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 1.118 by root, Mon Apr 2 17:53:27 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;
192 383
193#if !HAVE_PREADWRITE 384#if !HAVE_PREADWRITE
194/* 385/*
195 * make our pread/pwrite emulation safe against themselves, but not against 386 * make our pread/pwrite emulation safe against themselves, but not against
196 * 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,
197 * but that's your problem, not mine. 388 * but that's your problem, not mine.
198 */ 389 */
199static mutex_t preadwritelock = X_MUTEX_INIT; 390static xmutex_t preadwritelock;
200#endif 391#endif
201 392
202typedef struct etp_worker 393typedef struct etp_worker
203{ 394{
395 struct tmpbuf tmpbuf;
396
204 /* locked by wrklock */ 397 /* locked by wrklock */
205 struct etp_worker *prev, *next; 398 struct etp_worker *prev, *next;
206 399
207 thread_t tid; 400 xthread_t tid;
208 401
209 /* locked by reslock, reqlock or wrklock */ 402#ifdef ETP_WORKER_COMMON
210 ETP_REQ *req; /* currently processed request */
211
212 ETP_WORKER_COMMON 403 ETP_WORKER_COMMON
404#endif
213} etp_worker; 405} etp_worker;
214 406
215static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */ 407static etp_worker wrk_first; /* NOT etp */
216 408
217#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock) 409#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
218#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock) 410#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
219 411
220/* worker threads management */ 412/* worker threads management */
221 413
414static void
222static void etp_worker_clear (etp_worker *wrk) 415etp_worker_clear (etp_worker *wrk)
223{ 416{
224 ETP_WORKER_CLEAR (wrk);
225} 417}
226 418
419static void ecb_cold
227static void etp_worker_free (etp_worker *wrk) 420etp_worker_free (etp_worker *wrk)
228{ 421{
422 free (wrk->tmpbuf.ptr);
423
229 wrk->next->prev = wrk->prev; 424 wrk->next->prev = wrk->prev;
230 wrk->prev->next = wrk->next; 425 wrk->prev->next = wrk->next;
231 426
232 free (wrk); 427 free (wrk);
233} 428}
234 429
235static unsigned int etp_nreqs (void) 430static unsigned int
431etp_nreqs (void)
236{ 432{
237 int retval; 433 int retval;
238 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 434 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
239 retval = nreqs; 435 retval = nreqs;
240 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 436 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
241 return retval; 437 return retval;
242} 438}
243 439
244static unsigned int etp_nready (void) 440static unsigned int
441etp_nready (void)
245{ 442{
246 unsigned int retval; 443 unsigned int retval;
247 444
248 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 445 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
249 retval = nready; 446 retval = nready;
250 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 447 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
251 448
252 return retval; 449 return retval;
253} 450}
254 451
255static unsigned int etp_npending (void) 452static unsigned int
453etp_npending (void)
256{ 454{
257 unsigned int retval; 455 unsigned int retval;
258 456
259 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 457 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
260 retval = npending; 458 retval = npending;
261 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 459 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
262 460
263 return retval; 461 return retval;
264} 462}
265 463
266static unsigned int etp_nthreads (void) 464static unsigned int
465etp_nthreads (void)
267{ 466{
268 unsigned int retval; 467 unsigned int retval;
269 468
270 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 469 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
271 retval = started; 470 retval = started;
… …
285} etp_reqq; 484} etp_reqq;
286 485
287static etp_reqq req_queue; 486static etp_reqq req_queue;
288static etp_reqq res_queue; 487static etp_reqq res_queue;
289 488
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
290static int reqq_push (etp_reqq *q, ETP_REQ *req) 501reqq_push (etp_reqq *q, ETP_REQ *req)
291{ 502{
292 int pri = req->pri; 503 int pri = req->pri;
293 req->next = 0; 504 req->next = 0;
294 505
295 if (q->qe[pri]) 506 if (q->qe[pri])
… …
301 q->qe[pri] = q->qs[pri] = req; 512 q->qe[pri] = q->qs[pri] = req;
302 513
303 return q->size++; 514 return q->size++;
304} 515}
305 516
517static ETP_REQ * ecb_noinline
306static ETP_REQ *reqq_shift (etp_reqq *q) 518reqq_shift (etp_reqq *q)
307{ 519{
308 int pri; 520 int pri;
309 521
310 if (!q->size) 522 if (!q->size)
311 return 0; 523 return 0;
… …
326 } 538 }
327 539
328 abort (); 540 abort ();
329} 541}
330 542
331static void etp_atfork_prepare (void) 543static int ecb_cold
544etp_init (void (*want_poll)(void), void (*done_poll)(void))
332{ 545{
333 X_LOCK (wrklock); 546 X_MUTEX_CREATE (wrklock);
334 X_LOCK (reqlock); 547 X_MUTEX_CREATE (reslock);
335 X_LOCK (reslock); 548 X_MUTEX_CREATE (reqlock);
336#if !HAVE_PREADWRITE 549 X_COND_CREATE (reqwait);
337 X_LOCK (preadwritelock);
338#endif
339}
340 550
341static void etp_atfork_parent (void) 551 reqq_init (&req_queue);
342{ 552 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}
350 553
351static void etp_atfork_child (void) 554 wrk_first.next =
352{ 555 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 }
371 556
372 started = 0; 557 started = 0;
373 idle = 0; 558 idle = 0;
374 nreqs = 0; 559 nreqs = 0;
375 nready = 0; 560 nready = 0;
376 npending = 0; 561 npending = 0;
377 562
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; 563 want_poll_cb = want_poll;
395 done_poll_cb = done_poll; 564 done_poll_cb = done_poll;
396 565
397 return 0; 566 return 0;
398} 567}
399 568
400X_THREAD_PROC (etp_proc); 569X_THREAD_PROC (etp_proc);
401 570
571static void ecb_cold
402static void etp_start_thread (void) 572etp_start_thread (void)
403{ 573{
404 etp_worker *wrk = calloc (1, sizeof (etp_worker)); 574 etp_worker *wrk = calloc (1, sizeof (etp_worker));
405 575
406 /*TODO*/ 576 /*TODO*/
407 assert (("unable to allocate worker thread data", wrk)); 577 assert (("unable to allocate worker thread data", wrk));
… …
420 free (wrk); 590 free (wrk);
421 591
422 X_UNLOCK (wrklock); 592 X_UNLOCK (wrklock);
423} 593}
424 594
595static void
425static void etp_maybe_start_thread (void) 596etp_maybe_start_thread (void)
426{ 597{
427 if (expect_true (etp_nthreads () >= wanted)) 598 if (ecb_expect_true (etp_nthreads () >= wanted))
428 return; 599 return;
429 600
430 /* todo: maybe use idle here, but might be less exact */ 601 /* todo: maybe use idle here, but might be less exact */
431 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 ()))
432 return; 603 return;
433 604
434 etp_start_thread (); 605 etp_start_thread ();
435} 606}
436 607
608static void ecb_cold
437static void etp_end_thread (void) 609etp_end_thread (void)
438{ 610{
439 eio_req *req = calloc (1, sizeof (eio_req)); 611 eio_req *req = calloc (1, sizeof (eio_req)); /* will be freed by worker */
440 612
441 req->type = -1; 613 req->type = -1;
442 req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 614 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
443 615
444 X_LOCK (reqlock); 616 X_LOCK (reqlock);
… …
449 X_LOCK (wrklock); 621 X_LOCK (wrklock);
450 --started; 622 --started;
451 X_UNLOCK (wrklock); 623 X_UNLOCK (wrklock);
452} 624}
453 625
454static int etp_poll (void) 626static int
627etp_poll (void)
455{ 628{
456 unsigned int maxreqs; 629 unsigned int maxreqs;
457 unsigned int maxtime; 630 unsigned int maxtime;
458 struct timeval tv_start, tv_now; 631 struct timeval tv_start, tv_now;
459 632
… …
489 662
490 X_LOCK (reqlock); 663 X_LOCK (reqlock);
491 --nreqs; 664 --nreqs;
492 X_UNLOCK (reqlock); 665 X_UNLOCK (reqlock);
493 666
494 if (expect_false (req->type == EIO_GROUP && req->size)) 667 if (ecb_expect_false (req->type == EIO_GROUP && req->size))
495 { 668 {
496 req->int1 = 1; /* mark request as delayed */ 669 req->int1 = 1; /* mark request as delayed */
497 continue; 670 continue;
498 } 671 }
499 else 672 else
500 { 673 {
501 int res = ETP_FINISH (req); 674 int res = ETP_FINISH (req);
