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

Comparing cvsroot/libeio/eio.c (file contents):
Revision 1.38 by root, Fri Jun 12 20:01:42 2009 UTC vs.
Revision 1.116 by root, Tue Mar 27 19:27:35 2012 UTC

1/* 1/*
2 * libeio implementation 2 * libeio implementation
3 * 3 *
4 * Copyright (c) 2007,2008,2009 Marc Alexander Lehmann <libeio@schmorp.de> 4 * Copyright (c) 2007,2008,2009,2010,2011 Marc Alexander Lehmann <libeio@schmorp.de>
5 * All rights reserved. 5 * All rights reserved.
6 * 6 *
7 * Redistribution and use in source and binary forms, with or without modifica- 7 * Redistribution and use in source and binary forms, with or without modifica-
8 * tion, are permitted provided that the following conditions are met: 8 * tion, are permitted provided that the following conditions are met:
9 * 9 *
35 * and other provisions required by the GPL. If you do not delete the 35 * and other provisions required by the GPL. If you do not delete the
36 * provisions above, a recipient may use your version of this file under 36 * provisions above, a recipient may use your version of this file under
37 * either the BSD or the GPL. 37 * either the BSD or the GPL.
38 */ 38 */
39 39
40#ifndef _WIN32
41# include "config.h"
42#endif
43
40#include "eio.h" 44#include "eio.h"
45#include "ecb.h"
41 46
42#ifdef EIO_STACKSIZE 47#ifdef EIO_STACKSIZE
43# define XTHREAD_STACKSIZE EIO_STACKSIZE 48# define X_STACKSIZE EIO_STACKSIZE
44#endif 49#endif
45#include "xthread.h" 50#include "xthread.h"
46 51
47#include <errno.h> 52#include <errno.h>
48#include <stddef.h> 53#include <stddef.h>
53#include <sys/stat.h> 58#include <sys/stat.h>
54#include <limits.h> 59#include <limits.h>
55#include <fcntl.h> 60#include <fcntl.h>
56#include <assert.h> 61#include <assert.h>
57 62
63/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
64/* intptr_t only comes from stdint.h, says idiot openbsd coder */
65#if HAVE_STDINT_H
66# include <stdint.h>
67#endif
68
69#ifndef ECANCELED
70# define ECANCELED EDOM
71#endif
72#ifndef ELOOP
73# define ELOOP EDOM
74#endif
75
76#if !defined(ENOTSOCK) && defined(WSAENOTSOCK)
77# define ENOTSOCK WSAENOTSOCK
78#endif
79
80static void eio_destroy (eio_req *req);
81
58#ifndef EIO_FINISH 82#ifndef EIO_FINISH
59# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0 83# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
60#endif 84#endif
61 85
62#ifndef EIO_DESTROY 86#ifndef EIO_DESTROY
65 89
66#ifndef EIO_FEED 90#ifndef EIO_FEED
67# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0) 91# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)
68#endif 92#endif
69 93
94#ifndef EIO_FD_TO_WIN32_HANDLE
95# define EIO_FD_TO_WIN32_HANDLE(fd) _get_osfhandle (fd)
96#endif
97#ifndef EIO_WIN32_HANDLE_TO_FD
98# define EIO_WIN32_HANDLE_TO_FD(handle) _open_osfhandle (handle, 0)
99#endif
100
101#define EIO_ERRNO(errval,retval) ((errno = errval), retval)
102
103#define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1)
104
70#ifdef _WIN32 105#ifdef _WIN32
71 106
72 /*doh*/ 107 #undef PAGESIZE
108 #define PAGESIZE 4096 /* GetSystemInfo? */
109
110 /* TODO: look at how perl does stat (non-sloppy), unlink (ro-files), utime, link */
111
112 #ifdef EIO_STRUCT_STATI64
113 /* look at perl's non-sloppy stat */
114 #define stat(path,buf) _stati64 (path,buf)
115 #define fstat(fd,buf) _fstati64 (fd,buf)
116 #endif
117 #define lstat(path,buf) stat (path,buf)
118 #define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
119 #define mkdir(path,mode) _mkdir (path)
120 #define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
121
122 #define chmod(path,mode) _chmod (path, mode)
123 #define dup(fd) _dup (fd)
124 #define dup2(fd1,fd2) _dup2 (fd1, fd2)
125
126 #define fchmod(fd,mode) EIO_ENOSYS ()
127 #define chown(path,uid,gid) EIO_ENOSYS ()
128 #define fchown(fd,uid,gid) EIO_ENOSYS ()
129 #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
130 #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
131 #define mknod(path,mode,dev) EIO_ENOSYS ()
132 #define sync() EIO_ENOSYS ()
133 #define readlink(path,buf,s) EIO_ENOSYS ()
134 #define statvfs(path,buf) EIO_ENOSYS ()
135 #define fstatvfs(fd,buf) EIO_ENOSYS ()
136
137 /* rename() uses MoveFile, which fails to overwrite */
138 #define rename(old,neu) eio__rename (old, neu)
139
140 static int
141 eio__rename (const char *old, const char *neu)
142 {
143 if (MoveFileEx (old, neu, MOVEFILE_REPLACE_EXISTING))
144 return 0;
145
146 /* should steal _dosmaperr */
147 switch (GetLastError ())
148 {
149 case ERROR_FILE_NOT_FOUND:
150 case ERROR_PATH_NOT_FOUND:
151 case ERROR_INVALID_DRIVE:
152 case ERROR_NO_MORE_FILES:
153 case ERROR_BAD_NETPATH:
154 case ERROR_BAD_NET_NAME:
155 case ERROR_BAD_PATHNAME:
156 case ERROR_FILENAME_EXCED_RANGE:
157 errno = ENOENT;
158 break;
159
160 default:
161 errno = EACCES;
162 break;
163 }
164
165 return -1;
166 }
167
168 /* we could even stat and see if it exists */
169 static int
170 symlink (const char *old, const char *neu)
171 {
172 #if WINVER >= 0x0600
173 if (CreateSymbolicLink (neu, old, 1))
174 return 0;
175
176 if (CreateSymbolicLink (neu, old, 0))
177 return 0;
178 #endif
179
180 return EIO_ERRNO (ENOENT, -1);
181 }
182
183 /* POSIX API only */
184 #define CreateHardLink(neu,old,flags) 0
185 #define CreateSymbolicLink(neu,old,flags) 0
186
187 struct statvfs
188 {
189 int dummy;
190 };
191
192 #define DT_DIR EIO_DT_DIR
193 #define DT_REG EIO_DT_REG
194 #define D_NAME(entp) entp.cFileName
195 #define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
196
73#else 197#else
74 198
75# include "config.h"
76# include <sys/time.h> 199 #include <sys/time.h>
77# include <sys/select.h> 200 #include <sys/select.h>
78# include <sys/mman.h> 201 #include <sys/statvfs.h>
79# include <unistd.h> 202 #include <unistd.h>
203 #include <signal.h>
204 #include <dirent.h>
205
206 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
207 #include <sys/mman.h>
208 #endif
209
210 #define D_NAME(entp) entp->d_name
211
212 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
213 #if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
214 #define _DIRENT_HAVE_D_TYPE /* sigh */
215 #define D_INO(de) (de)->d_fileno
216 #define D_NAMLEN(de) (de)->d_namlen
217 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
218 #define D_INO(de) (de)->d_ino
219 #endif
220
221 #ifdef _D_EXACT_NAMLEN
222 #undef D_NAMLEN
223 #define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
224 #endif
225
226 #ifdef _DIRENT_HAVE_D_TYPE
227 #define D_TYPE(de) (de)->d_type
228 #endif
229
230 #ifndef EIO_STRUCT_DIRENT
231 #define EIO_STRUCT_DIRENT struct dirent
232 #endif
233
234#endif
235
236#if HAVE_UTIMES
80# include <utime.h> 237# include <utime.h>
81# include <signal.h>
82# include <dirent.h>
83
84/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
85# if defined(__FreeBSD__) || defined(__NetBSD__) || defined(__OpenBSD__)
86# define _DIRENT_HAVE_D_TYPE /* sigh */
87# define D_INO(de) (de)->d_fileno
88# define D_NAMLEN(de) (de)->d_namlen
89# elif defined(__linux) || defined(d_ino) || _XOPEN_SOURCE >= 600
90# define D_INO(de) (de)->d_ino
91# endif 238#endif
92 239
93#ifdef _D_EXACT_NAMLEN 240#if HAVE_SYS_SYSCALL_H
94# undef D_NAMLEN 241# include <sys/syscall.h>
95# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
96#endif
97
98# ifdef _DIRENT_HAVE_D_TYPE
99# define D_TYPE(de) (de)->d_type
100# endif 242#endif
101 243
102# ifndef EIO_STRUCT_DIRENT 244#if HAVE_SYS_PRCTL_H
103# define EIO_STRUCT_DIRENT struct dirent 245# include <sys/prctl.h>
104# endif
105
106#endif 246#endif
107 247
108#if HAVE_SENDFILE 248#if HAVE_SENDFILE
109# if __linux 249# if __linux
110# include <sys/sendfile.h> 250# include <sys/sendfile.h>
111# elif __freebsd 251# elif __FreeBSD__ || defined __APPLE__
112# include <sys/socket.h> 252# include <sys/socket.h>
113# include <sys/uio.h> 253# include <sys/uio.h>
114# elif __hpux 254# elif __hpux
115# include <sys/socket.h> 255# include <sys/socket.h>
116# elif __solaris /* not yet */ 256# elif __solaris
117# include <sys/sendfile.h> 257# include <sys/sendfile.h>
118# else 258# else
119# error sendfile support requested but not available 259# error sendfile support requested but not available
120# endif 260# endif
121#endif 261#endif
125#endif 265#endif
126#ifndef D_INO 266#ifndef D_INO
127# define D_INO(de) 0 267# define D_INO(de) 0
128#endif 268#endif
129#ifndef D_NAMLEN 269#ifndef D_NAMLEN
130# define D_NAMLEN(de) strlen ((de)->d_name) 270# define D_NAMLEN(entp) strlen (D_NAME (entp))
131#endif 271#endif
132
133/* number of seconds after which an idle threads exit */
134#define IDLE_TIMEOUT 10
135 272
136/* used for struct dirent, AIX doesn't provide it */ 273/* used for struct dirent, AIX doesn't provide it */
137#ifndef NAME_MAX 274#ifndef NAME_MAX
138# define NAME_MAX 4096 275# define NAME_MAX 4096
139#endif 276#endif
140 277
278/* used for readlink etc. */
279#ifndef PATH_MAX
280# define PATH_MAX 4096
281#endif
282
141/* buffer size for various temporary buffers */ 283/* buffer size for various temporary buffers */
142#define EIO_BUFSIZE 65536 284#define EIO_BUFSIZE 65536
143 285
144#define dBUF \ 286#define dBUF \
145 char *eio_buf; \
146 ETP_WORKER_LOCK (self); \
147 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 287 char *eio_buf = malloc (EIO_BUFSIZE); \
148 ETP_WORKER_UNLOCK (self); \
149 errno = ENOMEM; \ 288 errno = ENOMEM; \
150 if (!eio_buf) \ 289 if (!eio_buf) \
151 return -1; 290 return -1
291
292#define FUBd \
293 free (eio_buf)
152 294
153#define EIO_TICKS ((1000000 + 1023) >> 10) 295#define EIO_TICKS ((1000000 + 1023) >> 10)
154 296
