed */ 660 req->int1 = 1; /* mark request as delayed */
497 continue; 661 continue;
498 } 662 }
499 else 663 else
500 { 664 {
501 int res = ETP_FINISH (req); 665 int res = ETP_FINISH (req);
502 if (expect_false (res)) 666 if (ecb_expect_false (res))
503 return res; 667 return res;
504 } 668 }
505 669
506 if (expect_false (maxreqs && !--maxreqs)) 670 if (ecb_expect_false (maxreqs && !--maxreqs))
507 break; 671 break;
508 672
509 if (maxtime) 673 if (maxtime)
510 { 674 {
511 gettimeofday (&tv_now, 0); 675 gettimeofday (&tv_now, 0);
… …
517 681
518 errno = EAGAIN; 682 errno = EAGAIN;
519 return -1; 683 return -1;
520} 684}
521 685
686static void
522static void etp_cancel (ETP_REQ *req) 687etp_cancel (ETP_REQ *req)
523{ 688{
524 X_LOCK (wrklock); 689 req->cancelled = 1;
525 req->flags |= EIO_FLAG_CANCELLED;
526 X_UNLOCK (wrklock);
527 690
528 eio_grp_cancel (req); 691 eio_grp_cancel (req);
529} 692}
530 693
694static void
531static void etp_submit (ETP_REQ *req) 695etp_submit (ETP_REQ *req)
532{ 696{
533 req->pri -= ETP_PRI_MIN; 697 req->pri -= ETP_PRI_MIN;
534 698
535 if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN; 699 if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
536 if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 700 if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
537 701
538 if (expect_false (req->type == EIO_GROUP)) 702 if (ecb_expect_false (req->type == EIO_GROUP))
539 { 703 {
540 /* I hope this is worth it :/ */ 704 /* I hope this is worth it :/ */
541 X_LOCK (reqlock); 705 X_LOCK (reqlock);
542 ++nreqs; 706 ++nreqs;
543 X_UNLOCK (reqlock); 707 X_UNLOCK (reqlock);
… …
562 726
563 etp_maybe_start_thread (); 727 etp_maybe_start_thread ();
564 } 728 }
565} 729}
566 730
731static void ecb_cold
567static void etp_set_max_poll_time (double nseconds) 732etp_set_max_poll_time (double nseconds)
568{ 733{
569 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 734 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
570 max_poll_time = nseconds; 735 max_poll_time = nseconds * EIO_TICKS;
571 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 736 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
572} 737}
573 738
739static void ecb_cold
574static void etp_set_max_poll_reqs (unsigned int maxreqs) 740etp_set_max_poll_reqs (unsigned int maxreqs)
575{ 741{
576 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 742 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
577 max_poll_reqs = maxreqs; 743 max_poll_reqs = maxreqs;
578 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 744 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
579} 745}
580 746
747static void ecb_cold
581static void etp_set_max_idle (unsigned int nthreads) 748etp_set_max_idle (unsigned int nthreads)
582{ 749{
583 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 750 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
584 max_idle = nthreads <= 0 ? 1 : nthreads; 751 max_idle = nthreads;
585 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 752 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
586} 753}
587 754
755static void ecb_cold
756etp_set_idle_timeout (unsigned int seconds)
757{
758 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
759 idle_timeout = seconds;
760 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
761}
762
763static void ecb_cold
588static void etp_set_min_parallel (unsigned int nthreads) 764etp_set_min_parallel (unsigned int nthreads)
589{ 765{
590 if (wanted < nthreads) 766 if (wanted < nthreads)
591 wanted = nthreads; 767 wanted = nthreads;
592} 768}
593 769
770static void ecb_cold
594static void etp_set_max_parallel (unsigned int nthreads) 771etp_set_max_parallel (unsigned int nthreads)
595{ 772{
596 if (wanted > nthreads) 773 if (wanted > nthreads)
597 wanted = nthreads; 774 wanted = nthreads;
598 775
599 while (started > wanted) 776 while (started > wanted)
600 etp_end_thread (); 777 etp_end_thread ();
601} 778}
602 779
603/*****************************************************************************/ 780/*****************************************************************************/
604 781
782static void
605static void grp_try_feed (eio_req *grp) 783grp_try_feed (eio_req *grp)
606{ 784{
607 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 785 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
608 { 786 {
609 grp->flags &= ~EIO_FLAG_GROUPADD; 787 grp->flags &= ~EIO_FLAG_GROUPADD;
610 788
… …
617 break; 795 break;
618 } 796 }
619 } 797 }
620} 798}
621 799
800static int
622static int grp_dec (eio_req *grp) 801grp_dec (eio_req *grp)
623{ 802{
624 --grp->size; 803 --grp->size;
625 804
626 /* call feeder, if applicable */ 805 /* call feeder, if applicable */
627 grp_try_feed (grp); 806 grp_try_feed (grp);
… …
631 return eio_finish (grp); 810 return eio_finish (grp);
632 else 811 else
633 return 0; 812 return 0;
634} 813}
635 814
815static void
636void eio_destroy (eio_req *req) 816eio_destroy (eio_req *req)
637{ 817{
638 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); 818 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
639 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); 819 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
640 820
641 EIO_DESTROY (req); 821 EIO_DESTROY (req);
642} 822}
643 823
824static int
644static int eio_finish (eio_req *req) 825eio_finish (eio_req *req)
645{ 826{
646 int res = EIO_FINISH (req); 827 int res = EIO_FINISH (req);
647 828
648 if (req->grp) 829 if (req->grp)
649 { 830 {
… …
657 if (grp->grp_first == req) 838 if (grp->grp_first == req)
658 grp->grp_first = req->grp_next; 839 grp->grp_first = req->grp_next;
659 840
660 res2 = grp_dec (grp); 841 res2 = grp_dec (grp);
661 842
662 if (!res && res2) 843 if (!res)
663 res = res2; 844 res = res2;
664 } 845 }
665 846
666 eio_destroy (req); 847 eio_destroy (req);
667 848
668 return res; 849 return res;
669} 850}
670 851
852void
671void eio_grp_cancel (eio_req *grp) 853eio_grp_cancel (eio_req *grp)
672{ 854{
673 for (grp = grp->grp_first; grp; grp = grp->grp_next) 855 for (grp = grp->grp_first; grp; grp = grp->grp_next)
674 eio_cancel (grp); 856 eio_cancel (grp);
675} 857}
676 858
859void
677void eio_cancel (eio_req *req) 860eio_cancel (eio_req *req)
678{ 861{
679 etp_cancel (req); 862 etp_cancel (req);
680} 863}
681 864
865void
682void eio_submit (eio_req *req) 866eio_submit (eio_req *req)
683{ 867{
684 etp_submit (req); 868 etp_submit (req);
685} 869}
686 870
687unsigned int eio_nreqs (void) 871unsigned int
872eio_nreqs (void)
688{ 873{
689 return etp_nreqs (); 874 return etp_nreqs ();
690} 875}
691 876
692unsigned int eio_nready (void) 877unsigned int
878eio_nready (void)
693{ 879{
694 return etp_nready (); 880 return etp_nready ();
695} 881}
696 882
697unsigned int eio_npending (void) 883unsigned int
884eio_npending (void)
698{ 885{
699 return etp_npending (); 886 return etp_npending ();
700} 887}
701 888
702unsigned int eio_nthreads (void) 889unsigned int ecb_cold
890eio_nthreads (void)
703{ 891{
704 return etp_nthreads (); 892 return etp_nthreads ();
705} 893}
706 894