502 if (expect_false (res)) 675 if (ecb_expect_false (res))
503 return res; 676 return res;
504 } 677 }
505 678
506 if (expect_false (maxreqs && !--maxreqs)) 679 if (ecb_expect_false (maxreqs && !--maxreqs))
507 break; 680 break;
508 681
509 if (maxtime) 682 if (maxtime)
510 { 683 {
511 gettimeofday (&tv_now, 0); 684 gettimeofday (&tv_now, 0);
… …
517 690
518 errno = EAGAIN; 691 errno = EAGAIN;
519 return -1; 692 return -1;
520} 693}
521 694
695static void
522static void etp_cancel (ETP_REQ *req) 696etp_cancel (ETP_REQ *req)
523{ 697{
524 X_LOCK (wrklock); 698 req->cancelled = 1;
525 req->flags |= EIO_FLAG_CANCELLED;
526 X_UNLOCK (wrklock);
527 699
528 eio_grp_cancel (req); 700 eio_grp_cancel (req);
529} 701}
530 702
703static void
531static void etp_submit (ETP_REQ *req) 704etp_submit (ETP_REQ *req)
532{ 705{
533 req->pri -= ETP_PRI_MIN; 706 req->pri -= ETP_PRI_MIN;
534 707
535 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;
536 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;
537 710
538 if (expect_false (req->type == EIO_GROUP)) 711 if (ecb_expect_false (req->type == EIO_GROUP))
539 { 712 {
540 /* I hope this is worth it :/ */ 713 /* I hope this is worth it :/ */
541 X_LOCK (reqlock); 714 X_LOCK (reqlock);
542 ++nreqs; 715 ++nreqs;
543 X_UNLOCK (reqlock); 716 X_UNLOCK (reqlock);
… …
562 735
563 etp_maybe_start_thread (); 736 etp_maybe_start_thread ();
564 } 737 }
565} 738}
566 739
740static void ecb_cold
567static void etp_set_max_poll_time (double nseconds) 741etp_set_max_poll_time (double nseconds)
568{ 742{
569 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 743 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
570 max_poll_time = nseconds; 744 max_poll_time = nseconds * EIO_TICKS;
571 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 745 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
572} 746}
573 747
748static void ecb_cold
574static void etp_set_max_poll_reqs (unsigned int maxreqs) 749etp_set_max_poll_reqs (unsigned int maxreqs)
575{ 750{
576 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 751 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
577 max_poll_reqs = maxreqs; 752 max_poll_reqs = maxreqs;
578 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 753 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
579} 754}
580 755
756static void ecb_cold
581static void etp_set_max_idle (unsigned int nthreads) 757etp_set_max_idle (unsigned int nthreads)
582{ 758{
583 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 759 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
584 max_idle = nthreads <= 0 ? 1 : nthreads; 760 max_idle = nthreads;
585 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 761 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
586} 762}
587 763
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
588static void etp_set_min_parallel (unsigned int nthreads) 773etp_set_min_parallel (unsigned int nthreads)
589{ 774{
590 if (wanted < nthreads) 775 if (wanted < nthreads)
591 wanted = nthreads; 776 wanted = nthreads;
592} 777}
593 778
779static void ecb_cold
594static void etp_set_max_parallel (unsigned int nthreads) 780etp_set_max_parallel (unsigned int nthreads)
595{ 781{
596 if (wanted > nthreads) 782 if (wanted > nthreads)
597 wanted = nthreads; 783 wanted = nthreads;
598 784
599 while (started > wanted) 785 while (started > wanted)
600 etp_end_thread (); 786 etp_end_thread ();
601} 787}
602 788
603/*****************************************************************************/ 789/*****************************************************************************/
604 790
791static void
605static void grp_try_feed (eio_req *grp) 792grp_try_feed (eio_req *grp)
606{ 793{
607 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 794 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
608 { 795 {
609 grp->flags &= ~EIO_FLAG_GROUPADD; 796 grp->flags &= ~EIO_FLAG_GROUPADD;
610 797
… …
617 break; 804 break;
618 } 805 }
619 } 806 }
620} 807}
621 808
809static int
622static int grp_dec (eio_req *grp) 810grp_dec (eio_req *grp)
623{ 811{
624 --grp->size; 812 --grp->size;
625 813
626 /* call feeder, if applicable */ 814 /* call feeder, if applicable */
627 grp_try_feed (grp); 815 grp_try_feed (grp);
… …
631 return eio_finish (grp); 819 return eio_finish (grp);
632 else 820 else
633 return 0; 821 return 0;
634} 822}
635 823
824static void
636void eio_destroy (eio_req *req) 825eio_destroy (eio_req *req)
637{ 826{
638 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); 827 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
639 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); 828 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
640 829
641 EIO_DESTROY (req); 830 EIO_DESTROY (req);
642} 831}
643 832
833static int
644static int eio_finish (eio_req *req) 834eio_finish (eio_req *req)
645{ 835{
646 int res = EIO_FINISH (req); 836 int res = EIO_FINISH (req);
647 837
648 if (req->grp) 838 if (req->grp)
649 { 839 {
… …
657 if (grp->grp_first == req) 847 if (grp->grp_first == req)
658 grp->grp_first = req->grp_next; 848 grp->grp_first = req->grp_next;
659 849
660 res2 = grp_dec (grp); 850 res2 = grp_dec (grp);
661 851
662 if (!res && res2) 852 if (!res)
663 res = res2; 853 res = res2;
664 } 854 }
665 855
666 eio_destroy (req); 856 eio_destroy (req);
667 857
668 return res; 858 return res;
669} 859}
670 860
861void
671void eio_grp_cancel (eio_req *grp) 862eio_grp_cancel (eio_req *grp)
672{ 863{
673 for (grp = grp->grp_first; grp; grp = grp->grp_next) 864 for (grp = grp->grp_first; grp; grp = grp->grp_next)
674 eio_cancel (grp); 865 eio_cancel (grp);
675} 866}
676 867
868void
677void eio_cancel (eio_req *req) 869eio_cancel (eio_req *req)
678{ 870{
679 etp_cancel (req); 871 etp_cancel (req);
680} 872}
681 873
874void
682void eio_submit (eio_req *req) 875eio_submit (eio_req *req)
683{ 876{
684 etp_submit (req); 877 etp_submit (req);
685} 878}
686 879
687unsigned int eio_nreqs (void) 880unsigned int
881eio_nreqs (void)
688{ 882{
689 return etp_nreqs (); 883 return etp_nreqs ();
690} 884}
691 885
692unsigned int eio_nready (void) 886unsigned int
887eio_nready (void)
693{ 888{
694 return etp_nready (); 889 return etp_nready ();
695} 890}
696 891
697unsigned int eio_npending (void) 892unsigned int
893eio_npending (void)
698{ 894{
699 return etp_npending (); 895 return etp_npending ();
700} 896}
701 897
702unsigned int eio_nthreads (void) 898unsigned int ecb_cold
899eio_nthreads (void)
703{ 900{
704 return etp_nthreads (); 901 return etp_nthreads ();
705} 902}
706 903
904void ecb_cold
707void eio_set_max_poll_time (double nseconds) 905eio_set_max_poll_time (double nseconds)
708{ 906{
709 etp_set_max_poll_time (nseconds); 907 etp_set_max_poll_time (nseconds);
710} 908}
711 909
910void ecb_cold
712void eio_set_max_poll_reqs (unsigned int maxreqs) 911eio_set_max_poll_reqs (unsigned int maxreqs)
713{ 912{
714 etp_set_max_poll_reqs (maxreqs); 913 etp_set_max_poll_reqs (maxreqs);
715} 914}
716 915
916void ecb_cold
717void eio_set_max_idle (unsigned int nthreads) 917eio_set_max_idle (unsigned int nthreads)
718{ 918{
719 etp_set_max_idle (nthreads); 919 etp_set_max_idle (nthreads);
720} 920}
721 921
922void ecb_cold
923eio_set_idle_timeout (unsigned int seconds)
924{
925 etp_set_idle_timeout (seconds);
926}
927
928void ecb_cold
722void eio_set_min_parallel (unsigned int nthreads) 929eio_set_min_parallel (unsigned int nthreads)
723{ 930{
724 etp_set_min_parallel (nthreads); 931 etp_set_min_parallel (nthreads);
725} 932}
726 933
934void ecb_cold
727void eio_set_max_parallel (unsigned int nthreads) 935eio_set_max_parallel (unsigned int nthreads)
728{ 936{
729 etp_set_max_parallel (nthreads); 937 etp_set_max_parallel (nthreads);
730} 938}
731 939
732int eio_poll (void) 940int eio_poll (void)
… …
741# undef pread 949# undef pread
742# undef pwrite 950# undef pwrite
743# define pread eio__pread 951# define pread eio__pread
744# define pwrite eio__pwrite 952# define pwrite eio__pwrite
745 953
746static ssize_t 954static eio_ssize_t
747eio__pread (int fd, void *buf, size_t count, off_t offset) 955eio__pread (int fd, void *buf, size_t count, off_t offset)
748{ 956{
749 ssize_t res; 957 eio_ssize_t res;
750 off_t ooffset; 958 off_t ooffset;