155/*****************************************************************************/ 297/*****************************************************************************/
156 298
157#if __GNUC__ >= 3 299struct tmpbuf
158# 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
159#else 325#else
160# 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);
161#endif 328#endif
162 329
163#define expect_false(expr) expect ((expr) != 0, 0) 330struct eio_pwd
164#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};
165 338
166/*****************************************************************************/ 339/*****************************************************************************/
167 340
168#define ETP_PRI_MIN EIO_PRI_MIN 341#define ETP_PRI_MIN EIO_PRI_MIN
169#define ETP_PRI_MAX EIO_PRI_MAX 342#define ETP_PRI_MAX EIO_PRI_MAX
175static int eio_finish (eio_req *req); 348static int eio_finish (eio_req *req);
176#define ETP_FINISH(req) eio_finish (req) 349#define ETP_FINISH(req) eio_finish (req)
177static void eio_execute (struct etp_worker *self, eio_req *req); 350static void eio_execute (struct etp_worker *self, eio_req *req);
178#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) 351#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
179 352
180#define ETP_WORKER_CLEAR(req) \
181 if (wrk->dbuf) \
182 { \
183 free (wrk->dbuf); \
184 wrk->dbuf = 0; \
185 } \
186 \
187 if (wrk->dirp) \
188 { \
189 closedir (wrk->dirp); \
190 wrk->dirp = 0; \
191 }
192
193#define ETP_WORKER_COMMON \
194 void *dbuf; \
195 DIR *dirp;
196
197/*****************************************************************************/ 353/*****************************************************************************/
198 354
199#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) 355#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
200 356
201/* calculcate time difference in ~1/EIO_TICKS of a second */ 357/* calculate time difference in ~1/EIO_TICKS of a second */
358ecb_inline int
202static int tvdiff (struct timeval *tv1, struct timeval *tv2) 359tvdiff (struct timeval *tv1, struct timeval *tv2)
203{ 360{
204 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 361 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
205 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 362 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
206} 363}
207 364
211static void (*done_poll_cb) (void); 368static void (*done_poll_cb) (void);
212 369
213static unsigned int max_poll_time; /* reslock */ 370static unsigned int max_poll_time; /* reslock */
214static unsigned int max_poll_reqs; /* reslock */ 371static unsigned int max_poll_reqs; /* reslock */
215 372
216static volatile unsigned int nreqs; /* reqlock */ 373static unsigned int nreqs; /* reqlock */
217static volatile unsigned int nready; /* reqlock */ 374static unsigned int nready; /* reqlock */
218static volatile unsigned int npending; /* reqlock */ 375static unsigned int npending; /* reqlock */
219static 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 */
220 378
221static mutex_t wrklock = X_MUTEX_INIT; 379static xmutex_t wrklock;
222static mutex_t reslock = X_MUTEX_INIT; 380static xmutex_t reslock;
223static mutex_t reqlock = X_MUTEX_INIT; 381static xmutex_t reqlock;
224static cond_t reqwait = X_COND_INIT; 382static xcond_t reqwait;
225 383
226#if !HAVE_PREADWRITE 384#if !HAVE_PREADWRITE
227/* 385/*
228 * make our pread/pwrite emulation safe against themselves, but not against 386 * make our pread/pwrite emulation safe against themselves, but not against
229 * 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,
230 * but that's your problem, not mine. 388 * but that's your problem, not mine.
231 */ 389 */
232static mutex_t preadwritelock = X_MUTEX_INIT; 390static xmutex_t preadwritelock;
233#endif 391#endif
234 392
235typedef struct etp_worker 393typedef struct etp_worker
236{ 394{
395 struct tmpbuf tmpbuf;
396
237 /* locked by wrklock */ 397 /* locked by wrklock */
238 struct etp_worker *prev, *next; 398 struct etp_worker *prev, *next;
239 399
240 thread_t tid; 400 xthread_t tid;
241 401
242 /* locked by reslock, reqlock or wrklock */ 402#ifdef ETP_WORKER_COMMON
243 ETP_REQ *req; /* currently processed request */
244
245 ETP_WORKER_COMMON 403 ETP_WORKER_COMMON
404#endif
246} etp_worker; 405} etp_worker;
247 406
248static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */ 407static etp_worker wrk_first; /* NOT etp */
249 408
250#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock) 409#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
251#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock) 410#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
252 411
253/* worker threads management */ 412/* worker threads management */
254 413
414static void
255static void etp_worker_clear (etp_worker *wrk) 415etp_worker_clear (etp_worker *wrk)
256{ 416{
257 ETP_WORKER_CLEAR (wrk);
258} 417}
259 418
419static void ecb_cold
260static void etp_worker_free (etp_worker *wrk) 420etp_worker_free (etp_worker *wrk)
261{ 421{
422 free (wrk->tmpbuf.ptr);
423
262 wrk->next->prev = wrk->prev; 424 wrk->next->prev = wrk->prev;
263 wrk->prev->next = wrk->next; 425 wrk->prev->next = wrk->next;
264 426
265 free (wrk); 427 free (wrk);
266} 428}
267 429
268static unsigned int etp_nreqs (void) 430static unsigned int
431etp_nreqs (void)
269{ 432{
270 int retval; 433 int retval;
271 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 434 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
272 retval = nreqs; 435 retval = nreqs;
273 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 436 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
274 return retval; 437 return retval;
275} 438}
276 439
277static unsigned int etp_nready (void) 440static unsigned int
441etp_nready (void)
278{ 442{
279 unsigned int retval; 443 unsigned int retval;
280 444
281 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 445 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
282 retval = nready; 446 retval = nready;
283 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 447 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
284 448
285 return retval; 449 return retval;
286} 450}
287 451
288static unsigned int etp_npending (void) 452static unsigned int
453etp_npending (void)
289{ 454{
290 unsigned int retval; 455 unsigned int retval;
291 456
292 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 457 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
293 retval = npending; 458 retval = npending;
294 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 459 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
295 460
296 return retval; 461 return retval;
297} 462}
298 463
299static unsigned int etp_nthreads (void) 464static unsigned int
465etp_nthreads (void)
300{ 466{
301 unsigned int retval; 467 unsigned int retval;
302 468
303 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 469 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
304 retval = started; 470 retval = started;
318} etp_reqq; 484} etp_reqq;
319 485
320static etp_reqq req_queue; 486static etp_reqq req_queue;
321static etp_reqq res_queue; 487static etp_reqq res_queue;
322 488
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
323static int reqq_push (etp_reqq *q, ETP_REQ *req) 501reqq_push (etp_reqq *q, ETP_REQ *req)
324{ 502{
325 int pri = req->pri; 503 int pri = req->pri;
326 req->next = 0; 504 req->next = 0;
327 505
328 if (q->qe[pri]) 506 if (q->qe[pri])
334 q->qe[pri] = q->qs[pri] = req; 512 q->qe[pri] = q->qs[pri] = req;
335 513
336 return q->size++; 514 return q->size++;
337} 515}
338 516
517static ETP_REQ * ecb_noinline
339static ETP_REQ *reqq_shift (etp_reqq *q) 518reqq_shift (etp_reqq *q)
340{ 519{
341 int pri; 520 int pri;
342 521
343 if (!q->size) 522 if (!q->size)
344 return 0; 523 return 0;
359 } 538 }
360 539
361 abort (); 540 abort ();
362} 541}
363 542
364static void etp_atfork_prepare (void) 543static int ecb_cold
544etp_init (void (*want_poll)(void), void (*done_poll)(void))
365{ 545{
366 X_LOCK (wrklock); 546 X_MUTEX_CREATE (wrklock);
367 X_LOCK (reqlock); 547 X_MUTEX_CREATE (reslock);
368 X_LOCK (reslock); 548 X_MUTEX_CREATE (reqlock);
369#if !HAVE_PREADWRITE 549 X_COND_CREATE (reqwait);
370 X_LOCK (preadwritelock);
371#endif
372}
373 550
374static void etp_atfork_parent (void) 551 reqq_init (&req_queue);
375{ 552 reqq_init (&res_queue);
376#if !HAVE_PREADWRITE
377 X_UNLOCK (preadwritelock);
378#endif
379 X_UNLOCK (reslock);
380 X_UNLOCK (reqlock);
381 X_UNLOCK (wrklock);
382}
383 553
384static void etp_atfork_child (void) 554 wrk_first.next =
385{ 555 wrk_first.prev = &wrk_first;
386 ETP_REQ *prv;
387
388 while ((prv = reqq_shift (&req_queue)))
389 ETP_DESTROY (prv);
390
391 while ((prv = reqq_shift (&res_queue)))
392 ETP_DESTROY (prv);
393
394 while (wrk_first.next != &wrk_first)
395 {
396 etp_worker *wrk = wrk_first.next;
397
398 if (wrk->req)
399 ETP_DESTROY (wrk->req);
400
401 etp_worker_clear (wrk);
402 etp_worker_free (wrk);
403 }
404 556
405 started = 0; 557 started = 0;
406 idle = 0; 558 idle = 0;
407 nreqs = 0; 559 nreqs = 0;
408 nready = 0; 560 nready = 0;
409 npending = 0; 561 npending = 0;
410 562
411 etp_atfork_parent ();
412}
413
414static void
415etp_once_init (void)
416{
417 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
418}
419
420static int
421etp_init (void (*want_poll)(void), void (*done_poll)(void))
422{
423 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
424
425 pthread_once (&doinit, etp_once_init);
426
427 want_poll_cb = want_poll; 563 want_poll_cb = want_poll;
428 done_poll_cb = done_poll; 564 done_poll_cb = done_poll;