895void ecb_cold
707void eio_set_max_poll_time (double nseconds) 896eio_set_max_poll_time (double nseconds)
708{ 897{
709 etp_set_max_poll_time (nseconds); 898 etp_set_max_poll_time (nseconds);
710} 899}
711 900
901void ecb_cold
712void eio_set_max_poll_reqs (unsigned int maxreqs) 902eio_set_max_poll_reqs (unsigned int maxreqs)
713{ 903{
714 etp_set_max_poll_reqs (maxreqs); 904 etp_set_max_poll_reqs (maxreqs);
715} 905}
716 906
907void ecb_cold
717void eio_set_max_idle (unsigned int nthreads) 908eio_set_max_idle (unsigned int nthreads)
718{ 909{
719 etp_set_max_idle (nthreads); 910 etp_set_max_idle (nthreads);
720} 911}
721 912
913void ecb_cold
914eio_set_idle_timeout (unsigned int seconds)
915{
916 etp_set_idle_timeout (seconds);
917}
918
919void ecb_cold
722void eio_set_min_parallel (unsigned int nthreads) 920eio_set_min_parallel (unsigned int nthreads)
723{ 921{
724 etp_set_min_parallel (nthreads); 922 etp_set_min_parallel (nthreads);
725} 923}
726 924
925void ecb_cold
727void eio_set_max_parallel (unsigned int nthreads) 926eio_set_max_parallel (unsigned int nthreads)
728{ 927{
729 etp_set_max_parallel (nthreads); 928 etp_set_max_parallel (nthreads);
730} 929}
731 930
732int eio_poll (void) 931int eio_poll (void)
… …
735} 934}
736 935
737/*****************************************************************************/ 936/*****************************************************************************/
738/* work around various missing functions */ 937/* work around various missing functions */
739 938
740#if !HAVE_PREADWRITE
741# undef pread
742# undef pwrite
743# define pread eio__pread
744# define pwrite eio__pwrite
745
746static ssize_t
747eio__pread (int fd, void *buf, size_t count, off_t offset)
748{
749 ssize_t res;
750 off_t ooffset;
751
752 X_LOCK (preadwritelock);
753 ooffset = lseek (fd, 0, SEEK_CUR);
754 lseek (fd, offset, SEEK_SET);
755 res = read (fd, buf, count);
756 lseek (fd, ooffset, SEEK_SET);
757 X_UNLOCK (preadwritelock);
758
759 return res;
760}
761
762static ssize_t
763eio__pwrite (int fd, void *buf, size_t count, off_t offset)
764{
765 ssize_t res;
766 off_t ooffset;
767
768 X_LOCK (preadwritelock);
769 ooffset = lseek (fd, 0, SEEK_CUR);
770 lseek (fd, offset, SEEK_SET);
771 res = write (fd, buf, count);
772 lseek (fd, offset, SEEK_SET);
773 X_UNLOCK (preadwritelock);
774
775 return res;
776}
777#endif
778
779#ifndef HAVE_FUTIMES 939#ifndef HAVE_UTIMES
780 940
781# undef utimes 941# undef utimes
782# undef futimes
783# define utimes(path,times) eio__utimes (path, times) 942# define utimes(path,times) eio__utimes (path, times)
784# define futimes(fd,times) eio__futimes (fd, times)
785 943
786static int 944static int
787eio__utimes (const char *filename, const struct timeval times[2]) 945eio__utimes (const char *filename, const struct timeval times[2])
788{ 946{
789 if (times) 947 if (times)
… …
797 } 955 }
798 else 956 else
799 return utime (filename, 0); 957 return utime (filename, 0);
800} 958}
801 959
960#endif
961
962#ifndef HAVE_FUTIMES
963
964# undef futimes
965# define futimes(fd,times) eio__futimes (fd, times)
966
967static int
802static int eio__futimes (int fd, const struct timeval tv[2]) 968eio__futimes (int fd, const struct timeval tv[2])
803{ 969{
804 errno = ENOSYS; 970 errno = ENOSYS;
805 return -1; 971 return -1;
806} 972}
807 973
… …
810#if !HAVE_FDATASYNC 976#if !HAVE_FDATASYNC
811# undef fdatasync 977# undef fdatasync
812# define fdatasync(fd) fsync (fd) 978# define fdatasync(fd) fsync (fd)
813#endif 979#endif
814 980
981static int
982eio__syncfs (int fd)
983{
984 int res;
985
986#if HAVE_SYS_SYNCFS
987 res = (int)syscall (__NR_syncfs, (int)(fd));
988#else
989 res = EIO_ENOSYS ();
990#endif
991
992 if (res < 0 && errno == ENOSYS && fd >= 0)
993 sync ();
994
995 return res;
996}
997
815/* sync_file_range always needs emulation */ 998/* sync_file_range always needs emulation */
816int 999static int
817eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags) 1000eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
818{ 1001{
819#if HAVE_SYNC_FILE_RANGE 1002#if HAVE_SYNC_FILE_RANGE
820 int res; 1003 int res;
821 1004
… …
829 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0); 1012 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
830 } 1013 }
831 1014
832 res = sync_file_range (fd, offset, nbytes, flags); 1015 res = sync_file_range (fd, offset, nbytes, flags);
833 1016
834 if (res != ENOSYS) 1017 if (!res || errno != ENOSYS)
835 return res; 1018 return res;
836#endif 1019#endif
837 1020
838 /* even though we could play tricks with the flags, it's better to always 1021 /* even though we could play tricks with the flags, it's better to always
839 * call fdatasync, as thta matches the expectation of it's users best */ 1022 * call fdatasync, as that matches the expectation of its users best */
840 return fdatasync (fd); 1023 return fdatasync (fd);
1024}
1025
1026static int
1027eio__fallocate (int fd, int mode, off_t offset, size_t len)
1028{
1029#if HAVE_LINUX_FALLOCATE
1030 return fallocate (fd, mode, offset, len);
1031#else
1032 return EIO_ENOSYS ();
1033#endif
841} 1034}
842 1035
843#if !HAVE_READAHEAD 1036#if !HAVE_READAHEAD
844# undef readahead 1037# undef readahead
845# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 1038# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
846 1039
847static ssize_t 1040static eio_ssize_t
848eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) 1041eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
849{ 1042{
850 size_t todo = count; 1043 size_t todo = count;
851 dBUF; 1044 dBUF;
852 1045
… …
857 pread (fd, eio_buf, len, offset); 1050 pread (fd, eio_buf, len, offset);
858 offset += len; 1051 offset += len;
859 todo -= len; 1052 todo -= len;
860 } 1053 }
861 1054
862 errno = 0; 1055 FUBd;
1056
1057 /* linux's readahead basically only fails for EBADF or EINVAL (not mmappable) */
1058 /* but not for e.g. EIO or eof, so we also never fail */
863 return count; 1059 return 0;
864} 1060}
865 1061
866#endif 1062#endif
867 1063
868/* sendfile always needs emulation */ 1064/* sendfile always needs emulation */
869static ssize_t 1065static eio_ssize_t
870eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) 1066eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
871{ 1067{
1068 eio_ssize_t written = 0;
872 ssize_t res; 1069 eio_ssize_t res;
873 1070
874 if (!count) 1071 if (!count)
875 return 0; 1072 return 0;
876 1073
1074 for (;;)
1075 {
1076#ifdef __APPLE__
1077# undef HAVE_SENDFILE /* broken, as everything on os x */
1078#endif
877#if HAVE_SENDFILE 1079#if HAVE_SENDFILE
878# if __linux 1080# if __linux
1081 off_t soffset = offset;
879 res = sendfile (ofd, ifd, &offset, count); 1082 res = sendfile (ofd, ifd, &soffset, count);
880 1083
881# elif __freebsd 1084# elif __FreeBSD__
882 /* 1085 /*
883 * Of course, the freebsd sendfile is a dire hack with no thoughts 1086 * Of course, the freebsd sendfile is a dire hack with no thoughts
884 * wasted on making it similar to other I/O functions. 1087 * wasted on making it similar to other I/O functions.