751 959
752 X_LOCK (preadwritelock); 960 X_LOCK (preadwritelock);
753 ooffset = lseek (fd, 0, SEEK_CUR); 961 ooffset = lseek (fd, 0, SEEK_CUR);
754 lseek (fd, offset, SEEK_SET); 962 lseek (fd, offset, SEEK_SET);
… …
757 X_UNLOCK (preadwritelock); 965 X_UNLOCK (preadwritelock);
758 966
759 return res; 967 return res;
760} 968}
761 969
762static ssize_t 970static eio_ssize_t
763eio__pwrite (int fd, void *buf, size_t count, off_t offset) 971eio__pwrite (int fd, void *buf, size_t count, off_t offset)
764{ 972{
765 ssize_t res; 973 eio_ssize_t res;
766 off_t ooffset; 974 off_t ooffset;
767 975
768 X_LOCK (preadwritelock); 976 X_LOCK (preadwritelock);
769 ooffset = lseek (fd, 0, SEEK_CUR); 977 ooffset = lseek (fd, 0, SEEK_CUR);
770 lseek (fd, offset, SEEK_SET); 978 lseek (fd, offset, SEEK_SET);
771 res = write (fd, buf, count); 979 res = write (fd, buf, count);
772 lseek (fd, offset, SEEK_SET); 980 lseek (fd, ooffset, SEEK_SET);
773 X_UNLOCK (preadwritelock); 981 X_UNLOCK (preadwritelock);
774 982
775 return res; 983 return res;
776} 984}
777#endif 985#endif
778 986
779#ifndef HAVE_FUTIMES 987#ifndef HAVE_UTIMES
780 988
781# undef utimes 989# undef utimes
782# undef futimes
783# define utimes(path,times) eio__utimes (path, times) 990# define utimes(path,times) eio__utimes (path, times)
784# define futimes(fd,times) eio__futimes (fd, times)
785 991
786static int 992static int
787eio__utimes (const char *filename, const struct timeval times[2]) 993eio__utimes (const char *filename, const struct timeval times[2])
788{ 994{
789 if (times) 995 if (times)
… …
797 } 1003 }
798 else 1004 else
799 return utime (filename, 0); 1005 return utime (filename, 0);
800} 1006}
801 1007
1008#endif
1009
1010#ifndef HAVE_FUTIMES
1011
1012# undef futimes
1013# define futimes(fd,times) eio__futimes (fd, times)
1014
1015static int
802static int eio__futimes (int fd, const struct timeval tv[2]) 1016eio__futimes (int fd, const struct timeval tv[2])
803{ 1017{
804 errno = ENOSYS; 1018 errno = ENOSYS;
805 return -1; 1019 return -1;
806} 1020}
807 1021
… …
810#if !HAVE_FDATASYNC 1024#if !HAVE_FDATASYNC
811# undef fdatasync 1025# undef fdatasync
812# define fdatasync(fd) fsync (fd) 1026# define fdatasync(fd) fsync (fd)
813#endif 1027#endif
814 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
815/* sync_file_range always needs emulation */ 1047/* sync_file_range always needs emulation */
816int 1048static int
817eio__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)
818{ 1050{
819#if HAVE_SYNC_FILE_RANGE 1051#if HAVE_SYNC_FILE_RANGE
820 int res; 1052 int res;
821 1053
… …
829 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0); 1061 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
830 } 1062 }
831 1063
832 res = sync_file_range (fd, offset, nbytes, flags); 1064 res = sync_file_range (fd, offset, nbytes, flags);
833 1065
834 if (res != ENOSYS) 1066 if (!res || errno != ENOSYS)
835 return res; 1067 return res;
836#endif 1068#endif
837 1069
838 /* 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
839 * call fdatasync, as thta matches the expectation of it's users best */ 1071 * call fdatasync, as that matches the expectation of its users best */
840 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
841} 1084}
842 1085
843#if !HAVE_READAHEAD 1086#if !HAVE_READAHEAD
844# undef readahead 1087# undef readahead
845# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 1088# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
846 1089
847static ssize_t 1090static eio_ssize_t
848eio__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)
849{ 1092{
850 size_t todo = count; 1093 size_t todo = count;
851 dBUF; 1094 dBUF;
852 1095
… …
857 pread (fd, eio_buf, len, offset); 1100 pread (fd, eio_buf, len, offset);
858 offset += len; 1101 offset += len;
859 todo -= len; 1102 todo -= len;
860 } 1103 }
861 1104
862 errno = 0; 1105 FUBd;
1106
1107 /* linux's readahead basically only fails for EBADF or EINVAL (not mmappable) */
1108 /* but not for e.g. EIO or eof, so we also never fail */
863 return count; 1109 return 0;
864} 1110}
865 1111
866#endif 1112#endif
867 1113
868/* sendfile always needs emulation */ 1114/* sendfile always needs emulation */
869static ssize_t 1115static eio_ssize_t
870eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) 1116eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
871{ 1117{
1118 eio_ssize_t written = 0;
872 ssize_t res; 1119 eio_ssize_t res;
873 1120
874 if (!count) 1121 if (!count)
875 return 0; 1122 return 0;
876 1123
1124 for (;;)
1125 {
1126#ifdef __APPLE__
1127# undef HAVE_SENDFILE /* broken, as everything on os x */
1128#endif
877#if HAVE_SENDFILE 1129#if HAVE_SENDFILE
878# if __linux 1130# if __linux
1131 off_t soffset = offset;
879 res = sendfile (ofd, ifd, &offset, count); 1132 res = sendfile (ofd, ifd, &soffset, count);
880 1133
881# elif __freebsd 1134# elif __FreeBSD__
882 /* 1135 /*
883 * Of course, the freebsd sendfile is a dire hack with no thoughts 1136 * Of course, the freebsd sendfile is a dire hack with no thoughts
884 * wasted on making it similar to other I/O functions. 1137 * wasted on making it similar to other I/O functions.
885 */ 1138 */
886 {
887 off_t sbytes; 1139 off_t sbytes;
888 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 1140 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
889 1141
890 if (res < 0 && sbytes) 1142 #if 0 /* according to the manpage, this is correct, but broken behaviour */
891 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 1143 /* freebsd' sendfile will return 0 on success */
1144 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
1145 /* not on e.g. EIO or EPIPE - sounds broken */
1146 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
892 res = sbytes; 1147 res = sbytes;
893 } 1148 #endif
1149
1150 /* according to source inspection, this is correct, and useful behaviour */
1151 if (sbytes)
1152 res = sbytes;
1153
1154# elif defined __APPLE__
1155 off_t sbytes = count;
1156 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
1157
1158 /* according to the manpage, sbytes is always valid */
1159 if (sbytes)
1160 res = sbytes;
894 1161
895# elif __hpux 1162# elif __hpux
896 res = sendfile (ofd, ifd, offset, count, 0, 0); 1163 res = sendfile (ofd, ifd, offset, count, 0, 0);
897 1164
898# elif __solaris 1165# elif __solaris
899 {
900 struct sendfilevec vec; 1166 struct sendfilevec vec;
901 size_t sbytes; 1167 size_t sbytes;
902 1168
903 vec.sfv_fd = ifd; 1169 vec.sfv_fd = ifd;
904 vec.sfv_flag = 0; 1170 vec.sfv_flag = 0;
905 vec.sfv_off = offset; 1171 vec.sfv_off = offset;
906 vec.sfv_len = count; 1172 vec.sfv_len = count;
907 1173
908 res = sendfilev (ofd, &vec, 1, &sbytes); 1174 res = sendfilev (ofd, &vec, 1, &sbytes);
909 1175
910 if (res < 0 && sbytes) 1176 if (res < 0 && sbytes)
911 res = sbytes; 1177 res = sbytes;
912 }
913 1178
914# endif 1179# endif
1180
1181#elif defined (_WIN32) && 0
1182 /* does not work, just for documentation of what would need to be done */
1183 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
1184 /* libeio guarantees that the file offset does not change, and windows */
1185 /* has no way to get an independent handle to the same file description */
1186 HANDLE h = TO_SOCKET (ifd);
1187 SetFilePointer (h, offset, 0, FILE_BEGIN);
1188 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1189
915#else 1190#else
916 res = -1; 1191 res = -1;
917 errno = ENOSYS; 1192 errno = ENOSYS;
918#endif 1193#endif
919 1194
1195 /* we assume sendfile can copy at least 128mb in one go */
1196 if (res <= 128 * 1024 * 1024)
1197 {
1198 if (res > 0)
1199 written += res;
1200
1201 if (written)
1202 return written;
1203
1204 break;
1205 }
1206 else
1207 {
1208 /* if we requested more, then probably the kernel was lazy */
1209 written += res;
1210 offset += res;
1211 count -= res;
1212
1213 if (!count)
1214 return written;
1215 }
1216 }
1217
920 if (res < 0 1218 if (res < 0
921 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1219 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1220 /* BSDs */
1221#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1222 || errno == ENOTSUP
1223#endif
1224#ifdef EOPNOTSUPP /* windows */
1225 || errno == EOPNOTSUPP /* BSDs */
