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

Comparing libeio/eio.c (file contents):
Revision 1.30 by root, Wed Jun 3 12:24:49 2009 UTC vs.
Revision 1.109 by root, Tue Sep 27 02:16:49 2011 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 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"
41 46
42#ifdef EIO_STACKSIZE 47#ifdef EIO_STACKSIZE
43# define XTHREAD_STACKSIZE EIO_STACKSIZE 48# define X_STACKSIZE EIO_STACKSIZE
44#endif 49#endif
45#include "xthread.h" 50#include "xthread.h"
46 51
47#include <errno.h> 52#include <errno.h>
48#include <stddef.h> 53#include <stddef.h>
… …
53#include <sys/stat.h> 58#include <sys/stat.h>
54#include <limits.h> 59#include <limits.h>
55#include <fcntl.h> 60#include <fcntl.h>
56#include <assert.h> 61#include <assert.h>
57 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
58#ifndef EIO_FINISH 82#ifndef EIO_FINISH
59# 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
60#endif 84#endif
61 85
62#ifndef EIO_DESTROY 86#ifndef EIO_DESTROY
… …
65 89
66#ifndef EIO_FEED 90#ifndef EIO_FEED
67# 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)
68#endif 92#endif
69 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
70#ifdef _WIN32 105#ifdef _WIN32
71 106
72 /*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)
73 196
74#else 197#else
75 198
76# include "config.h"
77# include <sys/time.h> 199 #include <sys/time.h>
78# include <sys/select.h> 200 #include <sys/select.h>
201 #include <sys/statvfs.h>
79# 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__ || defined __NetBSD__ || defined __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
80# include <utime.h> 237# include <utime.h>
81# include <signal.h>
82# include <dirent.h>
83
84# ifndef EIO_STRUCT_DIRENT
85# define EIO_STRUCT_DIRENT struct dirent
86# endif 238#endif
87 239
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>
88#endif 246#endif
89 247
90#if HAVE_SENDFILE 248#if HAVE_SENDFILE
91# if __linux 249# if __linux
92# include <sys/sendfile.h> 250# include <sys/sendfile.h>
93# elif __freebsd 251# elif __FreeBSD__ || defined __APPLE__
94# include <sys/socket.h> 252# include <sys/socket.h>
95# include <sys/uio.h> 253# include <sys/uio.h>
96# elif __hpux 254# elif __hpux
97# include <sys/socket.h> 255# include <sys/socket.h>
98# elif __solaris /* not yet */ 256# elif __solaris
99# include <sys/sendfile.h> 257# include <sys/sendfile.h>
100# else 258# else
101# error sendfile support requested but not available 259# error sendfile support requested but not available
102# endif 260# endif
103#endif 261#endif
104 262
105/* number of seconds after which an idle threads exit */ 263#ifndef D_TYPE
106#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
107 272
108/* used for struct dirent, AIX doesn't provide it */ 273/* used for struct dirent, AIX doesn't provide it */
109#ifndef NAME_MAX 274#ifndef NAME_MAX
110# define NAME_MAX 4096 275# define NAME_MAX 4096
111#endif 276#endif
112 277
278/* used for readlink etc. */
279#ifndef PATH_MAX
280# define PATH_MAX 4096
281#endif
282
113/* buffer size for various temporary buffers */ 283/* buffer size for various temporary buffers */
114#define EIO_BUFSIZE 65536 284#define EIO_BUFSIZE 65536
115 285
116#define dBUF \ 286#define dBUF \
117 char *eio_buf; \
118 ETP_WORKER_LOCK (self); \
119 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 287 char *eio_buf = malloc (EIO_BUFSIZE); \
120 ETP_WORKER_UNLOCK (self); \
121 errno = ENOMEM; \ 288 errno = ENOMEM; \
122 if (!eio_buf) \ 289 if (!eio_buf) \
123 return -1; 290 return -1
291
292#define FUBd \
293 free (eio_buf)
124 294
125#define EIO_TICKS ((1000000 + 1023) >> 10) 295#define EIO_TICKS ((1000000 + 1023) >> 10)
126 296
127/*****************************************************************************/ 297/*****************************************************************************/
128 298
129#if __GNUC__ >= 3 299struct tmpbuf
130# 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
131#else 325#else
132# 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);
133#endif 328#endif
134 329
135#define expect_false(expr) expect ((expr) != 0, 0) 330struct eio_pwd
136#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};
137 338
138/*****************************************************************************/ 339/*****************************************************************************/
139 340
140#define ETP_PRI_MIN EIO_PRI_MIN 341#define ETP_PRI_MIN EIO_PRI_MIN
141#define ETP_PRI_MAX EIO_PRI_MAX 342#define ETP_PRI_MAX EIO_PRI_MAX
… …
147static int eio_finish (eio_req *req); 348static int eio_finish (eio_req *req);
148#define ETP_FINISH(req) eio_finish (req) 349#define ETP_FINISH(req) eio_finish (req)
149static void eio_execute (struct etp_worker *self, eio_req *req); 350static void eio_execute (struct etp_worker *self, eio_req *req);
150#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) 351#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
151 352
152#define ETP_WORKER_CLEAR(req) \
153 if (wrk->dbuf) \
154 { \
155 free (wrk->dbuf); \
156 wrk->dbuf = 0; \
157 } \
158 \
159 if (wrk->dirp) \
160 { \
161 closedir (wrk->dirp); \
162 wrk->dirp = 0; \
163 }
164#define ETP_WORKER_COMMON \
165 void *dbuf; \
166 DIR *dirp;
167
168/*****************************************************************************/ 353/*****************************************************************************/
169 354
170#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) 355#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
171 356
172/* calculcate time difference in ~1/EIO_TICKS of a second */ 357/* calculate time difference in ~1/EIO_TICKS of a second */
358ecb_inline int
173static int tvdiff (struct timeval *tv1, struct timeval *tv2) 359tvdiff (struct timeval *tv1, struct timeval *tv2)
174{ 360{
175 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 361 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
176 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 362 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
177} 363}
178 364
… …
182static void (*done_poll_cb) (void); 368static void (*done_poll_cb) (void);
183 369
184static unsigned int max_poll_time; /* reslock */ 370static unsigned int max_poll_time; /* reslock */
185static unsigned int max_poll_reqs; /* reslock */ 371static unsigned int max_poll_reqs; /* reslock */
186 372
187static volatile unsigned int nreqs; /* reqlock */ 373static unsigned int nreqs; /* reqlock */
188static volatile unsigned int nready; /* reqlock */ 374static unsigned int nready; /* reqlock */
189static volatile unsigned int npending; /* reqlock */ 375static unsigned int npending; /* reqlock */
190static 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 */
191 378
192static mutex_t wrklock = X_MUTEX_INIT; 379static xmutex_t wrklock;
193static mutex_t reslock = X_MUTEX_INIT; 380static xmutex_t reslock;
194static mutex_t reqlock = X_MUTEX_INIT; 381static xmutex_t reqlock;
195static cond_t reqwait = X_COND_INIT; 382static xcond_t reqwait;
196 383
197#if !HAVE_PREADWRITE 384#if !HAVE_PREADWRITE
198/* 385/*
199 * make our pread/pwrite emulation safe against themselves, but not against 386 * make our pread/pwrite emulation safe against themselves, but not against
200 * normal read/write by using a mutex. slows down execution a lot, 387 * normal read/write by using a mutex. slows down execution a lot,
201 * but that's your problem, not mine. 388 * but that's your problem, not mine.