429 565
430 return 0; 566 return 0;
431} 567}
432 568
433X_THREAD_PROC (etp_proc); 569X_THREAD_PROC (etp_proc);
434 570
571static void ecb_cold
435static void etp_start_thread (void) 572etp_start_thread (void)
436{ 573{
437 etp_worker *wrk = calloc (1, sizeof (etp_worker)); 574 etp_worker *wrk = calloc (1, sizeof (etp_worker));
438 575
439 /*TODO*/ 576 /*TODO*/
440 assert (("unable to allocate worker thread data", wrk)); 577 assert (("unable to allocate worker thread data", wrk));
453 free (wrk); 590 free (wrk);
454 591
455 X_UNLOCK (wrklock); 592 X_UNLOCK (wrklock);
456} 593}
457 594
595static void
458static void etp_maybe_start_thread (void) 596etp_maybe_start_thread (void)
459{ 597{
460 if (expect_true (etp_nthreads () >= wanted)) 598 if (ecb_expect_true (etp_nthreads () >= wanted))
461 return; 599 return;
462 600
463 /* todo: maybe use idle here, but might be less exact */ 601 /* todo: maybe use idle here, but might be less exact */
464 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 ()))
465 return; 603 return;
466 604
467 etp_start_thread (); 605 etp_start_thread ();
468} 606}
469 607
608static void ecb_cold
470static void etp_end_thread (void) 609etp_end_thread (void)
471{ 610{
472 eio_req *req = calloc (1, sizeof (eio_req)); 611 eio_req *req = calloc (1, sizeof (eio_req)); /* will be freed by worker */
473 612
474 req->type = -1; 613 req->type = -1;
475 req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 614 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
476 615
477 X_LOCK (reqlock); 616 X_LOCK (reqlock);
482 X_LOCK (wrklock); 621 X_LOCK (wrklock);
483 --started; 622 --started;
484 X_UNLOCK (wrklock); 623 X_UNLOCK (wrklock);
485} 624}
486 625
487static int etp_poll (void) 626static int
627etp_poll (void)
488{ 628{
489 unsigned int maxreqs; 629 unsigned int maxreqs;
490 unsigned int maxtime; 630 unsigned int maxtime;
491 struct timeval tv_start, tv_now; 631 struct timeval tv_start, tv_now;
492 632
522 662
523 X_LOCK (reqlock); 663 X_LOCK (reqlock);
524 --nreqs; 664 --nreqs;
525 X_UNLOCK (reqlock); 665 X_UNLOCK (reqlock);
526 666
527 if (expect_false (req->type == EIO_GROUP && req->size)) 667 if (ecb_expect_false (req->type == EIO_GROUP && req->size))
528 { 668 {
529 req->int1 = 1; /* mark request as delayed */ 669 req->int1 = 1; /* mark request as delayed */
530 continue; 670 continue;
531 } 671 }
532 else 672 else
533 { 673 {
534 int res = ETP_FINISH (req); 674 int res = ETP_FINISH (req);
535 if (expect_false (res)) 675 if (ecb_expect_false (res))
536 return res; 676 return res;
537 } 677 }
538 678
539 if (expect_false (maxreqs && !--maxreqs)) 679 if (ecb_expect_false (maxreqs && !--maxreqs))
540 break; 680 break;
541 681
542 if (maxtime) 682 if (maxtime)
543 { 683 {
544 gettimeofday (&tv_now, 0); 684 gettimeofday (&tv_now, 0);
550 690
551 errno = EAGAIN; 691 errno = EAGAIN;
552 return -1; 692 return -1;
553} 693}
554 694
695static void
555static void etp_cancel (ETP_REQ *req) 696etp_cancel (ETP_REQ *req)
556{ 697{
557 X_LOCK (wrklock); 698 req->cancelled = 1;
558 req->flags |= EIO_FLAG_CANCELLED;
559 X_UNLOCK (wrklock);
560 699
561 eio_grp_cancel (req); 700 eio_grp_cancel (req);
562} 701}
563 702
703static void
564static void etp_submit (ETP_REQ *req) 704etp_submit (ETP_REQ *req)
565{ 705{
566 req->pri -= ETP_PRI_MIN; 706 req->pri -= ETP_PRI_MIN;
567 707
568 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;
569 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;
570 710
571 if (expect_false (req->type == EIO_GROUP)) 711 if (ecb_expect_false (req->type == EIO_GROUP))
572 { 712 {
573 /* I hope this is worth it :/ */ 713 /* I hope this is worth it :/ */
574 X_LOCK (reqlock); 714 X_LOCK (reqlock);
575 ++nreqs; 715 ++nreqs;
576 X_UNLOCK (reqlock); 716 X_UNLOCK (reqlock);
595 735
596 etp_maybe_start_thread (); 736 etp_maybe_start_thread ();
597 } 737 }
598} 738}
599 739
740static void ecb_cold
600static void etp_set_max_poll_time (double nseconds) 741etp_set_max_poll_time (double nseconds)
601{ 742{
602 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 743 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
603 max_poll_time = nseconds; 744 max_poll_time = nseconds * EIO_TICKS;
604 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 745 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
605} 746}
606 747
748static void ecb_cold
607static void etp_set_max_poll_reqs (unsigned int maxreqs) 749etp_set_max_poll_reqs (unsigned int maxreqs)
608{ 750{
609 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 751 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
610 max_poll_reqs = maxreqs; 752 max_poll_reqs = maxreqs;
611 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 753 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
612} 754}
613 755
756static void ecb_cold
614static void etp_set_max_idle (unsigned int nthreads) 757etp_set_max_idle (unsigned int nthreads)
615{ 758{
616 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 759 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
617 max_idle = nthreads <= 0 ? 1 : nthreads; 760 max_idle = nthreads;
618 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 761 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
619} 762}
620 763
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
621static void etp_set_min_parallel (unsigned int nthreads) 773etp_set_min_parallel (unsigned int nthreads)
622{ 774{
623 if (wanted < nthreads) 775 if (wanted < nthreads)
624 wanted = nthreads; 776 wanted = nthreads;
625} 777}
626 778
779static void ecb_cold
627static void etp_set_max_parallel (unsigned int nthreads) 780etp_set_max_parallel (unsigned int nthreads)
628{ 781{
629 if (wanted > nthreads) 782 if (wanted > nthreads)
630 wanted = nthreads; 783 wanted = nthreads;
631 784
632 while (started > wanted) 785 while (started > wanted)
633 etp_end_thread (); 786 etp_end_thread ();
634} 787}
635 788
636/*****************************************************************************/ 789/*****************************************************************************/
637 790
791static void
638static void grp_try_feed (eio_req *grp) 792grp_try_feed (eio_req *grp)
639{ 793{
640 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 794 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
641 { 795 {
642 grp->flags &= ~EIO_FLAG_GROUPADD; 796 grp->flags &= ~EIO_FLAG_GROUPADD;
643 797
650 break; 804 break;
651 } 805 }
652 } 806 }
653} 807}
654 808
809static int
655static int grp_dec (eio_req *grp) 810grp_dec (eio_req *grp)
656{ 811{
657 --grp->size; 812 --grp->size;
658 813
659 /* call feeder, if applicable */ 814 /* call feeder, if applicable */
660 grp_try_feed (grp); 815 grp_try_feed (grp);
664 return eio_finish (grp); 819 return eio_finish (grp);
665 else 820 else
666 return 0; 821 return 0;
667} 822}
668 823
824static void
669void eio_destroy (eio_req *req) 825eio_destroy (eio_req *req)
670{ 826{
671 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); 827 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
672 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); 828 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
673 829
674 EIO_DESTROY (req); 830 EIO_DESTROY (req);
675} 831}
676 832
833static int
677static int eio_finish (eio_req *req) 834eio_finish (eio_req *req)
678{ 835{
679 int res = EIO_FINISH (req); 836 int res = EIO_FINISH (req);
680 837
681 if (req->grp) 838 if (req->grp)
682 { 839 {
690 if (grp->grp_first == req) 847 if (grp->grp_first == req)
691 grp->grp_first = req->grp_next; 848 grp->grp_first = req->grp_next;
692 849
693 res2 = grp_dec (grp); 850 res2 = grp_dec (grp);
694 851
695 if (!res && res2) 852 if (!res)
696 res = res2; 853 res = res2;
697 } 854 }
698 855
699 eio_destroy (req); 856 eio_destroy (req);
700 857
701 return res; 858 return res;
702} 859}
703 860
861void
704void eio_grp_cancel (eio_req *grp) 862eio_grp_cancel (eio_req *grp)
705{ 863{
706 for (grp = grp->grp_first; grp; grp = grp->grp_next) 864 for (grp = grp->grp_first; grp; grp = grp->grp_next)
707 eio_cancel (grp); 865 eio_cancel (grp);
708} 866}
709 867
868void
710void eio_cancel (eio_req *req) 869eio_cancel (eio_req *req)
711{ 870{
712 etp_cancel (req); 871 etp_cancel (req);
713} 872}
714 873
874void
715void eio_submit (eio_req *req) 875eio_submit (eio_req *req)
716{ 876{
717 etp_submit (req); 877 etp_submit (req);
718} 878}
719 879
720unsigned int eio_nreqs (void) 880unsigned int
881eio_nreqs (void)
721{ 882{
722 return etp_nreqs (); 883 return etp_nreqs ();
723} 884}
724 885
725unsigned int eio_nready (void) 886unsigned int
887eio_nready (void)
726{ 888{
727 return etp_nready (); 889 return etp_nready ();
728} 890}
729 891
730unsigned int eio_npending (void) 892unsigned int
893eio_npending (void)
731{ 894{
732 return etp_npending (); 895 return etp_npending ();
733} 896}
734 897
735unsigned int eio_nthreads (void) 898unsigned int ecb_cold
899eio_nthreads (void)
736{ 900{
737 return etp_nthreads (); 901 return etp_nthreads ();
738} 902}
739 903
904void ecb_cold
740void eio_set_max_poll_time (double nseconds) 905eio_set_max_poll_time (double nseconds)
741{ 906{
742 etp_set_max_poll_time (nseconds); 907 etp_set_max_poll_time (nseconds);
743} 908}
744 909
910void ecb_cold
745void eio_set_max_poll_reqs (unsigned int maxreqs) 911eio_set_max_poll_reqs (unsigned int maxreqs)
746{ 912{
747 etp_set_max_poll_reqs (maxreqs); 913 etp_set_max_poll_reqs (maxreqs);
748} 914}
749 915
916void ecb_cold
750void eio_set_max_idle (unsigned int nthreads) 917eio_set_max_idle (unsigned int nthreads)
751{ 918{