1226#endif
922#if __solaris 1227#if __solaris
923 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1228 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
924#endif 1229#endif
925 ) 1230 )
926 ) 1231 )
… …
930 1235
931 res = 0; 1236 res = 0;
932 1237
933 while (count) 1238 while (count)
934 { 1239 {
935 ssize_t cnt; 1240 eio_ssize_t cnt;
936 1241
937 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 1242 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
938 1243
939 if (cnt <= 0) 1244 if (cnt <= 0)
940 { 1245 {
… …
952 1257
953 offset += cnt; 1258 offset += cnt;
954 res += cnt; 1259 res += cnt;
955 count -= cnt; 1260 count -= cnt;
956 } 1261 }
1262
1263 FUBd;
957 } 1264 }
958 1265
959 return res; 1266 return res;
1267}
1268
1269#ifdef PAGESIZE
1270# define eio_pagesize() PAGESIZE
1271#else
1272static intptr_t
1273eio_pagesize (void)
1274{
1275 static intptr_t page;
1276
1277 if (!page)
1278 page = sysconf (_SC_PAGESIZE);
1279
1280 return page;
1281}
1282#endif
1283
1284static void
1285eio_page_align (void **addr, size_t *length)
1286{
1287 intptr_t mask = eio_pagesize () - 1;
1288
1289 /* round down addr */
1290 intptr_t adj = mask & (intptr_t)*addr;
1291
1292 *addr = (void *)((intptr_t)*addr - adj);
1293 *length += adj;
1294
1295 /* round up length */
1296 *length = (*length + mask) & ~mask;
1297}
1298
1299#if !_POSIX_MEMLOCK
1300# define eio__mlockall(a) EIO_ENOSYS ()
1301#else
1302
1303static int
1304eio__mlockall (int flags)
1305{
1306 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1307 extern int mallopt (int, int);
1308 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1309 #endif
1310
1311 if (EIO_MCL_CURRENT != MCL_CURRENT
1312 || EIO_MCL_FUTURE != MCL_FUTURE)
1313 {
1314 flags = 0
1315 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1316 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1317 }
1318
1319 return mlockall (flags);
1320}
1321#endif
1322
1323#if !_POSIX_MEMLOCK_RANGE
1324# define eio__mlock(a,b) EIO_ENOSYS ()
1325#else
1326
1327static int
1328eio__mlock (void *addr, size_t length)
1329{
1330 eio_page_align (&addr, &length);
1331
1332 return mlock (addr, length);
1333}
1334
1335#endif
1336
1337#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1338# define eio__msync(a,b,c) EIO_ENOSYS ()
1339#else
1340
1341static int
1342eio__msync (void *mem, size_t len, int flags)
1343{
1344 eio_page_align (&mem, &len);
1345
1346 if (EIO_MS_ASYNC != MS_SYNC
1347 || EIO_MS_INVALIDATE != MS_INVALIDATE
1348 || EIO_MS_SYNC != MS_SYNC)
1349 {
1350 flags = 0
1351 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1352 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1353 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1354 }
1355
1356 return msync (mem, len, flags);
1357}
1358
1359#endif
1360
1361static int
1362eio__mtouch (eio_req *req)
1363{
1364 void *mem = req->ptr2;
1365 size_t len = req->size;
1366 int flags = req->int1;
1367
1368 eio_page_align (&mem, &len);
1369
1370 {
1371 intptr_t addr = (intptr_t)mem;
1372 intptr_t end = addr + len;
1373 intptr_t page = eio_pagesize ();
1374
1375 if (addr < end)
1376 if (flags & EIO_MT_MODIFY) /* modify */
1377 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1378 else
1379 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1380 }
1381
1382 return 0;
1383}
1384
1385/*****************************************************************************/
1386/* requests implemented outside eio_execute, because they are so large */
1387
1388static void
1389eio__lseek (eio_req *req)
1390{
1391 /* this usually gets optimised away completely, or your compiler sucks, */
1392 /* or the whence constants really are not 0, 1, 2 */
1393 int whence = req->int2 == EIO_SEEK_SET ? SEEK_SET
1394 : req->int2 == EIO_SEEK_CUR ? SEEK_CUR
1395 : req->int2 == EIO_SEEK_END ? SEEK_END
1396 : req->int2;
1397
1398 req->offs = lseek (req->int1, req->offs, whence);
1399 req->result = req->offs == (off_t)-1 ? -1 : 0;
1400}
1401
1402/* result will always end up in tmpbuf, there is always space for adding a 0-byte */
1403static int
1404eio__realpath (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
1405{
1406 const char *rel = path;
1407 char *res;
1408 char *tmp1, *tmp2;
1409#if SYMLOOP_MAX > 32
1410 int symlinks = SYMLOOP_MAX;
1411#else
1412 int symlinks = 32;
1413#endif
1414
1415 errno = EINVAL;
1416 if (!rel)
1417 return -1;
1418
1419 errno = ENOENT;
1420 if (!*rel)
1421 return -1;
1422
1423 res = tmpbuf_get (tmpbuf, PATH_MAX * 3);
1424 tmp1 = res + PATH_MAX;
1425 tmp2 = tmp1 + PATH_MAX;
1426
1427#if 0 /* disabled, the musl way to do things is just too racy */
1428#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1429 /* on linux we may be able to ask the kernel */
1430 {
1431 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1432
1433 if (fd >= 0)
1434 {
1435 sprintf (tmp1, "/proc/self/fd/%d", fd);
1436 req->result = readlink (tmp1, res, PATH_MAX);
1437 close (fd);
1438
1439 /* here we should probably stat the open file and the disk file, to make sure they still match */
1440
1441 if (req->result > 0)
1442 goto done;
1443 }
1444 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1445 return;
1446 }
1447#endif
1448#endif
1449
1450 if (*rel != '/')
1451 {
1452 int len;
1453
1454 errno = ENOENT;
1455 if (wd == EIO_INVALID_WD)
1456 return -1;
1457
1458 if (wd == EIO_CWD)
1459 {
1460 if (!getcwd (res, PATH_MAX))
1461 return -1;
1462
1463 len = strlen (res);
1464 }
1465 else
1466 memcpy (res, wd->str, len = wd->len);
1467
1468 if (res [1]) /* only use if not / */
1469 res += len;
1470 }
1471
1472 while (*rel)
1473 {
1474 eio_ssize_t len, linklen;
1475 const char *beg = rel;
1476
1477 while (*rel && *rel != '/')
1478 ++rel;
1479
1480 len = rel - beg;
1481
1482 if (!len) /* skip slashes */
1483 {
1484 ++rel;
1485 continue;
1486 }
1487
1488 if (beg [0] == '.')
1489 {
1490 if (len == 1)
1491 continue; /* . - nop */
1492
1493 if (beg [1] == '.' && len == 2)
1494 {
1495 /* .. - back up one component, if possible */
1496
1497 while (res != tmpbuf->ptr)
1498 if (*--res == '/')
1499 break;
1500
1501 continue;
1502 }
1503 }
1504
1505 errno = ENAMETOOLONG;
1506 if (res + 1 + len + 1 >= tmp1)
1507 return -1;
1508
1509 /* copy one component */
1510 *res = '/';
1511 memcpy (res + 1, beg, len);
1512
1513 /* zero-terminate, for readlink */
1514 res [len + 1] = 0;
1515
1516 /* now check if it's a symlink */
1517 linklen = readlink (tmpbuf->ptr, tmp1, PATH_MAX);
1518
1519 if (linklen < 0)
1520 {
1521 if (errno != EINVAL)
1522 return -1;
1523
1524 /* it's a normal directory. hopefully */
1525 res += len + 1;
1526 }
1527 else
1528 {
1529 /* yay, it was a symlink - build new path in tmp2 */
1530 int rellen = strlen (rel);
1531
1532 errno = ENAMETOOLONG;
1533 if (linklen + 1 + rellen >= PATH_MAX)
1534 return -1;
1535
1536 errno = ELOOP;
1537 if (!--symlinks)
1538 return -1;
1539
1540 if (*tmp1 == '/')
1541 res = tmpbuf->ptr; /* symlink resolves to an absolute path */
1542
1543 /* we need to be careful, as rel might point into tmp2 already */
1544 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1545 tmp2 [linklen] = '/';
1546 memcpy (tmp2, tmp1, linklen);
1547
1548 rel = tmp2;
1549 }
1550 }
1551
1552 /* special case for the lone root path */
1553 if (res == tmpbuf->ptr)
1554 *res++ = '/';
1555
1556 return res - (char *)tmpbuf->ptr;
1557}
1558
1559static signed char
1560eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1561{
1562 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1563 : a->inode < b->inode ? -1
1564 : a->inode > b->inode ? 1
1565 : 0;
1566}
1567
1568#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1569
1570#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1571#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1572
1573static void
1574eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1575{
1576 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1577 unsigned char *bit = bits;
1578
1579 assert (CHAR_BIT == 8);
1580 assert (sizeof (eio_dirent) * 8 < 256);
1581 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1582 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1583
1584 if (size <= EIO_SORT_FAST)
1585 return;
1586
1587 /* first prepare an array of bits to test in our radix sort */