202 */ 389 */
203static mutex_t preadwritelock = X_MUTEX_INIT; 390static xmutex_t preadwritelock;
204#endif 391#endif
205 392
206typedef struct etp_worker 393typedef struct etp_worker
207{ 394{
395 struct tmpbuf tmpbuf;
396
208 /* locked by wrklock */ 397 /* locked by wrklock */
209 struct etp_worker *prev, *next; 398 struct etp_worker *prev, *next;
210 399
211 thread_t tid; 400 xthread_t tid;
212 401
213 /* locked by reslock, reqlock or wrklock */ 402#ifdef ETP_WORKER_COMMON
214 ETP_REQ *req; /* currently processed request */
215
216 ETP_WORKER_COMMON 403 ETP_WORKER_COMMON
404#endif
217} etp_worker; 405} etp_worker;
218 406
219static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */ 407static etp_worker wrk_first; /* NOT etp */
220 408
221#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock) 409#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
222#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock) 410#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
223 411
224/* worker threads management */ 412/* worker threads management */
225 413
414static void
226static void etp_worker_clear (etp_worker *wrk) 415etp_worker_clear (etp_worker *wrk)
227{ 416{
228 ETP_WORKER_CLEAR (wrk);
229} 417}
230 418
419static void ecb_cold
231static void etp_worker_free (etp_worker *wrk) 420etp_worker_free (etp_worker *wrk)
232{ 421{
422 free (wrk->tmpbuf.ptr);
423
233 wrk->next->prev = wrk->prev; 424 wrk->next->prev = wrk->prev;
234 wrk->prev->next = wrk->next; 425 wrk->prev->next = wrk->next;
235 426
236 free (wrk); 427 free (wrk);
237} 428}
238 429
239static unsigned int etp_nreqs (void) 430static unsigned int
431etp_nreqs (void)
240{ 432{
241 int retval; 433 int retval;
242 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 434 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
243 retval = nreqs; 435 retval = nreqs;
244 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 436 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
245 return retval; 437 return retval;
246} 438}
247 439
248static unsigned int etp_nready (void) 440static unsigned int
441etp_nready (void)
249{ 442{
250 unsigned int retval; 443 unsigned int retval;
251 444
252 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 445 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
253 retval = nready; 446 retval = nready;
254 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 447 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
255 448
256 return retval; 449 return retval;
257} 450}
258 451
259static unsigned int etp_npending (void) 452static unsigned int
453etp_npending (void)
260{ 454{
261 unsigned int retval; 455 unsigned int retval;
262 456
263 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 457 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
264 retval = npending; 458 retval = npending;
265 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 459 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
266 460
267 return retval; 461 return retval;
268} 462}
269 463
270static unsigned int etp_nthreads (void) 464static unsigned int
465etp_nthreads (void)
271{ 466{
272 unsigned int retval; 467 unsigned int retval;
273 468
274 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 469 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
275 retval = started; 470 retval = started;
… …
289} etp_reqq; 484} etp_reqq;
290 485
291static etp_reqq req_queue; 486static etp_reqq req_queue;
292static etp_reqq res_queue; 487static etp_reqq res_queue;
293 488
489static void ecb_noinline ecb_cold
490reqq_init (etp_reqq *q)
491{
492 int pri;
493
494 for (pri = 0; pri < ETP_NUM_PRI; ++pri)
495 q->qs[pri] = q->qe[pri] = 0;
496
497 q->size = 0;
498}
499
500static int ecb_noinline
294static int reqq_push (etp_reqq *q, ETP_REQ *req) 501reqq_push (etp_reqq *q, ETP_REQ *req)
295{ 502{
296 int pri = req->pri; 503 int pri = req->pri;
297 req->next = 0; 504 req->next = 0;
298 505
299 if (q->qe[pri]) 506 if (q->qe[pri])
… …
305 q->qe[pri] = q->qs[pri] = req; 512 q->qe[pri] = q->qs[pri] = req;
306 513
307 return q->size++; 514 return q->size++;
308} 515}
309 516
517static ETP_REQ * ecb_noinline
310static ETP_REQ *reqq_shift (etp_reqq *q) 518reqq_shift (etp_reqq *q)
311{ 519{
312 int pri; 520 int pri;
313 521
314 if (!q->size) 522 if (!q->size)
315 return 0; 523 return 0;
… …
330 } 538 }
331 539
332 abort (); 540 abort ();
333} 541}
334 542
335static void etp_atfork_prepare (void) 543static int ecb_cold
544etp_init (void (*want_poll)(void), void (*done_poll)(void))
336{ 545{
337 X_LOCK (wrklock); 546 X_MUTEX_CREATE (wrklock);
338 X_LOCK (reqlock); 547 X_MUTEX_CREATE (reslock);
339 X_LOCK (reslock); 548 X_MUTEX_CREATE (reqlock);
340#if !HAVE_PREADWRITE 549 X_COND_CREATE (reqwait);
341 X_LOCK (preadwritelock);
342#endif
343}
344 550
345static void etp_atfork_parent (void) 551 reqq_init (&req_queue);
346{ 552 reqq_init (&res_queue);
347#if !HAVE_PREADWRITE
348 X_UNLOCK (preadwritelock);
349#endif
350 X_UNLOCK (reslock);
351 X_UNLOCK (reqlock);
352 X_UNLOCK (wrklock);
353}
354 553
355static void etp_atfork_child (void) 554 wrk_first.next =
356{ 555 wrk_first.prev = &wrk_first;
357 ETP_REQ *prv;
358
359 while ((prv = reqq_shift (&req_queue)))
360 ETP_DESTROY (prv);
361
362 while ((prv = reqq_shift (&res_queue)))
363 ETP_DESTROY (prv);
364
365 while (wrk_first.next != &wrk_first)
366 {
367 etp_worker *wrk = wrk_first.next;
368
369 if (wrk->req)
370 ETP_DESTROY (wrk->req);
371
372 etp_worker_clear (wrk);
373 etp_worker_free (wrk);
374 }
375 556
376 started = 0; 557 started = 0;
377 idle = 0; 558 idle = 0;
378 nreqs = 0; 559 nreqs = 0;
379 nready = 0; 560 nready = 0;
380 npending = 0; 561 npending = 0;
381 562
382 etp_atfork_parent ();
383}
384
385static void
386etp_once_init (void)
387{
388 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
389}
390
391static int
392etp_init (void (*want_poll)(void), void (*done_poll)(void))
393{
394 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
395
396 pthread_once (&doinit, etp_once_init);
397
398 want_poll_cb = want_poll; 563 want_poll_cb = want_poll;
399 done_poll_cb = done_poll; 564 done_poll_cb = done_poll;
400 565
401 return 0; 566 return 0;
402} 567}
403 568
404X_THREAD_PROC (etp_proc); 569X_THREAD_PROC (etp_proc);
405 570
571static void ecb_cold
406static void etp_start_thread (void) 572etp_start_thread (void)
407{ 573{
408 etp_worker *wrk = calloc (1, sizeof (etp_worker)); 574 etp_worker *wrk = calloc (1, sizeof (etp_worker));
409 575
410 /*TODO*/ 576 /*TODO*/
411 assert (("unable to allocate worker thread data", wrk)); 577 assert (("unable to allocate worker thread data", wrk));
… …
424 free (wrk); 590 free (wrk);
425 591
426 X_UNLOCK (wrklock); 592 X_UNLOCK (wrklock);
427} 593}
428 594
595static void
429static void etp_maybe_start_thread (void) 596etp_maybe_start_thread (void)
430{ 597{
431 if (expect_true (etp_nthreads () >= wanted)) 598 if (ecb_expect_true (etp_nthreads () >= wanted))
432 return; 599 return;
433 600
434 /* todo: maybe use idle here, but might be less exact */ 601 /* todo: maybe use idle here, but might be less exact */
435 if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ())) 602 if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
436 return; 603 return;
437 604
438 etp_start_thread (); 605 etp_start_thread ();
439} 606}
440 607
608static void ecb_cold
441static void etp_end_thread (void) 609etp_end_thread (void)
442{ 610{
443 eio_req *req = calloc (1, sizeof (eio_req)); 611 eio_req *req = calloc (1, sizeof (eio_req)); /* will be freed by worker */
444 612
445 req->type = -1; 613 req->type = -1;
446 req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 614 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
447 615
448 X_LOCK (reqlock); 616 X_LOCK (reqlock);
… …
453 X_LOCK (wrklock); 621 X_LOCK (wrklock);
454 --started; 622 --started;
455 X_UNLOCK (wrklock); 623 X_UNLOCK (wrklock);
456} 624}
457 625
458static int etp_poll (void) 626static int
627etp_poll (void)
459{ 628{
460 unsigned int maxreqs; 629 unsigned int maxreqs;
461 unsigned int maxtime; 630 unsigned int maxtime;
462 struct timeval tv_start, tv_now; 631 struct timeval tv_start, tv_now;
463 632
… …
493 662
494 X_LOCK (reqlock); 663 X_LOCK (reqlock);
495 --nreqs; 664 --nreqs;
496 X_UNLOCK (reqlock); 665 X_UNLOCK (reqlock);
497 666
498 if (expect_false (req->type == EIO_GROUP && req->size)) 667 if (ecb_expect_false (req->type == EIO_GROUP && req->size))
499 { 668 {
500 req->int1 = 1; /* mark request as delayed */ 669 req->int1 = 1; /* mark request as delayed */
501 continue; 670 continue;
502 } 671 }
503 else 672 else
504 { 673 {
505 int res = ETP_FINISH (req); 674 int res = ETP_FINISH (req);
506 if (expect_false (res)) 675 if (ecb_expect_false (res))
507 return res; 676 return res;
508 } 677 }
509 678
510 if (expect_false (maxreqs && !--maxreqs)) 679 if (ecb_expect_false (maxreqs && !--maxreqs))
511 break; 680 break;
512 681
513 if (maxtime) 682 if (maxtime)
514 { 683 {
515 gettimeofday (&tv_now, 0); 684 gettimeofday (&tv_now, 0);
… …
521 690
522 errno = EAGAIN; 691 errno = EAGAIN;
523 return -1; 692 return -1;
524} 693}
525 694
695static void
526static void etp_cancel (ETP_REQ *req) 696etp_cancel (ETP_REQ *req)
527{ 697{
528 X_LOCK (wrklock); 698 req->cancelled = 1;
529 req->flags |= EIO_FLAG_CANCELLED;
530 X_UNLOCK (wrklock);
531 699
532 eio_grp_cancel (req); 700 eio_grp_cancel (req);
533} 701}
534 702
703static void
535static void etp_submit (ETP_REQ *req) 704etp_submit (ETP_REQ *req)
536{ 705{
537 req->pri -= ETP_PRI_MIN; 706 req->pri -= ETP_PRI_MIN;
538 707
539 if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN; 708 if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
540 if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 709 if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
541 710
542 if (expect_false (req->type == EIO_GROUP)) 711 if (ecb_expect_false (req->type == EIO_GROUP))
543 { 712 {
544 /* I hope this is worth it :/ */ 713 /* I hope this is worth it :/ */
545 X_LOCK (reqlock); 714 X_LOCK (reqlock);
546 ++nreqs; 715 ++nreqs;
547 X_UNLOCK (reqlock); 716 X_UNLOCK (reqlock);
… …
566 735
567 etp_maybe_start_thread (); 736 etp_maybe_start_thread ();
568 } 737 }
569} 738}
570 739
740static void ecb_cold
571static void etp_set_max_poll_time (double nseconds) 741etp_set_max_poll_time (double nseconds)
572{ 742{
573 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 743 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
574 max_poll_time = nseconds; 744 max_poll_time = nseconds * EIO_TICKS;
575 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 745 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
576} 746}
577 747
748static void ecb_cold
578static void etp_set_max_poll_reqs (unsigned int maxreqs) 749etp_set_max_poll_reqs (unsigned int maxreqs)
579{ 750{
580 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 751 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
581 max_poll_reqs = maxreqs; 752 max_poll_reqs = maxreqs;
582 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 753 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
583} 754}
584 755
756static void ecb_cold
585static void etp_set_max_idle (unsigned int nthreads) 757etp_set_max_idle (unsigned int nthreads)
586{ 758{