885 */ 1088 */
886 {
887 off_t sbytes; 1089 off_t sbytes;
888 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 1090 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
889 1091
890 if (res < 0 && sbytes) 1092 #if 0 /* according to the manpage, this is correct, but broken behaviour */
891 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 1093 /* freebsd' sendfile will return 0 on success */
1094 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
1095 /* not on e.g. EIO or EPIPE - sounds broken */
1096 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
892 res = sbytes; 1097 res = sbytes;
893 } 1098 #endif
1099
1100 /* according to source inspection, this is correct, and useful behaviour */
1101 if (sbytes)
1102 res = sbytes;
1103
1104# elif defined __APPLE__
1105 off_t sbytes = count;
1106 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
1107
1108 /* according to the manpage, sbytes is always valid */
1109 if (sbytes)
1110 res = sbytes;
894 1111
895# elif __hpux 1112# elif __hpux
896 res = sendfile (ofd, ifd, offset, count, 0, 0); 1113 res = sendfile (ofd, ifd, offset, count, 0, 0);
897 1114
898# elif __solaris 1115# elif __solaris
899 {
900 struct sendfilevec vec; 1116 struct sendfilevec vec;
901 size_t sbytes; 1117 size_t sbytes;
902 1118
903 vec.sfv_fd = ifd; 1119 vec.sfv_fd = ifd;
904 vec.sfv_flag = 0; 1120 vec.sfv_flag = 0;
905 vec.sfv_off = offset; 1121 vec.sfv_off = offset;
906 vec.sfv_len = count; 1122 vec.sfv_len = count;
907 1123
908 res = sendfilev (ofd, &vec, 1, &sbytes); 1124 res = sendfilev (ofd, &vec, 1, &sbytes);
909 1125
910 if (res < 0 && sbytes) 1126 if (res < 0 && sbytes)
911 res = sbytes; 1127 res = sbytes;
912 }
913 1128
914# endif 1129# endif
1130
1131#elif defined (_WIN32) && 0
1132 /* does not work, just for documentation of what would need to be done */
1133 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
1134 /* libeio guarantees that the file offset does not change, and windows */
1135 /* has no way to get an independent handle to the same file description */
1136 HANDLE h = TO_SOCKET (ifd);
1137 SetFilePointer (h, offset, 0, FILE_BEGIN);
1138 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1139
915#else 1140#else
916 res = -1; 1141 res = EIO_ENOSYS ();
917 errno = ENOSYS;
918#endif 1142#endif
919 1143
1144 /* we assume sendfile can copy at least 128mb in one go */
1145 if (res <= 128 * 1024 * 1024)
1146 {
1147 if (res > 0)
1148 written += res;
1149
1150 if (written)
1151 return written;
1152
1153 break;
1154 }
1155 else
1156 {
1157 /* if we requested more, then probably the kernel was lazy */
1158 written += res;
1159 offset += res;
1160 count -= res;
1161
1162 if (!count)
1163 return written;
1164 }
1165 }
1166
920 if (res < 0 1167 if (res < 0
921 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1168 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1169 /* BSDs */
1170#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1171 || errno == ENOTSUP
1172#endif
1173#ifdef EOPNOTSUPP /* windows */
1174 || errno == EOPNOTSUPP /* BSDs */
1175#endif
922#if __solaris 1176#if __solaris
923 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1177 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
924#endif 1178#endif
925 ) 1179 )
926 ) 1180 )
… …
930 1184
931 res = 0; 1185 res = 0;
932 1186
933 while (count) 1187 while (count)
934 { 1188 {
935 ssize_t cnt; 1189 eio_ssize_t cnt;
936 1190
937 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 1191 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
938 1192
939 if (cnt <= 0) 1193 if (cnt <= 0)
940 { 1194 {
… …
952 1206
953 offset += cnt; 1207 offset += cnt;
954 res += cnt; 1208 res += cnt;
955 count -= cnt; 1209 count -= cnt;
956 } 1210 }
1211
1212 FUBd;
957 } 1213 }
958 1214
959 return res; 1215 return res;
1216}
1217
1218#ifdef PAGESIZE
1219# define eio_pagesize() PAGESIZE
1220#else
1221static intptr_t
1222eio_pagesize (void)
1223{
1224 static intptr_t page;
1225
1226 if (!page)
1227 page = sysconf (_SC_PAGESIZE);
1228
1229 return page;
1230}
1231#endif
1232
1233static void
1234eio_page_align (void **addr, size_t *length)
1235{
1236 intptr_t mask = eio_pagesize () - 1;
1237
1238 /* round down addr */
1239 intptr_t adj = mask & (intptr_t)*addr;
1240
1241 *addr = (void *)((intptr_t)*addr - adj);
1242 *length += adj;
1243
1244 /* round up length */
1245 *length = (*length + mask) & ~mask;
1246}
1247
1248#if !_POSIX_MEMLOCK
1249# define eio__mlockall(a) EIO_ENOSYS ()
1250#else
1251
1252static int
1253eio__mlockall (int flags)
1254{
1255 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1256 extern int mallopt (int, int);
1257 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1258 #endif
1259
1260 if (EIO_MCL_CURRENT != MCL_CURRENT
1261 || EIO_MCL_FUTURE != MCL_FUTURE)
1262 {
1263 flags = 0
1264 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1265 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1266 }
1267
1268 return mlockall (flags);
1269}
1270#endif
1271
1272#if !_POSIX_MEMLOCK_RANGE
1273# define eio__mlock(a,b) EIO_ENOSYS ()
1274#else
1275
1276static int
1277eio__mlock (void *addr, size_t length)
1278{
1279 eio_page_align (&addr, &length);
1280
1281 return mlock (addr, length);
1282}
1283
1284#endif
1285
1286#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1287# define eio__msync(a,b,c) EIO_ENOSYS ()
1288#else
1289
1290static int
1291eio__msync (void *mem, size_t len, int flags)
1292{
1293 eio_page_align (&mem, &len);
1294
1295 if (EIO_MS_ASYNC != MS_SYNC
1296 || EIO_MS_INVALIDATE != MS_INVALIDATE
1297 || EIO_MS_SYNC != MS_SYNC)
1298 {
1299 flags = 0
1300 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1301 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1302 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1303 }
1304
1305 return msync (mem, len, flags);
1306}
1307
1308#endif
1309
1310static int
1311eio__mtouch (eio_req *req)
1312{
1313 void *mem = req->ptr2;
1314 size_t len = req->size;
1315 int flags = req->int1;
1316
1317 eio_page_align (&mem, &len);
1318
1319 {
1320 intptr_t addr = (intptr_t)mem;
1321 intptr_t end = addr + len;
1322 intptr_t page = eio_pagesize ();
1323
1324 if (addr < end)
1325 if (flags & EIO_MT_MODIFY) /* modify */
1326 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1327 else
1328 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1329 }
1330
1331 return 0;
1332}
1333
1334/*****************************************************************************/
1335/* requests implemented outside eio_execute, because they are so large */
1336
1337static void
1338eio__lseek (eio_req *req)
1339{
1340 /* this usually gets optimised away completely, or your compiler sucks, */
1341 /* or the whence constants really are not 0, 1, 2 */
1342 int whence = req->int2 == EIO_SEEK_SET ? SEEK_SET
1343 : req->int2 == EIO_SEEK_CUR ? SEEK_CUR
1344 : req->int2 == EIO_SEEK_END ? SEEK_END
1345 : req->int2;
1346
1347 req->offs = lseek (req->int1, req->offs, whence);
1348 req->result = req->offs == (off_t)-1 ? -1 : 0;
1349}
1350
1351/* result will always end up in tmpbuf, there is always space for adding a 0-byte */
1352static int
1353eio__realpath (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
1354{
1355 const char *rel = path;
1356 char *res;
1357 char *tmp1, *tmp2;
1358#if SYMLOOP_MAX > 32
1359 int symlinks = SYMLOOP_MAX;
1360#else
1361 int symlinks = 32;
1362#endif
1363
1364 errno = EINVAL;
1365 if (!rel)
1366 return -1;
1367
1368 errno = ENOENT;
1369 if (!*rel)
1370 return -1;
1371
1372 res = tmpbuf_get (tmpbuf, PATH_MAX * 3);
1373 tmp1 = res + PATH_MAX;
1374 tmp2 = tmp1 + PATH_MAX;
1375
1376#if 0 /* disabled, the musl way to do things is just too racy */
1377#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1378 /* on linux we may be able to ask the kernel */
1379 {
1380 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1381
1382 if (fd >= 0)
1383 {
1384 sprintf (tmp1, "/proc/self/fd/%d", fd);
1385 req->result = readlink (tmp1, res, PATH_MAX);
1386 close (fd);
1387
1388 /* here we should probably stat the open file and the disk file, to make sure they still match */
1389
1390 if (req->result > 0)
1391 goto done;
1392 }
1393 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1394 return;
1395 }
1396#endif
1397#endif
1398
1399 if (*rel != '/')
1400 {
1401 int len;
1402
1403 errno = ENOENT;
1404 if (wd == EIO_INVALID_WD)
1405 return -1;
1406
1407 if (wd == EIO_CWD)
1408 {
1409 if (!getcwd (res, PATH_MAX))
1410 return -1;
1411
1412 len = strlen (res);
1413 }
1414 else
1415 memcpy (res, wd->str, len = wd->len);
1416
1417 if (res [1]) /* only use if not / */
1418 res += len;
1419 }
1420
1421 while (*rel)
1422 {
1423 eio_ssize_t len, linklen;
1424 const char *beg = rel;
1425
1426 while (*rel && *rel != '/')
1427 ++rel;
1428
1429 len = rel - beg;
1430
1431 if (!len) /* skip slashes */
1432 {
1433 ++rel;
1434 continue;
1435 }
1436
1437 if (beg [0] == '.')