1588 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1589 /* inode_bits must contain all inodes ORed together */
1590 /* which is used to skip bits that are 0 everywhere, which is very common */
1591 {
1592 eio_ino_t endianness;
1593 int i, j;
1594
1595 /* we store the byte offset of byte n into byte n of "endianness" */
1596 for (i = 0; i < sizeof (eio_ino_t); ++i)
1597 ((unsigned char *)&endianness)[i] = i;
1598
1599 *bit++ = 0;
1600
1601 for (i = 0; i < sizeof (eio_ino_t); ++i)
1602 {
1603 /* shifting off the byte offsets out of "endianness" */
1604 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1605 endianness >>= 8;
1606
1607 for (j = 0; j < 8; ++j)
1608 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1609 *bit++ = offs + j;
1610 }
1611
1612 for (j = 0; j < 8; ++j)
1613 if (score_bits & (1 << j))
1614 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1615 }
1616
1617 /* now actually do the sorting (a variant of MSD radix sort) */
1618 {
1619 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1620 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1621 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1622 int stk_idx = 0;
1623
1624 base_stk [stk_idx] = dents;
1625 end_stk [stk_idx] = dents + size;
1626 bit_stk [stk_idx] = bit - 1;
1627
1628 do
1629 {
1630 base = base_stk [stk_idx];
1631 end = end_stk [stk_idx];
1632 bit = bit_stk [stk_idx];
1633
1634 for (;;)
1635 {
1636 unsigned char O = *bit >> 3;
1637 unsigned char M = 1 << (*bit & 7);
1638
1639 eio_dirent *a = base;
1640 eio_dirent *b = end;
1641
1642 if (b - a < EIO_SORT_CUTOFF)
1643 break;
1644
1645 /* now bit-partition the array on the bit */
1646 /* this ugly asymmetric loop seems to perform much better than typical */
1647 /* partition algos found in the literature */
1648 do
1649 if (!(((unsigned char *)a)[O] & M))
1650 ++a;
1651 else if (!(((unsigned char *)--b)[O] & M))
1652 {
1653 eio_dirent tmp = *a; *a = *b; *b = tmp;
1654 ++a;
1655 }
1656 while (b > a);
1657
1658 /* next bit, or stop, if no bits left in this path */
1659 if (!*--bit)
1660 break;
1661
1662 base_stk [stk_idx] = a;
1663 end_stk [stk_idx] = end;
1664 bit_stk [stk_idx] = bit;
1665 ++stk_idx;
1666
1667 end = a;
1668 }
1669 }
1670 while (stk_idx--);
1671 }
1672}
1673
1674static void
1675eio_dent_insertion_sort (eio_dirent *dents, int size)
1676{
1677 /* first move the smallest element to the front, to act as a sentinel */
1678 {
1679 int i;
1680 eio_dirent *min = dents;
1681
1682 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1683 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1684 if (EIO_DENT_CMP (dents [i], <, *min))
1685 min = &dents [i];
1686
1687 /* swap elements 0 and j (minimum) */
1688 {
1689 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1690 }
1691 }
1692
1693 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1694 {
1695 eio_dirent *i, *j;
1696
1697 for (i = dents + 1; i < dents + size; ++i)
1698 {
1699 eio_dirent value = *i;
1700
1701 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1702 j [1] = j [0];
1703
1704 j [1] = value;
1705 }
1706 }
1707}
1708
1709static void
1710eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1711{
1712 if (size <= 1)
1713 return; /* our insertion sort relies on size > 0 */
1714
1715 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1716 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1717 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1718
1719 /* use an insertion sort at the end, or for small arrays, */
1720 /* as insertion sort is more efficient for small partitions */
1721 eio_dent_insertion_sort (dents, size);
960} 1722}
961 1723
962/* read a full directory */ 1724/* read a full directory */
963static void 1725static void
964eio__scandir (eio_req *req, etp_worker *self) 1726eio__scandir (eio_req *req, etp_worker *self)
965{ 1727{
1728 char *name, *names;
1729 int namesalloc = 4096 - sizeof (void *) * 4;
1730 int namesoffs = 0;
1731 int flags = req->int1;
1732 eio_dirent *dents = 0;
1733 int dentalloc = 128;
1734 int dentoffs = 0;
1735 eio_ino_t inode_bits = 0;
1736#ifdef _WIN32
1737 HANDLE dirp;
1738 WIN32_FIND_DATA entp;
1739#else
966 DIR *dirp; 1740 DIR *dirp;
967 EIO_STRUCT_DIRENT *entp; 1741 EIO_STRUCT_DIRENT *entp;
968 char *name, *names; 1742#endif
969 int memlen = 4096;
970 int memofs = 0;
971 int res = 0;
972 1743
973 X_LOCK (wrklock); 1744 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 1745
980 if (dirp && names) 1746 if (!(flags & EIO_READDIR_DENTS))
981 for (;;) 1747 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1748
1749#ifdef _WIN32
1750 {
1751 int len = strlen ((const char *)req->ptr1);
1752 char *path = malloc (MAX_PATH);
1753 const char *fmt;
1754 const char *reqpath = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1755
1756 if (!len)
1757 fmt = "./*";
1758 else if (reqpath[len - 1] == '/' || reqpath[len - 1] == '\\')
1759 fmt = "%s*";
1760 else
1761 fmt = "%s/*";
1762
1763 _snprintf (path, MAX_PATH, fmt, reqpath);
1764 dirp = FindFirstFile (path, &entp);
1765 free (path);
1766
1767 if (dirp == INVALID_HANDLE_VALUE)
1768 {
1769 /* should steal _dosmaperr */
1770 switch (GetLastError ())
1771 {
1772 case ERROR_FILE_NOT_FOUND:
1773 req->result = 0;
1774 break;
1775
1776 case ERROR_INVALID_NAME:
1777 case ERROR_PATH_NOT_FOUND:
1778 case ERROR_NO_MORE_FILES:
1779 errno = ENOENT;
1780 break;
1781
1782 case ERROR_NOT_ENOUGH_MEMORY:
1783 errno = ENOMEM;
1784 break;
1785
1786 default:
1787 errno = EINVAL;
1788 break;
1789 }
1790
1791 return;
1792 }
1793 }
1794#else
1795 #if HAVE_AT
1796 if (req->wd)
982 { 1797 {
1798 int fd = openat (WD2FD (req->wd), req->ptr1, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1799
1800 if (fd < 0)
1801 return;
1802
1803 dirp = fdopendir (fd);
1804
1805 if (!dirp)
1806 close (fd);
1807 }
1808 else
1809 dirp = opendir (req->ptr1);
1810 #else
1811 dirp = opendir (wd_expand (&self->tmpbuf, req->wd, req->ptr1));
1812 #endif
1813
1814 if (!dirp)
1815 return;
1816#endif
1817
1818 if (req->flags & EIO_FLAG_PTR1_FREE)
1819 free (req->ptr1);
1820
1821 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1822 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1823 req->ptr2 = names = malloc (namesalloc);
1824
1825 if (!names || (flags && !dents))
1826 return;
1827
1828 for (;;)
1829 {
1830 int done;
1831
1832#ifdef _WIN32
1833 done = !dirp;
1834#else
983 errno = 0; 1835 errno = 0;
984 entp = readdir (dirp); 1836 entp = readdir (dirp);
1837 done = !entp;
1838#endif
985 1839
1840 if (done)
1841 {
1842#ifndef _WIN32
1843 int old_errno = errno;
1844 closedir (dirp);
1845 errno = old_errno;
1846
986 if (!entp) 1847 if (errno)
1848 break;
1849#endif
1850
1851 /* sort etc. */
1852 req->int1 = flags;
1853 req->result = dentoffs;
1854
1855 if (flags & EIO_READDIR_STAT_ORDER)
1856 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1857 else if (flags & EIO_READDIR_DIRS_FIRST)
1858 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1859 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1860 else
1861 {
1862 /* in this case, all is known, and we just put dirs first and sort them */
1863 eio_dirent *oth = dents + dentoffs;
1864 eio_dirent *dir = dents;
1865
1866 /* now partition dirs to the front, and non-dirs to the back */
1867 /* by walking from both sides and swapping if necessary */
1868 while (oth > dir)
1869 {
1870 if (dir->type == EIO_DT_DIR)
1871 ++dir;
1872 else if ((--oth)->type == EIO_DT_DIR)
1873 {
1874 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1875
1876 ++dir;
1877 }
1878 }
1879
1880 /* now sort the dirs only (dirs all have the same score) */
1881 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1882 }
1883
987 break; 1884 break;
1885 }
988 1886
989 name = entp->d_name; 1887 /* now add the entry to our list(s) */
1888 name = D_NAME (entp);
990 1889
1890 /* skip . and .. entries */
991 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1891 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
992 { 1892 {
993 int len = strlen (name) + 1; 1893 int len = D_NAMLEN (entp) + 1;
994 1894
995 res++; 1895 while (ecb_expect_false (namesoffs + len > namesalloc))
996
997 while (memofs + len > memlen)
998 { 1896 {