587 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 759 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
588 max_idle = nthreads <= 0 ? 1 : nthreads; 760 max_idle = nthreads;
589 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 761 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
590} 762}
591 763
764static void ecb_cold
765etp_set_idle_timeout (unsigned int seconds)
766{
767 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
768 idle_timeout = seconds;
769 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
770}
771
772static void ecb_cold
592static void etp_set_min_parallel (unsigned int nthreads) 773etp_set_min_parallel (unsigned int nthreads)
593{ 774{
594 if (wanted < nthreads) 775 if (wanted < nthreads)
595 wanted = nthreads; 776 wanted = nthreads;
596} 777}
597 778
779static void ecb_cold
598static void etp_set_max_parallel (unsigned int nthreads) 780etp_set_max_parallel (unsigned int nthreads)
599{ 781{
600 if (wanted > nthreads) 782 if (wanted > nthreads)
601 wanted = nthreads; 783 wanted = nthreads;
602 784
603 while (started > wanted) 785 while (started > wanted)
604 etp_end_thread (); 786 etp_end_thread ();
605} 787}
606 788
607/*****************************************************************************/ 789/*****************************************************************************/
608 790
791static void
609static void grp_try_feed (eio_req *grp) 792grp_try_feed (eio_req *grp)
610{ 793{
611 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 794 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
612 { 795 {
613 grp->flags &= ~EIO_FLAG_GROUPADD; 796 grp->flags &= ~EIO_FLAG_GROUPADD;
614 797
… …
621 break; 804 break;
622 } 805 }
623 } 806 }
624} 807}
625 808
809static int
626static int grp_dec (eio_req *grp) 810grp_dec (eio_req *grp)
627{ 811{
628 --grp->size; 812 --grp->size;
629 813
630 /* call feeder, if applicable */ 814 /* call feeder, if applicable */
631 grp_try_feed (grp); 815 grp_try_feed (grp);
… …
635 return eio_finish (grp); 819 return eio_finish (grp);
636 else 820 else
637 return 0; 821 return 0;
638} 822}
639 823
824static void
640void eio_destroy (eio_req *req) 825eio_destroy (eio_req *req)
641{ 826{
642 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); 827 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
643 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); 828 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
644 829
645 EIO_DESTROY (req); 830 EIO_DESTROY (req);
646} 831}
647 832
833static int
648static int eio_finish (eio_req *req) 834eio_finish (eio_req *req)
649{ 835{
650 int res = EIO_FINISH (req); 836 int res = EIO_FINISH (req);
651 837
652 if (req->grp) 838 if (req->grp)
653 { 839 {
… …
661 if (grp->grp_first == req) 847 if (grp->grp_first == req)
662 grp->grp_first = req->grp_next; 848 grp->grp_first = req->grp_next;
663 849
664 res2 = grp_dec (grp); 850 res2 = grp_dec (grp);
665 851
666 if (!res && res2) 852 if (!res)
667 res = res2; 853 res = res2;
668 } 854 }
669 855
670 eio_destroy (req); 856 eio_destroy (req);
671 857
672 return res; 858 return res;
673} 859}
674 860
861void
675void eio_grp_cancel (eio_req *grp) 862eio_grp_cancel (eio_req *grp)
676{ 863{
677 for (grp = grp->grp_first; grp; grp = grp->grp_next) 864 for (grp = grp->grp_first; grp; grp = grp->grp_next)
678 eio_cancel (grp); 865 eio_cancel (grp);
679} 866}
680 867
868void
681void eio_cancel (eio_req *req) 869eio_cancel (eio_req *req)
682{ 870{
683 etp_cancel (req); 871 etp_cancel (req);
684} 872}
685 873
874void
686void eio_submit (eio_req *req) 875eio_submit (eio_req *req)
687{ 876{
688 etp_submit (req); 877 etp_submit (req);
689} 878}
690 879
691unsigned int eio_nreqs (void) 880unsigned int
881eio_nreqs (void)
692{ 882{
693 return etp_nreqs (); 883 return etp_nreqs ();
694} 884}
695 885
696unsigned int eio_nready (void) 886unsigned int
887eio_nready (void)
697{ 888{
698 return etp_nready (); 889 return etp_nready ();
699} 890}
700 891
701unsigned int eio_npending (void) 892unsigned int
893eio_npending (void)
702{ 894{
703 return etp_npending (); 895 return etp_npending ();
704} 896}
705 897
706unsigned int eio_nthreads (void) 898unsigned int ecb_cold
899eio_nthreads (void)
707{ 900{
708 return etp_nthreads (); 901 return etp_nthreads ();
709} 902}
710 903
904void ecb_cold
711void eio_set_max_poll_time (double nseconds) 905eio_set_max_poll_time (double nseconds)
712{ 906{
713 etp_set_max_poll_time (nseconds); 907 etp_set_max_poll_time (nseconds);
714} 908}
715 909
910void ecb_cold
716void eio_set_max_poll_reqs (unsigned int maxreqs) 911eio_set_max_poll_reqs (unsigned int maxreqs)
717{ 912{
718 etp_set_max_poll_reqs (maxreqs); 913 etp_set_max_poll_reqs (maxreqs);
719} 914}
720 915
916void ecb_cold
721void eio_set_max_idle (unsigned int nthreads) 917eio_set_max_idle (unsigned int nthreads)
722{ 918{
723 etp_set_max_idle (nthreads); 919 etp_set_max_idle (nthreads);
724} 920}
725 921
922void ecb_cold
923eio_set_idle_timeout (unsigned int seconds)
924{
925 etp_set_idle_timeout (seconds);
926}
927
928void ecb_cold
726void eio_set_min_parallel (unsigned int nthreads) 929eio_set_min_parallel (unsigned int nthreads)
727{ 930{
728 etp_set_min_parallel (nthreads); 931 etp_set_min_parallel (nthreads);
729} 932}
730 933
934void ecb_cold
731void eio_set_max_parallel (unsigned int nthreads) 935eio_set_max_parallel (unsigned int nthreads)
732{ 936{
733 etp_set_max_parallel (nthreads); 937 etp_set_max_parallel (nthreads);
734} 938}
735 939
736int eio_poll (void) 940int eio_poll (void)
… …
745# undef pread 949# undef pread
746# undef pwrite 950# undef pwrite
747# define pread eio__pread 951# define pread eio__pread
748# define pwrite eio__pwrite 952# define pwrite eio__pwrite
749 953
750static ssize_t 954static eio_ssize_t
751eio__pread (int fd, void *buf, size_t count, off_t offset) 955eio__pread (int fd, void *buf, size_t count, off_t offset)
752{ 956{
753 ssize_t res; 957 eio_ssize_t res;
754 off_t ooffset; 958 off_t ooffset;
755 959
756 X_LOCK (preadwritelock); 960 X_LOCK (preadwritelock);
757 ooffset = lseek (fd, 0, SEEK_CUR); 961 ooffset = lseek (fd, 0, SEEK_CUR);
758 lseek (fd, offset, SEEK_SET); 962 lseek (fd, offset, SEEK_SET);
… …
761 X_UNLOCK (preadwritelock); 965 X_UNLOCK (preadwritelock);
762 966
763 return res; 967 return res;
764} 968}
765 969
766static ssize_t 970static eio_ssize_t
767eio__pwrite (int fd, void *buf, size_t count, off_t offset) 971eio__pwrite (int fd, void *buf, size_t count, off_t offset)
768{ 972{
769 ssize_t res; 973 eio_ssize_t res;
770 off_t ooffset; 974 off_t ooffset;
771 975
772 X_LOCK (preadwritelock); 976 X_LOCK (preadwritelock);
773 ooffset = lseek (fd, 0, SEEK_CUR); 977 ooffset = lseek (fd, 0, SEEK_CUR);
774 lseek (fd, offset, SEEK_SET); 978 lseek (fd, offset, SEEK_SET);
… …
778 982
779 return res; 983 return res;
780} 984}
781#endif 985#endif
782 986
783#ifndef HAVE_FUTIMES 987#ifndef HAVE_UTIMES
784 988
785# undef utimes 989# undef utimes
786# undef futimes
787# define utimes(path,times) eio__utimes (path, times) 990# define utimes(path,times) eio__utimes (path, times)
788# define futimes(fd,times) eio__futimes (fd, times)
789 991
790static int 992static int
791eio__utimes (const char *filename, const struct timeval times[2]) 993eio__utimes (const char *filename, const struct timeval times[2])
792{ 994{
793 if (times) 995 if (times)
… …
801 } 1003 }
802 else 1004 else
803 return utime (filename, 0); 1005 return utime (filename, 0);
804} 1006}
805 1007
1008#endif
1009
1010#ifndef HAVE_FUTIMES
1011
1012# undef futimes
1013# define futimes(fd,times) eio__futimes (fd, times)
1014
1015static int
806static int eio__futimes (int fd, const struct timeval tv[2]) 1016eio__futimes (int fd, const struct timeval tv[2])
807{ 1017{
808 errno = ENOSYS; 1018 errno = ENOSYS;
809 return -1; 1019 return -1;
810} 1020}
811 1021
… …
814#if !HAVE_FDATASYNC 1024#if !HAVE_FDATASYNC
815# undef fdatasync 1025# undef fdatasync
816# define fdatasync(fd) fsync (fd) 1026# define fdatasync(fd) fsync (fd)
817#endif 1027#endif
818 1028
1029static int
1030eio__syncfs (int fd)
1031{
1032 int res;
1033
1034#if HAVE_SYS_SYNCFS
1035 res = (int)syscall (__NR_syncfs, (int)(fd));
1036#else
1037 res = -1;
1038 errno = ENOSYS;
1039#endif
1040
1041 if (res < 0 && errno == ENOSYS && fd >= 0)
1042 sync ();
1043
1044 return res;
1045}
1046
819/* sync_file_range always needs emulation */ 1047/* sync_file_range always needs emulation */
820int 1048static int
821eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags) 1049eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
822{ 1050{
823#if HAVE_SYNC_FILE_RANGE 1051#if HAVE_SYNC_FILE_RANGE
824 int res; 1052 int res;
825 1053
… …
838 if (!res || errno != ENOSYS) 1066 if (!res || errno != ENOSYS)
839 return res; 1067 return res;
840#endif 1068#endif
841 1069
842 /* even though we could play tricks with the flags, it's better to always 1070 /* even though we could play tricks with the flags, it's better to always
843 * call fdatasync, as thta matches the expectation of it's users best */ 1071 * call fdatasync, as that matches the expectation of its users best */
844 return fdatasync (fd); 1072 return fdatasync (fd);
1073}
1074
1075static int
1076eio__fallocate (int fd, int mode, off_t offset, size_t len)
1077{
1078#if HAVE_FALLOCATE
1079 return fallocate (fd, mode, offset, len);
1080#else
1081 errno = ENOSYS;
1082 return -1;
1083#endif
845} 1084}
846 1085
847#if !HAVE_READAHEAD 1086#if !HAVE_READAHEAD
848# undef readahead 1087# undef readahead
849# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 1088# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
850 1089
851static ssize_t 1090static eio_ssize_t
852eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) 1091eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
853{ 1092{
854 size_t todo = count; 1093 size_t todo = count;
855 dBUF; 1094 dBUF;
856 1095
… …
861 pread (fd, eio_buf, len, offset); 1100 pread (fd, eio_buf, len, offset);
862 offset += len; 1101 offset += len;
863 todo -= len; 1102 todo -= len;
864 } 1103 }
865 1104
1105 FUBd;
1106
866 errno = 0; 1107 errno = 0;
867 return count; 1108 return count;
868} 1109}
869 1110
870#endif 1111#endif
871 1112
872/* sendfile always needs emulation */ 1113/* sendfile always needs emulation */
873static ssize_t 1114static eio_ssize_t
874eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) 1115eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
875{ 1116{
1117 eio_ssize_t written = 0;
876 ssize_t res; 1118 eio_ssize_t res;
877 1119
878 if (!count) 1120 if (!count)
879 return 0; 1121 return 0;
880 1122
1123 for (;;)
1124 {
1125#ifdef __APPLE__
1126# undef HAVE_SENDFILE /* broken, as everything on os x */
1127#endif
881#if HAVE_SENDFILE 1128#if HAVE_SENDFILE
882# if __linux 1129# if __linux
1130 off_t soffset = offset;
883 res = sendfile (ofd, ifd, &offset, count); 1131 res = sendfile (ofd, ifd, &soffset, count);
884 1132
885# elif __freebsd 1133# elif __FreeBSD__
886 /* 1134 /*
887 * Of course, the freebsd sendfile is a dire hack with no thoughts 1135 * Of course, the freebsd sendfile is a dire hack with no thoughts
888 * wasted on making it similar to other I/O functions. 1136 * wasted on making it similar to other I/O functions.
889 */ 1137 */
890 {
891 off_t sbytes; 1138 off_t sbytes;
892 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 1139 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
893 1140
894 if (res < 0 && sbytes) 1141 #if 0 /* according to the manpage, this is correct, but broken behaviour */
895 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 1142 /* freebsd' sendfile will return 0 on success */
1143 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
1144 /* not on e.g. EIO or EPIPE - sounds broken */
1145 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
896 res = sbytes; 1146 res = sbytes;
897 } 1147 #endif
1148
1149 /* according to source inspection, this is correct, and useful behaviour */
1150 if (sbytes)
1151 res = sbytes;
1152
1153# elif defined (__APPLE__)
1154 off_t sbytes = count;
1155 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
1156
1157 /* according to the manpage, sbytes is always valid */
1158 if (sbytes)
1159 res = sbytes;
898 1160
899# elif __hpux 1161# elif __hpux
900 res = sendfile (ofd, ifd, offset, count, 0, 0); 1162 res = sendfile (ofd, ifd, offset, count, 0, 0);
901 1163
902# elif __solaris 1164# elif __solaris
903 {
904 struct sendfilevec vec; 1165 struct sendfilevec vec;
905 size_t sbytes; 1166 size_t sbytes;
906 1167
907 vec.sfv_fd = ifd; 1168 vec.sfv_fd = ifd;
908 vec.sfv_flag = 0; 1169 vec.sfv_flag = 0;
909 vec.sfv_off = offset; 1170 vec.sfv_off = offset;
910 vec.sfv_len = count; 1171 vec.sfv_len = count;
911 1172
912 res = sendfilev (ofd, &vec, 1, &sbytes); 1173 res = sendfilev (ofd, &vec, 1, &sbytes);
913 1174
914 if (res < 0 && sbytes) 1175 if (res < 0 && sbytes)
915 res = sbytes; 1176 res = sbytes;
916 }
917 1177
918# endif 1178# endif
1179
1180#elif defined (_WIN32) && 0
1181 /* does not work, just for documentation of what would need to be done */
1182 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
1183 /* libeio guarantees that the file offset does not change, and windows */
1184 /* has no way to get an independent handle to the same file description */
1185 HANDLE h = TO_SOCKET (ifd);
1186 SetFilePointer (h, offset, 0, FILE_BEGIN);
1187 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1188
919#else 1189#else
920 res = -1; 1190 res = -1;
921 errno = ENOSYS; 1191 errno = ENOSYS;
922#endif 1192#endif
923 1193
1194 /* we assume sendfile can copy at least 128mb in one go */
1195 if (res <= 128 * 1024 * 1024)
1196 {
1197 if (res > 0)
1198 written += res;
1199
1200 if (written)
1201 return written;
1202
1203 break;
1204 }
1205 else
1206 {
1207 /* if we requested more, then probably the kernel was lazy */
1208 written += res;
1209 offset += res;
1210 count -= res;
1211
1212 if (!count)
1213 return written;
1214 }
1215 }
1216
924 if (res < 0 1217 if (res < 0
925 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1218 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1219 /* BSDs */
1220#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1221 || errno == ENOTSUP
1222#endif
1223#ifdef EOPNOTSUPP /* windows */
1224 || errno == EOPNOTSUPP /* BSDs */
1225#endif
926#if __solaris 1226#if __solaris
927 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1227 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
928#endif 1228#endif
929 ) 1229 )
930 ) 1230 )
… …
934 1234
935 res = 0; 1235 res = 0;
936 1236
937 while (count) 1237 while (count)
938 { 1238 {
939 ssize_t cnt; 1239 eio_ssize_t cnt;
940 1240
941 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 1241 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
942 1242
943 if (cnt <= 0) 1243 if (cnt <= 0)
944 { 1244 {
… …
956 1256
957 offset += cnt; 1257 offset += cnt;
958 res += cnt; 1258 res += cnt;
959 count -= cnt; 1259 count -= cnt;
960 } 1260 }
1261
1262 FUBd;
961 } 1263 }
962 1264
963 return res; 1265 return res;
1266}
1267
1268#ifdef PAGESIZE
1269# define eio_pagesize() PAGESIZE
1270#else
1271static intptr_t
1272eio_pagesize (void)
1273{
1274 static intptr_t page;
1275
1276 if (!page)
1277 page = sysconf (_SC_PAGESIZE);
1278
1279 return page;
1280}
1281#endif
1282
1283static void
1284eio_page_align (void **addr, size_t *length)
1285{
1286 intptr_t mask = eio_pagesize () - 1;
1287
1288 /* round down addr */
1289 intptr_t adj = mask & (intptr_t)*addr;
1290
1291 *addr = (void *)((intptr_t)*addr - adj);
1292 *length += adj;
1293
1294 /* round up length */
1295 *length = (*length + mask) & ~mask;
1296}
1297
1298#if !_POSIX_MEMLOCK
1299# define eio__mlockall(a) EIO_ENOSYS ()
1300#else
1301
1302static int
1303eio__mlockall (int flags)
1304{
1305 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1306 extern int mallopt (int, int);
1307 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1308 #endif
1309
1310 if (EIO_MCL_CURRENT != MCL_CURRENT
1311 || EIO_MCL_FUTURE != MCL_FUTURE)
1312 {
1313 flags = 0
1314 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1315 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1316 }
1317
1318 return mlockall (flags);
1319}
1320#endif
1321
1322#if !_POSIX_MEMLOCK_RANGE
1323# define eio__mlock(a,b) EIO_ENOSYS ()
1324#else
1325
1326static int
1327eio__mlock (void *addr, size_t length)
1328{
1329 eio_page_align (&addr, &length);
1330
1331 return mlock (addr, length);
1332}
1333
1334#endif
1335
1336#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1337# define eio__msync(a,b,c) EIO_ENOSYS ()
1338#else
1339
1340static int
1341eio__msync (void *mem, size_t len, int flags)
1342{
1343 eio_page_align (&mem, &len);
1344
1345 if (EIO_MS_ASYNC != MS_SYNC
1346 || EIO_MS_INVALIDATE != MS_INVALIDATE
1347 || EIO_MS_SYNC != MS_SYNC)
1348 {
1349 flags = 0
1350 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1351 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1352 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1353 }
1354
1355 return msync (mem, len, flags);
1356}
1357
1358#endif
1359
1360static int
1361eio__mtouch (eio_req *req)
1362{
1363 void *mem = req->ptr2;
1364 size_t len = req->size;
1365 int flags = req->int1;
1366
1367 eio_page_align (&mem, &len);
1368
1369 {
1370 intptr_t addr = (intptr_t)mem;
1371 intptr_t end = addr + len;
1372 intptr_t page = eio_pagesize ();
1373
1374 if (addr < end)
1375 if (flags & EIO_MT_MODIFY) /* modify */
1376 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1377 else
1378 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1379 }
1380
1381 return 0;
1382}
1383
1384/*****************************************************************************/
1385/* requests implemented outside eio_execute, because they are so large */
1386
1387/* result will always end up in tmpbuf, there is always space for adding a 0-byte */
1388static int
1389eio__realpath (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
1390{
1391 const char *rel = path;
1392 char *res;
1393 char *tmp1, *tmp2;
1394#if SYMLOOP_MAX > 32
1395 int symlinks = SYMLOOP_MAX;
1396#else
1397 int symlinks = 32;
1398#endif
1399
1400 errno = EINVAL;
1401 if (!rel)
1402 return -1;
1403
1404 errno = ENOENT;
1405 if (!*rel)
1406 return -1;
1407
1408 res = tmpbuf_get (tmpbuf, PATH_MAX * 3);
1409 tmp1 = res + PATH_MAX;
1410 tmp2 = tmp1 + PATH_MAX;
1411
1412#if 0 /* disabled, the musl way to do things is just too racy */
1413#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1414 /* on linux we may be able to ask the kernel */
1415 {
1416 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1417
1418 if (fd >= 0)
1419 {
1420 sprintf (tmp1, "/proc/self/fd/%d", fd);
1421 req->result = readlink (tmp1, res, PATH_MAX);
1422 close (fd);
1423
1424 /* here we should probably stat the open file and the disk file, to make sure they still match */
1425
1426 if (req->result > 0)
1427 goto done;
1428 }
1429 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1430 return;
1431 }
1432#endif
1433#endif
1434
1435 if (*rel != '/')
1436 {
1437 int len;
1438
1439 errno = ENOENT;
1440 if (wd == EIO_INVALID_WD)
1441 return -1;
1442
1443 if (wd == EIO_CWD)
1444 {
1445 if (!getcwd (res, PATH_MAX))
1446 return -1;
1447
1448 len = strlen (res);
1449 }
1450 else
1451 memcpy (res, wd->str, len = wd->len);
1452
1453 if (res [1]) /* only use if not / */
1454 res += len;
1455 }
1456
1457 while (*rel)
1458 {
1459 eio_ssize_t len, linklen;
1460 const char *beg = rel;
1461
1462 while (*rel && *rel != '/')
1463 ++rel;
1464
1465 len = rel - beg;
1466
1467 if (!len) /* skip slashes */
1468 {
1469 ++rel;
1470 continue;
1471 }
1472
1473 if (beg [0] == '.')
1474 {
1475 if (len == 1)
1476 continue; /* . - nop */
1477
1478 if (beg [1] == '.' && len == 2)
1479 {
1480 /* .. - back up one component, if possible */
1481
1482 while (res != tmpbuf->ptr)
1483 if (*--res == '/')
1484 break;
1485
1486 continue;
1487 }
1488 }
1489
1490 errno = ENAMETOOLONG;
1491 if (res + 1 + len + 1 >= tmp1)
1492 return;
1493
1494 /* copy one component */
1495 *res = '/';
1496 memcpy (res + 1, beg, len);
1497
1498 /* zero-terminate, for readlink */
1499 res [len + 1] = 0;
1500
1501 /* now check if it's a symlink */
1502 linklen = readlink (tmpbuf->ptr, tmp1, PATH_MAX);
1503
1504 if (linklen < 0)
1505 {
1506 if (errno != EINVAL)
1507 return -1;
1508
1509 /* it's a normal directory. hopefully */
1510 res += len + 1;
1511 }
1512 else
1513 {
1514 /* yay, it was a symlink - build new path in tmp2 */
1515 int rellen = strlen (rel);
1516
1517 errno = ENAMETOOLONG;
1518 if (linklen + 1 + rellen >= PATH_MAX)
1519 return -1;
1520
1521 errno = ELOOP;
1522 if (!--symlinks)
1523 return -1;
1524
1525 if (*tmp1 == '/')
1526 res = tmpbuf->ptr; /* symlink resolves to an absolute path */
1527
1528 /* we need to be careful, as rel might point into tmp2 already */
1529 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1530 tmp2 [linklen] = '/';
1531 memcpy (tmp2, tmp1, linklen);
1532
1533 rel = tmp2;
1534 }
1535 }
1536
1537 /* special case for the lone root path */
1538 if (res == tmpbuf->ptr)
1539 *res++ = '/';
1540
1541 return res - (char *)tmpbuf->ptr;
1542}
1543
1544static signed char
1545eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1546{
1547 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1548 : a->inode < b->inode ? -1
1549 : a->inode > b->inode ? 1
1550 : 0;
1551}
1552
1553#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1554
1555#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1556#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1557
1558static void
1559eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1560{
1561 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1562 unsigned char *bit = bits;
1563
1564 assert (CHAR_BIT == 8);
1565 assert (sizeof (eio_dirent) * 8 < 256);
1566 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1567 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1568
1569 if (size <= EIO_SORT_FAST)
1570 return;
1571
1572 /* first prepare an array of bits to test in our radix sort */
1573 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1574 /* inode_bits must contain all inodes ORed together */
1575 /* which is used to skip bits that are 0 everywhere, which is very common */
1576 {
1577 eio_ino_t endianness;
1578 int i, j;
1579
1580 /* we store the byte offset of byte n into byte n of "endianness" */
1581 for (i = 0; i < sizeof (eio_ino_t); ++i)
1582 ((unsigned char *)&endianness)[i] = i;
1583
1584 *bit++ = 0;
1585
1586 for (i = 0; i < sizeof (eio_ino_t); ++i)
1587 {
1588 /* shifting off the byte offsets out of "endianness" */
1589 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1590 endianness >>= 8;
1591
1592 for (j = 0; j < 8; ++j)
1593 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1594 *bit++ = offs + j;
1595 }
1596
1597 for (j = 0; j < 8; ++j)
1598 if (score_bits & (1 << j))
1599 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1600 }
1601
1602 /* now actually do the sorting (a variant of MSD radix sort) */
1603 {
1604 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1605 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1606 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1607 int stk_idx = 0;
1608
1609 base_stk [stk_idx] = dents;
1610 end_stk [stk_idx] = dents + size;
1611 bit_stk [stk_idx] = bit - 1;
1612
1613 do
1614 {
1615 base = base_stk [stk_idx];
1616 end = end_stk [stk_idx];
1617 bit = bit_stk [stk_idx];
1618
1619 for (;;)
1620 {
1621 unsigned char O = *bit >> 3;
1622 unsigned char M = 1 << (*bit & 7);
1623
1624 eio_dirent *a = base;
1625 eio_dirent *b = end;
1626
1627 if (b - a < EIO_SORT_CUTOFF)
1628 break;
1629
1630 /* now bit-partition the array on the bit */
1631 /* this ugly asymmetric loop seems to perform much better than typical */
1632 /* partition algos found in the literature */
1633 do
1634 if (!(((unsigned char *)a)[O] & M))
1635 ++a;
1636 else if (!(((unsigned char *)--b)[O] & M))
1637 {
1638 eio_dirent tmp = *a; *a = *b; *b = tmp;
1639 ++a;
1640 }
1641 while (b > a);
1642
1643 /* next bit, or stop, if no bits left in this path */
1644 if (!*--bit)
1645 break;
1646
1647 base_stk [stk_idx] = a;
1648 end_stk [stk_idx] = end;
1649 bit_stk [stk_idx] = bit;
1650 ++stk_idx;
1651
1652 end = a;
1653 }
1654 }
1655 while (stk_idx--);
1656 }
1657}
1658
1659static void
1660eio_dent_insertion_sort (eio_dirent *dents, int size)
1661{
1662 /* first move the smallest element to the front, to act as a sentinel */
1663 {
1664 int i;
1665 eio_dirent *min = dents;
1666
1667 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1668 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1669 if (EIO_DENT_CMP (dents [i], <, *min))
1670 min = &dents [i];
1671
1672 /* swap elements 0 and j (minimum) */
1673 {
1674 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1675 }
1676 }
1677
1678 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1679 {
1680 eio_dirent *i, *j;
1681
1682 for (i = dents + 1; i < dents + size; ++i)
1683 {
1684 eio_dirent value = *i;
1685
1686 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1687 j [1] = j [0];
1688
1689 j [1] = value;
1690 }
1691 }
1692}
1693
1694static void
1695eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1696{
1697 if (size <= 1)
1698 return; /* our insertion sort relies on size > 0 */
1699
1700 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1701 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1702 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1703
1704 /* use an insertion sort at the end, or for small arrays, */
1705 /* as insertion sort is more efficient for small partitions */
1706 eio_dent_insertion_sort (dents, size);
964} 1707}
965 1708
966/* read a full directory */ 1709/* read a full directory */
967static void 1710static void
968eio__scandir (eio_req *req, etp_worker *self) 1711eio__scandir (eio_req *req, etp_worker *self)
969{ 1712{
1713 char *name, *names;
1714 int namesalloc = 4096 - sizeof (void *) * 4;
1715 int namesoffs = 0;
1716 int flags = req->int1;
1717 eio_dirent *dents = 0;
1718 int dentalloc = 128;
1719 int dentoffs = 0;
1720 eio_ino_t inode_bits = 0;
1721#ifdef _WIN32
1722 HANDLE dirp;
1723 WIN32_FIND_DATA entp;
1724#else
970 DIR *dirp; 1725 DIR *dirp;
971 EIO_STRUCT_DIRENT *entp; 1726 EIO_STRUCT_DIRENT *entp;
972 char *name, *names; 1727#endif
973 int memlen = 4096;
974 int memofs = 0;
975 int res = 0;
976 1728
977 X_LOCK (wrklock); 1729 req->result = -1;
978 /* the corresponding closedir is in ETP_WORKER_CLEAR */
979 self->dirp = dirp = opendir (req->ptr1);
980 req->flags |= EIO_FLAG_PTR2_FREE;
981 req->ptr2 = names = malloc (memlen);
982 X_UNLOCK (wrklock);
983 1730
984 if (dirp && names) 1731 if (!(flags & EIO_READDIR_DENTS))
985 for (;;) 1732 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1733
1734#ifdef _WIN32
1735 {
1736 int len = strlen ((const char *)req->ptr1);
1737 char *path = malloc (MAX_PATH);
1738 const char *fmt;
1739 const char *reqpath = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1740
1741 if (!len)
1742 fmt = "./*";
1743 else if (reqpath[len - 1] == '/' || reqpath[len - 1] == '\\')
1744 fmt = "%s*";
1745 else
1746 fmt = "%s/*";
1747
1748 _snprintf (path, MAX_PATH, fmt, reqpath);
1749 dirp = FindFirstFile (path, &entp);
1750 free (path);
1751
1752 if (dirp == INVALID_HANDLE_VALUE)
1753 {
1754 /* should steal _dosmaperr */
1755 switch (GetLastError ())
1756 {
1757 case ERROR_FILE_NOT_FOUND:
1758 req->result = 0;
1759 break;
1760
1761 case ERROR_INVALID_NAME:
1762 case ERROR_PATH_NOT_FOUND:
1763 case ERROR_NO_MORE_FILES:
1764 errno = ENOENT;
1765 break;
1766
1767 case ERROR_NOT_ENOUGH_MEMORY:
1768 errno = ENOMEM;
1769 break;
1770
1771 default:
1772 errno = EINVAL;
1773 break;
1774 }
1775
1776 return;
1777 }
1778 }
1779#else
1780 #if HAVE_AT
1781 if (req->wd)
986 { 1782 {
1783 int fd = openat (WD2FD (req->wd), req->ptr1, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1784
1785 if (fd < 0)
1786 return;
1787
1788 dirp = fdopendir (fd);
1789
1790 if (!dirp)
1791 close (fd);
1792 }
1793 else
1794 dirp = opendir (req->ptr1);
1795 #else
1796 dirp = opendir (wd_expand (&self->tmpbuf, req->wd, req->ptr1));
1797 #endif
1798
1799 if (!dirp)
1800 return;
1801#endif
1802
1803 if (req->flags & EIO_FLAG_PTR1_FREE)
1804 free (req->ptr1);
1805
1806 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1807 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1808 req->ptr2 = names = malloc (namesalloc);
1809
1810 if (!names || (flags && !dents))
1811 return;
1812
1813 for (;;)
1814 {
1815 int done;
1816
1817#ifdef _WIN32
1818 done = !dirp;
1819#else
987 errno = 0; 1820 errno = 0;
988 entp = readdir (dirp); 1821 entp = readdir (dirp);
1822 done = !entp;
1823#endif
989 1824
1825 if (done)
1826 {
1827#ifndef _WIN32
1828 int old_errno = errno;
1829 closedir (dirp);
1830 errno = old_errno;
1831
990 if (!entp) 1832 if (errno)
1833 break;
1834#endif
1835
1836 /* sort etc. */
1837 req->int1 = flags;
1838 req->result = dentoffs;
1839
1840 if (flags & EIO_READDIR_STAT_ORDER)
1841 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1842 else if (flags & EIO_READDIR_DIRS_FIRST)
1843 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1844 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1845 else
1846 {
1847 /* in this case, all is known, and we just put dirs first and sort them */
1848 eio_dirent *oth = dents + dentoffs;
1849 eio_dirent *dir = dents;
1850
1851 /* now partition dirs to the front, and non-dirs to the back */
1852 /* by walking from both sides and swapping if necessary */
1853 while (oth > dir)
1854 {
1855 if (dir->type == EIO_DT_DIR)
1856 ++dir;
1857 else if ((--oth)->type == EIO_DT_DIR)
1858 {
1859 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1860
1861 ++dir;
1862 }
1863 }
1864
1865 /* now sort the dirs only (dirs all have the same score) */
1866 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1867 }
1868
991 break; 1869 break;
1870 }
992 1871
993 name = entp->d_name; 1872 /* now add the entry to our list(s) */
1873 name = D_NAME (entp);
994 1874
1875 /* skip . and .. entries */
995 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1876 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
996 { 1877 {
997 int len = strlen (name) + 1; 1878 int len = D_NAMLEN (entp) + 1;
998 1879
999 res++; 1880 while (ecb_expect_false (namesoffs + len > namesalloc))
1000
1001 while (memofs + len > memlen)
1002 { 1881 {
1003 memlen *= 2; 1882 namesalloc *= 2;
1004 X_LOCK (wrklock);
1005 req->ptr2 = names = realloc (names, memlen); 1883 req->ptr2 = names = realloc (names, namesalloc);
1006 X_UNLOCK (wrklock);
1007 1884
1008 if (!names) 1885 if (!names)
1886 break;
1887 }
1888
1889 memcpy (names + namesoffs, name, len);
1890
1891 if (dents)
1892 {
1893 struct eio_dirent *ent;
1894
1895 if (ecb_expect_false (dentoffs == dentalloc))
1896 {
1897 dentalloc *= 2;
1898 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1899
1900 if (!dents)
1009 break; 1901 break;
1010 } 1902 }
1011 1903
1012 memcpy (names + memofs, name, len); 1904 ent = dents + dentoffs;
1013 memofs += len; 1905
1906 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1907 ent->namelen = len - 1;
1908 ent->inode = D_INO (entp);
1909
1910 inode_bits |= ent->inode;
1911
1912 switch (D_TYPE (entp))
1913 {
1914 default:
1915 ent->type = EIO_DT_UNKNOWN;
1916 flags |= EIO_READDIR_FOUND_UNKNOWN;
1917 break;
1918
1919 #ifdef DT_FIFO
1920 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1921 #endif
1922 #ifdef DT_CHR
1923 case DT_CHR: ent->type = EIO_DT_CHR; break;
1924 #endif
1925 #ifdef DT_MPC
1926 case DT_MPC: ent->type = EIO_DT_MPC; break;
1927 #endif
1928 #ifdef DT_DIR
1929 case DT_DIR: ent->type = EIO_DT_DIR; break;
1930 #endif
1931 #ifdef DT_NAM
1932 case DT_NAM: ent->type = EIO_DT_NAM; break;
1933 #endif
1934 #ifdef DT_BLK
1935 case DT_BLK: ent->type = EIO_DT_BLK; break;
1936 #endif
1937 #ifdef DT_MPB
1938 case DT_MPB: ent->type = EIO_DT_MPB; break;
1939 #endif
1940 #ifdef DT_REG
1941 case DT_REG: ent->type = EIO_DT_REG; break;
1942 #endif
1943 #ifdef DT_NWK
1944 case DT_NWK: ent->type = EIO_DT_NWK; break;
1945 #endif
1946 #ifdef DT_CMP
1947 case DT_CMP: ent->type = EIO_DT_CMP; break;
1948 #endif
1949 #ifdef DT_LNK
1950 case DT_LNK: ent->type = EIO_DT_LNK; break;
1951 #endif
1952 #ifdef DT_SOCK
1953 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1954 #endif
1955 #ifdef DT_DOOR
1956 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1957 #endif
1958 #ifdef DT_WHT
1959 case DT_WHT: ent->type = EIO_DT_WHT; break;
1960 #endif
1961 }
1962
1963 ent->score = 7;
1964
1965 if (flags & EIO_READDIR_DIRS_FIRST)
1966 {
1967 if (ent->type == EIO_DT_UNKNOWN)
1968 {
1969 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1970 ent->score = 1;
1971 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1972 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1973 }
1974 else if (ent->type == EIO_DT_DIR)
1975 ent->score = 0;
1976 }
1014 } 1977 }
1978
1979 namesoffs += len;
1980 ++dentoffs;
1015 } 1981 }
1016 1982
1017 if (errno) 1983 if (EIO_CANCELLED (req))
1018 res = -1; 1984 {
1985 errno = ECANCELED;
1986 break;
1987 }
1988
1989#ifdef _WIN32
1990 if (!FindNextFile (dirp, &entp))
1991 {
1992 FindClose (dirp);
1993 dirp = 0;
1994 }
1995#endif
1996 }
1997}
1998
1999/*****************************************************************************/
2000/* working directory stuff */
2001/* various deficiencies in the posix 2008 api force us to */
2002/* keep the absolute path in string form at all times */
2003/* fuck yeah. */
2004
2005#if !HAVE_AT
2006
2007/* a bit like realpath, but usually faster because it doesn'T have to return */
2008/* an absolute or canonical path */
2009static const char *
2010wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
2011{
2012 if (!wd || *path == '/')
2013 return path;
2014
2015 if (path [0] == '.' && !path [1])
2016 return wd->str;
2017
1019 2018 {
1020 req->result = res; 2019 int l1 = wd->len;
1021} 2020 int l2 = strlen (path);
1022 2021
1023#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) 2022 char *res = tmpbuf_get (tmpbuf, l1 + l2 + 2);
1024# undef msync
1025# define msync(a,b,c) ((errno = ENOSYS), -1)
1026#endif
1027 2023
1028int 2024 memcpy (res, wd->str, l1);
1029eio__mtouch (void *mem, size_t len, int flags) 2025 res [l1] = '/';
1030{ 2026 memcpy (res + l1 + 1, path, l2 + 1);
1031 intptr_t addr = (intptr_t)mem;
1032 intptr_t end = addr + len;
1033#ifdef PAGESIZE
1034 const intptr_t page = PAGESIZE;
1035#else
1036 static intptr_t page;
1037 2027
1038 if (!page) 2028 return res;
1039 page = sysconf (_SC_PAGESIZE); 2029 }
1040#endif 2030}
1041 2031
1042 addr &= ~(page - 1); /* assume page size is always a power of two */ 2032#endif
1043 2033
1044 if (addr < end) 2034static eio_wd
1045 if (flags) /* modify */ 2035eio__wd_open_sync (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
1046 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len); 2036{
1047 else 2037 int fd;
1048 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len); 2038 eio_wd res;
2039 int len = eio__realpath (tmpbuf, wd, path);
1049 2040
2041 if (len < 0)
2042 return EIO_INVALID_WD;
2043
2044#if HAVE_AT
2045 fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
2046
2047 if (fd < 0)
2048 return EIO_INVALID_WD;
2049#endif
2050
2051 res = malloc (sizeof (*res) + len); /* one extra 0-byte */
2052
2053#if HAVE_AT
2054 res->fd = fd;
2055#endif
2056
2057 res->len = len;
2058 memcpy (res->str, tmpbuf->ptr, len);
2059 res->str [len] = 0;
2060
2061 return res;
2062}
2063
2064eio_wd
2065eio_wd_open_sync (eio_wd wd, const char *path)
2066{
2067 struct tmpbuf tmpbuf = { 0 };
2068 wd = eio__wd_open_sync (&tmpbuf, wd, path);
2069 free (tmpbuf.ptr);
2070
1050 return 0; 2071 return wd;
1051} 2072}
2073
2074void
2075eio_wd_close_sync (eio_wd wd)
2076{
2077 if (wd != EIO_INVALID_WD && wd != EIO_CWD)
2078 {
2079 #if HAVE_AT
2080 close (wd->fd);
2081 #endif
2082 free (wd);
2083 }
2084}
2085
2086#if HAVE_AT
2087
2088/* they forgot these */
2089
2090static int
2091eio__truncateat (int dirfd, const char *path, off_t length)
2092{
2093 int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
2094 int res;
2095
2096 if (fd < 0)
2097 return fd;
2098
2099 res = ftruncate (fd, length);
2100 close (fd);
2101 return res;
2102}
2103
2104static int
2105eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
2106{
2107 int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
2108 int res;
2109
2110 if (fd < 0)
2111 return fd;
2112
2113 res = fstatvfs (fd, buf);
2114 close (fd);
2115 return res;
2116
2117}
2118
2119#endif
1052 2120
1053/*****************************************************************************/ 2121/*****************************************************************************/
1054 2122
1055#define ALLOC(len) \ 2123#define ALLOC(len) \
1056 if (!req->ptr2) \ 2124 if (!req->ptr2) \
… …
1071{ 2139{
1072 ETP_REQ *req; 2140 ETP_REQ *req;
1073 struct timespec ts; 2141 struct timespec ts;
1074 etp_worker *self = (etp_worker *)thr_arg; 2142 etp_worker *self = (etp_worker *)thr_arg;
1075 2143
2144#if HAVE_PRCTL_SET_NAME
2145 prctl (PR_SET_NAME, (unsigned long)"eio_thread", 0, 0, 0);
2146#endif
2147