1438 {
1439 if (len == 1)
1440 continue; /* . - nop */
1441
1442 if (beg [1] == '.' && len == 2)
1443 {
1444 /* .. - back up one component, if possible */
1445
1446 while (res != tmpbuf->ptr)
1447 if (*--res == '/')
1448 break;
1449
1450 continue;
1451 }
1452 }
1453
1454 errno = ENAMETOOLONG;
1455 if (res + 1 + len + 1 >= tmp1)
1456 return -1;
1457
1458 /* copy one component */
1459 *res = '/';
1460 memcpy (res + 1, beg, len);
1461
1462 /* zero-terminate, for readlink */
1463 res [len + 1] = 0;
1464
1465 /* now check if it's a symlink */
1466 linklen = readlink (tmpbuf->ptr, tmp1, PATH_MAX);
1467
1468 if (linklen < 0)
1469 {
1470 if (errno != EINVAL)
1471 return -1;
1472
1473 /* it's a normal directory. hopefully */
1474 res += len + 1;
1475 }
1476 else
1477 {
1478 /* yay, it was a symlink - build new path in tmp2 */
1479 int rellen = strlen (rel);
1480
1481 errno = ENAMETOOLONG;
1482 if (linklen + 1 + rellen >= PATH_MAX)
1483 return -1;
1484
1485 errno = ELOOP;
1486 if (!--symlinks)
1487 return -1;
1488
1489 if (*tmp1 == '/')
1490 res = tmpbuf->ptr; /* symlink resolves to an absolute path */
1491
1492 /* we need to be careful, as rel might point into tmp2 already */
1493 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1494 tmp2 [linklen] = '/';
1495 memcpy (tmp2, tmp1, linklen);
1496
1497 rel = tmp2;
1498 }
1499 }
1500
1501 /* special case for the lone root path */
1502 if (res == tmpbuf->ptr)
1503 *res++ = '/';
1504
1505 return res - (char *)tmpbuf->ptr;
1506}
1507
1508static signed char
1509eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1510{
1511 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1512 : a->inode < b->inode ? -1
1513 : a->inode > b->inode ? 1
1514 : 0;
1515}
1516
1517#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1518
1519#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1520#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1521
1522static void
1523eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1524{
1525 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1526 unsigned char *bit = bits;
1527
1528 assert (CHAR_BIT == 8);
1529 assert (sizeof (eio_dirent) * 8 < 256);
1530 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1531 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1532
1533 if (size <= EIO_SORT_FAST)
1534 return;
1535
1536 /* first prepare an array of bits to test in our radix sort */
1537 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1538 /* inode_bits must contain all inodes ORed together */
1539 /* which is used to skip bits that are 0 everywhere, which is very common */
1540 {
1541 eio_ino_t endianness;
1542 int i, j;
1543
1544 /* we store the byte offset of byte n into byte n of "endianness" */
1545 for (i = 0; i < sizeof (eio_ino_t); ++i)
1546 ((unsigned char *)&endianness)[i] = i;
1547
1548 *bit++ = 0;
1549
1550 for (i = 0; i < sizeof (eio_ino_t); ++i)
1551 {
1552 /* shifting off the byte offsets out of "endianness" */
1553 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1554 endianness >>= 8;
1555
1556 for (j = 0; j < 8; ++j)
1557 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1558 *bit++ = offs + j;
1559 }
1560
1561 for (j = 0; j < 8; ++j)
1562 if (score_bits & (1 << j))
1563 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1564 }
1565
1566 /* now actually do the sorting (a variant of MSD radix sort) */
1567 {
1568 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1569 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1570 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1571 int stk_idx = 0;
1572
1573 base_stk [stk_idx] = dents;
1574 end_stk [stk_idx] = dents + size;
1575 bit_stk [stk_idx] = bit - 1;
1576
1577 do
1578 {
1579 base = base_stk [stk_idx];
1580 end = end_stk [stk_idx];
1581 bit = bit_stk [stk_idx];
1582
1583 for (;;)
1584 {
1585 unsigned char O = *bit >> 3;
1586 unsigned char M = 1 << (*bit & 7);
1587
1588 eio_dirent *a = base;
1589 eio_dirent *b = end;
1590
1591 if (b - a < EIO_SORT_CUTOFF)
1592 break;
1593
1594 /* now bit-partition the array on the bit */
1595 /* this ugly asymmetric loop seems to perform much better than typical */
1596 /* partition algos found in the literature */
1597 do
1598 if (!(((unsigned char *)a)[O] & M))
1599 ++a;
1600 else if (!(((unsigned char *)--b)[O] & M))
1601 {
1602 eio_dirent tmp = *a; *a = *b; *b = tmp;
1603 ++a;
1604 }
1605 while (b > a);
1606
1607 /* next bit, or stop, if no bits left in this path */
1608 if (!*--bit)
1609 break;
1610
1611 base_stk [stk_idx] = a;
1612 end_stk [stk_idx] = end;
1613 bit_stk [stk_idx] = bit;
1614 ++stk_idx;
1615
1616 end = a;
1617 }
1618 }
1619 while (stk_idx--);
1620 }
1621}
1622
1623static void
1624eio_dent_insertion_sort (eio_dirent *dents, int size)
1625{
1626 /* first move the smallest element to the front, to act as a sentinel */
1627 {
1628 int i;
1629 eio_dirent *min = dents;
1630
1631 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1632 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1633 if (EIO_DENT_CMP (dents [i], <, *min))
1634 min = &dents [i];
1635
1636 /* swap elements 0 and j (minimum) */
1637 {
1638 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1639 }
1640 }
1641
1642 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1643 {
1644 eio_dirent *i, *j;
1645
1646 for (i = dents + 1; i < dents + size; ++i)
1647 {
1648 eio_dirent value = *i;
1649
1650 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1651 j [1] = j [0];
1652
1653 j [1] = value;
1654 }
1655 }
1656}
1657
1658static void
1659eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1660{
1661 if (size <= 1)
1662 return; /* our insertion sort relies on size > 0 */
1663
1664 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1665 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1666 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1667
1668 /* use an insertion sort at the end, or for small arrays, */
1669 /* as insertion sort is more efficient for small partitions */
1670 eio_dent_insertion_sort (dents, size);
960} 1671}
961 1672
962/* read a full directory */ 1673/* read a full directory */
963static void 1674static void
964eio__scandir (eio_req *req, etp_worker *self) 1675eio__scandir (eio_req *req, etp_worker *self)
965{ 1676{
1677 char *name, *names;
1678 int namesalloc = 4096 - sizeof (void *) * 4;
1679 int namesoffs = 0;
1680 int flags = req->int1;
1681 eio_dirent *dents = 0;
1682 int dentalloc = 128;
1683 int dentoffs = 0;
1684 eio_ino_t inode_bits = 0;
1685#ifdef _WIN32
1686 HANDLE dirp;
1687 WIN32_FIND_DATA entp;
1688#else
966 DIR *dirp; 1689 DIR *dirp;
967 EIO_STRUCT_DIRENT *entp; 1690 EIO_STRUCT_DIRENT *entp;
968 char *name, *names; 1691#endif
969 int memlen = 4096;
970 int memofs = 0;
971 int res = 0;
972 1692
973 X_LOCK (wrklock); 1693 req->result = -1;
974 /* the corresponding closedir is in ETP_WORKER_CLEAR */
975 self->dirp = dirp = opendir (req->ptr1);
976 req->flags |= EIO_FLAG_PTR2_FREE;
977 req->ptr2 = names = malloc (memlen);
978 X_UNLOCK (wrklock);
979 1694
980 if (dirp && names) 1695 if (!(flags & EIO_READDIR_DENTS))
981 for (;;) 1696 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1697
1698#ifdef _WIN32
1699 {
1700 int len = strlen ((const char *)req->ptr1);
1701 char *path = malloc (MAX_PATH);
1702 const char *fmt;
1703 const char *reqpath = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1704
1705 if (!len)
1706 fmt = "./*";
1707 else if (reqpath[len - 1] == '/' || reqpath[len - 1] == '\\')
1708 fmt = "%s*";
1709 else
1710 fmt = "%s/*";
1711
1712 _snprintf (path, MAX_PATH, fmt, reqpath);
1713 dirp = FindFirstFile (path, &entp);
1714 free (path);
1715
1716 if (dirp == INVALID_HANDLE_VALUE)
1717 {
1718 /* should steal _dosmaperr */
1719 switch (GetLastError ())
1720 {
1721 case ERROR_FILE_NOT_FOUND:
1722 req->result = 0;
1723 break;
1724
1725 case ERROR_INVALID_NAME:
1726 case ERROR_PATH_NOT_FOUND:
1727 case ERROR_NO_MORE_FILES:
1728 errno = ENOENT;
1729 break;
1730
1731 case ERROR_NOT_ENOUGH_MEMORY:
1732 errno = ENOMEM;
1733 break;
1734
1735 default:
1736 errno = EINVAL;
1737 break;
1738 }
1739
1740 return;
1741 }
1742 }
1743#else
1744 #if HAVE_AT
1745 if (req->wd)
982 { 1746 {
1747 int fd = openat (WD2FD (req->wd), req->ptr1, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1748
1749 if (fd < 0)
1750 return;
1751
1752 dirp = fdopendir (fd);
1753
1754 if (!dirp)
1755 close (fd);
1756 }
1757 else
1758 dirp = opendir (req->ptr1);
1759 #else
1760 dirp = opendir (wd_expand (&self->tmpbuf, req->wd, req->ptr1));
1761 #endif
1762
1763 if (!dirp)
1764 return;
1765#endif
1766
1767 if (req->flags & EIO_FLAG_PTR1_FREE)
1768 free (req->ptr1);
1769
1770 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1771 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1772 req->ptr2 = names = malloc (namesalloc);
1773
1774 if (!names || (flags && !dents))
1775 return;
1776
1777 for (;;)
1778 {
1779 int done;
1780
1781#ifdef _WIN32
1782 done = !dirp;
1783#else
983 errno = 0; 1784 errno = 0;
984 entp = readdir (dirp); 1785 entp = readdir (dirp);
1786 done = !entp;
1787#endif
985 1788
1789 if (done)
1790 {
1791#ifndef _WIN32
1792 int old_errno = errno;
1793 closedir (dirp);
1794 errno = old_errno;
1795
986 if (!entp) 1796 if (errno)
1797 break;
1798#endif
1799
1800 /* sort etc. */
1801 req->int1 = flags;
1802 req->result = dentoffs;
1803
1804 if (flags & EIO_READDIR_STAT_ORDER)
1805 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1806 else if (flags & EIO_READDIR_DIRS_FIRST)
1807 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1808 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1809 else
1810 {
1811 /* in this case, all is known, and we just put dirs first and sort them */
1812 eio_dirent *oth = dents + dentoffs;
1813 eio_dirent *dir = dents;
1814
1815 /* now partition dirs to the front, and non-dirs to the back */
1816 /* by walking from both sides and swapping if necessary */
1817 while (oth > dir)
1818 {
1819 if (dir->type == EIO_DT_DIR)
1820 ++dir;
1821 else if ((--oth)->type == EIO_DT_DIR)
1822 {
1823 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1824
1825 ++dir;
1826 }
1827 }
1828
1829 /* now sort the dirs only (dirs all have the same score) */
1830 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1831 }
1832
987 break; 1833 break;
1834 }
988 1835
989 name = entp->d_name; 1836 /* now add the entry to our list(s) */
1837 name = D_NAME (entp);
990 1838
1839 /* skip . and .. entries */
991 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1840 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
992 { 1841 {
993 int len = strlen (name) + 1; 1842 int len = D_NAMLEN (entp) + 1;
994 1843
995 res++; 1844 while (ecb_expect_false (namesoffs + len > namesalloc))
996
997 while (memofs + len > memlen)
998 { 1845 {
999 memlen *= 2; 1846 namesalloc *= 2;
1000 X_LOCK (wrklock);
1001 req->ptr2 = names = realloc (names, memlen); 1847 req->ptr2 = names = realloc (names, namesalloc);
1002 X_UNLOCK (wrklock);
1003 1848
1004 if (!names) 1849 if (!names)
1850 break;
1851 }
1852
1853 memcpy (names + namesoffs, name, len);
1854
1855 if (dents)
1856 {
1857 struct eio_dirent *ent;
1858
1859 if (ecb_expect_false (dentoffs == dentalloc))
1860 {
1861 dentalloc *= 2;
1862 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1863
1864 if (!dents)
1005 break; 1865 break;
1006 } 1866 }
1007 1867
1008 memcpy (names + memofs, name, len); 1868 ent = dents + dentoffs;
1009 memofs += len; 1869
1870 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1871 ent->namelen = len - 1;
1872 ent->inode = D_INO (entp);
1873
1874 inode_bits |= ent->inode;
1875
1876 switch (D_TYPE (entp))
1877 {
1878 default:
1879 ent->type = EIO_DT_UNKNOWN;
1880 flags |= EIO_READDIR_FOUND_UNKNOWN;
1881 break;
1882
1883 #ifdef DT_FIFO
1884 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1885 #endif
1886 #ifdef DT_CHR
1887 case DT_CHR: ent->type = EIO_DT_CHR; break;
1888 #endif
1889 #ifdef DT_MPC
1890 case DT_MPC: ent->type = EIO_DT_MPC; break;
1891 #endif
1892 #ifdef DT_DIR
1893 case DT_DIR: ent->type = EIO_DT_DIR; break;
1894 #endif
1895 #ifdef DT_NAM
1896 case DT_NAM: ent->type = EIO_DT_NAM; break;
1897 #endif
1898 #ifdef DT_BLK
1899 case DT_BLK: ent->type = EIO_DT_BLK; break;
1900 #endif
1901 #ifdef DT_MPB
1902 case DT_MPB: ent->type = EIO_DT_MPB; break;
1903 #endif
1904 #ifdef DT_REG
1905 case DT_REG: ent->type = EIO_DT_REG; break;
1906 #endif
1907 #ifdef DT_NWK
1908 case DT_NWK: ent->type = EIO_DT_NWK; break;
1909 #endif
1910 #ifdef DT_CMP
1911 case DT_CMP: ent->type = EIO_DT_CMP; break;
1912 #endif
1913 #ifdef DT_LNK
1914 case DT_LNK: ent->type = EIO_DT_LNK; break;
1915 #endif
1916 #ifdef DT_SOCK
1917 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1918 #endif
1919 #ifdef DT_DOOR
1920 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1921 #endif
1922 #ifdef DT_WHT
1923 case DT_WHT: ent->type = EIO_DT_WHT; break;
1924 #endif
1925 }
1926
1927 ent->score = 7;
1928
1929 if (flags & EIO_READDIR_DIRS_FIRST)
1930 {
1931 if (ent->type == EIO_DT_UNKNOWN)
1932 {
1933 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1934 ent->score = 1;
1935 else if (!strchr (name, '.')) /* absence of dots indicate likely dirs */
1936 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1937 }
1938 else if (ent->type == EIO_DT_DIR)
1939 ent->score = 0;
1940 }
1010 } 1941 }
1942
1943 namesoffs += len;
1944 ++dentoffs;
1011 } 1945 }
1012 1946
1013 if (errno) 1947 if (EIO_CANCELLED (req))
1014 res = -1; 1948 {
1949 errno = ECANCELED;
1950 break;
1951 }
1952
1953#ifdef _WIN32
1954 if (!FindNextFile (dirp, &entp))
1955 {
1956 FindClose (dirp);
1957 dirp = 0;
1958 }
1959#endif
1960 }
1961}
1962
1963/*****************************************************************************/
1964/* working directory stuff */
1965/* various deficiencies in the posix 2008 api force us to */
1966/* keep the absolute path in string form at all times */
1967/* fuck yeah. */
1968
1969#if !HAVE_AT
1970
1971/* a bit like realpath, but usually faster because it doesn'T have to return */
1972/* an absolute or canonical path */
1973static const char *
1974wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
1975{
1976 if (!wd || *path == '/')
1977 return path;
1978
1979 if (path [0] == '.' && !path [1])
1980 return wd->str;
1981
1015 1982 {
1016 req->result = res; 1983 int l1 = wd->len;
1017} 1984 int l2 = strlen (path);
1018 1985
1019#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) 1986 char *res = tmpbuf_get (tmpbuf, l1 + l2 + 2);
1020# undef msync
1021# define msync(a,b,c) ENOSYS
1022#endif
1023 1987
1024int 1988 memcpy (res, wd->str, l1);