999 memlen *= 2; 1897 namesalloc *= 2;
1000 X_LOCK (wrklock);
1001 req->ptr2 = names = realloc (names, memlen); 1898 req->ptr2 = names = realloc (names, namesalloc);
1002 X_UNLOCK (wrklock);
1003 1899
1004 if (!names) 1900 if (!names)
1901 break;
1902 }
1903
1904 memcpy (names + namesoffs, name, len);
1905
1906 if (dents)
1907 {
1908 struct eio_dirent *ent;
1909
1910 if (ecb_expect_false (dentoffs == dentalloc))
1911 {
1912 dentalloc *= 2;
1913 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1914
1915 if (!dents)
1005 break; 1916 break;
1006 } 1917 }
1007 1918
1008 memcpy (names + memofs, name, len); 1919 ent = dents + dentoffs;
1009 memofs += len; 1920
1921 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1922 ent->namelen = len - 1;
1923 ent->inode = D_INO (entp);
1924
1925 inode_bits |= ent->inode;
1926
1927 switch (D_TYPE (entp))
1928 {
1929 default:
1930 ent->type = EIO_DT_UNKNOWN;
1931 flags |= EIO_READDIR_FOUND_UNKNOWN;
1932 break;
1933
1934 #ifdef DT_FIFO
1935 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1936 #endif
1937 #ifdef DT_CHR
1938 case DT_CHR: ent->type = EIO_DT_CHR; break;
1939 #endif
1940 #ifdef DT_MPC
1941 case DT_MPC: ent->type = EIO_DT_MPC; break;
1942 #endif
1943 #ifdef DT_DIR
1944 case DT_DIR: ent->type = EIO_DT_DIR; break;
1945 #endif
1946 #ifdef DT_NAM
1947 case DT_NAM: ent->type = EIO_DT_NAM; break;
1948 #endif
1949 #ifdef DT_BLK
1950 case DT_BLK: ent->type = EIO_DT_BLK; break;
1951 #endif
1952 #ifdef DT_MPB
1953 case DT_MPB: ent->type = EIO_DT_MPB; break;
1954 #endif
1955 #ifdef DT_REG
1956 case DT_REG: ent->type = EIO_DT_REG; break;
1957 #endif
1958 #ifdef DT_NWK
1959 case DT_NWK: ent->type = EIO_DT_NWK; break;
1960 #endif
1961 #ifdef DT_CMP
1962 case DT_CMP: ent->type = EIO_DT_CMP; break;
1963 #endif
1964 #ifdef DT_LNK
1965 case DT_LNK: ent->type = EIO_DT_LNK; break;
1966 #endif
1967 #ifdef DT_SOCK
1968 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1969 #endif
1970 #ifdef DT_DOOR
1971 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1972 #endif
1973 #ifdef DT_WHT
1974 case DT_WHT: ent->type = EIO_DT_WHT; break;
1975 #endif
1976 }
1977
1978 ent->score = 7;
1979
1980 if (flags & EIO_READDIR_DIRS_FIRST)
1981 {
1982 if (ent->type == EIO_DT_UNKNOWN)
1983 {
1984 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1985 ent->score = 1;
1986 else if (!strchr (name, '.')) /* absence of dots indicate likely dirs */
1987 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1988 }
1989 else if (ent->type == EIO_DT_DIR)
1990 ent->score = 0;
1991 }
1010 } 1992 }
1993
1994 namesoffs += len;
1995 ++dentoffs;
1011 } 1996 }
1012 1997
1013 if (errno) 1998 if (EIO_CANCELLED (req))
1014 res = -1; 1999 {
2000 errno = ECANCELED;
2001 break;
2002 }
2003
2004#ifdef _WIN32
2005 if (!FindNextFile (dirp, &entp))
2006 {
2007 FindClose (dirp);
2008 dirp = 0;
2009 }
2010#endif
2011 }
2012}
2013
2014/*****************************************************************************/
2015/* working directory stuff */
2016/* various deficiencies in the posix 2008 api force us to */
2017/* keep the absolute path in string form at all times */
2018/* fuck yeah. */
2019
2020#if !HAVE_AT
2021
2022/* a bit like realpath, but usually faster because it doesn'T have to return */
2023/* an absolute or canonical path */
2024static const char *
2025wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
2026{
2027 if (!wd || *path == '/')
2028 return path;
2029
2030 if (path [0] == '.' && !path [1])
2031 return wd->str;
2032
1015 2033 {
1016 req->result = res; 2034 int l1 = wd->len;
1017} 2035 int l2 = strlen (path);
1018 2036
1019#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) 2037 char *res = tmpbuf_get (tmpbuf, l1 + l2 + 2);
1020# undef msync
1021# define msync(a,b,c) ENOSYS
1022#endif
1023 2038
1024int 2039 memcpy (res, wd->str, l1);
1025eio__mtouch (void *mem, size_t len, int flags) 2040 res [l1] = '/';
1026{ 2041 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 2042
1034 if (!page) 2043 return res;
1035 page = sysconf (_SC_PAGESIZE); 2044 }
1036#endif 2045}
1037 2046
1038 addr &= ~(page - 1); /* assume page size is always a power of two */ 2047#endif
1039 2048
1040 if (addr < end) 2049static eio_wd
1041 if (flags) /* modify */ 2050eio__wd_open_sync (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
1042 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len); 2051{
1043 else 2052 int fd;
1044 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len); 2053 eio_wd res;
2054 int len = eio__realpath (tmpbuf, wd, path);
1045 2055
2056 if (len < 0)
2057 return EIO_INVALID_WD;
2058
2059#if HAVE_AT
2060 fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
2061
2062 if (fd < 0)
2063 return EIO_INVALID_WD;
2064#endif
2065
2066 res = malloc (sizeof (*res) + len); /* one extra 0-byte */
2067
2068#if HAVE_AT
2069 res->fd = fd;
2070#endif
2071
2072 res->len = len;
2073 memcpy (res->str, tmpbuf->ptr, len);
2074 res->str [len] = 0;
2075
2076 return res;
2077}
2078
2079eio_wd
2080eio_wd_open_sync (eio_wd wd, const char *path)
2081{
2082 struct tmpbuf tmpbuf = { 0 };
2083 wd = eio__wd_open_sync (&tmpbuf, wd, path);
2084 free (tmpbuf.ptr);
2085
1046 return 0; 2086 return wd;
1047} 2087}
2088
2089void
2090eio_wd_close_sync (eio_wd wd)
2091{
2092 if (wd != EIO_INVALID_WD && wd != EIO_CWD)
2093 {
2094 #if HAVE_AT
2095 close (wd->fd);
2096 #endif
2097 free (wd);
2098 }
2099}
2100
2101#if HAVE_AT
2102
2103/* they forgot these */
2104
2105static int
2106eio__truncateat (int dirfd, const char *path, off_t length)
2107{
2108 int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
2109 int res;
2110
2111 if (fd < 0)
2112 return fd;
2113
2114 res = ftruncate (fd, length);
2115 close (fd);
2116 return res;
2117}
2118
2119static int
2120eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
2121{
2122 int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
2123 int res;
2124
2125 if (fd < 0)
2126 return fd;
2127
2128 res = fstatvfs (fd, buf);
2129 close (fd);
2130 return res;
2131
2132}
2133
2134#endif
1048 2135
1049/*****************************************************************************/ 2136/*****************************************************************************/
1050 2137
1051#define ALLOC(len) \ 2138#define ALLOC(len) \
1052 if (!req->ptr2) \ 2139 if (!req->ptr2) \
… …
1061 req->result = -1; \ 2148 req->result = -1; \
1062 break; \ 2149 break; \
1063 } \ 2150 } \
1064 } 2151 }
1065 2152
2153static void ecb_noinline ecb_cold
2154etp_proc_init (void)
2155{
2156#if HAVE_PRCTL_SET_NAME
2157 /* provide a more sensible "thread name" */
2158 char name[16 + 1];
2159 const int namelen = sizeof (name) - 1;
2160 int len;
2161
2162 prctl (PR_GET_NAME, (unsigned long)name, 0, 0, 0);
2163 name [namelen] = 0;
2164 len = strlen (name);
2165 strcpy (name + (len <= namelen - 4 ? len : namelen - 4), "/eio");
2166 prctl (PR_SET_NAME, (unsigned long)name, 0, 0, 0);
2167#endif
2168}
2169
1066X_THREAD_PROC (etp_proc) 2170X_THREAD_PROC (etp_proc)
1067{ 2171{
1068 ETP_REQ *req; 2172 ETP_REQ *req;
1069 struct timespec ts; 2173 struct timespec ts;
1070 etp_worker *self = (etp_worker *)thr_arg; 2174 etp_worker *self = (etp_worker *)thr_arg;
1071 2175
2176 etp_proc_init ();
2177
1072 /* try to distribute timeouts somewhat randomly */ 2178 /* try to distribute timeouts somewhat evenly */
1073 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 2179 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1074 2180
1075 for (;;) 2181 for (;;)
1076 { 2182 {
2183 ts.tv_sec = 0;
2184
1077 X_LOCK (reqlock); 2185 X_LOCK (reqlock);
1078 2186
1079 for (;;) 2187 for (;;)
1080 { 2188 {
1081 self->req = req = reqq_shift (&req_queue); 2189 req = reqq_shift (&req_queue);
1082 2190
1083 if (req) 2191 if (req)
1084 break; 2192 break;
1085 2193
2194 if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
2195 {
2196 X_UNLOCK (reqlock);
2197 X_LOCK (wrklock);
2198 --started;
2199 X_UNLOCK (wrklock);
2200 goto quit;
2201 }
2202
1086 ++idle; 2203 ++idle;
1087 2204
1088 ts.tv_sec = time (0) + IDLE_TIMEOUT; 2205 if (idle <= max_idle)
1089 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 2206 /* we are allowed to idle, so do so without any timeout */
2207 X_COND_WAIT (reqwait, reqlock);
2208 else
1090 { 2209 {
1091 if (idle > max_idle) 2210 /* initialise timeout once */
1092 { 2211 if (!ts.tv_sec)
1093 --idle; 2212 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 2213
1101 /* we are allowed to idle, so do so without any timeout */
1102 X_COND_WAIT (reqwait, reqlock); 2214 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
2215 ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1103 } 2216 }
1104 2217
1105 --idle; 2218 --idle;
1106 } 2219 }
1107 2220
… …
1110 X_UNLOCK (reqlock); 2223 X_UNLOCK (reqlock);
1111 2224
1112 if (req->type < 0) 2225 if (req->type < 0)
1113 goto quit; 2226 goto quit;
1114 2227
1115 if (!EIO_CANCELLED (req))
1116 ETP_EXECUTE (self, req); 2228 ETP_EXECUTE (self, req);
1117 2229
1118 X_LOCK (reslock); 2230 X_LOCK (reslock);
1119 2231
1120 ++npending; 2232 ++npending;
1121 2233
1122 if (!reqq_push (&res_queue, req) && want_poll_cb) 2234 if (!reqq_push (&res_queue, req) && want_poll_cb)
1123 want_poll_cb (); 2235 want_poll_cb ();
1124 2236
1125 self->req = 0;
1126 etp_worker_clear (self); 2237 etp_worker_clear (self);
1127 2238
1128 X_UNLOCK (reslock); 2239 X_UNLOCK (reslock);
1129 } 2240 }
1130 2241
1131quit: 2242quit:
2243 free (req);
2244
1132 X_LOCK (wrklock); 2245 X_LOCK (wrklock);
1133 etp_worker_free (self); 2246 etp_worker_free (self);
1134 X_UNLOCK (wrklock); 2247 X_UNLOCK (wrklock);
1135 2248
1136 return 0; 2249 return 0;
1137} 2250}
1138 2251
1139/*****************************************************************************/ 2252/*****************************************************************************/
1140 2253
2254int ecb_cold
1141int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 2255eio_init (void (*want_poll)(void), void (*done_poll)(void))
1142{ 2256{
2257#if !HAVE_PREADWRITE
2258 X_MUTEX_CREATE (preadwritelock);
2259#endif
2260
1143 return etp_init (want_poll, done_poll); 2261 return etp_init (want_poll, done_poll);
1144} 2262}
1145 2263
2264ecb_inline void
1146static void eio_api_destroy (eio_req *req) 2265eio_api_destroy (eio_req *req)
1147{ 2266{
1148 free (req); 2267 free (req);
1149} 2268}
1150 2269
1151#define REQ(rtype) \ 2270#define REQ(rtype) \
… …
1170 { \ 2289 { \
1171 eio_api_destroy (req); \ 2290 eio_api_destroy (req); \
1172 return 0; \ 2291 return 0; \
1173 } 2292 }
1174 2293
2294static void
1175static void eio_execute (etp_worker *self, eio_req *req) 2295eio_execute (etp_worker *self, eio_req *req)
1176{ 2296{
1177 errno = 0; 2297#if HAVE_AT
2298 int dirfd;
2299#else
2300 const char *path;
2301#endif
2302
2303 if (ecb_expect_false (EIO_CANCELLED (req)))
2304 {
2305 req->result = -1;
2306 req->errorno = ECANCELED;
2307 return;
2308 }
2309
2310 if (ecb_expect_false (req->wd == EIO_INVALID_WD))
2311 {
2312 req->result = -1;
2313 req->errorno = ENOENT;
2314 return;
2315 }
2316
2317 if (req->type >= EIO_OPEN)
2318 {
2319 #if HAVE_AT
2320 dirfd = WD2FD (req->wd);
2321 #else
2322 path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
2323 #endif
2324 }
1178 2325
1179 switch (req->type) 2326 switch (req->type)
1180 { 2327 {
2328 case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
2329 req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
2330 break;
2331 case EIO_WD_CLOSE: req->result = 0;
2332 eio_wd_close_sync (req->wd); break;
2333
2334 case EIO_SEEK: eio__lseek (req); break;
1181 case EIO_READ: ALLOC (req->size); 2335 case EIO_READ: ALLOC (req->size);
1182 req->result = req->offs >= 0 2336 req->result = req->offs >= 0
1183 ? pread (req->int1, req->ptr2, req->size, req->offs) 2337 ? pread (req->int1, req->ptr2, req->size, req->offs)
1184 : read (req->int1, req->ptr2, req->size); break; 2338 : read (req->int1, req->ptr2, req->size); break;
1185 case EIO_WRITE: req->result = req->offs >= 0 2339 case EIO_WRITE: req->result = req->offs >= 0
1186 ? pwrite (req->int1, req->ptr2, req->size, req->offs) 2340 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1187 : write (req->int1, req->ptr2, req->size); break; 2341 : write (req->int1, req->ptr2, req->size); break;
1188 2342
1189 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break; 2343 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; 2344 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
2345
2346#if HAVE_AT
1191 2347
1192 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2348 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1193 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2349 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break;
1194 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2350 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1195 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2351 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; 2352 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; 2353 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; 2354 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; 2355 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; 2356
1208 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1209 case EIO_UNLINK: req->result = unlink (req->ptr1); break; 2357 case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
1210 case EIO_RMDIR: req->result = rmdir (req->ptr1); break; 2358 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; 2359 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; 2360 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; 2361 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; 2362 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; 2363 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); 2364 case EIO_READLINK: ALLOC (PATH_MAX);
1218 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; 2365 req->result = readlinkat (dirfd, req->ptr1, req->ptr2, PATH_MAX); break;
1219 2366 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1220 case EIO_SYNC: req->result = 0; sync (); break; 2367 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: 2368 case EIO_UTIME:
1230#ifdef _WIN32 2369 case EIO_FUTIME:
1231 Sleep (req->nv1 * 1000.); 2370 {
2371 struct timespec ts[2];
2372 struct timespec *times;
2373
2374 if (req->nv1 != -1. || req->nv2 != -1.)
2375 {
2376 ts[0].tv_sec = req->nv1;
2377 ts[0].tv_nsec = (req->nv1 - ts[0].tv_sec) * 1e9;
2378 ts[1].tv_sec = req->nv2;
2379 ts[1].tv_nsec = (req->nv2 - ts[1].tv_sec) * 1e9;
2380
2381 times = ts;
2382 }
2383 else
2384 times = 0;
2385
2386 req->result = req->type == EIO_FUTIME
2387 ? futimens (req->int1, times)
2388 : utimensat (dirfd, req->ptr1, times, 0);
2389 }
2390 break;
2391
1232#else 2392#else
1233 {
1234 struct timeval tv;
1235 2393
1236 tv.tv_sec = req->nv1; 2394 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1237 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; 2395 req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
2396 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2397 req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
2398 case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
2399 case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
2400 case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
2401 case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
1238 2402
1239 req->result = select (0, 0, 0, 0, &tv); 2403 case EIO_UNLINK: req->result = unlink (path ); break;
1240 } 2404 case EIO_RMDIR: req->result = rmdir (path ); break;
1241#endif 2405 case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
1242 break; 2406 case EIO_RENAME: req->result = rename (path , req->ptr2); break;
2407 case EIO_LINK: req->result = link (path , req->ptr2); break;
2408 case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
2409 case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
2410 case EIO_READLINK: ALLOC (PATH_MAX);
2411 req->result = readlink (path, req->ptr2, PATH_MAX); break;
2412 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2413 req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
1243 2414
1244 case EIO_UTIME: 2415 case EIO_UTIME:
1245 case EIO_FUTIME: 2416 case EIO_FUTIME:
1246 { 2417 {
1247 struct timeval tv[2]; 2418 struct timeval tv[2];
1248 struct timeval *times; 2419 struct timeval *times;
1249 2420
1250 if (req->nv1 != -1. || req->nv2 != -1.) 2421 if (req->nv1 != -1. || req->nv2 != -1.)