1076 /* try to distribute timeouts somewhat randomly */ 2148 /* try to distribute timeouts somewhat evenly */
1077 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 2149 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1078 2150
1079 for (;;) 2151 for (;;)
1080 { 2152 {
2153 ts.tv_sec = 0;
2154
1081 X_LOCK (reqlock); 2155 X_LOCK (reqlock);
1082 2156
1083 for (;;) 2157 for (;;)
1084 { 2158 {
1085 self->req = req = reqq_shift (&req_queue); 2159 req = reqq_shift (&req_queue);
1086 2160
1087 if (req) 2161 if (req)
1088 break; 2162 break;
1089 2163
2164 if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
2165 {
2166 X_UNLOCK (reqlock);
2167 X_LOCK (wrklock);
2168 --started;
2169 X_UNLOCK (wrklock);
2170 goto quit;
2171 }
2172
1090 ++idle; 2173 ++idle;
1091 2174
1092 ts.tv_sec = time (0) + IDLE_TIMEOUT; 2175 if (idle <= max_idle)
1093 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 2176 /* we are allowed to idle, so do so without any timeout */
2177 X_COND_WAIT (reqwait, reqlock);
2178 else
1094 { 2179 {
1095 if (idle > max_idle) 2180 /* initialise timeout once */
1096 { 2181 if (!ts.tv_sec)
1097 --idle; 2182 ts.tv_sec = time (0) + idle_timeout;
1098 X_UNLOCK (reqlock);
1099 X_LOCK (wrklock);
1100 --started;
1101 X_UNLOCK (wrklock);
1102 goto quit;
1103 }
1104 2183
1105 /* we are allowed to idle, so do so without any timeout */
1106 X_COND_WAIT (reqwait, reqlock); 2184 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
2185 ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1107 } 2186 }
1108 2187
1109 --idle; 2188 --idle;
1110 } 2189 }
1111 2190
… …
1114 X_UNLOCK (reqlock); 2193 X_UNLOCK (reqlock);
1115 2194
1116 if (req->type < 0) 2195 if (req->type < 0)
1117 goto quit; 2196 goto quit;
1118 2197
1119 if (!EIO_CANCELLED (req))
1120 ETP_EXECUTE (self, req); 2198 ETP_EXECUTE (self, req);
1121 2199
1122 X_LOCK (reslock); 2200 X_LOCK (reslock);
1123 2201
1124 ++npending; 2202 ++npending;
1125 2203
1126 if (!reqq_push (&res_queue, req) && want_poll_cb) 2204 if (!reqq_push (&res_queue, req) && want_poll_cb)
1127 want_poll_cb (); 2205 want_poll_cb ();
1128 2206
1129 self->req = 0;
1130 etp_worker_clear (self); 2207 etp_worker_clear (self);
1131 2208
1132 X_UNLOCK (reslock); 2209 X_UNLOCK (reslock);
1133 } 2210 }
1134 2211
1135quit: 2212quit:
2213 free (req);
2214
1136 X_LOCK (wrklock); 2215 X_LOCK (wrklock);
1137 etp_worker_free (self); 2216 etp_worker_free (self);
1138 X_UNLOCK (wrklock); 2217 X_UNLOCK (wrklock);
1139 2218
1140 return 0; 2219 return 0;
1141} 2220}
1142 2221
1143/*****************************************************************************/ 2222/*****************************************************************************/
1144 2223
2224int ecb_cold
1145int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 2225eio_init (void (*want_poll)(void), void (*done_poll)(void))
1146{ 2226{
2227#if !HAVE_PREADWRITE
2228 X_MUTEX_CREATE (preadwritelock);
2229#endif
2230
1147 return etp_init (want_poll, done_poll); 2231 return etp_init (want_poll, done_poll);
1148} 2232}
1149 2233
2234ecb_inline void
1150static void eio_api_destroy (eio_req *req) 2235eio_api_destroy (eio_req *req)
1151{ 2236{
1152 free (req); 2237 free (req);
1153} 2238}
1154 2239
1155#define REQ(rtype) \ 2240#define REQ(rtype) \
… …
1174 { \ 2259 { \
1175 eio_api_destroy (req); \ 2260 eio_api_destroy (req); \
1176 return 0; \ 2261 return 0; \
1177 } 2262 }
1178 2263
2264static void
1179static void eio_execute (etp_worker *self, eio_req *req) 2265eio_execute (etp_worker *self, eio_req *req)
1180{ 2266{
1181 errno = 0; 2267#if HAVE_AT
2268 int dirfd;
2269#else
2270 const char *path;
2271#endif
2272
2273 if (ecb_expect_false (EIO_CANCELLED (req)))
2274 {
2275 req->result = -1;
2276 req->errorno = ECANCELED;
2277 return;
2278 }
2279
2280 if (ecb_expect_false (req->wd == EIO_INVALID_WD))
2281 {
2282 req->result = -1;
2283 req->errorno = ENOENT;
2284 return;
2285 }
2286
2287 if (req->type >= EIO_OPEN)
2288 {
2289 #if HAVE_AT
2290 dirfd = WD2FD (req->wd);
2291 #else
2292 path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
2293 #endif
2294 }
1182 2295
1183 switch (req->type) 2296 switch (req->type)
1184 { 2297 {
2298 case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
2299 req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
2300 break;
2301 case EIO_WD_CLOSE: req->result = 0;
2302 eio_wd_close_sync (req->wd); break;
2303
1185 case EIO_READ: ALLOC (req->size); 2304 case EIO_READ: ALLOC (req->size);
1186 req->result = req->offs >= 0 2305 req->result = req->offs >= 0
1187 ? pread (req->int1, req->ptr2, req->size, req->offs) 2306 ? pread (req->int1, req->ptr2, req->size, req->offs)
1188 : read (req->int1, req->ptr2, req->size); break; 2307 : read (req->int1, req->ptr2, req->size); break;
1189 case EIO_WRITE: req->result = req->offs >= 0 2308 case EIO_WRITE: req->result = req->offs >= 0
1190 ? pwrite (req->int1, req->ptr2, req->size, req->offs) 2309 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1191 : write (req->int1, req->ptr2, req->size); break; 2310 : write (req->int1, req->ptr2, req->size); break;
1192 2311
1193 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break; 2312 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1194 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break; 2313 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
2314
2315#if HAVE_AT
1195 2316
1196 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2317 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1197 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2318 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break;
1198 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2319 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1199 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2320 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, AT_SYMLINK_NOFOLLOW); break;
1200 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1201 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1202
1203 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; 2321 case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break;
1204 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1205 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break; 2322 case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break;
1206 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1207 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break; 2323 case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break;
1208 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1209
1210 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break; 2324 case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break;
1211 case EIO_CLOSE: req->result = close (req->int1); break; 2325
1212 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1213 case EIO_UNLINK: req->result = unlink (req->ptr1); break; 2326 case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
1214 case EIO_RMDIR: req->result = rmdir (req->ptr1); break; 2327 case EIO_RMDIR: req->result = unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break;
1215 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; 2328 case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); break;
1216 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; 2329 case EIO_RENAME: req->result = renameat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2); break;
1217 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; 2330 case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2, 0); break;
1218 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; 2331 case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
1219 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break; 2332 case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1220
1221 case EIO_READLINK: ALLOC (NAME_MAX); 2333 case EIO_READLINK: ALLOC (PATH_MAX);
1222 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; 2334 req->result = readlinkat (dirfd, req->ptr1, req->ptr2, PATH_MAX); break;
1223 2335 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1224 case EIO_SYNC: req->result = 0; sync (); break; 2336 req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1225 case EIO_FSYNC: req->result = fsync (req->int1); break;
1226 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1227 case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break;
1228 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;
1229 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1230
1231 case EIO_READDIR: eio__scandir (req, self); break;
1232
1233 case EIO_BUSY: 2337 case EIO_UTIME:
1234#ifdef _WIN32 2338 case EIO_FUTIME:
1235 Sleep (req->nv1 * 1000.); 2339 {
2340 struct timespec ts[2];
2341 struct timespec *times;
2342
2343 if (req->nv1 != -1. || req->nv2 != -1.)