752 etp_set_max_idle (nthreads); 919 etp_set_max_idle (nthreads);
753} 920}
754 921
922void ecb_cold
923eio_set_idle_timeout (unsigned int seconds)
924{
925 etp_set_idle_timeout (seconds);
926}
927
928void ecb_cold
755void eio_set_min_parallel (unsigned int nthreads) 929eio_set_min_parallel (unsigned int nthreads)
756{ 930{
757 etp_set_min_parallel (nthreads); 931 etp_set_min_parallel (nthreads);
758} 932}
759 933
934void ecb_cold
760void eio_set_max_parallel (unsigned int nthreads) 935eio_set_max_parallel (unsigned int nthreads)
761{ 936{
762 etp_set_max_parallel (nthreads); 937 etp_set_max_parallel (nthreads);
763} 938}
764 939
765int eio_poll (void) 940int eio_poll (void)
774# undef pread 949# undef pread
775# undef pwrite 950# undef pwrite
776# define pread eio__pread 951# define pread eio__pread
777# define pwrite eio__pwrite 952# define pwrite eio__pwrite
778 953
779static ssize_t 954static eio_ssize_t
780eio__pread (int fd, void *buf, size_t count, off_t offset) 955eio__pread (int fd, void *buf, size_t count, off_t offset)
781{ 956{
782 ssize_t res; 957 eio_ssize_t res;
783 off_t ooffset; 958 off_t ooffset;
784 959
785 X_LOCK (preadwritelock); 960 X_LOCK (preadwritelock);
786 ooffset = lseek (fd, 0, SEEK_CUR); 961 ooffset = lseek (fd, 0, SEEK_CUR);
787 lseek (fd, offset, SEEK_SET); 962 lseek (fd, offset, SEEK_SET);
790 X_UNLOCK (preadwritelock); 965 X_UNLOCK (preadwritelock);
791 966
792 return res; 967 return res;
793} 968}
794 969
795static ssize_t 970static eio_ssize_t
796eio__pwrite (int fd, void *buf, size_t count, off_t offset) 971eio__pwrite (int fd, void *buf, size_t count, off_t offset)
797{ 972{
798 ssize_t res; 973 eio_ssize_t res;
799 off_t ooffset; 974 off_t ooffset;
800 975
801 X_LOCK (preadwritelock); 976 X_LOCK (preadwritelock);
802 ooffset = lseek (fd, 0, SEEK_CUR); 977 ooffset = lseek (fd, 0, SEEK_CUR);
803 lseek (fd, offset, SEEK_SET); 978 lseek (fd, offset, SEEK_SET);
807 982
808 return res; 983 return res;
809} 984}
810#endif 985#endif
811 986
812#ifndef HAVE_FUTIMES 987#ifndef HAVE_UTIMES
813 988
814# undef utimes 989# undef utimes
815# undef futimes
816# define utimes(path,times) eio__utimes (path, times) 990# define utimes(path,times) eio__utimes (path, times)
817# define futimes(fd,times) eio__futimes (fd, times)
818 991
819static int 992static int
820eio__utimes (const char *filename, const struct timeval times[2]) 993eio__utimes (const char *filename, const struct timeval times[2])
821{ 994{
822 if (times) 995 if (times)
830 } 1003 }
831 else 1004 else
832 return utime (filename, 0); 1005 return utime (filename, 0);
833} 1006}
834 1007
1008#endif
1009
1010#ifndef HAVE_FUTIMES
1011
1012# undef futimes
1013# define futimes(fd,times) eio__futimes (fd, times)
1014
1015static int
835static int eio__futimes (int fd, const struct timeval tv[2]) 1016eio__futimes (int fd, const struct timeval tv[2])
836{ 1017{
837 errno = ENOSYS; 1018 errno = ENOSYS;
838 return -1; 1019 return -1;
839} 1020}
840 1021
843#if !HAVE_FDATASYNC 1024#if !HAVE_FDATASYNC
844# undef fdatasync 1025# undef fdatasync
845# define fdatasync(fd) fsync (fd) 1026# define fdatasync(fd) fsync (fd)
846#endif 1027#endif
847 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
848/* sync_file_range always needs emulation */ 1047/* sync_file_range always needs emulation */
849int 1048static int
850eio__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)
851{ 1050{
852#if HAVE_SYNC_FILE_RANGE 1051#if HAVE_SYNC_FILE_RANGE
853 int res; 1052 int res;
854 1053
867 if (!res || errno != ENOSYS) 1066 if (!res || errno != ENOSYS)
868 return res; 1067 return res;
869#endif 1068#endif
870 1069
871 /* 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
872 * call fdatasync, as thta matches the expectation of it's users best */ 1071 * call fdatasync, as that matches the expectation of its users best */
873 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
874} 1084}
875 1085
876#if !HAVE_READAHEAD 1086#if !HAVE_READAHEAD
877# undef readahead 1087# undef readahead
878# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 1088# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
879 1089
880static ssize_t 1090static eio_ssize_t
881eio__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)
882{ 1092{
883 size_t todo = count; 1093 size_t todo = count;
884 dBUF; 1094 dBUF;
885 1095
890 pread (fd, eio_buf, len, offset); 1100 pread (fd, eio_buf, len, offset);
891 offset += len; 1101 offset += len;
892 todo -= len; 1102 todo -= len;
893 } 1103 }
894 1104
895 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 */
896 return count; 1109 return 0;
897} 1110}
898 1111
899#endif 1112#endif
900 1113
901/* sendfile always needs emulation */ 1114/* sendfile always needs emulation */
902static ssize_t 1115static eio_ssize_t
903eio__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)
904{ 1117{
1118 eio_ssize_t written = 0;
905 ssize_t res; 1119 eio_ssize_t res;
906 1120
907 if (!count) 1121 if (!count)
908 return 0; 1122 return 0;
909 1123
1124 for (;;)
1125 {
1126#ifdef __APPLE__
1127# undef HAVE_SENDFILE /* broken, as everything on os x */
1128#endif
910#if HAVE_SENDFILE 1129#if HAVE_SENDFILE
911# if __linux 1130# if __linux
1131 off_t soffset = offset;
912 res = sendfile (ofd, ifd, &offset, count); 1132 res = sendfile (ofd, ifd, &soffset, count);
913 1133
914# elif __freebsd 1134# elif __FreeBSD__
915 /* 1135 /*
916 * 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
917 * wasted on making it similar to other I/O functions. 1137 * wasted on making it similar to other I/O functions.
918 */ 1138 */
919 {
920 off_t sbytes; 1139 off_t sbytes;
921 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 1140 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
922 1141
923 if (res < 0 && sbytes) 1142 #if 0 /* according to the manpage, this is correct, but broken behaviour */
924 /* 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)
925 res = sbytes; 1147 res = sbytes;
926 } 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;
927 1161
928# elif __hpux 1162# elif __hpux
929 res = sendfile (ofd, ifd, offset, count, 0, 0); 1163 res = sendfile (ofd, ifd, offset, count, 0, 0);
930 1164
931# elif __solaris 1165# elif __solaris
932 {
933 struct sendfilevec vec; 1166 struct sendfilevec vec;
934 size_t sbytes; 1167 size_t sbytes;
935 1168
936 vec.sfv_fd = ifd; 1169 vec.sfv_fd = ifd;
937 vec.sfv_flag = 0; 1170 vec.sfv_flag = 0;
938 vec.sfv_off = offset; 1171 vec.sfv_off = offset;
939 vec.sfv_len = count; 1172 vec.sfv_len = count;
940 1173
941 res = sendfilev (ofd, &vec, 1, &sbytes); 1174 res = sendfilev (ofd, &vec, 1, &sbytes);
942 1175
943 if (res < 0 && sbytes) 1176 if (res < 0 && sbytes)
944 res = sbytes; 1177 res = sbytes;
945 }
946 1178
947# 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
948#else 1190#else
949 res = -1; 1191 res = -1;
950 errno = ENOSYS; 1192 errno = ENOSYS;
951#endif 1193#endif
952 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
953 if (res < 0 1218 if (res < 0
954 && (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
955#if __solaris 1227#if __solaris
956 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1228 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
957#endif 1229#endif
958 ) 1230 )
959 ) 1231 )
963 1235
964 res = 0; 1236 res = 0;
965 1237
966 while (count) 1238 while (count)
967 { 1239 {
968 ssize_t cnt; 1240 eio_ssize_t cnt;
969 1241
970 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);
971 1243
972 if (cnt <= 0) 1244 if (cnt <= 0)
973 { 1245 {
985 1257
986 offset += cnt; 1258 offset += cnt;
987 res += cnt; 1259 res += cnt;
988 count -= cnt; 1260 count -= cnt;
989 } 1261 }
1262
1263 FUBd;
990 } 1264 }
991 1265
992 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
1388/* result will always end up in tmpbuf, there is always space for adding a 0-byte */
1389static int
1390eio__realpath (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
1391{
1392 const char *rel = path;
1393 char *res;
1394 char *tmp1, *tmp2;
1395#if SYMLOOP_MAX > 32
1396 int symlinks = SYMLOOP_MAX;
1397#else
1398 int symlinks = 32;
1399#endif
1400
1401 errno = EINVAL;
1402 if (!rel)
1403 return -1;
1404
1405 errno = ENOENT;
1406 if (!*rel)
1407 return -1;
1408
1409 res = tmpbuf_get (tmpbuf, PATH_MAX * 3);
1410 tmp1 = res + PATH_MAX;
1411 tmp2 = tmp1 + PATH_MAX;
1412
1413#if 0 /* disabled, the musl way to do things is just too racy */
1414#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1415 /* on linux we may be able to ask the kernel */
1416 {
1417 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1418
1419 if (fd >= 0)
1420 {
1421 sprintf (tmp1, "/proc/self/fd/%d", fd);
1422 req->result = readlink (tmp1, res, PATH_MAX);
1423 close (fd);
1424
1425 /* here we should probably stat the open file and the disk file, to make sure they still match */
1426
1427 if (req->result > 0)
1428 goto done;
1429 }
1430 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1431 return;
1432 }
1433#endif
1434#endif
1435
1436 if (*rel != '/')
1437 {
1438 int len;
1439
1440 errno = ENOENT;
1441 if (wd == EIO_INVALID_WD)
1442 return -1;
1443
1444 if (wd == EIO_CWD)
1445 {
1446 if (!getcwd (res, PATH_MAX))
1447 return -1;
1448
1449 len = strlen (res);
1450 }
1451 else
1452 memcpy (res, wd->str, len = wd->len);
1453
1454 if (res [1]) /* only use if not / */
1455 res += len;
1456 }
1457
1458 while (*rel)
1459 {
1460 eio_ssize_t len, linklen;
1461 const char *beg = rel;
1462
1463 while (*rel && *rel != '/')
1464 ++rel;
1465
1466 len = rel - beg;
1467
1468 if (!len) /* skip slashes */
1469 {
1470 ++rel;
1471 continue;
1472 }
1473
1474 if (beg [0] == '.')