1251 { 2422 {
1252 tv[0].tv_sec = req->nv1; 2423 tv[0].tv_sec = req->nv1;
1253 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.; 2424 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
1254 tv[1].tv_sec = req->nv2; 2425 tv[1].tv_sec = req->nv2;
1255 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.; 2426 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
1256 2427
1257 times = tv; 2428 times = tv;
1258 } 2429 }
1259 else 2430 else
1260 times = 0; 2431 times = 0;
1261
1262 2432
1263 req->result = req->type == EIO_FUTIME 2433 req->result = req->type == EIO_FUTIME
1264 ? futimes (req->int1, times) 2434 ? futimes (req->int1, times)
1265 : utimes (req->ptr1, times); 2435 : utimes (req->ptr1, times);
1266 } 2436 }
1267 break; 2437 break;
1268 2438
2439#endif
2440
2441 case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
2442 {
2443 ALLOC (req->result);
2444 memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
2445 }
2446 break;
2447
2448 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2449 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
2450
2451 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2452 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2453
2454 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
2455 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
2456 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
2457
2458 case EIO_CLOSE: req->result = close (req->int1); break;
2459 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
2460 case EIO_SYNC: req->result = 0; sync (); break;
2461 case EIO_FSYNC: req->result = fsync (req->int1); break;
2462 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
2463 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
2464 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
2465 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
2466 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
2467 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
2468 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
2469 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
2470
2471 case EIO_READDIR: eio__scandir (req, self); break;
2472
2473 case EIO_BUSY:
2474#ifdef _WIN32
2475 Sleep (req->nv1 * 1e3);
2476#else
2477 {
2478 struct timeval tv;
2479
2480 tv.tv_sec = req->nv1;
2481 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
2482
2483 req->result = select (0, 0, 0, 0, &tv);
2484 }
2485#endif
2486 break;
2487
1269 case EIO_GROUP: 2488 case EIO_GROUP:
1270 abort (); /* handled in eio_request */ 2489 abort (); /* handled in eio_request */
1271 2490
1272 case EIO_NOP: 2491 case EIO_NOP:
1273 req->result = 0; 2492 req->result = 0;
1274 break; 2493 break;
1275 2494
1276 case EIO_CUSTOM: 2495 case EIO_CUSTOM:
1277 ((void (*)(eio_req *))req->feed) (req); 2496 req->feed (req);
1278 break; 2497 break;
1279 2498
1280 default: 2499 default:
2500 errno = ENOSYS;
1281 req->result = -1; 2501 req->result = -1;
1282 break; 2502 break;
1283 } 2503 }
1284 2504
1285 req->errorno = errno; 2505 req->errorno = errno;
1286} 2506}
1287 2507
1288#ifndef EIO_NO_WRAPPERS 2508#ifndef EIO_NO_WRAPPERS
1289 2509
2510eio_req *eio_wd_open (const char *path, int pri, eio_cb cb, void *data)
2511{
2512 REQ (EIO_WD_OPEN); PATH; SEND;
2513}
2514
2515eio_req *eio_wd_close (eio_wd wd, int pri, eio_cb cb, void *data)
2516{
2517 REQ (EIO_WD_CLOSE); req->wd = wd; SEND;
2518}
2519
1290eio_req *eio_nop (int pri, eio_cb cb, void *data) 2520eio_req *eio_nop (int pri, eio_cb cb, void *data)
1291{ 2521{
1292 REQ (EIO_NOP); SEND; 2522 REQ (EIO_NOP); SEND;
1293} 2523}
1294 2524
… …
1310eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2540eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1311{ 2541{
1312 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2542 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1313} 2543}
1314 2544
2545eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2546{
2547 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2548}
2549
2550eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
2551{
2552 REQ (EIO_SYNCFS); req->int1 = fd; SEND;
2553}
2554
2555eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2556{
2557 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2558}
2559
1315eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2560eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1316{ 2561{
1317 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2562 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1318} 2563}
1319 2564
1320eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) 2565eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
1321{ 2566{
1322 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; 2567 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
1323} 2568}
1324 2569
1325eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) 2570eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
1326{ 2571{
1327 REQ (EIO_FDATASYNC); req->int1 = fd; SEND; 2572 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2573}
2574
2575eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2576{
2577 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
1328} 2578}
1329 2579
1330eio_req *eio_close (int fd, int pri, eio_cb cb, void *data) 2580eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
1331{ 2581{
1332 REQ (EIO_CLOSE); req->int1 = fd; SEND; 2582 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) 2585eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
1336{ 2586{
1337 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND; 2587 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
1338} 2588}
1339 2589
2590eio_req *eio_seek (int fd, off_t offset, int whence, int pri, eio_cb cb, void *data)
2591{
2592 REQ (EIO_SEEK); req->int1 = fd; req->offs = offset; req->int2 = whence; SEND;
2593}
2594
1340eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data) 2595eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1341{ 2596{
1342 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND; 2597 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1343} 2598}
1344 2599
… …
1350eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) 2605eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1351{ 2606{
1352 REQ (EIO_FSTAT); req->int1 = fd; SEND; 2607 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1353} 2608}
1354 2609
2610eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
2611{
2612 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
2613}
2614
1355eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) 2615eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1356{ 2616{
1357 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; 2617 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1358} 2618}
1359 2619
… …
1365eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) 2625eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1366{ 2626{
1367 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2627 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1368} 2628}
1369 2629
1370eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2630eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1371{ 2631{
1372 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2632 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1373} 2633}
1374 2634
1375eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) 2635eio_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) 2655eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1396{ 2656{
1397 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2657 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1398} 2658}
1399 2659
1400eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2660eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1401{ 2661{
1402 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2662 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1403} 2663}
1404 2664
1405eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) 2665eio_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) 2681eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1422{ 2682{
1423 return eio__1path (EIO_READLINK, path, pri, cb, data); 2683 return eio__1path (EIO_READLINK, path, pri, cb, data);
1424} 2684}
1425 2685
2686eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2687{
2688 return eio__1path (EIO_REALPATH, path, pri, cb, data);
2689}
2690
1426eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) 2691eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1427{ 2692{
1428 return eio__1path (EIO_STAT, path, pri, cb, data); 2693 return eio__1path (EIO_STAT, path, pri, cb, data);
1429} 2694}
1430 2695
1431eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) 2696eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1432{ 2697{
1433 return eio__1path (EIO_LSTAT, path, pri, cb, data); 2698 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1434} 2699}
1435 2700
2701eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
2702{
2703 return eio__1path (EIO_STATVFS, path, pri, cb, data);
2704}
2705
1436eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) 2706eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1437{ 2707{
1438 return eio__1path (EIO_UNLINK, path, pri, cb, data); 2708 return eio__1path (EIO_UNLINK, path, pri, cb, data);
1439} 2709}
1440 2710
1441eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data) 2711eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1442{ 2712{
1443 return eio__1path (EIO_RMDIR, path, pri, cb, data); 2713 return eio__1path (EIO_RMDIR, path, pri, cb, data);
1444} 2714}
1445 2715
1446eio_req *eio_readdir (const char *path, int pri, eio_cb cb, void *data) 2716eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1447{ 2717{
1448 return eio__1path (EIO_READDIR, path, pri, cb, data); 2718 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1449} 2719}
1450 2720
1451eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) 2721eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1452{ 2722{
1453 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; 2723 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1454} 2724}
1455 2725
1456static eio_req * 2726static eio_req *
1457eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2727eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1458{ 2728{
… …
1482eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2752eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1483{ 2753{
1484 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2754 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1485} 2755}
1486 2756
1487eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) 2757eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
1488{ 2758{
1489 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND; 2759 REQ (EIO_CUSTOM); req->feed = execute; SEND;
1490} 2760}
1491 2761
1492#endif 2762#endif
1493 2763
1494eio_req *eio_grp (eio_cb cb, void *data) 2764eio_req *eio_grp (eio_cb cb, void *data)
… …
1503#undef SEND 2773#undef SEND
1504 2774
1505/*****************************************************************************/ 2775/*****************************************************************************/
1506/* grp functions */ 2776/* grp functions */
1507 2777
2778void
1508void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2779eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1509{ 2780{
1510 grp->int2 = limit; 2781 grp->int2 = limit;
1511 grp->feed = feed; 2782 grp->feed = feed;
1512 2783
1513 grp_try_feed (grp); 2784 grp_try_feed (grp);
1514} 2785}
1515 2786
2787void
1516void eio_grp_limit (eio_req *grp, int limit) 2788eio_grp_limit (eio_req *grp, int limit)
1517{ 2789{
1518 grp->int2 = limit; 2790 grp->int2 = limit;
1519 2791
1520 grp_try_feed (grp); 2792 grp_try_feed (grp);
1521} 2793}
1522 2794
2795void
1523void eio_grp_add (eio_req *grp, eio_req *req) 2796eio_grp_add (eio_req *grp, eio_req *req)
1524{ 2797{
1525 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2798 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1526 2799
1527 grp->flags |= EIO_FLAG_GROUPADD; 2800 grp->flags |= EIO_FLAG_GROUPADD;
1528 2801
… …
1539} 2812}
1540 2813
1541/*****************************************************************************/ 2814/*****************************************************************************/
1542/* misc garbage */ 2815/* misc garbage */
1543 2816
2817eio_ssize_t
1544ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2818eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1545{ 2819{
1546 etp_worker wrk;
1547
1548 wrk.dbuf = 0;
1549
1550 eio__sendfile (ofd, ifd, offset, count, &wrk); 2820 return eio__sendfile (ofd, ifd, offset, count);
1551
1552 if (wrk.dbuf)
1553 free (wrk.dbuf);
1554} 2821}
1555 2822

Diff Legend

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