1025eio__mtouch (void *mem, size_t len, int flags) 1989 res [l1] = '/';
1026{ 1990 memcpy (res + l1 + 1, path, l2 + 1);
1027 intptr_t addr = (intptr_t)mem;
1028 intptr_t end = addr + len;
1029#ifdef PAGESIZE
1030 const intptr_t page = PAGESIZE;
1031#else
1032 static intptr_t page;
1033 1991
1034 if (!page) 1992 return res;
1035 page = sysconf (_SC_PAGESIZE); 1993 }
1036#endif 1994}
1037 1995
1038 addr &= ~(page - 1); /* assume page size is always a power of two */ 1996#endif
1039 1997
1040 if (addr < end) 1998static eio_wd
1041 if (flags) /* modify */ 1999eio__wd_open_sync (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
1042 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len); 2000{
1043 else 2001 int fd;
1044 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len); 2002 eio_wd res;
2003 int len = eio__realpath (tmpbuf, wd, path);
1045 2004
2005 if (len < 0)
2006 return EIO_INVALID_WD;
2007
2008#if HAVE_AT
2009 fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
2010
2011 if (fd < 0)
2012 return EIO_INVALID_WD;
2013#endif
2014
2015 res = malloc (sizeof (*res) + len); /* one extra 0-byte */
2016
2017#if HAVE_AT
2018 res->fd = fd;
2019#endif
2020
2021 res->len = len;
2022 memcpy (res->str, tmpbuf->ptr, len);
2023 res->str [len] = 0;
2024
2025 return res;
2026}
2027
2028eio_wd
2029eio_wd_open_sync (eio_wd wd, const char *path)
2030{
2031 struct tmpbuf tmpbuf = { 0 };
2032 wd = eio__wd_open_sync (&tmpbuf, wd, path);
2033 free (tmpbuf.ptr);
2034
1046 return 0; 2035 return wd;
1047} 2036}
2037
2038void
2039eio_wd_close_sync (eio_wd wd)
2040{
2041 if (wd != EIO_INVALID_WD && wd != EIO_CWD)
2042 {
2043 #if HAVE_AT
2044 close (wd->fd);
2045 #endif
2046 free (wd);
2047 }
2048}
2049
2050#if HAVE_AT
2051
2052/* they forgot these */
2053
2054static int
2055eio__truncateat (int dirfd, const char *path, off_t length)
2056{
2057 int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
2058 int res;
2059
2060 if (fd < 0)
2061 return fd;
2062
2063 res = ftruncate (fd, length);
2064 close (fd);
2065 return res;
2066}
2067
2068static int
2069eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
2070{
2071 int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
2072 int res;
2073
2074 if (fd < 0)
2075 return fd;
2076
2077 res = fstatvfs (fd, buf);
2078 close (fd);
2079 return res;
2080
2081}
2082
2083#endif
1048 2084
1049/*****************************************************************************/ 2085/*****************************************************************************/
1050 2086
1051#define ALLOC(len) \ 2087#define ALLOC(len) \
1052 if (!req->ptr2) \ 2088 if (!req->ptr2) \
… …
1061 req->result = -1; \ 2097 req->result = -1; \
1062 break; \ 2098 break; \
1063 } \ 2099 } \
1064 } 2100 }
1065 2101
2102static void ecb_noinline ecb_cold
2103etp_proc_init (void)
2104{
2105#if HAVE_PRCTL_SET_NAME
2106 /* provide a more sensible "thread name" */
2107 char name[16 + 1];
2108 const int namelen = sizeof (name) - 1;
2109 int len;
2110
2111 prctl (PR_GET_NAME, (unsigned long)name, 0, 0, 0);
2112 name [namelen] = 0;
2113 len = strlen (name);
2114 strcpy (name + (len <= namelen - 4 ? len : namelen - 4), "/eio");
2115 prctl (PR_SET_NAME, (unsigned long)name, 0, 0, 0);
2116#endif
2117}
2118
1066X_THREAD_PROC (etp_proc) 2119X_THREAD_PROC (etp_proc)
1067{ 2120{
1068 ETP_REQ *req; 2121 ETP_REQ *req;
1069 struct timespec ts; 2122 struct timespec ts;
1070 etp_worker *self = (etp_worker *)thr_arg; 2123 etp_worker *self = (etp_worker *)thr_arg;
1071 2124
2125 etp_proc_init ();
2126
1072 /* try to distribute timeouts somewhat randomly */ 2127 /* try to distribute timeouts somewhat evenly */
1073 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 2128 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1074 2129
1075 for (;;) 2130 for (;;)
1076 { 2131 {
2132 ts.tv_sec = 0;
2133
1077 X_LOCK (reqlock); 2134 X_LOCK (reqlock);
1078 2135
1079 for (;;) 2136 for (;;)
1080 { 2137 {
1081 self->req = req = reqq_shift (&req_queue); 2138 req = reqq_shift (&req_queue);
1082 2139
1083 if (req) 2140 if (req)
1084 break; 2141 break;
1085 2142
2143 if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
2144 {
2145 X_UNLOCK (reqlock);
2146 X_LOCK (wrklock);
2147 --started;
2148 X_UNLOCK (wrklock);
2149 goto quit;
2150 }
2151
1086 ++idle; 2152 ++idle;
1087 2153
1088 ts.tv_sec = time (0) + IDLE_TIMEOUT; 2154 if (idle <= max_idle)
1089 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 2155 /* we are allowed to idle, so do so without any timeout */
2156 X_COND_WAIT (reqwait, reqlock);
2157 else
1090 { 2158 {
1091 if (idle > max_idle) 2159 /* initialise timeout once */
1092 { 2160 if (!ts.tv_sec)
1093 --idle; 2161 ts.tv_sec = time (0) + idle_timeout;
1094 X_UNLOCK (reqlock);
1095 X_LOCK (wrklock);
1096 --started;
1097 X_UNLOCK (wrklock);
1098 goto quit;
1099 }
1100 2162
1101 /* we are allowed to idle, so do so without any timeout */
1102 X_COND_WAIT (reqwait, reqlock); 2163 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
2164 ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1103 } 2165 }
1104 2166
1105 --idle; 2167 --idle;
1106 } 2168 }
1107 2169
… …
1110 X_UNLOCK (reqlock); 2172 X_UNLOCK (reqlock);
1111 2173
1112 if (req->type < 0) 2174 if (req->type < 0)
1113 goto quit; 2175 goto quit;
1114 2176
1115 if (!EIO_CANCELLED (req))
1116 ETP_EXECUTE (self, req); 2177 ETP_EXECUTE (self, req);
1117 2178
1118 X_LOCK (reslock); 2179 X_LOCK (reslock);
1119 2180
1120 ++npending; 2181 ++npending;
1121 2182
1122 if (!reqq_push (&res_queue, req) && want_poll_cb) 2183 if (!reqq_push (&res_queue, req) && want_poll_cb)
1123 want_poll_cb (); 2184 want_poll_cb ();
1124 2185
1125 self->req = 0;
1126 etp_worker_clear (self); 2186 etp_worker_clear (self);
1127 2187
1128 X_UNLOCK (reslock); 2188 X_UNLOCK (reslock);
1129 } 2189 }
1130 2190
1131quit: 2191quit:
2192 free (req);
2193
1132 X_LOCK (wrklock); 2194 X_LOCK (wrklock);
1133 etp_worker_free (self); 2195 etp_worker_free (self);
1134 X_UNLOCK (wrklock); 2196 X_UNLOCK (wrklock);
1135
1136 return 0;
1137} 2197}
1138 2198
1139/*****************************************************************************/ 2199/*****************************************************************************/
1140 2200
2201int ecb_cold
1141int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 2202eio_init (void (*want_poll)(void), void (*done_poll)(void))
1142{ 2203{
1143 return etp_init (want_poll, done_poll); 2204 return etp_init (want_poll, done_poll);
1144} 2205}
1145 2206
2207ecb_inline void
1146static void eio_api_destroy (eio_req *req) 2208eio_api_destroy (eio_req *req)
1147{ 2209{
1148 free (req); 2210 free (req);
1149} 2211}
1150 2212
1151#define REQ(rtype) \ 2213#define REQ(rtype) \
1152 eio_req *req; \ 2214 eio_req *req; \
1153 \ 2215 \
1154 req = (eio_req *)calloc (1, sizeof *req); \ 2216 req = (eio_req *)calloc (1, sizeof *req); \
1155 if (!req) \ 2217 if (!req) \
1156 return 0; \ 2218 return 0; \
… …
1170 { \ 2232 { \
1171 eio_api_destroy (req); \ 2233 eio_api_destroy (req); \
1172 return 0; \ 2234 return 0; \
1173 } 2235 }
1174 2236
2237static void
1175static void eio_execute (etp_worker *self, eio_req *req) 2238eio_execute (etp_worker *self, eio_req *req)
1176{ 2239{
1177 errno = 0; 2240#if HAVE_AT
2241 int dirfd;
2242#else
2243 const char *path;
2244#endif
2245