1475 {
1476 if (len == 1)
1477 continue; /* . - nop */
1478
1479 if (beg [1] == '.' && len == 2)
1480 {
1481 /* .. - back up one component, if possible */
1482
1483 while (res != tmpbuf->ptr)
1484 if (*--res == '/')
1485 break;
1486
1487 continue;
1488 }
1489 }
1490
1491 errno = ENAMETOOLONG;
1492 if (res + 1 + len + 1 >= tmp1)
1493 return -1;
1494
1495 /* copy one component */
1496 *res = '/';
1497 memcpy (res + 1, beg, len);
1498
1499 /* zero-terminate, for readlink */
1500 res [len + 1] = 0;
1501
1502 /* now check if it's a symlink */
1503 linklen = readlink (tmpbuf->ptr, tmp1, PATH_MAX);
1504
1505 if (linklen < 0)
1506 {
1507 if (errno != EINVAL)
1508 return -1;
1509
1510 /* it's a normal directory. hopefully */
1511 res += len + 1;
1512 }
1513 else
1514 {
1515 /* yay, it was a symlink - build new path in tmp2 */
1516 int rellen = strlen (rel);
1517
1518 errno = ENAMETOOLONG;
1519 if (linklen + 1 + rellen >= PATH_MAX)
1520 return -1;
1521
1522 errno = ELOOP;
1523 if (!--symlinks)
1524 return -1;
1525
1526 if (*tmp1 == '/')
1527 res = tmpbuf->ptr; /* symlink resolves to an absolute path */
1528
1529 /* we need to be careful, as rel might point into tmp2 already */
1530 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1531 tmp2 [linklen] = '/';
1532 memcpy (tmp2, tmp1, linklen);
1533
1534 rel = tmp2;
1535 }
1536 }
1537
1538 /* special case for the lone root path */
1539 if (res == tmpbuf->ptr)
1540 *res++ = '/';
1541
1542 return res - (char *)tmpbuf->ptr;
993} 1543}
994 1544
995static signed char 1545static signed char
996eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) 1546eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
997{ 1547{
998 return b->score - a->score ? b->score - a->score /* works because our signed char is always 0..100 */ 1548 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
999 : a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0; 1549 : a->inode < b->inode ? -1
1550 : a->inode > b->inode ? 1
1551 : 0;
1000} 1552}
1001 1553
1554#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1555
1002#define EIO_QSORT_CUTOFF 20 /* quite high, but performs well on many filesystems */ 1556#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1557#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1003 1558
1004static void 1559static void
1005eio_dent_sort (eio_dirent *dents, int size) 1560eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1006{ 1561{
1562 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1563 unsigned char *bit = bits;
1564
1565 assert (CHAR_BIT == 8);
1566 assert (sizeof (eio_dirent) * 8 < 256);
1567 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1568 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1569
1570 if (size <= EIO_SORT_FAST)
1571 return;
1572
1573 /* first prepare an array of bits to test in our radix sort */
1574 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1575 /* inode_bits must contain all inodes ORed together */
1576 /* which is used to skip bits that are 0 everywhere, which is very common */
1577 {
1578 eio_ino_t endianness;
1007 int i, j; 1579 int i, j;
1008 1580
1009 if (size > EIO_QSORT_CUTOFF * 3) /* skip quicksort for small directories */ 1581 /* we store the byte offset of byte n into byte n of "endianness" */
1582 for (i = 0; i < sizeof (eio_ino_t); ++i)
1583 ((unsigned char *)&endianness)[i] = i;
1584
1585 *bit++ = 0;
1586
1587 for (i = 0; i < sizeof (eio_ino_t); ++i)
1588 {
1589 /* shifting off the byte offsets out of "endianness" */
1590 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1591 endianness >>= 8;
1592
1593 for (j = 0; j < 8; ++j)
1594 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1595 *bit++ = offs + j;
1596 }
1597
1598 for (j = 0; j < 8; ++j)
1599 if (score_bits & (1 << j))
1600 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1601 }
1602
1603 /* now actually do the sorting (a variant of MSD radix sort) */
1604 {
1605 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1606 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1607 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1608 int stk_idx = 0;
1609
1610 base_stk [stk_idx] = dents;
1611 end_stk [stk_idx] = dents + size;
1612 bit_stk [stk_idx] = bit - 1;
1613
1614 do
1615 {
1616 base = base_stk [stk_idx];
1617 end = end_stk [stk_idx];
1618 bit = bit_stk [stk_idx];
1619
1620 for (;;)
1621 {
1622 unsigned char O = *bit >> 3;
1623 unsigned char M = 1 << (*bit & 7);
1624
1625 eio_dirent *a = base;
1626 eio_dirent *b = end;
1627
1628 if (b - a < EIO_SORT_CUTOFF)
1629 break;
1630
1631 /* now bit-partition the array on the bit */
1632 /* this ugly asymmetric loop seems to perform much better than typical */
1633 /* partition algos found in the literature */
1634 do
1635 if (!(((unsigned char *)a)[O] & M))
1636 ++a;
1637 else if (!(((unsigned char *)--b)[O] & M))
1638 {
1639 eio_dirent tmp = *a; *a = *b; *b = tmp;
1640 ++a;
1641 }
1642 while (b > a);
1643
1644 /* next bit, or stop, if no bits left in this path */
1645 if (!*--bit)
1646 break;
1647
1648 base_stk [stk_idx] = a;
1649 end_stk [stk_idx] = end;
1650 bit_stk [stk_idx] = bit;
1651 ++stk_idx;
1652
1653 end = a;
1654 }
1655 }
1656 while (stk_idx--);
1657 }
1658}
1659
1660static void
1661eio_dent_insertion_sort (eio_dirent *dents, int size)
1662{
1663 /* first move the smallest element to the front, to act as a sentinel */
1664 {
1665 int i;
1666 eio_dirent *min = dents;
1667
1668 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1669 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1670 if (EIO_DENT_CMP (dents [i], <, *min))
1671 min = &dents [i];
1672
1673 /* swap elements 0 and j (minimum) */
1010 { 1674 {
1011 /* first, use quicksort */ 1675 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1012 /* should be good for 2**31 entries */
1013 struct rng { int l, r; } rng [32];
1014
1015 i = 0;
1016 rng[0].l = 0;
1017 rng[0].r = size;
1018
1019 while (expect_true (i >= 0))
1020 {
1021 int L = rng [i].l;
1022 int R = rng [i].r - 1;
1023
1024 if (expect_false (L + EIO_QSORT_CUTOFF < R))
1025 {
1026 eio_dirent piv = dents [L];
1027
1028 while (L < R)
1029 {
1030 while (eio_dent_cmp (&dents [R], &piv) >= 0 && L < R)
1031 --R;
1032
1033 if (L < R)
1034 dents [L++] = dents [R];
1035
1036 while (eio_dent_cmp (&dents [L], &piv) <= 0 && L < R)
1037 ++L;
1038
1039 if (L < R)
1040 dents [R--] = dents [L];
1041 }
1042
1043 dents [L] = piv;
1044
1045 ++i;
1046 rng [i].l = L + 1;
1047 rng [i].r = rng [i - 1].r;
1048 rng [i - 1].r = L;
1049
1050 if (rng [i].r - rng [i].l > rng [i - 1].r - rng [i - 1].l)
1051 {
1052 struct rng t;
1053
1054 t = rng [i]; rng [i] = rng [i - 1]; rng [i - 1] = t;
1055 }
1056 }
1057 else
1058 --i;
1059 }
1060 } 1676 }
1677 }
1061 1678
1062 /* use a simple insertion sort at the end */ 1679 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1680 {
1681 eio_dirent *i, *j;
1682
1063 for (i = 1; i < size; ++i) 1683 for (i = dents + 1; i < dents + size; ++i)
1064 { 1684 {
1065 eio_dirent value = dents [i]; 1685 eio_dirent value = *i;
1066 1686
1067 for (j = i - 1; j >= 0 && eio_dent_cmp (&dents [j], &value) > 0; --j) 1687 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1068 dents [j + 1] = dents [j]; 1688 j [1] = j [0];
1069 1689
1070 dents [j + 1] = value; 1690 j [1] = value;
1071 } 1691 }
1692 }
1693}
1694
1695static void
1696eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1697{
1698 if (size <= 1)
1699 return; /* our insertion sort relies on size > 0 */
1700
1701 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1702 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1703 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1704
1705 /* use an insertion sort at the end, or for small arrays, */
1706 /* as insertion sort is more efficient for small partitions */
1707 eio_dent_insertion_sort (dents, size);
1072} 1708}
1073 1709
1074/* read a full directory */ 1710/* read a full directory */
1075static void 1711static void
1076eio__scandir (eio_req *req, etp_worker *self) 1712eio__scandir (eio_req *req, etp_worker *self)
1077{ 1713{
1078 DIR *dirp;
1079 EIO_STRUCT_DIRENT *entp;
1080 char *name, *names; 1714 char *name, *names;
1081 int namesalloc = 4096; 1715 int namesalloc = 4096 - sizeof (void *) * 4;
1082 int namesoffs = 0; 1716 int namesoffs = 0;
1083 int flags = req->int1; 1717 int flags = req->int1;
1084 eio_dirent *dents = 0; 1718 eio_dirent *dents = 0;
1085 int dentalloc = 128; 1719 int dentalloc = 128;
1086 int dentoffs = 0; 1720 int dentoffs = 0;
1721 eio_ino_t inode_bits = 0;
1722#ifdef _WIN32
1723 HANDLE dirp;
1724 WIN32_FIND_DATA entp;
1725#else
1726 DIR *dirp;
1727 EIO_STRUCT_DIRENT *entp;
1728#endif
1087 1729
1088 req->result = -1; 1730 req->result = -1;
1089 1731
1090 if (!(flags & EIO_READDIR_DENTS)) 1732 if (!(flags & EIO_READDIR_DENTS))
1091 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); 1733 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1092 1734
1093 X_LOCK (wrklock); 1735#ifdef _WIN32
1094 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 1736 {
1737 int len = strlen ((const char *)req->ptr1);
1738 char *path = malloc (MAX_PATH);
1739 const char *fmt;
1740 const char *reqpath = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1741
1742 if (!len)
1743 fmt = "./*";
1744 else if (reqpath[len - 1] == '/' || reqpath[len - 1] == '\\')
1745 fmt = "%s*";
1746 else
1747 fmt = "%s/*";
1748
1749 _snprintf (path, MAX_PATH, fmt, reqpath);
1750 dirp = FindFirstFile (path, &entp);
1751 free (path);
1752
1753 if (dirp == INVALID_HANDLE_VALUE)
1754 {
1755 /* should steal _dosmaperr */
1756 switch (GetLastError ())
1757 {
1758 case ERROR_FILE_NOT_FOUND:
1759 req->result = 0;
1760 break;
1761
1762 case ERROR_INVALID_NAME:
1763 case ERROR_PATH_NOT_FOUND:
1764 case ERROR_NO_MORE_FILES:
1765 errno = ENOENT;
1766 break;
1767
1768 case ERROR_NOT_ENOUGH_MEMORY:
1769 errno = ENOMEM;
1770 break;
1771
1772 default:
1773 errno = EINVAL;
1774 break;
1775 }
1776
1777 return;
1778 }
1779 }
1780#else
1781 #if HAVE_AT
1782 if (req->wd)
1783 {
1784 int fd = openat (WD2FD (req->wd), req->ptr1, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1785
1786 if (fd < 0)
1787 return;
1788
1789 dirp = fdopendir (fd);
1790
1791 if (!dirp)
1792 close (fd);
1793 }
1794 else
1095 self->dirp = dirp = opendir (req->ptr1); 1795 dirp = opendir (req->ptr1);
1796 #else
1797 dirp = opendir (wd_expand (&self->tmpbuf, req->wd, req->ptr1));
1798 #endif
1799
1800 if (!dirp)
1801 return;
1802#endif
1803
1804 if (req->flags & EIO_FLAG_PTR1_FREE)
1805 free (req->ptr1);
1806
1096 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1807 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1097 req->ptr1 = names = malloc (namesalloc);
1098 req->ptr2 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; 1808 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1099 X_UNLOCK (wrklock); 1809 req->ptr2 = names = malloc (namesalloc);
1100 1810
1101 if (dirp && names && (!flags || dents)) 1811 if (!names || (flags && !dents))