2344 {
2345 ts[0].tv_sec = req->nv1;
2346 ts[0].tv_nsec = (req->nv1 - ts[0].tv_sec) * 1e9;
2347 ts[1].tv_sec = req->nv2;
2348 ts[1].tv_nsec = (req->nv2 - ts[1].tv_sec) * 1e9;
2349
2350 times = ts;
2351 }
2352 else
2353 times = 0;
2354
2355 req->result = req->type == EIO_FUTIME
2356 ? futimens (req->int1, times)
2357 : utimensat (dirfd, req->ptr1, times, 0);
2358 }
2359 break;
2360
1236#else 2361#else
1237 {
1238 struct timeval tv;
1239 2362
1240 tv.tv_sec = req->nv1; 2363 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1241 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; 2364 req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
2365 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2366 req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
2367 case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
2368 case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
2369 case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
2370 case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
1242 2371
1243 req->result = select (0, 0, 0, 0, &tv); 2372 case EIO_UNLINK: req->result = unlink (path ); break;
1244 } 2373 case EIO_RMDIR: req->result = rmdir (path ); break;
1245#endif 2374 case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
1246 break; 2375 case EIO_RENAME: req->result = rename (path , req->ptr2); break;
2376 case EIO_LINK: req->result = link (path , req->ptr2); break;
2377 case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
2378 case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
2379 case EIO_READLINK: ALLOC (PATH_MAX);
2380 req->result = readlink (path, req->ptr2, PATH_MAX); break;
2381 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2382 req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
1247 2383
1248 case EIO_UTIME: 2384 case EIO_UTIME:
1249 case EIO_FUTIME: 2385 case EIO_FUTIME:
1250 { 2386 {
1251 struct timeval tv[2]; 2387 struct timeval tv[2];
1252 struct timeval *times; 2388 struct timeval *times;
1253 2389
1254 if (req->nv1 != -1. || req->nv2 != -1.) 2390 if (req->nv1 != -1. || req->nv2 != -1.)
1255 { 2391 {
1256 tv[0].tv_sec = req->nv1; 2392 tv[0].tv_sec = req->nv1;
1257 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.; 2393 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
1258 tv[1].tv_sec = req->nv2; 2394 tv[1].tv_sec = req->nv2;
1259 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.; 2395 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
1260 2396
1261 times = tv; 2397 times = tv;
1262 } 2398 }
1263 else 2399 else
1264 times = 0; 2400 times = 0;
1265
1266 2401
1267 req->result = req->type == EIO_FUTIME 2402 req->result = req->type == EIO_FUTIME
1268 ? futimes (req->int1, times) 2403 ? futimes (req->int1, times)
1269 : utimes (req->ptr1, times); 2404 : utimes (req->ptr1, times);
1270 } 2405 }
1271 break; 2406 break;
1272 2407
2408#endif
2409
2410 case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
2411 {
2412 ALLOC (req->result);
2413 memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
2414 }
2415 break;
2416
2417 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2418 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
2419
2420 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2421 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2422
2423 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
2424 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
2425 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
2426
2427 case EIO_CLOSE: req->result = close (req->int1); break;
2428 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
2429 case EIO_SYNC: req->result = 0; sync (); break;
2430 case EIO_FSYNC: req->result = fsync (req->int1); break;
2431 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
2432 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
2433 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
2434 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
2435 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
2436 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
2437 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
2438 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
2439
2440 case EIO_READDIR: eio__scandir (req, self); break;
2441
2442 case EIO_BUSY:
2443#ifdef _WIN32
2444 Sleep (req->nv1 * 1e3);
2445#else
2446 {
2447 struct timeval tv;
2448
2449 tv.tv_sec = req->nv1;
2450 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
2451
2452 req->result = select (0, 0, 0, 0, &tv);
2453 }
2454#endif
2455 break;
2456
1273 case EIO_GROUP: 2457 case EIO_GROUP:
1274 abort (); /* handled in eio_request */ 2458 abort (); /* handled in eio_request */
1275 2459
1276 case EIO_NOP: 2460 case EIO_NOP:
1277 req->result = 0; 2461 req->result = 0;
1278 break; 2462 break;
1279 2463
1280 case EIO_CUSTOM: 2464 case EIO_CUSTOM:
1281 ((void (*)(eio_req *))req->feed) (req); 2465 req->feed (req);
1282 break; 2466 break;
1283 2467
1284 default: 2468 default:
2469 errno = ENOSYS;
1285 req->result = -1; 2470 req->result = -1;
1286 break; 2471 break;
1287 } 2472 }
1288 2473
1289 req->errorno = errno; 2474 req->errorno = errno;
… …
1314eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2499eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1315{ 2500{
1316 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2501 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1317} 2502}
1318 2503
2504eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2505{
2506 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2507}
2508
2509eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
2510{
2511 REQ (EIO_SYNCFS); req->int1 = fd; SEND;
2512}
2513
2514eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2515{
2516 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2517}
2518
1319eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2519eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1320{ 2520{
1321 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2521 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1322} 2522}
1323 2523
1324eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) 2524eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
1325{ 2525{
1326 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; 2526 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
1327} 2527}
1328 2528
1329eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) 2529eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
1330{ 2530{
1331 REQ (EIO_FDATASYNC); req->int1 = fd; SEND; 2531 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2532}
2533
2534eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2535{
2536 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
1332} 2537}
1333 2538
1334eio_req *eio_close (int fd, int pri, eio_cb cb, void *data) 2539eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
1335{ 2540{
1336 REQ (EIO_CLOSE); req->int1 = fd; SEND; 2541 REQ (EIO_CLOSE); req->int1 = fd; SEND;
… …
1354eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) 2559eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1355{ 2560{
1356 REQ (EIO_FSTAT); req->int1 = fd; SEND; 2561 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1357} 2562}
1358 2563
2564eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
2565{
2566 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
2567}
2568
1359eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) 2569eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1360{ 2570{
1361 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; 2571 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1362} 2572}
1363 2573
… …
1369eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) 2579eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1370{ 2580{
1371 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2581 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1372} 2582}
1373 2583
1374eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2584eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1375{ 2585{
1376 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2586 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1377} 2587}
1378 2588
1379eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) 2589eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
… …
1399eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data) 2609eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1400{ 2610{
1401 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2611 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1402} 2612}
1403 2613
1404eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2614eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1405{ 2615{
1406 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2616 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1407} 2617}
1408 2618
1409eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) 2619eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
… …
1425eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data) 2635eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1426{ 2636{
1427 return eio__1path (EIO_READLINK, path, pri, cb, data); 2637 return eio__1path (EIO_READLINK, path, pri, cb, data);
1428} 2638}
1429 2639
2640eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2641{
2642 return eio__1path (EIO_REALPATH, path, pri, cb, data);
2643}
2644
1430eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) 2645eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1431{ 2646{
1432 return eio__1path (EIO_STAT, path, pri, cb, data); 2647 return eio__1path (EIO_STAT, path, pri, cb, data);
1433} 2648}
1434 2649
1435eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) 2650eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1436{ 2651{
1437 return eio__1path (EIO_LSTAT, path, pri, cb, data); 2652 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1438} 2653}
1439 2654
2655eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
2656{
2657 return eio__1path (EIO_STATVFS, path, pri, cb, data);
2658}
2659
1440eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) 2660eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1441{ 2661{
1442 return eio__1path (EIO_UNLINK, path, pri, cb, data); 2662 return eio__1path (EIO_UNLINK, path, pri, cb, data);
1443} 2663}
1444 2664
1445eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data) 2665eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1446{ 2666{
1447 return eio__1path (EIO_RMDIR, path, pri, cb, data); 2667 return eio__1path (EIO_RMDIR, path, pri, cb, data);
1448} 2668}
1449 2669
1450eio_req *eio_readdir (const char *path, int pri, eio_cb cb, void *data) 2670eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1451{ 2671{
1452 return eio__1path (EIO_READDIR, path, pri, cb, data); 2672 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1453} 2673}
1454 2674
1455eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) 2675eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1456{ 2676{
1457 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; 2677 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1458} 2678}
1459 2679
1460static eio_req * 2680static eio_req *
1461eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2681eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1462{ 2682{
… …
1486eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2706eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1487{ 2707{
1488 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2708 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1489} 2709}
1490 2710
1491eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) 2711eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
1492{ 2712{
1493 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND; 2713 REQ (EIO_CUSTOM); req->feed = execute; SEND;
1494} 2714}
1495 2715
1496#endif 2716#endif
1497 2717
1498eio_req *eio_grp (eio_cb cb, void *data) 2718eio_req *eio_grp (eio_cb cb, void *data)
… …
1507#undef SEND 2727#undef SEND
1508 2728
1509/*****************************************************************************/ 2729/*****************************************************************************/
1510/* grp functions */ 2730/* grp functions */
1511 2731
2732void
1512void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2733eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1513{ 2734{
1514 grp->int2 = limit; 2735 grp->int2 = limit;
1515 grp->feed = feed; 2736 grp->feed = feed;
1516 2737
1517 grp_try_feed (grp); 2738 grp_try_feed (grp);
1518} 2739}
1519 2740
2741void
1520void eio_grp_limit (eio_req *grp, int limit) 2742eio_grp_limit (eio_req *grp, int limit)
1521{ 2743{
1522 grp->int2 = limit; 2744 grp->int2 = limit;
1523 2745
1524 grp_try_feed (grp); 2746 grp_try_feed (grp);
1525} 2747}
1526 2748
2749void
1527void eio_grp_add (eio_req *grp, eio_req *req) 2750eio_grp_add (eio_req *grp, eio_req *req)
1528{ 2751{
1529 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2752 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1530 2753
1531 grp->flags |= EIO_FLAG_GROUPADD; 2754 grp->flags |= EIO_FLAG_GROUPADD;
1532 2755
… …
1543} 2766}
1544 2767
1545/*****************************************************************************/ 2768/*****************************************************************************/
1546/* misc garbage */ 2769/* misc garbage */
1547 2770
2771eio_ssize_t
1548ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2772eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1549{ 2773{
1550 etp_worker wrk;
1551
1552 wrk.dbuf = 0;
1553
1554 eio__sendfile (ofd, ifd, offset, count, &wrk); 2774 return eio__sendfile (ofd, ifd, offset, count);
1555
1556 if (wrk.dbuf)
1557 free (wrk.dbuf);
1558} 2775}
1559 2776

Diff Legend

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