2246 if (ecb_expect_false (EIO_CANCELLED (req)))
2247 {
2248 req->result = -1;
2249 req->errorno = ECANCELED;
2250 return;
2251 }
2252
2253 if (ecb_expect_false (req->wd == EIO_INVALID_WD))
2254 {
2255 req->result = -1;
2256 req->errorno = ENOENT;
2257 return;
2258 }
2259
2260 if (req->type >= EIO_OPEN)
2261 {
2262 #if HAVE_AT
2263 dirfd = WD2FD (req->wd);
2264 #else
2265 path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
2266 #endif
2267 }
1178 2268
1179 switch (req->type) 2269 switch (req->type)
1180 { 2270 {
2271 case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
2272 req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
2273 break;
2274 case EIO_WD_CLOSE: req->result = 0;
2275 eio_wd_close_sync (req->wd); break;
2276
2277 case EIO_SEEK: eio__lseek (req); break;
1181 case EIO_READ: ALLOC (req->size); 2278 case EIO_READ: ALLOC (req->size);
1182 req->result = req->offs >= 0 2279 req->result = req->offs >= 0
1183 ? pread (req->int1, req->ptr2, req->size, req->offs) 2280 ? pread (req->int1, req->ptr2, req->size, req->offs)
1184 : read (req->int1, req->ptr2, req->size); break; 2281 : read (req->int1, req->ptr2, req->size); break;
1185 case EIO_WRITE: req->result = req->offs >= 0 2282 case EIO_WRITE: req->result = req->offs >= 0
1186 ? pwrite (req->int1, req->ptr2, req->size, req->offs) 2283 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1187 : write (req->int1, req->ptr2, req->size); break; 2284 : write (req->int1, req->ptr2, req->size); break;
1188 2285
1189 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break; 2286 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1190 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break; 2287 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
2288
2289#if HAVE_AT
1191 2290
1192 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2291 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1193 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2292 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break;
1194 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2293 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1195 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2294 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, AT_SYMLINK_NOFOLLOW); break;
1196 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1197 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1198
1199 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; 2295 case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break;
1200 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1201 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break; 2296 case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break;
1202 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1203 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break; 2297 case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break;
1204 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1205
1206 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break; 2298 case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break;
1207 case EIO_CLOSE: req->result = close (req->int1); break; 2299
1208 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1209 case EIO_UNLINK: req->result = unlink (req->ptr1); break; 2300 case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
1210 case EIO_RMDIR: req->result = rmdir (req->ptr1); break; 2301 case EIO_RMDIR: req->result = unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break;
1211 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; 2302 case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); break;
1212 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; 2303 case EIO_RENAME: req->result = renameat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2); break;
1213 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; 2304 case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2, 0); break;
1214 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; 2305 case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
1215 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break; 2306 case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1216
1217 case EIO_READLINK: ALLOC (NAME_MAX); 2307 case EIO_READLINK: ALLOC (PATH_MAX);
1218 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; 2308 req->result = readlinkat (dirfd, req->ptr1, req->ptr2, PATH_MAX); break;
1219 2309 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1220 case EIO_SYNC: req->result = 0; sync (); break; 2310 req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1221 case EIO_FSYNC: req->result = fsync (req->int1); break;
1222 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1223 case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break;
1224 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;
1225 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1226
1227 case EIO_READDIR: eio__scandir (req, self); break;
1228
1229 case EIO_BUSY: 2311 case EIO_UTIME:
1230#ifdef _WIN32 2312 case EIO_FUTIME:
1231 Sleep (req->nv1 * 1000.); 2313 {
2314 struct timespec ts[2];
2315 struct timespec *times;
2316
2317 if (req->nv1 != -1. || req->nv2 != -1.)
2318 {
2319 ts[0].tv_sec = req->nv1;
2320 ts[0].tv_nsec = (req->nv1 - ts[0].tv_sec) * 1e9;
2321 ts[1].tv_sec = req->nv2;
2322 ts[1].tv_nsec = (req->nv2 - ts[1].tv_sec) * 1e9;
2323
2324 times = ts;
2325 }
2326 else
2327 times = 0;
2328
2329 req->result = req->type == EIO_FUTIME
2330 ? futimens (req->int1, times)
2331 : utimensat (dirfd, req->ptr1, times, 0);
2332 }
2333 break;
2334
1232#else 2335#else
1233 {
1234 struct timeval tv;
1235 2336
1236 tv.tv_sec = req->nv1; 2337 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1237 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; 2338 req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
2339 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2340 req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
2341 case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
2342 case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
2343 case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
2344 case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
1238 2345
1239 req->result = select (0, 0, 0, 0, &tv); 2346 case EIO_UNLINK: req->result = unlink (path ); break;
1240 } 2347 case EIO_RMDIR: req->result = rmdir (path ); break;
1241#endif 2348 case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
1242 break; 2349 case EIO_RENAME: req->result = rename (path , req->ptr2); break;
2350 case EIO_LINK: req->result = link (path , req->ptr2); break;
2351 case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
2352 case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
2353 case EIO_READLINK: ALLOC (PATH_MAX);
2354 req->result = readlink (path, req->ptr2, PATH_MAX); break;
2355 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2356 req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
1243 2357
1244 case EIO_UTIME: 2358 case EIO_UTIME:
1245 case EIO_FUTIME: 2359 case EIO_FUTIME:
1246 { 2360 {
1247 struct timeval tv[2]; 2361 struct timeval tv[2];
1248 struct timeval *times; 2362 struct timeval *times;
1249 2363
1250 if (req->nv1 != -1. || req->nv2 != -1.) 2364 if (req->nv1 != -1. || req->nv2 != -1.)
1251 { 2365 {
1252 tv[0].tv_sec = req->nv1; 2366 tv[0].tv_sec = req->nv1;
1253 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.; 2367 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
1254 tv[1].tv_sec = req->nv2; 2368 tv[1].tv_sec = req->nv2;
1255 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.; 2369 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
1256 2370
1257 times = tv; 2371 times = tv;
1258 } 2372 }
1259 else 2373 else
1260 times = 0; 2374 times = 0;
1261
1262 2375
1263 req->result = req->type == EIO_FUTIME 2376 req->result = req->type == EIO_FUTIME
1264 ? futimes (req->int1, times) 2377 ? futimes (req->int1, times)
1265 : utimes (req->ptr1, times); 2378 : utimes (req->ptr1, times);
1266 } 2379 }
1267 break; 2380 break;
1268 2381
2382#endif
2383
2384 case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
2385 {
2386 ALLOC (req->result);
2387 memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
2388 }
2389 break;
2390
2391 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2392 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
2393
2394 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2395 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2396
2397 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
2398 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
2399 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
2400
2401 case EIO_CLOSE: req->result = close (req->int1); break;
2402 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
2403 case EIO_SYNC: req->result = 0; sync (); break;
2404 case EIO_FSYNC: req->result = fsync (req->int1); break;
2405 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
2406 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
2407 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
2408 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
2409 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
2410 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
2411 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
2412 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
2413
2414 case EIO_READDIR: eio__scandir (req, self); break;
2415
2416 case EIO_BUSY:
2417#ifdef _WIN32
2418 Sleep (req->nv1 * 1e3);
2419#else
2420 {
2421 struct timeval tv;
2422
2423 tv.tv_sec = req->nv1;
2424 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
2425
2426 req->result = select (0, 0, 0, 0, &tv);
2427 }
2428#endif
2429 break;
2430
1269 case EIO_GROUP: 2431 case EIO_GROUP:
1270 abort (); /* handled in eio_request */ 2432 abort (); /* handled in eio_request */
1271 2433
1272 case EIO_NOP: 2434 case EIO_NOP:
1273 req->result = 0; 2435 req->result = 0;
1274 break; 2436 break;
1275 2437
1276 case EIO_CUSTOM: 2438 case EIO_CUSTOM:
1277 ((void (*)(eio_req *))req->feed) (req); 2439 req->feed (req);
1278 break; 2440 break;
1279 2441
1280 default: 2442 default:
1281 req->result = -1; 2443 req->result = EIO_ENOSYS ();
1282 break; 2444 break;
1283 } 2445 }
1284 2446
1285 req->errorno = errno; 2447 req->errorno = errno;
1286} 2448}
1287 2449
1288#ifndef EIO_NO_WRAPPERS 2450#ifndef EIO_NO_WRAPPERS
1289 2451
2452eio_req *eio_wd_open (const char *path, int pri, eio_cb cb, void *data)
2453{
2454 REQ (EIO_WD_OPEN); PATH; SEND;
2455}
2456
2457eio_req *eio_wd_close (eio_wd wd, int pri, eio_cb cb, void *data)
2458{
2459 REQ (EIO_WD_CLOSE); req->wd = wd; SEND;
2460}
2461
1290eio_req *eio_nop (int pri, eio_cb cb, void *data) 2462eio_req *eio_nop (int pri, eio_cb cb, void *data)
1291{ 2463{
1292 REQ (EIO_NOP); SEND; 2464 REQ (EIO_NOP); SEND;
1293} 2465}
1294 2466
… …
1310eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2482eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1311{ 2483{
1312 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2484 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1313} 2485}
1314 2486
2487eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2488{
2489 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2490}
2491
2492eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
2493{
2494 REQ (EIO_SYNCFS); req->int1 = fd; SEND;
2495}
2496
2497eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2498{
2499 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2500}
2501
1315eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2502eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1316{ 2503{
1317 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2504 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1318} 2505}
1319 2506
1320eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) 2507eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
1321{ 2508{
1322 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; 2509 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
1323} 2510}
1324 2511
1325eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) 2512eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
1326{ 2513{
1327 REQ (EIO_FDATASYNC); req->int1 = fd; SEND; 2514 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2515}
2516
2517eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2518{
2519 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
1328} 2520}
1329 2521
1330eio_req *eio_close (int fd, int pri, eio_cb cb, void *data) 2522eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
1331{ 2523{
1332 REQ (EIO_CLOSE); req->int1 = fd; SEND; 2524 REQ (EIO_CLOSE); req->int1 = fd; SEND;
… …
1335eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data) 2527eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
1336{ 2528{
1337 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND; 2529 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
1338} 2530}
1339 2531
2532eio_req *eio_seek (int fd, off_t offset, int whence, int pri, eio_cb cb, void *data)
2533{
2534 REQ (EIO_SEEK); req->int1 = fd; req->offs = offset; req->int2 = whence; SEND;
2535}
2536
1340eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data) 2537eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1341{ 2538{
1342 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND; 2539 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1343} 2540}
1344 2541
… …
1350eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) 2547eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1351{ 2548{
1352 REQ (EIO_FSTAT); req->int1 = fd; SEND; 2549 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1353} 2550}
1354 2551
2552eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
2553{
2554 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
2555}
2556
1355eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) 2557eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1356{ 2558{
1357 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; 2559 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1358} 2560}
1359 2561
… …
1365eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) 2567eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1366{ 2568{
1367 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2569 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1368} 2570}
1369 2571
1370eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2572eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1371{ 2573{
1372 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2574 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1373} 2575}
1374 2576
1375eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) 2577eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
… …
1395eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data) 2597eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1396{ 2598{
1397 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2599 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1398} 2600}
1399 2601
1400eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2602eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1401{ 2603{
1402 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2604 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1403} 2605}
1404 2606
1405eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) 2607eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
… …
1421eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data) 2623eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1422{ 2624{
1423 return eio__1path (EIO_READLINK, path, pri, cb, data); 2625 return eio__1path (EIO_READLINK, path, pri, cb, data);
1424} 2626}
1425 2627
2628eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2629{
2630 return eio__1path (EIO_REALPATH, path, pri, cb, data);
2631}
2632
1426eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) 2633eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1427{ 2634{
1428 return eio__1path (EIO_STAT, path, pri, cb, data); 2635 return eio__1path (EIO_STAT, path, pri, cb, data);
1429} 2636}
1430 2637
1431eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) 2638eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1432{ 2639{
1433 return eio__1path (EIO_LSTAT, path, pri, cb, data); 2640 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1434} 2641}
1435 2642
2643eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
2644{
2645 return eio__1path (EIO_STATVFS, path, pri, cb, data);
2646}
2647
1436eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) 2648eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1437{ 2649{
1438 return eio__1path (EIO_UNLINK, path, pri, cb, data); 2650 return eio__1path (EIO_UNLINK, path, pri, cb, data);
1439} 2651}
1440 2652
1441eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data) 2653eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1442{ 2654{
1443 return eio__1path (EIO_RMDIR, path, pri, cb, data); 2655 return eio__1path (EIO_RMDIR, path, pri, cb, data);
1444} 2656}
1445 2657
1446eio_req *eio_readdir (const char *path, int pri, eio_cb cb, void *data) 2658eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1447{ 2659{
1448 return eio__1path (EIO_READDIR, path, pri, cb, data); 2660 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1449} 2661}
1450 2662
1451eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) 2663eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1452{ 2664{
1453 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; 2665 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1454} 2666}
1455 2667
1456static eio_req * 2668static eio_req *
1457eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2669eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1458{ 2670{
… …
1482eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2694eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1483{ 2695{
1484 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2696 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1485} 2697}
1486 2698
1487eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) 2699eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
1488{ 2700{
1489 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND; 2701 REQ (EIO_CUSTOM); req->feed = execute; SEND;
1490} 2702}
1491 2703
1492#endif 2704#endif
1493 2705
1494eio_req *eio_grp (eio_cb cb, void *data) 2706eio_req *eio_grp (eio_cb cb, void *data)
… …
1503#undef SEND 2715#undef SEND
1504 2716
1505/*****************************************************************************/ 2717/*****************************************************************************/
1506/* grp functions */ 2718/* grp functions */
1507 2719
2720void
1508void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2721eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1509{ 2722{
1510 grp->int2 = limit; 2723 grp->int2 = limit;
1511 grp->feed = feed; 2724 grp->feed = feed;
1512 2725
1513 grp_try_feed (grp); 2726 grp_try_feed (grp);
1514} 2727}
1515 2728
2729void
1516void eio_grp_limit (eio_req *grp, int limit) 2730eio_grp_limit (eio_req *grp, int limit)
1517{ 2731{
1518 grp->int2 = limit; 2732 grp->int2 = limit;
1519 2733
1520 grp_try_feed (grp); 2734 grp_try_feed (grp);
1521} 2735}
1522 2736
2737void
1523void eio_grp_add (eio_req *grp, eio_req *req) 2738eio_grp_add (eio_req *grp, eio_req *req)
1524{ 2739{
1525 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2740 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1526 2741
1527 grp->flags |= EIO_FLAG_GROUPADD; 2742 grp->flags |= EIO_FLAG_GROUPADD;
1528 2743
… …
1539} 2754}
1540 2755
1541/*****************************************************************************/ 2756/*****************************************************************************/
1542/* misc garbage */ 2757/* misc garbage */
1543 2758
2759eio_ssize_t
1544ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2760eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1545{ 2761{
1546 etp_worker wrk;
1547
1548 wrk.dbuf = 0;
1549
1550 eio__sendfile (ofd, ifd, offset, count, &wrk); 2762 return eio__sendfile (ofd, ifd, offset, count);
1551
1552 if (wrk.dbuf)
1553 free (wrk.dbuf);
1554} 2763}
1555 2764

Diff Legend

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