1812 return;
1813
1102 for (;;) 1814 for (;;)
1103 { 1815 {
1816 int done;
1817
1818#ifdef _WIN32
1819 done = !dirp;
1820#else
1104 errno = 0; 1821 errno = 0;
1105 entp = readdir (dirp); 1822 entp = readdir (dirp);
1823 done = !entp;
1824#endif
1106 1825
1107 if (!entp) 1826 if (done)
1108 { 1827 {
1828#ifndef _WIN32
1829 int old_errno = errno;
1830 closedir (dirp);
1831 errno = old_errno;
1832
1109 if (errno) 1833 if (errno)
1110 break; 1834 break;
1835#endif
1111 1836
1112 /* sort etc. */ 1837 /* sort etc. */
1113 req->int1 = flags; 1838 req->int1 = flags;
1114 req->result = dentoffs; 1839 req->result = dentoffs;
1115 1840
1116 if (dents)
1117 {
1118 eio_dirent *ent = dents + dentoffs;
1119
1120 while (ent > dents)
1121 (--ent)->name = names + (size_t)ent->name;
1122 }
1123
1124 if (flags & EIO_READDIR_STAT_ORDER 1841 if (flags & EIO_READDIR_STAT_ORDER)
1125 || !(~flags & (EIO_READDIR_DIRS_FIRST | EIO_READDIR_FOUND_UNKNOWN))) 1842 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1126 eio_dent_sort (dents, dentoffs); /* score depends of DIRS_FIRST */
1127 else if (flags & EIO_READDIR_DIRS_FIRST) 1843 else if (flags & EIO_READDIR_DIRS_FIRST)
1844 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1845 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1846 else
1128 { 1847 {
1129 /* in this case, all is known, and we just put dirs first and sort them */ 1848 /* in this case, all is known, and we just put dirs first and sort them */
1130 eio_dirent *ent = dents + dentoffs; 1849 eio_dirent *oth = dents + dentoffs;
1131 eio_dirent *dir = dents; 1850 eio_dirent *dir = dents;
1132 1851
1133 /* now move dirs to the front, and non-dirs to the back */ 1852 /* now partition dirs to the front, and non-dirs to the back */
1134 /* by walking from both sides and swapping if necessary */ 1853 /* by walking from both sides and swapping if necessary */
1135 while (ent > dir) 1854 while (oth > dir)
1136 { 1855 {
1137 if (dir->type == DT_DIR) 1856 if (dir->type == EIO_DT_DIR)
1138 ++dir; 1857 ++dir;
1139 else 1858 else if ((--oth)->type == EIO_DT_DIR)
1140 { 1859 {
1141 --ent;
1142
1143 if (ent->type == DT_DIR)
1144 {
1145 eio_dirent tmp = *dir; 1860 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1146 *dir = *ent;
1147 *ent = tmp;
1148 1861
1149 ++dir; 1862 ++dir;
1150 }
1151 } 1863 }
1152 } 1864 }
1153 1865
1154 /* now sort the dirs only */ 1866 /* now sort the dirs only (dirs all have the same score) */
1155 eio_dent_sort (dents, dir - dents); 1867 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1156 } 1868 }
1157 1869
1158 /* only provide the names array unless DENTS is specified */
1159 if (!(flags & EIO_READDIR_DENTS))
1160 {
1161 X_LOCK (wrklock);
1162 assert (!dents);
1163 req->ptr1 = 0;
1164 req->ptr2 = names;
1165 X_UNLOCK (wrklock);
1166 }
1167
1168 break; 1870 break;
1169 } 1871 }
1170 1872
1171 /* now add the entry to our list(s) */ 1873 /* now add the entry to our list(s) */
1172 name = entp->d_name; 1874 name = D_NAME (entp);
1173 1875
1174 /* skip . and .. entries */ 1876 /* skip . and .. entries */
1175 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1877 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1176 { 1878 {
1177 int len = D_NAMLEN (entp) + 1; 1879 int len = D_NAMLEN (entp) + 1;
1178 1880
1179 while (expect_false (namesoffs + len > namesalloc)) 1881 while (ecb_expect_false (namesoffs + len > namesalloc))
1180 { 1882 {
1181 namesalloc *= 2; 1883 namesalloc *= 2;
1182 X_LOCK (wrklock);
1183 req->ptr1 = names = realloc (names, namesalloc); 1884 req->ptr2 = names = realloc (names, namesalloc);
1184 X_UNLOCK (wrklock);
1185 1885
1186 if (!names) 1886 if (!names)
1187 break; 1887 break;
1188 } 1888 }
1189 1889
1190 memcpy (names + namesoffs, name, len); 1890 memcpy (names + namesoffs, name, len);
1191 1891
1192 if (dents) 1892 if (dents)
1193 { 1893 {
1194 struct eio_dirent *ent; 1894 struct eio_dirent *ent;
1195 1895
1196 if (expect_false (dentoffs == dentalloc)) 1896 if (ecb_expect_false (dentoffs == dentalloc))
1197 { 1897 {
1198 dentalloc *= 2; 1898 dentalloc *= 2;
1199 X_LOCK (wrklock);
1200 req->ptr2 = dents = realloc (dents, dentalloc * sizeof (eio_dirent)); 1899 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1201 X_UNLOCK (wrklock);
1202 1900
1203 if (!dents) 1901 if (!dents)
1204 break; 1902 break;
1205 } 1903 }
1206 1904
1207 ent = dents + dentoffs; 1905 ent = dents + dentoffs;
1208 1906
1209 ent->name = (char *)(size_t)namesoffs; /* rather dirtily we store the offset in the pointer */ 1907 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1210 ent->namelen = len - 1; 1908 ent->namelen = len - 1;
1211 ent->inode = D_INO (entp); 1909 ent->inode = D_INO (entp);
1212 1910
1911 inode_bits |= ent->inode;
1912
1213 switch (D_TYPE (entp)) 1913 switch (D_TYPE (entp))
1214 { 1914 {
1215 default: 1915 default:
1216 ent->type = EIO_DT_UNKNOWN; 1916 ent->type = EIO_DT_UNKNOWN;
1217 flags |= EIO_READDIR_FOUND_UNKNOWN; 1917 flags |= EIO_READDIR_FOUND_UNKNOWN;
1218 break; 1918 break;
1219 1919
1220 #ifdef DT_FIFO 1920 #ifdef DT_FIFO
1221 case DT_FIFO: ent->type = EIO_DT_FIFO; break; 1921 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1222 #endif 1922 #endif
1223 #ifdef DT_CHR 1923 #ifdef DT_CHR
1224 case DT_CHR: ent->type = EIO_DT_CHR; break; 1924 case DT_CHR: ent->type = EIO_DT_CHR; break;
1225 #endif 1925 #endif
1226 #ifdef DT_MPC 1926 #ifdef DT_MPC
1227 case DT_MPC: ent->type = EIO_DT_MPC; break; 1927 case DT_MPC: ent->type = EIO_DT_MPC; break;
1228 #endif 1928 #endif
1229 #ifdef DT_DIR 1929 #ifdef DT_DIR
1230 case DT_DIR: ent->type = EIO_DT_DIR; break; 1930 case DT_DIR: ent->type = EIO_DT_DIR; break;
1231 #endif 1931 #endif
1232 #ifdef DT_NAM 1932 #ifdef DT_NAM
1233 case DT_NAM: ent->type = EIO_DT_NAM; break; 1933 case DT_NAM: ent->type = EIO_DT_NAM; break;
1234 #endif 1934 #endif
1235 #ifdef DT_BLK 1935 #ifdef DT_BLK
1236 case DT_BLK: ent->type = EIO_DT_BLK; break; 1936 case DT_BLK: ent->type = EIO_DT_BLK; break;
1237 #endif 1937 #endif
1238 #ifdef DT_MPB 1938 #ifdef DT_MPB
1239 case DT_MPB: ent->type = EIO_DT_MPB; break; 1939 case DT_MPB: ent->type = EIO_DT_MPB; break;
1240 #endif 1940 #endif
1241 #ifdef DT_REG 1941 #ifdef DT_REG
1242 case DT_REG: ent->type = EIO_DT_REG; break; 1942 case DT_REG: ent->type = EIO_DT_REG; break;
1243 #endif 1943 #endif
1244 #ifdef DT_NWK 1944 #ifdef DT_NWK
1245 case DT_NWK: ent->type = EIO_DT_NWK; break; 1945 case DT_NWK: ent->type = EIO_DT_NWK; break;
1246 #endif 1946 #endif
1247 #ifdef DT_CMP 1947 #ifdef DT_CMP
1248 case DT_CMP: ent->type = EIO_DT_CMP; break; 1948 case DT_CMP: ent->type = EIO_DT_CMP; break;
1249 #endif 1949 #endif
1250 #ifdef DT_LNK 1950 #ifdef DT_LNK
1251 case DT_LNK: ent->type = EIO_DT_LNK; break; 1951 case DT_LNK: ent->type = EIO_DT_LNK; break;
1252 #endif 1952 #endif
1253 #ifdef DT_SOCK 1953 #ifdef DT_SOCK
1254 case DT_SOCK: ent->type = EIO_DT_SOCK; break; 1954 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1255 #endif 1955 #endif
1256 #ifdef DT_DOOR 1956 #ifdef DT_DOOR
1257 case DT_DOOR: ent->type = EIO_DT_DOOR; break; 1957 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1258 #endif 1958 #endif
1259 #ifdef DT_WHT 1959 #ifdef DT_WHT
1260 case DT_WHT: ent->type = EIO_DT_WHT; break; 1960 case DT_WHT: ent->type = EIO_DT_WHT; break;
1261 #endif 1961 #endif
1262 } 1962 }
1263 1963
1264 ent->score = 0; 1964 ent->score = 7;
1265 1965
1266 if (flags & EIO_READDIR_DIRS_FIRST) 1966 if (flags & EIO_READDIR_DIRS_FIRST)
1267 { 1967 {
1268 if (ent->type == EIO_DT_UNKNOWN) 1968 if (ent->type == EIO_DT_UNKNOWN)
1269 { 1969 {
1270 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */ 1970 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1271 ent->score = 98; 1971 ent->score = 1;
1272 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */ 1972 else if (!strchr (name, '.')) /* absence of dots indicate likely dirs */
1273 ent->score = len <= 2 ? len + 6 : len <= 4 ? 5 : len <= 7 ? 4 : 1; /* shorter == more likely dir, but avoid too many classes */ 1973 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1274 } 1974 }
1275 else if (ent->type == EIO_DT_DIR) 1975 else if (ent->type == EIO_DT_DIR)
1276 ent->score = 100; 1976 ent->score = 0;
1277 } 1977 }
1278 } 1978 }
1279 1979
1280 namesoffs += len; 1980 namesoffs += len;
1281 ++dentoffs; 1981 ++dentoffs;
1282 } 1982 }
1283 1983
1284 if (EIO_CANCELLED (req)) 1984 if (EIO_CANCELLED (req))
1285 { 1985 {
1286 errno = ECANCELED; 1986 errno = ECANCELED;
1287 break; 1987 break;
1288 } 1988 }
1989
1990#ifdef _WIN32
1991 if (!FindNextFile (dirp, &entp))
1992 {
1993 FindClose (dirp);
1994 dirp = 0;
1289 } 1995 }
1290}
1291
1292#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1293# undef msync
1294# define msync(a,b,c) ((errno = ENOSYS), -1)
1295#endif 1996#endif
1997 }
1998}
1296 1999
1297int 2000/*****************************************************************************/
1298eio__mtouch (void *mem, size_t len, int flags) 2001/* working directory stuff */
1299{ 2002/* various deficiencies in the posix 2008 api force us to */
1300 intptr_t addr = (intptr_t)mem; 2003/* keep the absolute path in string form at all times */
1301 intptr_t end = addr + len; 2004/* fuck yeah. */
1302#ifdef PAGESIZE
1303 const intptr_t page = PAGESIZE;
1304#else
1305 static intptr_t page;
1306 2005
1307 if (!page) 2006#if !HAVE_AT
1308 page = sysconf (_SC_PAGESIZE);
1309#endif
1310 2007
1311 addr &= ~(page - 1); /* assume page size is always a power of two */ 2008/* a bit like realpath, but usually faster because it doesn'T have to return */
2009/* an absolute or canonical path */
2010static const char *
2011wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
2012{
2013 if (!wd || *path == '/')
2014 return path;
1312 2015
1313 if (addr < end) 2016 if (path [0] == '.' && !path [1])
1314 if (flags) /* modify */ 2017 return wd->str;
1315 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
1316 else
1317 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
1318 2018
2019 {
2020 int l1 = wd->len;
2021 int l2 = strlen (path);
2022
2023 char *res = tmpbuf_get (tmpbuf, l1 + l2 + 2);
2024
2025 memcpy (res, wd->str, l1);
2026 res [l1] = '/';
2027 memcpy (res + l1 + 1, path, l2 + 1);
2028
2029 return res;
2030 }
2031}
2032
2033#endif
2034
2035static eio_wd
2036eio__wd_open_sync (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
2037{
2038 int fd;
2039 eio_wd res;
2040 int len = eio__realpath (tmpbuf, wd, path);
2041
2042 if (len < 0)
2043 return EIO_INVALID_WD;
2044
2045#if HAVE_AT
2046 fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
2047
2048 if (fd < 0)
2049 return EIO_INVALID_WD;
2050#endif
2051
2052 res = malloc (sizeof (*res) + len); /* one extra 0-byte */
2053
2054#if HAVE_AT
2055 res->fd = fd;
2056#endif
2057
2058 res->len = len;
2059 memcpy (res->str, tmpbuf->ptr, len);
2060 res->str [len] = 0;
2061
2062 return res;
2063}
2064
2065eio_wd
2066eio_wd_open_sync (eio_wd wd, const char *path)
2067{
2068 struct tmpbuf tmpbuf = { 0 };
2069 wd = eio__wd_open_sync (&tmpbuf, wd, path);
2070 free (tmpbuf.ptr);
2071
1319 return 0; 2072 return wd;
1320} 2073}
2074
2075void
2076eio_wd_close_sync (eio_wd wd)
2077{
2078 if (wd != EIO_INVALID_WD && wd != EIO_CWD)
2079 {
2080 #if HAVE_AT
2081 close (wd->fd);
2082 #endif
2083 free (wd);
2084 }
2085}
2086
2087#if HAVE_AT
2088
2089/* they forgot these */
2090
2091static int
2092eio__truncateat (int dirfd, const char *path, off_t length)
2093{
2094 int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
2095 int res;
2096
2097 if (fd < 0)
2098 return fd;
2099
2100 res = ftruncate (fd, length);
2101 close (fd);
2102 return res;
2103}
2104
2105static int
2106eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
2107{
2108 int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
2109 int res;
2110
2111 if (fd < 0)
2112 return fd;
2113
2114 res = fstatvfs (fd, buf);
2115 close (fd);
2116 return res;
2117
2118}
2119
2120#endif
1321 2121
1322/*****************************************************************************/ 2122/*****************************************************************************/
1323 2123
1324#define ALLOC(len) \ 2124#define ALLOC(len) \
1325 if (!req->ptr2) \ 2125 if (!req->ptr2) \
1334 req->result = -1; \ 2134 req->result = -1; \
1335 break; \ 2135 break; \
1336 } \ 2136 } \
1337 } 2137 }
1338 2138
2139static void ecb_noinline ecb_cold
2140etp_proc_init (void)
2141{
2142#if HAVE_PRCTL_SET_NAME
2143 /* provide a more sensible "thread name" */
2144 char name[16 + 1];
2145 const int namelen = sizeof (name) - 1;
2146 int len;
2147
2148 prctl (PR_GET_NAME, (unsigned long)name, 0, 0, 0);
2149 name [namelen] = 0;
2150 len = strlen (name);
2151 strcpy (name + (len <= namelen - 4 ? len : namelen - 4), "/eio");
2152 prctl (PR_SET_NAME, (unsigned long)name, 0, 0, 0);
2153#endif
2154}
2155
1339X_THREAD_PROC (etp_proc) 2156X_THREAD_PROC (etp_proc)
1340{ 2157{
1341 ETP_REQ *req; 2158 ETP_REQ *req;
1342 struct timespec ts; 2159 struct timespec ts;
1343 etp_worker *self = (etp_worker *)thr_arg; 2160 etp_worker *self = (etp_worker *)thr_arg;
1344 2161
2162 etp_proc_init ();
2163
1345 /* try to distribute timeouts somewhat randomly */ 2164 /* try to distribute timeouts somewhat evenly */
1346 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 2165 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1347 2166
1348 for (;;) 2167 for (;;)
1349 { 2168 {
2169 ts.tv_sec = 0;
2170
1350 X_LOCK (reqlock); 2171 X_LOCK (reqlock);
1351 2172
1352 for (;;) 2173 for (;;)
1353 { 2174 {
1354 self->req = req = reqq_shift (&req_queue); 2175 req = reqq_shift (&req_queue);
1355 2176
1356 if (req) 2177 if (req)
1357 break; 2178 break;
1358 2179
2180 if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
2181 {
2182 X_UNLOCK (reqlock);
2183 X_LOCK (wrklock);
2184 --started;
2185 X_UNLOCK (wrklock);
2186 goto quit;
2187 }
2188
1359 ++idle; 2189 ++idle;
1360 2190
1361 ts.tv_sec = time (0) + IDLE_TIMEOUT; 2191 if (idle <= max_idle)
1362 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 2192 /* we are allowed to idle, so do so without any timeout */
2193 X_COND_WAIT (reqwait, reqlock);
2194 else
1363 { 2195 {
1364 if (idle > max_idle) 2196 /* initialise timeout once */
1365 { 2197 if (!ts.tv_sec)
1366 --idle; 2198 ts.tv_sec = time (0) + idle_timeout;
1367 X_UNLOCK (reqlock);
1368 X_LOCK (wrklock);
1369 --started;
1370 X_UNLOCK (wrklock);
1371 goto quit;
1372 }
1373 2199
1374 /* we are allowed to idle, so do so without any timeout */
1375 X_COND_WAIT (reqwait, reqlock); 2200 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
2201 ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1376 } 2202 }
1377 2203
1378 --idle; 2204 --idle;
1379 } 2205 }
1380 2206
1383 X_UNLOCK (reqlock); 2209 X_UNLOCK (reqlock);
1384 2210
1385 if (req->type < 0) 2211 if (req->type < 0)
1386 goto quit; 2212 goto quit;
1387 2213
1388 if (!EIO_CANCELLED (req))
1389 ETP_EXECUTE (self, req); 2214 ETP_EXECUTE (self, req);
1390 2215
1391 X_LOCK (reslock); 2216 X_LOCK (reslock);
1392 2217
1393 ++npending; 2218 ++npending;
1394 2219
1395 if (!reqq_push (&res_queue, req) && want_poll_cb) 2220 if (!reqq_push (&res_queue, req) && want_poll_cb)
1396 want_poll_cb (); 2221 want_poll_cb ();
1397 2222
1398 self->req = 0;
1399 etp_worker_clear (self); 2223 etp_worker_clear (self);
1400 2224
1401 X_UNLOCK (reslock); 2225 X_UNLOCK (reslock);
1402 } 2226 }
1403 2227
1404quit: 2228quit:
2229 free (req);
2230
1405 X_LOCK (wrklock); 2231 X_LOCK (wrklock);
1406 etp_worker_free (self); 2232 etp_worker_free (self);
1407 X_UNLOCK (wrklock); 2233 X_UNLOCK (wrklock);
1408 2234
1409 return 0; 2235 return 0;
1410} 2236}
1411 2237
1412/*****************************************************************************/ 2238/*****************************************************************************/
1413 2239
2240int ecb_cold
1414int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 2241eio_init (void (*want_poll)(void), void (*done_poll)(void))
1415{ 2242{
2243#if !HAVE_PREADWRITE
2244 X_MUTEX_CREATE (preadwritelock);
2245#endif
2246
1416 return etp_init (want_poll, done_poll); 2247 return etp_init (want_poll, done_poll);
1417} 2248}
1418 2249
2250ecb_inline void
1419static void eio_api_destroy (eio_req *req) 2251eio_api_destroy (eio_req *req)
1420{ 2252{
1421 free (req); 2253 free (req);
1422} 2254}
1423 2255
1424#define REQ(rtype) \ 2256#define REQ(rtype) \
1443 { \ 2275 { \
1444 eio_api_destroy (req); \ 2276 eio_api_destroy (req); \
1445 return 0; \ 2277 return 0; \
1446 } 2278 }
1447 2279
2280static void
1448static void eio_execute (etp_worker *self, eio_req *req) 2281eio_execute (etp_worker *self, eio_req *req)
1449{ 2282{
1450 errno = 0; 2283#if HAVE_AT
2284 int dirfd;
2285#else
2286 const char *path;
2287#endif
2288
2289 if (ecb_expect_false (EIO_CANCELLED (req)))
2290 {
2291 req->result = -1;
2292 req->errorno = ECANCELED;
2293 return;
2294 }
2295
2296 if (ecb_expect_false (req->wd == EIO_INVALID_WD))
2297 {
2298 req->result = -1;
2299 req->errorno = ENOENT;
2300 return;
2301 }
2302
2303 if (req->type >= EIO_OPEN)
2304 {
2305 #if HAVE_AT
2306 dirfd = WD2FD (req->wd);
2307 #else
2308 path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
2309 #endif
2310 }
1451 2311
1452 switch (req->type) 2312 switch (req->type)
1453 { 2313 {
2314 case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
2315 req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
2316 break;
2317 case EIO_WD_CLOSE: req->result = 0;
2318 eio_wd_close_sync (req->wd); break;
2319
1454 case EIO_READ: ALLOC (req->size); 2320 case EIO_READ: ALLOC (req->size);
1455 req->result = req->offs >= 0 2321 req->result = req->offs >= 0
1456 ? pread (req->int1, req->ptr2, req->size, req->offs) 2322 ? pread (req->int1, req->ptr2, req->size, req->offs)
1457 : read (req->int1, req->ptr2, req->size); break; 2323 : read (req->int1, req->ptr2, req->size); break;
1458 case EIO_WRITE: req->result = req->offs >= 0 2324 case EIO_WRITE: req->result = req->offs >= 0
1459 ? pwrite (req->int1, req->ptr2, req->size, req->offs) 2325 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1460 : write (req->int1, req->ptr2, req->size); break; 2326 : write (req->int1, req->ptr2, req->size); break;
1461 2327
1462 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break; 2328 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1463 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break; 2329 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
2330
2331#if HAVE_AT
1464 2332
1465 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2333 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1466 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2334 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break;
1467 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2335 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1468 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2336 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, AT_SYMLINK_NOFOLLOW); break;
1469 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1470 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1471
1472 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; 2337 case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break;
1473 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1474 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break; 2338 case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break;
1475 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1476 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break; 2339 case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break;
1477 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1478
1479 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break; 2340 case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break;
1480 case EIO_CLOSE: req->result = close (req->int1); break; 2341
1481 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1482 case EIO_UNLINK: req->result = unlink (req->ptr1); break; 2342 case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
1483 case EIO_RMDIR: req->result = rmdir (req->ptr1); break; 2343 case EIO_RMDIR: req->result = unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break;
1484 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; 2344 case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); break;
1485 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; 2345 case EIO_RENAME: req->result = renameat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2); break;
1486 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; 2346 case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2, 0); break;
1487 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; 2347 case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
1488 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break; 2348 case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1489
1490 case EIO_READLINK: ALLOC (NAME_MAX); 2349 case EIO_READLINK: ALLOC (PATH_MAX);
1491 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; 2350 req->result = readlinkat (dirfd, req->ptr1, req->ptr2, PATH_MAX); break;
1492 2351 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1493 case EIO_SYNC: req->result = 0; sync (); break; 2352 req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1494 case EIO_FSYNC: req->result = fsync (req->int1); break;
1495 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1496 case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break;
1497 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;
1498 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1499
1500 case EIO_READDIR: eio__scandir (req, self); break;
1501
1502 case EIO_BUSY: 2353 case EIO_UTIME:
1503#ifdef _WIN32 2354 case EIO_FUTIME:
1504 Sleep (req->nv1 * 1000.); 2355 {
2356 struct timespec ts[2];
2357 struct timespec *times;
2358
2359 if (req->nv1 != -1. || req->nv2 != -1.)
2360 {
2361 ts[0].tv_sec = req->nv1;
2362 ts[0].tv_nsec = (req->nv1 - ts[0].tv_sec) * 1e9;
2363 ts[1].tv_sec = req->nv2;
2364 ts[1].tv_nsec = (req->nv2 - ts[1].tv_sec) * 1e9;
2365
2366 times = ts;
2367 }
2368 else
2369 times = 0;
2370
2371 req->result = req->type == EIO_FUTIME
2372 ? futimens (req->int1, times)
2373 : utimensat (dirfd, req->ptr1, times, 0);
2374 }
2375 break;
2376
1505#else 2377#else
1506 {
1507 struct timeval tv;
1508 2378
1509 tv.tv_sec = req->nv1; 2379 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1510 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; 2380 req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
2381 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2382 req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
2383 case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
2384 case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
2385 case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
2386 case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
1511 2387
1512 req->result = select (0, 0, 0, 0, &tv); 2388 case EIO_UNLINK: req->result = unlink (path ); break;
1513 } 2389 case EIO_RMDIR: req->result = rmdir (path ); break;
1514#endif 2390 case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
1515 break; 2391 case EIO_RENAME: req->result = rename (path , req->ptr2); break;
2392 case EIO_LINK: req->result = link (path , req->ptr2); break;
2393 case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
2394 case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
2395 case EIO_READLINK: ALLOC (PATH_MAX);
2396 req->result = readlink (path, req->ptr2, PATH_MAX); break;
2397 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2398 req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
1516 2399
1517 case EIO_UTIME: 2400 case EIO_UTIME:
1518 case EIO_FUTIME: 2401 case EIO_FUTIME:
1519 { 2402 {
1520 struct timeval tv[2]; 2403 struct timeval tv[2];
1521 struct timeval *times; 2404 struct timeval *times;
1522 2405
1523 if (req->nv1 != -1. || req->nv2 != -1.) 2406 if (req->nv1 != -1. || req->nv2 != -1.)
1524 { 2407 {
1525 tv[0].tv_sec = req->nv1; 2408 tv[0].tv_sec = req->nv1;
1526 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.; 2409 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
1527 tv[1].tv_sec = req->nv2; 2410 tv[1].tv_sec = req->nv2;
1528 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.; 2411 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
1529 2412
1530 times = tv; 2413 times = tv;
1531 } 2414 }
1532 else 2415 else
1533 times = 0; 2416 times = 0;
1534
1535 2417
1536 req->result = req->type == EIO_FUTIME 2418 req->result = req->type == EIO_FUTIME
1537 ? futimes (req->int1, times) 2419 ? futimes (req->int1, times)
1538 : utimes (req->ptr1, times); 2420 : utimes (req->ptr1, times);
1539 } 2421 }
1540 break; 2422 break;
1541 2423
2424#endif
2425
2426 case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
2427 {
2428 ALLOC (req->result);
2429 memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
2430 }
2431 break;
2432
2433 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2434 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
2435
2436 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2437 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2438
2439 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
2440 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
2441 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
2442
2443 case EIO_CLOSE: req->result = close (req->int1); break;
2444 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
2445 case EIO_SYNC: req->result = 0; sync (); break;
2446 case EIO_FSYNC: req->result = fsync (req->int1); break;
2447 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
2448 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
2449 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
2450 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
2451 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
2452 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
2453 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
2454 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
2455
2456 case EIO_READDIR: eio__scandir (req, self); break;
2457
2458 case EIO_BUSY:
2459#ifdef _WIN32
2460 Sleep (req->nv1 * 1e3);
2461#else
2462 {
2463 struct timeval tv;
2464
2465 tv.tv_sec = req->nv1;
2466 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
2467
2468 req->result = select (0, 0, 0, 0, &tv);
2469 }
2470#endif
2471 break;
2472
1542 case EIO_GROUP: 2473 case EIO_GROUP:
1543 abort (); /* handled in eio_request */ 2474 abort (); /* handled in eio_request */
1544 2475
1545 case EIO_NOP: 2476 case EIO_NOP:
1546 req->result = 0; 2477 req->result = 0;
1547 break; 2478 break;
1548 2479
1549 case EIO_CUSTOM: 2480 case EIO_CUSTOM:
1550 ((void (*)(eio_req *))req->feed) (req); 2481 req->feed (req);
1551 break; 2482 break;
1552 2483
1553 default: 2484 default:
2485 errno = ENOSYS;
1554 req->result = -1; 2486 req->result = -1;
1555 break; 2487 break;
1556 } 2488 }
1557 2489
1558 req->errorno = errno; 2490 req->errorno = errno;
1559} 2491}
1560 2492
1561#ifndef EIO_NO_WRAPPERS 2493#ifndef EIO_NO_WRAPPERS
1562 2494
2495eio_req *eio_wd_open (const char *path, int pri, eio_cb cb, void *data)
2496{
2497 REQ (EIO_WD_OPEN); PATH; SEND;
2498}
2499
2500eio_req *eio_wd_close (eio_wd wd, int pri, eio_cb cb, void *data)
2501{
2502 REQ (EIO_WD_CLOSE); req->wd = wd; SEND;
2503}
2504
1563eio_req *eio_nop (int pri, eio_cb cb, void *data) 2505eio_req *eio_nop (int pri, eio_cb cb, void *data)
1564{ 2506{
1565 REQ (EIO_NOP); SEND; 2507 REQ (EIO_NOP); SEND;
1566} 2508}
1567 2509
1583eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2525eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1584{ 2526{
1585 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2527 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1586} 2528}
1587 2529
2530eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2531{
2532 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2533}
2534
2535eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
2536{
2537 REQ (EIO_SYNCFS); req->int1 = fd; SEND;
2538}
2539
2540eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2541{
2542 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2543}
2544
1588eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2545eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1589{ 2546{
1590 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2547 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1591} 2548}
1592 2549
1593eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) 2550eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
1594{ 2551{
1595 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; 2552 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
1596} 2553}
1597 2554
1598eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) 2555eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
1599{ 2556{
1600 REQ (EIO_FDATASYNC); req->int1 = fd; SEND; 2557 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2558}
2559
2560eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2561{
2562 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
1601} 2563}
1602 2564
1603eio_req *eio_close (int fd, int pri, eio_cb cb, void *data) 2565eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
1604{ 2566{
1605 REQ (EIO_CLOSE); req->int1 = fd; SEND; 2567 REQ (EIO_CLOSE); req->int1 = fd; SEND;
1623eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) 2585eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1624{ 2586{
1625 REQ (EIO_FSTAT); req->int1 = fd; SEND; 2587 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1626} 2588}
1627 2589
2590eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
2591{
2592 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
2593}
2594
1628eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) 2595eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1629{ 2596{
1630 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; 2597 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1631} 2598}
1632 2599
1638eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) 2605eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1639{ 2606{
1640 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2607 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1641} 2608}
1642 2609
1643eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2610eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1644{ 2611{
1645 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2612 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1646} 2613}
1647 2614
1648eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) 2615eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
1668eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data) 2635eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1669{ 2636{
1670 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2637 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1671} 2638}
1672 2639
1673eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2640eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1674{ 2641{
1675 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2642 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1676} 2643}
1677 2644
1678eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) 2645eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
1694eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data) 2661eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1695{ 2662{
1696 return eio__1path (EIO_READLINK, path, pri, cb, data); 2663 return eio__1path (EIO_READLINK, path, pri, cb, data);
1697} 2664}
1698 2665
2666eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2667{
2668 return eio__1path (EIO_REALPATH, path, pri, cb, data);
2669}
2670
1699eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) 2671eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1700{ 2672{
1701 return eio__1path (EIO_STAT, path, pri, cb, data); 2673 return eio__1path (EIO_STAT, path, pri, cb, data);
1702} 2674}
1703 2675
1704eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) 2676eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1705{ 2677{
1706 return eio__1path (EIO_LSTAT, path, pri, cb, data); 2678 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1707} 2679}
1708 2680
2681eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
2682{
2683 return eio__1path (EIO_STATVFS, path, pri, cb, data);
2684}
2685
1709eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) 2686eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1710{ 2687{
1711 return eio__1path (EIO_UNLINK, path, pri, cb, data); 2688 return eio__1path (EIO_UNLINK, path, pri, cb, data);
1712} 2689}
1713 2690
1721 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND; 2698 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1722} 2699}
1723 2700
1724eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) 2701eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1725{ 2702{
1726 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; 2703 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1727} 2704}
1728 2705
1729static eio_req * 2706static eio_req *
1730eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2707eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1731{ 2708{
1755eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2732eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1756{ 2733{
1757 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2734 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1758} 2735}
1759 2736
1760eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) 2737eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
1761{ 2738{
1762 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND; 2739 REQ (EIO_CUSTOM); req->feed = execute; SEND;
1763} 2740}
1764 2741
1765#endif 2742#endif
1766 2743
1767eio_req *eio_grp (eio_cb cb, void *data) 2744eio_req *eio_grp (eio_cb cb, void *data)
1776#undef SEND 2753#undef SEND
1777 2754
1778/*****************************************************************************/ 2755/*****************************************************************************/
1779/* grp functions */ 2756/* grp functions */
1780 2757
2758void
1781void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2759eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1782{ 2760{
1783 grp->int2 = limit; 2761 grp->int2 = limit;
1784 grp->feed = feed; 2762 grp->feed = feed;
1785 2763
1786 grp_try_feed (grp); 2764 grp_try_feed (grp);
1787} 2765}
1788 2766
2767void
1789void eio_grp_limit (eio_req *grp, int limit) 2768eio_grp_limit (eio_req *grp, int limit)
1790{ 2769{
1791 grp->int2 = limit; 2770 grp->int2 = limit;
1792 2771
1793 grp_try_feed (grp); 2772 grp_try_feed (grp);
1794} 2773}
1795 2774
2775void
1796void eio_grp_add (eio_req *grp, eio_req *req) 2776eio_grp_add (eio_req *grp, eio_req *req)
1797{ 2777{
1798 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2778 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1799 2779
1800 grp->flags |= EIO_FLAG_GROUPADD; 2780 grp->flags |= EIO_FLAG_GROUPADD;
1801 2781
1812} 2792}
1813 2793
1814/*****************************************************************************/ 2794/*****************************************************************************/
1815/* misc garbage */ 2795/* misc garbage */
1816 2796
2797eio_ssize_t
1817ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2798eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1818{ 2799{
1819 etp_worker wrk;
1820 ssize_t ret;
1821
1822 wrk.dbuf = 0;
1823
1824 ret = eio__sendfile (ofd, ifd, offset, count, &wrk); 2800 return eio__sendfile (ofd, ifd, offset, count);
1825
1826 if (wrk.dbuf)
1827 free (wrk.dbuf);
1828
1829 return ret;
1830} 2801}
1831 2802

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines