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

Comparing libeio/eio.c (file contents):
Revision 1.53 by root, Sat Jan 9 10:03:09 2010 UTC vs.
Revision 1.105 by root, Mon Sep 26 18:23:18 2011 UTC

1/* 1/*
2 * libeio implementation 2 * libeio implementation
3 * 3 *
4 * Copyright (c) 2007,2008,2009,2010 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>
49#include <stdlib.h> 54#include <stdlib.h>
50#include <string.h> 55#include <string.h>
51#include <errno.h> 56#include <errno.h>
52#include <sys/types.h> 57#include <sys/types.h>
53#include <sys/stat.h> 58#include <sys/stat.h>
54#include <sys/statvfs.h>
55#include <limits.h> 59#include <limits.h>
56#include <fcntl.h> 60#include <fcntl.h>
57#include <assert.h> 61#include <assert.h>
58 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
59#ifndef EIO_FINISH 82#ifndef EIO_FINISH
60# 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
61#endif 84#endif
62 85
63#ifndef EIO_DESTROY 86#ifndef EIO_DESTROY
66 89
67#ifndef EIO_FEED 90#ifndef EIO_FEED
68# 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)
69#endif 92#endif
70 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
71#ifdef _WIN32 105#ifdef _WIN32
72 106
73 /*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
74#else 197#else
75 198
76# include "config.h"
77# include <sys/time.h> 199 #include <sys/time.h>
78# include <sys/select.h> 200 #include <sys/select.h>
79# include <sys/mman.h> 201 #include <sys/statvfs.h>
80# 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
81# include <utime.h> 237# include <utime.h>
82# include <signal.h>
83# include <dirent.h>
84
85/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
86# if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
87# define _DIRENT_HAVE_D_TYPE /* sigh */
88# define D_INO(de) (de)->d_fileno
89# define D_NAMLEN(de) (de)->d_namlen
90# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
91# define D_INO(de) (de)->d_ino
92# endif 238#endif
93 239
94#ifdef _D_EXACT_NAMLEN 240#if HAVE_SYS_SYSCALL_H
95# undef D_NAMLEN 241# include <sys/syscall.h>
96# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
97#endif
98
99# ifdef _DIRENT_HAVE_D_TYPE
100# define D_TYPE(de) (de)->d_type
101# endif 242#endif
102 243
103# ifndef EIO_STRUCT_DIRENT 244#if HAVE_SYS_PRCTL_H
104# define EIO_STRUCT_DIRENT struct dirent 245# include <sys/prctl.h>
105# endif
106
107#endif 246#endif
108 247
109#if HAVE_SENDFILE 248#if HAVE_SENDFILE
110# if __linux 249# if __linux
111# include <sys/sendfile.h> 250# include <sys/sendfile.h>
126#endif 265#endif
127#ifndef D_INO 266#ifndef D_INO
128# define D_INO(de) 0 267# define D_INO(de) 0
129#endif 268#endif
130#ifndef D_NAMLEN 269#ifndef D_NAMLEN
131# define D_NAMLEN(de) strlen ((de)->d_name) 270# define D_NAMLEN(entp) strlen (D_NAME (entp))
132#endif 271#endif
133
134/* number of seconds after which an idle threads exit */
135#define IDLE_TIMEOUT 10
136 272
137/* used for struct dirent, AIX doesn't provide it */ 273/* used for struct dirent, AIX doesn't provide it */
138#ifndef NAME_MAX 274#ifndef NAME_MAX
139# define NAME_MAX 4096 275# define NAME_MAX 4096
140#endif 276#endif
141 277
278/* used for readlink etc. */
279#ifndef PATH_MAX
280# define PATH_MAX 4096
281#endif
282
142/* buffer size for various temporary buffers */ 283/* buffer size for various temporary buffers */
143#define EIO_BUFSIZE 65536 284#define EIO_BUFSIZE 65536
144 285
145#define dBUF \ 286#define dBUF \
146 char *eio_buf; \
147 ETP_WORKER_LOCK (self); \
148 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 287 char *eio_buf = malloc (EIO_BUFSIZE); \
149 ETP_WORKER_UNLOCK (self); \
150 errno = ENOMEM; \ 288 errno = ENOMEM; \
151 if (!eio_buf) \ 289 if (!eio_buf) \
152 return -1; 290 return -1
291
292#define FUBd \
293 free (eio_buf)
153 294
154#define EIO_TICKS ((1000000 + 1023) >> 10) 295#define EIO_TICKS ((1000000 + 1023) >> 10)
155 296
156/*****************************************************************************/ 297/*****************************************************************************/
157 298
158#if __GNUC__ >= 3 299struct tmpbuf
159# define expect(expr,value) __builtin_expect ((expr),(value)) 300{
160#else 301 void *ptr;
161# define expect(expr,value) (expr) 302 int len;
162#endif 303};
163 304
164#define expect_false(expr) expect ((expr) != 0, 0) 305static void *
165#define expect_true(expr) expect ((expr) != 0, 1) 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}
166 316
167/*****************************************************************************/ 317/*****************************************************************************/
168 318
169#define ETP_PRI_MIN EIO_PRI_MIN 319#define ETP_PRI_MIN EIO_PRI_MIN
170#define ETP_PRI_MAX EIO_PRI_MAX 320#define ETP_PRI_MAX EIO_PRI_MAX
176static int eio_finish (eio_req *req); 326static int eio_finish (eio_req *req);
177#define ETP_FINISH(req) eio_finish (req) 327#define ETP_FINISH(req) eio_finish (req)
178static void eio_execute (struct etp_worker *self, eio_req *req); 328static void eio_execute (struct etp_worker *self, eio_req *req);
179#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) 329#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
180 330
181#define ETP_WORKER_CLEAR(req) \
182 if (wrk->dbuf) \
183 { \
184 free (wrk->dbuf); \
185 wrk->dbuf = 0; \
186 } \
187 \
188 if (wrk->dirp) \
189 { \
190 closedir (wrk->dirp); \
191 wrk->dirp = 0; \
192 }
193
194#define ETP_WORKER_COMMON \
195 void *dbuf; \
196 DIR *dirp;
197
198/*****************************************************************************/ 331/*****************************************************************************/
199 332
200#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) 333#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
201 334
202/* calculate time difference in ~1/EIO_TICKS of a second */ 335/* calculate time difference in ~1/EIO_TICKS of a second */
336ecb_inline int
203static int tvdiff (struct timeval *tv1, struct timeval *tv2) 337tvdiff (struct timeval *tv1, struct timeval *tv2)
204{ 338{
205 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 339 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
206 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 340 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
207} 341}
208 342
212static void (*done_poll_cb) (void); 346static void (*done_poll_cb) (void);
213 347
214static unsigned int max_poll_time; /* reslock */ 348static unsigned int max_poll_time; /* reslock */
215static unsigned int max_poll_reqs; /* reslock */ 349static unsigned int max_poll_reqs; /* reslock */
216 350
217static volatile unsigned int nreqs; /* reqlock */ 351static unsigned int nreqs; /* reqlock */
218static volatile unsigned int nready; /* reqlock */ 352static unsigned int nready; /* reqlock */
219static volatile unsigned int npending; /* reqlock */ 353static unsigned int npending; /* reqlock */
220static volatile unsigned int max_idle = 4; 354static unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
355static unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
221 356
222static mutex_t wrklock = X_MUTEX_INIT; 357static xmutex_t wrklock;
223static mutex_t reslock = X_MUTEX_INIT; 358static xmutex_t reslock;
224static mutex_t reqlock = X_MUTEX_INIT; 359static xmutex_t reqlock;
225static cond_t reqwait = X_COND_INIT; 360static xcond_t reqwait;
226 361
227#if !HAVE_PREADWRITE 362#if !HAVE_PREADWRITE
228/* 363/*
229 * make our pread/pwrite emulation safe against themselves, but not against 364 * make our pread/pwrite emulation safe against themselves, but not against
230 * normal read/write by using a mutex. slows down execution a lot, 365 * normal read/write by using a mutex. slows down execution a lot,
231 * but that's your problem, not mine. 366 * but that's your problem, not mine.
232 */ 367 */
233static mutex_t preadwritelock = X_MUTEX_INIT; 368static xmutex_t preadwritelock;
234#endif 369#endif
235 370
236typedef struct etp_worker 371typedef struct etp_worker
237{ 372{
373 struct tmpbuf tmpbuf;
374
238 /* locked by wrklock */ 375 /* locked by wrklock */
239 struct etp_worker *prev, *next; 376 struct etp_worker *prev, *next;
240 377
241 thread_t tid; 378 xthread_t tid;
242 379
243 /* locked by reslock, reqlock or wrklock */ 380#ifdef ETP_WORKER_COMMON
244 ETP_REQ *req; /* currently processed request */
245
246 ETP_WORKER_COMMON 381 ETP_WORKER_COMMON
382#endif
247} etp_worker; 383} etp_worker;
248 384
249static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */ 385static etp_worker wrk_first; /* NOT etp */
250 386
251#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock) 387#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
252#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock) 388#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
253 389
254/* worker threads management */ 390/* worker threads management */
255 391
392static void ecb_cold
256static void etp_worker_clear (etp_worker *wrk) 393etp_worker_clear (etp_worker *wrk)
257{ 394{
258 ETP_WORKER_CLEAR (wrk);
259} 395}
260 396
397static void ecb_cold
261static void etp_worker_free (etp_worker *wrk) 398etp_worker_free (etp_worker *wrk)
262{ 399{
400 free (wrk->tmpbuf.ptr);
401
263 wrk->next->prev = wrk->prev; 402 wrk->next->prev = wrk->prev;
264 wrk->prev->next = wrk->next; 403 wrk->prev->next = wrk->next;
265 404
266 free (wrk); 405 free (wrk);
267} 406}
268 407
269static unsigned int etp_nreqs (void) 408static unsigned int
409etp_nreqs (void)
270{ 410{
271 int retval; 411 int retval;
272 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 412 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
273 retval = nreqs; 413 retval = nreqs;
274 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 414 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
275 return retval; 415 return retval;
276} 416}
277 417
278static unsigned int etp_nready (void) 418static unsigned int
419etp_nready (void)
279{ 420{
280 unsigned int retval; 421 unsigned int retval;
281 422
282 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 423 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
283 retval = nready; 424 retval = nready;
284 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 425 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
285 426
286 return retval; 427 return retval;
287} 428}
288 429
289static unsigned int etp_npending (void) 430static unsigned int
431etp_npending (void)
290{ 432{
291 unsigned int retval; 433 unsigned int retval;
292 434
293 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 435 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
294 retval = npending; 436 retval = npending;
295 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 437 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
296 438
297 return retval; 439 return retval;
298} 440}
299 441
300static unsigned int etp_nthreads (void) 442static unsigned int
443etp_nthreads (void)
301{ 444{
302 unsigned int retval; 445 unsigned int retval;
303 446
304 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 447 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
305 retval = started; 448 retval = started;
319} etp_reqq; 462} etp_reqq;
320 463
321static etp_reqq req_queue; 464static etp_reqq req_queue;
322static etp_reqq res_queue; 465static etp_reqq res_queue;
323 466
467static void ecb_noinline ecb_cold
468reqq_init (etp_reqq *q)
469{
470 int pri;
471
472 for (pri = 0; pri < ETP_NUM_PRI; ++pri)
473 q->qs[pri] = q->qe[pri] = 0;
474
475 q->size = 0;
476}
477
478static int ecb_noinline
324static int reqq_push (etp_reqq *q, ETP_REQ *req) 479reqq_push (etp_reqq *q, ETP_REQ *req)
325{ 480{
326 int pri = req->pri; 481 int pri = req->pri;
327 req->next = 0; 482 req->next = 0;
328 483
329 if (q->qe[pri]) 484 if (q->qe[pri])
335 q->qe[pri] = q->qs[pri] = req; 490 q->qe[pri] = q->qs[pri] = req;
336 491
337 return q->size++; 492 return q->size++;
338} 493}
339 494
495static ETP_REQ * ecb_noinline
340static ETP_REQ *reqq_shift (etp_reqq *q) 496reqq_shift (etp_reqq *q)
341{ 497{
342 int pri; 498 int pri;
343 499
344 if (!q->size) 500 if (!q->size)
345 return 0; 501 return 0;
360 } 516 }
361 517
362 abort (); 518 abort ();
363} 519}
364 520
365static void etp_atfork_prepare (void) 521static int ecb_cold
522etp_init (void (*want_poll)(void), void (*done_poll)(void))
366{ 523{
367 X_LOCK (wrklock); 524 X_MUTEX_CREATE (wrklock);
368 X_LOCK (reqlock); 525 X_MUTEX_CREATE (reslock);
369 X_LOCK (reslock); 526 X_MUTEX_CREATE (reqlock);
370#if !HAVE_PREADWRITE 527 X_COND_CREATE (reqwait);
371 X_LOCK (preadwritelock);
372#endif
373}
374 528
375static void etp_atfork_parent (void) 529 reqq_init (&req_queue);
376{ 530 reqq_init (&res_queue);
377#if !HAVE_PREADWRITE
378 X_UNLOCK (preadwritelock);
379#endif
380 X_UNLOCK (reslock);
381 X_UNLOCK (reqlock);
382 X_UNLOCK (wrklock);
383}
384 531
385static void etp_atfork_child (void) 532 wrk_first.next =
386{ 533 wrk_first.prev = &wrk_first;
387 ETP_REQ *prv;
388
389 while ((prv = reqq_shift (&req_queue)))
390 ETP_DESTROY (prv);
391
392 while ((prv = reqq_shift (&res_queue)))
393 ETP_DESTROY (prv);
394
395 while (wrk_first.next != &wrk_first)
396 {
397 etp_worker *wrk = wrk_first.next;
398
399 if (wrk->req)
400 ETP_DESTROY (wrk->req);
401
402 etp_worker_clear (wrk);
403 etp_worker_free (wrk);
404 }
405 534
406 started = 0; 535 started = 0;
407 idle = 0; 536 idle = 0;
408 nreqs = 0; 537 nreqs = 0;
409 nready = 0; 538 nready = 0;
410 npending = 0; 539 npending = 0;
411 540
412 etp_atfork_parent ();
413}
414
415static void
416etp_once_init (void)
417{
418 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
419}
420
421static int
422etp_init (void (*want_poll)(void), void (*done_poll)(void))
423{
424 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
425
426 pthread_once (&doinit, etp_once_init);
427
428 want_poll_cb = want_poll; 541 want_poll_cb = want_poll;
429 done_poll_cb = done_poll; 542 done_poll_cb = done_poll;
430 543
431 return 0; 544 return 0;
432} 545}
433 546
434X_THREAD_PROC (etp_proc); 547X_THREAD_PROC (etp_proc);
435 548
549static void ecb_cold
436static void etp_start_thread (void) 550etp_start_thread (void)
437{ 551{
438 etp_worker *wrk = calloc (1, sizeof (etp_worker)); 552 etp_worker *wrk = calloc (1, sizeof (etp_worker));
439 553
440 /*TODO*/ 554 /*TODO*/
441 assert (("unable to allocate worker thread data", wrk)); 555 assert (("unable to allocate worker thread data", wrk));
454 free (wrk); 568 free (wrk);
455 569
456 X_UNLOCK (wrklock); 570 X_UNLOCK (wrklock);
457} 571}
458 572
573static void
459static void etp_maybe_start_thread (void) 574etp_maybe_start_thread (void)
460{ 575{
461 if (expect_true (etp_nthreads () >= wanted)) 576 if (ecb_expect_true (etp_nthreads () >= wanted))
462 return; 577 return;
463 578
464 /* todo: maybe use idle here, but might be less exact */ 579 /* todo: maybe use idle here, but might be less exact */
465 if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ())) 580 if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
466 return; 581 return;
467 582
468 etp_start_thread (); 583 etp_start_thread ();
469} 584}
470 585
586static void ecb_cold
471static void etp_end_thread (void) 587etp_end_thread (void)
472{ 588{
473 eio_req *req = calloc (1, sizeof (eio_req)); 589 eio_req *req = calloc (1, sizeof (eio_req)); /* will be freed by worker */
474 590
475 req->type = -1; 591 req->type = -1;
476 req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 592 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
477 593
478 X_LOCK (reqlock); 594 X_LOCK (reqlock);
483 X_LOCK (wrklock); 599 X_LOCK (wrklock);
484 --started; 600 --started;
485 X_UNLOCK (wrklock); 601 X_UNLOCK (wrklock);
486} 602}
487 603
488static int etp_poll (void) 604static int
605etp_poll (void)
489{ 606{
490 unsigned int maxreqs; 607 unsigned int maxreqs;
491 unsigned int maxtime; 608 unsigned int maxtime;
492 struct timeval tv_start, tv_now; 609 struct timeval tv_start, tv_now;
493 610
523 640
524 X_LOCK (reqlock); 641 X_LOCK (reqlock);
525 --nreqs; 642 --nreqs;
526 X_UNLOCK (reqlock); 643 X_UNLOCK (reqlock);
527 644
528 if (expect_false (req->type == EIO_GROUP && req->size)) 645 if (ecb_expect_false (req->type == EIO_GROUP && req->size))
529 { 646 {
530 req->int1 = 1; /* mark request as delayed */ 647 req->int1 = 1; /* mark request as delayed */
531 continue; 648 continue;
532 } 649 }
533 else 650 else
534 { 651 {
535 int res = ETP_FINISH (req); 652 int res = ETP_FINISH (req);
536 if (expect_false (res)) 653 if (ecb_expect_false (res))
537 return res; 654 return res;
538 } 655 }
539 656
540 if (expect_false (maxreqs && !--maxreqs)) 657 if (ecb_expect_false (maxreqs && !--maxreqs))
541 break; 658 break;
542 659
543 if (maxtime) 660 if (maxtime)
544 { 661 {
545 gettimeofday (&tv_now, 0); 662 gettimeofday (&tv_now, 0);
551 668
552 errno = EAGAIN; 669 errno = EAGAIN;
553 return -1; 670 return -1;
554} 671}
555 672
673static void
556static void etp_cancel (ETP_REQ *req) 674etp_cancel (ETP_REQ *req)
557{ 675{
558 X_LOCK (wrklock); 676 req->cancelled = 1;
559 req->flags |= EIO_FLAG_CANCELLED;
560 X_UNLOCK (wrklock);
561 677
562 eio_grp_cancel (req); 678 eio_grp_cancel (req);
563} 679}
564 680
681static void
565static void etp_submit (ETP_REQ *req) 682etp_submit (ETP_REQ *req)
566{ 683{
567 req->pri -= ETP_PRI_MIN; 684 req->pri -= ETP_PRI_MIN;
568 685
569 if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN; 686 if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
570 if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 687 if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
571 688
572 if (expect_false (req->type == EIO_GROUP)) 689 if (ecb_expect_false (req->type == EIO_GROUP))
573 { 690 {
574 /* I hope this is worth it :/ */ 691 /* I hope this is worth it :/ */
575 X_LOCK (reqlock); 692 X_LOCK (reqlock);
576 ++nreqs; 693 ++nreqs;
577 X_UNLOCK (reqlock); 694 X_UNLOCK (reqlock);
596 713
597 etp_maybe_start_thread (); 714 etp_maybe_start_thread ();
598 } 715 }
599} 716}
600 717
718static void ecb_cold
601static void etp_set_max_poll_time (double nseconds) 719etp_set_max_poll_time (double nseconds)
602{ 720{
603 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 721 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
604 max_poll_time = nseconds * EIO_TICKS; 722 max_poll_time = nseconds * EIO_TICKS;
605 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 723 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
606} 724}
607 725
726static void ecb_cold
608static void etp_set_max_poll_reqs (unsigned int maxreqs) 727etp_set_max_poll_reqs (unsigned int maxreqs)
609{ 728{
610 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 729 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
611 max_poll_reqs = maxreqs; 730 max_poll_reqs = maxreqs;
612 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 731 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
613} 732}
614 733
734static void ecb_cold
615static void etp_set_max_idle (unsigned int nthreads) 735etp_set_max_idle (unsigned int nthreads)
616{ 736{
617 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 737 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
618 max_idle = nthreads <= 0 ? 1 : nthreads; 738 max_idle = nthreads;
619 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 739 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
620} 740}
621 741
742static void ecb_cold
743etp_set_idle_timeout (unsigned int seconds)
744{
745 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
746 idle_timeout = seconds;
747 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
748}
749
750static void ecb_cold
622static void etp_set_min_parallel (unsigned int nthreads) 751etp_set_min_parallel (unsigned int nthreads)
623{ 752{
624 if (wanted < nthreads) 753 if (wanted < nthreads)
625 wanted = nthreads; 754 wanted = nthreads;
626} 755}
627 756
757static void ecb_cold
628static void etp_set_max_parallel (unsigned int nthreads) 758etp_set_max_parallel (unsigned int nthreads)
629{ 759{
630 if (wanted > nthreads) 760 if (wanted > nthreads)
631 wanted = nthreads; 761 wanted = nthreads;
632 762
633 while (started > wanted) 763 while (started > wanted)
634 etp_end_thread (); 764 etp_end_thread ();
635} 765}
636 766
637/*****************************************************************************/ 767/*****************************************************************************/
638 768
769static void
639static void grp_try_feed (eio_req *grp) 770grp_try_feed (eio_req *grp)
640{ 771{
641 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 772 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
642 { 773 {
643 grp->flags &= ~EIO_FLAG_GROUPADD; 774 grp->flags &= ~EIO_FLAG_GROUPADD;
644 775
651 break; 782 break;
652 } 783 }
653 } 784 }
654} 785}
655 786
787static int
656static int grp_dec (eio_req *grp) 788grp_dec (eio_req *grp)
657{ 789{
658 --grp->size; 790 --grp->size;
659 791
660 /* call feeder, if applicable */ 792 /* call feeder, if applicable */
661 grp_try_feed (grp); 793 grp_try_feed (grp);
665 return eio_finish (grp); 797 return eio_finish (grp);
666 else 798 else
667 return 0; 799 return 0;
668} 800}
669 801
802static void
670void eio_destroy (eio_req *req) 803eio_destroy (eio_req *req)
671{ 804{
672 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); 805 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
673 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); 806 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
674 807
675 EIO_DESTROY (req); 808 EIO_DESTROY (req);
676} 809}
677 810
811static int
678static int eio_finish (eio_req *req) 812eio_finish (eio_req *req)
679{ 813{
680 int res = EIO_FINISH (req); 814 int res = EIO_FINISH (req);
681 815
682 if (req->grp) 816 if (req->grp)
683 { 817 {
691 if (grp->grp_first == req) 825 if (grp->grp_first == req)
692 grp->grp_first = req->grp_next; 826 grp->grp_first = req->grp_next;
693 827
694 res2 = grp_dec (grp); 828 res2 = grp_dec (grp);
695 829
696 if (!res && res2) 830 if (!res)
697 res = res2; 831 res = res2;
698 } 832 }
699 833
700 eio_destroy (req); 834 eio_destroy (req);
701 835
702 return res; 836 return res;
703} 837}
704 838
839void
705void eio_grp_cancel (eio_req *grp) 840eio_grp_cancel (eio_req *grp)
706{ 841{
707 for (grp = grp->grp_first; grp; grp = grp->grp_next) 842 for (grp = grp->grp_first; grp; grp = grp->grp_next)
708 eio_cancel (grp); 843 eio_cancel (grp);
709} 844}
710 845
846void
711void eio_cancel (eio_req *req) 847eio_cancel (eio_req *req)
712{ 848{
713 etp_cancel (req); 849 etp_cancel (req);
714} 850}
715 851
852void
716void eio_submit (eio_req *req) 853eio_submit (eio_req *req)
717{ 854{
718 etp_submit (req); 855 etp_submit (req);
719} 856}
720 857
721unsigned int eio_nreqs (void) 858unsigned int
859eio_nreqs (void)
722{ 860{
723 return etp_nreqs (); 861 return etp_nreqs ();
724} 862}
725 863
726unsigned int eio_nready (void) 864unsigned int
865eio_nready (void)
727{ 866{
728 return etp_nready (); 867 return etp_nready ();
729} 868}
730 869
731unsigned int eio_npending (void) 870unsigned int
871eio_npending (void)
732{ 872{
733 return etp_npending (); 873 return etp_npending ();
734} 874}
735 875
736unsigned int eio_nthreads (void) 876unsigned int ecb_cold
877eio_nthreads (void)
737{ 878{
738 return etp_nthreads (); 879 return etp_nthreads ();
739} 880}
740 881
882void ecb_cold
741void eio_set_max_poll_time (double nseconds) 883eio_set_max_poll_time (double nseconds)
742{ 884{
743 etp_set_max_poll_time (nseconds); 885 etp_set_max_poll_time (nseconds);
744} 886}
745 887
888void ecb_cold
746void eio_set_max_poll_reqs (unsigned int maxreqs) 889eio_set_max_poll_reqs (unsigned int maxreqs)
747{ 890{
748 etp_set_max_poll_reqs (maxreqs); 891 etp_set_max_poll_reqs (maxreqs);
749} 892}
750 893
894void ecb_cold
751void eio_set_max_idle (unsigned int nthreads) 895eio_set_max_idle (unsigned int nthreads)
752{ 896{
753 etp_set_max_idle (nthreads); 897 etp_set_max_idle (nthreads);
754} 898}
755 899
900void ecb_cold
901eio_set_idle_timeout (unsigned int seconds)
902{
903 etp_set_idle_timeout (seconds);
904}
905
906void ecb_cold
756void eio_set_min_parallel (unsigned int nthreads) 907eio_set_min_parallel (unsigned int nthreads)
757{ 908{
758 etp_set_min_parallel (nthreads); 909 etp_set_min_parallel (nthreads);
759} 910}
760 911
912void ecb_cold
761void eio_set_max_parallel (unsigned int nthreads) 913eio_set_max_parallel (unsigned int nthreads)
762{ 914{
763 etp_set_max_parallel (nthreads); 915 etp_set_max_parallel (nthreads);
764} 916}
765 917
766int eio_poll (void) 918int eio_poll (void)
775# undef pread 927# undef pread
776# undef pwrite 928# undef pwrite
777# define pread eio__pread 929# define pread eio__pread
778# define pwrite eio__pwrite 930# define pwrite eio__pwrite
779 931
780static ssize_t 932static eio_ssize_t
781eio__pread (int fd, void *buf, size_t count, off_t offset) 933eio__pread (int fd, void *buf, size_t count, off_t offset)
782{ 934{
783 ssize_t res; 935 eio_ssize_t res;
784 off_t ooffset; 936 off_t ooffset;
785 937
786 X_LOCK (preadwritelock); 938 X_LOCK (preadwritelock);
787 ooffset = lseek (fd, 0, SEEK_CUR); 939 ooffset = lseek (fd, 0, SEEK_CUR);
788 lseek (fd, offset, SEEK_SET); 940 lseek (fd, offset, SEEK_SET);
791 X_UNLOCK (preadwritelock); 943 X_UNLOCK (preadwritelock);
792 944
793 return res; 945 return res;
794} 946}
795 947
796static ssize_t 948static eio_ssize_t
797eio__pwrite (int fd, void *buf, size_t count, off_t offset) 949eio__pwrite (int fd, void *buf, size_t count, off_t offset)
798{ 950{
799 ssize_t res; 951 eio_ssize_t res;
800 off_t ooffset; 952 off_t ooffset;
801 953
802 X_LOCK (preadwritelock); 954 X_LOCK (preadwritelock);
803 ooffset = lseek (fd, 0, SEEK_CUR); 955 ooffset = lseek (fd, 0, SEEK_CUR);
804 lseek (fd, offset, SEEK_SET); 956 lseek (fd, offset, SEEK_SET);
808 960
809 return res; 961 return res;
810} 962}
811#endif 963#endif
812 964
813#ifndef HAVE_FUTIMES 965#ifndef HAVE_UTIMES
814 966
815# undef utimes 967# undef utimes
816# undef futimes
817# define utimes(path,times) eio__utimes (path, times) 968# define utimes(path,times) eio__utimes (path, times)
818# define futimes(fd,times) eio__futimes (fd, times)
819 969
820static int 970static int
821eio__utimes (const char *filename, const struct timeval times[2]) 971eio__utimes (const char *filename, const struct timeval times[2])
822{ 972{
823 if (times) 973 if (times)
831 } 981 }
832 else 982 else
833 return utime (filename, 0); 983 return utime (filename, 0);
834} 984}
835 985
986#endif
987
988#ifndef HAVE_FUTIMES
989
990# undef futimes
991# define futimes(fd,times) eio__futimes (fd, times)
992
993static int
836static int eio__futimes (int fd, const struct timeval tv[2]) 994eio__futimes (int fd, const struct timeval tv[2])
837{ 995{
838 errno = ENOSYS; 996 errno = ENOSYS;
839 return -1; 997 return -1;
840} 998}
841 999
844#if !HAVE_FDATASYNC 1002#if !HAVE_FDATASYNC
845# undef fdatasync 1003# undef fdatasync
846# define fdatasync(fd) fsync (fd) 1004# define fdatasync(fd) fsync (fd)
847#endif 1005#endif
848 1006
1007static int
1008eio__syncfs (int fd)
1009{
1010 int res;
1011
1012#if HAVE_SYS_SYNCFS
1013 res = (int)syscall (__NR_syncfs, (int)(fd));
1014#else
1015 res = -1;
1016 errno = ENOSYS;
1017#endif
1018
1019 if (res < 0 && errno == ENOSYS && fd >= 0)
1020 sync ();
1021
1022 return res;
1023}
1024
849/* sync_file_range always needs emulation */ 1025/* sync_file_range always needs emulation */
850int 1026static int
851eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags) 1027eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
852{ 1028{
853#if HAVE_SYNC_FILE_RANGE 1029#if HAVE_SYNC_FILE_RANGE
854 int res; 1030 int res;
855 1031
872 /* even though we could play tricks with the flags, it's better to always 1048 /* even though we could play tricks with the flags, it's better to always
873 * call fdatasync, as that matches the expectation of its users best */ 1049 * call fdatasync, as that matches the expectation of its users best */
874 return fdatasync (fd); 1050 return fdatasync (fd);
875} 1051}
876 1052
1053static int
1054eio__fallocate (int fd, int mode, off_t offset, size_t len)
1055{
1056#if HAVE_FALLOCATE
1057 return fallocate (fd, mode, offset, len);
1058#else
1059 errno = ENOSYS;
1060 return -1;
1061#endif
1062}
1063
877#if !HAVE_READAHEAD 1064#if !HAVE_READAHEAD
878# undef readahead 1065# undef readahead
879# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 1066# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
880 1067
881static ssize_t 1068static eio_ssize_t
882eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) 1069eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
883{ 1070{
884 size_t todo = count; 1071 size_t todo = count;
885 dBUF; 1072 dBUF;
886 1073
891 pread (fd, eio_buf, len, offset); 1078 pread (fd, eio_buf, len, offset);
892 offset += len; 1079 offset += len;
893 todo -= len; 1080 todo -= len;
894 } 1081 }
895 1082
1083 FUBd;
1084
896 errno = 0; 1085 errno = 0;
897 return count; 1086 return count;
898} 1087}
899 1088
900#endif 1089#endif
901 1090
902/* sendfile always needs emulation */ 1091/* sendfile always needs emulation */
903static ssize_t 1092static eio_ssize_t
904eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) 1093eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
905{ 1094{
1095 eio_ssize_t written = 0;
906 ssize_t res; 1096 eio_ssize_t res;
907 1097
908 if (!count) 1098 if (!count)
909 return 0; 1099 return 0;
910 1100
1101 for (;;)
1102 {
1103#ifdef __APPLE__
1104# undef HAVE_SENDFILE /* broken, as everything on os x */
1105#endif
911#if HAVE_SENDFILE 1106#if HAVE_SENDFILE
912# if __linux 1107# if __linux
1108 off_t soffset = offset;
913 res = sendfile (ofd, ifd, &offset, count); 1109 res = sendfile (ofd, ifd, &soffset, count);
914 1110
915# elif __FreeBSD__ 1111# elif __FreeBSD__
916 /* 1112 /*
917 * Of course, the freebsd sendfile is a dire hack with no thoughts 1113 * Of course, the freebsd sendfile is a dire hack with no thoughts
918 * wasted on making it similar to other I/O functions. 1114 * wasted on making it similar to other I/O functions.
919 */ 1115 */
920 {
921 off_t sbytes; 1116 off_t sbytes;
922 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 1117 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
923 1118
924 #if 0 /* according to the manpage, this is correct, but broken behaviour */ 1119 #if 0 /* according to the manpage, this is correct, but broken behaviour */
925 /* freebsd' sendfile will return 0 on success */ 1120 /* freebsd' sendfile will return 0 on success */
926 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */ 1121 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
927 /* not on e.g. EIO or EPIPE - sounds broken */ 1122 /* not on e.g. EIO or EPIPE - sounds broken */
928 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0) 1123 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
929 res = sbytes; 1124 res = sbytes;
930 #endif 1125 #endif
931 1126
932 /* according to source inspection, this is correct, and useful behaviour */ 1127 /* according to source inspection, this is correct, and useful behaviour */
933 if (sbytes) 1128 if (sbytes)
934 res = sbytes; 1129 res = sbytes;
935 }
936 1130
937# elif defined (__APPLE__) 1131# elif defined (__APPLE__)
938
939 {
940 off_t sbytes = count; 1132 off_t sbytes = count;
941 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0); 1133 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
942 1134
943 /* according to the manpage, sbytes is always valid */ 1135 /* according to the manpage, sbytes is always valid */
944 if (sbytes) 1136 if (sbytes)
945 res = sbytes; 1137 res = sbytes;
946 }
947 1138
948# elif __hpux 1139# elif __hpux
949 res = sendfile (ofd, ifd, offset, count, 0, 0); 1140 res = sendfile (ofd, ifd, offset, count, 0, 0);
950 1141
951# elif __solaris 1142# elif __solaris
952 {
953 struct sendfilevec vec; 1143 struct sendfilevec vec;
954 size_t sbytes; 1144 size_t sbytes;
955 1145
956 vec.sfv_fd = ifd; 1146 vec.sfv_fd = ifd;
957 vec.sfv_flag = 0; 1147 vec.sfv_flag = 0;
958 vec.sfv_off = offset; 1148 vec.sfv_off = offset;
959 vec.sfv_len = count; 1149 vec.sfv_len = count;
960 1150
961 res = sendfilev (ofd, &vec, 1, &sbytes); 1151 res = sendfilev (ofd, &vec, 1, &sbytes);
962 1152
963 if (res < 0 && sbytes) 1153 if (res < 0 && sbytes)
964 res = sbytes; 1154 res = sbytes;
965 }
966 1155
967# endif 1156# endif
968 1157
969#elif defined (_WIN32) 1158#elif defined (_WIN32) && 0
970
971 /* does not work, just for documentation of what would need to be done */ 1159 /* does not work, just for documentation of what would need to be done */
972 { 1160 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
1161 /* libeio guarantees that the file offset does not change, and windows */
1162 /* has no way to get an independent handle to the same file description */
973 HANDLE h = TO_SOCKET (ifd); 1163 HANDLE h = TO_SOCKET (ifd);
974 SetFilePointer (h, offset, 0, FILE_BEGIN); 1164 SetFilePointer (h, offset, 0, FILE_BEGIN);
975 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0); 1165 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
976 }
977 1166
978#else 1167#else
979 res = -1; 1168 res = -1;
980 errno = ENOSYS; 1169 errno = ENOSYS;
981#endif 1170#endif
982 1171
1172 /* we assume sendfile can copy at least 128mb in one go */
1173 if (res <= 128 * 1024 * 1024)
1174 {
1175 if (res > 0)
1176 written += res;
1177
1178 if (written)
1179 return written;
1180
1181 break;
1182 }
1183 else
1184 {
1185 /* if we requested more, then probably the kernel was lazy */
1186 written += res;
1187 offset += res;
1188 count -= res;
1189
1190 if (!count)
1191 return written;
1192 }
1193 }
1194
983 if (res < 0 1195 if (res < 0
984 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1196 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
985 /* BSDs */ 1197 /* BSDs */
986#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */ 1198#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
987 || errno == ENOTSUP 1199 || errno == ENOTSUP
988#endif 1200#endif
1201#ifdef EOPNOTSUPP /* windows */
989 || errno == EOPNOTSUPP /* BSDs */ 1202 || errno == EOPNOTSUPP /* BSDs */
1203#endif
990#if __solaris 1204#if __solaris
991 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1205 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
992#endif 1206#endif
993 ) 1207 )
994 ) 1208 )
998 1212
999 res = 0; 1213 res = 0;
1000 1214
1001 while (count) 1215 while (count)
1002 { 1216 {
1003 ssize_t cnt; 1217 eio_ssize_t cnt;
1004 1218
1005 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 1219 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
1006 1220
1007 if (cnt <= 0) 1221 if (cnt <= 0)
1008 { 1222 {
1020 1234
1021 offset += cnt; 1235 offset += cnt;
1022 res += cnt; 1236 res += cnt;
1023 count -= cnt; 1237 count -= cnt;
1024 } 1238 }
1239
1240 FUBd;
1025 } 1241 }
1026 1242
1027 return res; 1243 return res;
1244}
1245
1246#ifdef PAGESIZE
1247# define eio_pagesize() PAGESIZE
1248#else
1249static intptr_t
1250eio_pagesize (void)
1251{
1252 static intptr_t page;
1253
1254 if (!page)
1255 page = sysconf (_SC_PAGESIZE);
1256
1257 return page;
1258}
1259#endif
1260
1261static void
1262eio_page_align (void **addr, size_t *length)
1263{
1264 intptr_t mask = eio_pagesize () - 1;
1265
1266 /* round down addr */
1267 intptr_t adj = mask & (intptr_t)*addr;
1268
1269 *addr = (void *)((intptr_t)*addr - adj);
1270 *length += adj;
1271
1272 /* round up length */
1273 *length = (*length + mask) & ~mask;
1274}
1275
1276#if !_POSIX_MEMLOCK
1277# define eio__mlockall(a) EIO_ENOSYS ()
1278#else
1279
1280static int
1281eio__mlockall (int flags)
1282{
1283 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1284 extern int mallopt (int, int);
1285 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1286 #endif
1287
1288 if (EIO_MCL_CURRENT != MCL_CURRENT
1289 || EIO_MCL_FUTURE != MCL_FUTURE)
1290 {
1291 flags = 0
1292 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1293 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1294 }
1295
1296 return mlockall (flags);
1297}
1298#endif
1299
1300#if !_POSIX_MEMLOCK_RANGE
1301# define eio__mlock(a,b) EIO_ENOSYS ()
1302#else
1303
1304static int
1305eio__mlock (void *addr, size_t length)
1306{
1307 eio_page_align (&addr, &length);
1308
1309 return mlock (addr, length);
1310}
1311
1312#endif
1313
1314#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1315# define eio__msync(a,b,c) EIO_ENOSYS ()
1316#else
1317
1318static int
1319eio__msync (void *mem, size_t len, int flags)
1320{
1321 eio_page_align (&mem, &len);
1322
1323 if (EIO_MS_ASYNC != MS_SYNC
1324 || EIO_MS_INVALIDATE != MS_INVALIDATE
1325 || EIO_MS_SYNC != MS_SYNC)
1326 {
1327 flags = 0
1328 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1329 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1330 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1331 }
1332
1333 return msync (mem, len, flags);
1334}
1335
1336#endif
1337
1338static int
1339eio__mtouch (eio_req *req)
1340{
1341 void *mem = req->ptr2;
1342 size_t len = req->size;
1343 int flags = req->int1;
1344
1345 eio_page_align (&mem, &len);
1346
1347 {
1348 intptr_t addr = (intptr_t)mem;
1349 intptr_t end = addr + len;
1350 intptr_t page = eio_pagesize ();
1351
1352 if (addr < end)
1353 if (flags & EIO_MT_MODIFY) /* modify */
1354 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1355 else
1356 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1357 }
1358
1359 return 0;
1360}
1361
1362/*****************************************************************************/
1363/* requests implemented outside eio_execute, because they are so large */
1364
1365/* copies some absolute path to tmpbuf */
1366static char *
1367eio__getwd (struct tmpbuf *tmpbuf, eio_wd wd)
1368{
1369 if (wd == EIO_CWD)
1370 return getcwd (tmpbuf->ptr, PATH_MAX);
1371
1372#if HAVE_AT
1373 abort (); /*TODO*/
1374#else
1375 strcpy (tmpbuf->ptr, wd);
1376#endif
1377 return tmpbuf->ptr;
1378}
1379
1380/* result will always end up in tmpbuf, there is always space for adding a 0-byte */
1381static int
1382eio__realpath (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
1383{
1384 const char *rel = path;
1385 char *res;
1386 char *tmp1, *tmp2;
1387#if SYMLOOP_MAX > 32
1388 int symlinks = SYMLOOP_MAX;
1389#else
1390 int symlinks = 32;
1391#endif
1392
1393 /*D*/ /*TODO: wd ignored */
1394
1395 errno = EINVAL;
1396 if (!rel)
1397 return -1;
1398
1399 errno = ENOENT;
1400 if (!*rel)
1401 return -1;
1402
1403 res = tmpbuf_get (tmpbuf, PATH_MAX * 3);
1404 tmp1 = res + PATH_MAX;
1405 tmp2 = tmp1 + PATH_MAX;
1406
1407#if 0 /* disabled, the musl way to do things is just too racy */
1408#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1409 /* on linux we may be able to ask the kernel */
1410 {
1411 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1412
1413 if (fd >= 0)
1414 {
1415 sprintf (tmp1, "/proc/self/fd/%d", fd);
1416 req->result = readlink (tmp1, res, PATH_MAX);
1417 close (fd);
1418
1419 /* here we should probably stat the open file and the disk file, to make sure they still match */
1420
1421 if (req->result > 0)
1422 goto done;
1423 }
1424 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1425 return;
1426 }
1427#endif
1428#endif
1429
1430 if (*rel != '/')
1431 {
1432 if (!eio__getwd (tmpbuf, wd))
1433 return -1;
1434
1435 if (res [1]) /* only use if not / */
1436 res += strlen (res);
1437 }
1438
1439 while (*rel)
1440 {
1441 eio_ssize_t len, linklen;
1442 const char *beg = rel;
1443
1444 while (*rel && *rel != '/')
1445 ++rel;
1446
1447 len = rel - beg;
1448
1449 if (!len) /* skip slashes */
1450 {
1451 ++rel;
1452 continue;
1453 }
1454
1455 if (beg [0] == '.')
1456 {
1457 if (len == 1)
1458 continue; /* . - nop */
1459
1460 if (beg [1] == '.' && len == 2)
1461 {
1462 /* .. - back up one component, if possible */
1463
1464 while (res != tmpbuf->ptr)
1465 if (*--res == '/')
1466 break;
1467
1468 continue;
1469 }
1470 }
1471
1472 errno = ENAMETOOLONG;
1473 if (res + 1 + len + 1 >= tmp1)
1474 return;
1475
1476 /* copy one component */
1477 *res = '/';
1478 memcpy (res + 1, beg, len);
1479
1480 /* zero-terminate, for readlink */
1481 res [len + 1] = 0;
1482
1483 /* now check if it's a symlink */
1484 linklen = readlink (tmpbuf->ptr, tmp1, PATH_MAX);
1485
1486 if (linklen < 0)
1487 {
1488 if (errno != EINVAL)
1489 return -1;
1490
1491 /* it's a normal directory. hopefully */
1492 res += len + 1;
1493 }
1494 else
1495 {
1496 /* yay, it was a symlink - build new path in tmp2 */
1497 int rellen = strlen (rel);
1498
1499 errno = ENAMETOOLONG;
1500 if (linklen + 1 + rellen >= PATH_MAX)
1501 return -1;
1502
1503 errno = ELOOP;
1504 if (!--symlinks)
1505 return -1;
1506
1507 if (*tmp1 == '/')
1508 res = tmpbuf->ptr; /* symlink resolves to an absolute path */
1509
1510 /* we need to be careful, as rel might point into tmp2 already */
1511 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1512 tmp2 [linklen] = '/';
1513 memcpy (tmp2, tmp1, linklen);
1514
1515 rel = tmp2;
1516 }
1517 }
1518
1519 /* special case for the lone root path */
1520 if (res == tmpbuf->ptr)
1521 *res++ = '/';
1522
1523 return res - (char *)tmpbuf->ptr;
1028} 1524}
1029 1525
1030static signed char 1526static signed char
1031eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) 1527eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1032{ 1528{
1033 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */ 1529 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1034 : a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0; 1530 : a->inode < b->inode ? -1
1531 : a->inode > b->inode ? 1
1532 : 0;
1035} 1533}
1036 1534
1037#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0 1535#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1038 1536
1039#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */ 1537#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1040#define EIO_SORT_FAST 60 /* when to only use insertion sort */ 1538#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1041 1539
1042static void 1540static void
1043eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1541eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1044{ 1542{
1045 unsigned char bits [9 + sizeof (ino_t) * 8]; 1543 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1046 unsigned char *bit = bits; 1544 unsigned char *bit = bits;
1047 1545
1048 assert (CHAR_BIT == 8); 1546 assert (CHAR_BIT == 8);
1049 assert (sizeof (eio_dirent) * 8 < 256); 1547 assert (sizeof (eio_dirent) * 8 < 256);
1050 assert (offsetof (eio_dirent, inode)); /* we use 0 as sentinel */ 1548 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1051 assert (offsetof (eio_dirent, score)); /* we use 0 as sentinel */ 1549 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1052 1550
1053 if (size <= EIO_SORT_FAST) 1551 if (size <= EIO_SORT_FAST)
1054 return; 1552 return;
1055 1553
1056 /* first prepare an array of bits to test in our radix sort */ 1554 /* first prepare an array of bits to test in our radix sort */
1057 /* try to take endianness into account, as well as differences in ino_t sizes */ 1555 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1058 /* inode_bits must contain all inodes ORed together */ 1556 /* inode_bits must contain all inodes ORed together */
1059 /* which is used to skip bits that are 0 everywhere, which is very common */ 1557 /* which is used to skip bits that are 0 everywhere, which is very common */
1060 { 1558 {
1061 ino_t endianness; 1559 eio_ino_t endianness;
1062 int i, j; 1560 int i, j;
1063 1561
1064 /* we store the byte offset of byte n into byte n of "endianness" */ 1562 /* we store the byte offset of byte n into byte n of "endianness" */
1065 for (i = 0; i < sizeof (ino_t); ++i) 1563 for (i = 0; i < sizeof (eio_ino_t); ++i)
1066 ((unsigned char *)&endianness)[i] = i; 1564 ((unsigned char *)&endianness)[i] = i;
1067 1565
1068 *bit++ = 0; 1566 *bit++ = 0;
1069 1567
1070 for (i = 0; i < sizeof (ino_t); ++i) 1568 for (i = 0; i < sizeof (eio_ino_t); ++i)
1071 { 1569 {
1072 /* shifting off the byte offsets out of "endianness" */ 1570 /* shifting off the byte offsets out of "endianness" */
1073 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8; 1571 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1074 endianness >>= 8; 1572 endianness >>= 8;
1075 1573
1076 for (j = 0; j < 8; ++j) 1574 for (j = 0; j < 8; ++j)
1077 if (inode_bits & (((ino_t)1) << (i * 8 + j))) 1575 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1078 *bit++ = offs + j; 1576 *bit++ = offs + j;
1079 } 1577 }
1080 1578
1081 for (j = 0; j < 8; ++j) 1579 for (j = 0; j < 8; ++j)
1082 if (score_bits & (1 << j)) 1580 if (score_bits & (1 << j))
1083 *bit++ = offsetof (eio_dirent, score) * 8 + j; 1581 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1084 } 1582 }
1085 1583
1086 /* now actually do the sorting (a variant of MSD radix sort) */ 1584 /* now actually do the sorting (a variant of MSD radix sort) */
1087 { 1585 {
1088 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base; 1586 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1089 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end; 1587 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1090 unsigned char *bit_stk [9 + sizeof (ino_t) * 8]; 1588 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1091 int stk_idx = 0; 1589 int stk_idx = 0;
1092 1590
1093 base_stk [stk_idx] = dents; 1591 base_stk [stk_idx] = dents;
1094 end_stk [stk_idx] = dents + size; 1592 end_stk [stk_idx] = dents + size;
1095 bit_stk [stk_idx] = bit - 1; 1593 bit_stk [stk_idx] = bit - 1;
1174 } 1672 }
1175 } 1673 }
1176} 1674}
1177 1675
1178static void 1676static void
1179eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1677eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1180{ 1678{
1181 if (size <= 1) 1679 if (size <= 1)
1182 return; /* our insertion sort relies on size > 0 */ 1680 return; /* our insertion sort relies on size > 0 */
1183 1681
1184 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */ 1682 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1190 eio_dent_insertion_sort (dents, size); 1688 eio_dent_insertion_sort (dents, size);
1191} 1689}
1192 1690
1193/* read a full directory */ 1691/* read a full directory */
1194static void 1692static void
1195eio__scandir (eio_req *req, etp_worker *self) 1693eio__scandir (eio_req *req)
1196{ 1694{
1197 DIR *dirp;
1198 EIO_STRUCT_DIRENT *entp;
1199 char *name, *names; 1695 char *name, *names;
1200 int namesalloc = 4096; 1696 int namesalloc = 4096 - sizeof (void *) * 4;
1201 int namesoffs = 0; 1697 int namesoffs = 0;
1202 int flags = req->int1; 1698 int flags = req->int1;
1203 eio_dirent *dents = 0; 1699 eio_dirent *dents = 0;
1204 int dentalloc = 128; 1700 int dentalloc = 128;
1205 int dentoffs = 0; 1701 int dentoffs = 0;
1206 ino_t inode_bits = 0; 1702 eio_ino_t inode_bits = 0;
1703#ifdef _WIN32
1704 HANDLE dirp;
1705 WIN32_FIND_DATA entp;
1706#else
1707 DIR *dirp;
1708 EIO_STRUCT_DIRENT *entp;
1709#endif
1207 1710
1208 req->result = -1; 1711 req->result = -1;
1209 1712
1210 if (!(flags & EIO_READDIR_DENTS)) 1713 if (!(flags & EIO_READDIR_DENTS))
1211 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); 1714 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1212 1715
1213 X_LOCK (wrklock); 1716#ifdef _WIN32
1214 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 1717 {
1718 int len = strlen ((const char *)req->ptr1);
1719 char *path = malloc (MAX_PATH);
1720 const char *fmt;
1721
1722 if (!len)
1723 fmt = "./*";
1724 else if (((const char *)req->ptr1)[len - 1] == '/' || ((const char *)req->ptr1)[len - 1] == '\\')
1725 fmt = "%s*";
1726 else
1727 fmt = "%s/*";
1728
1729 _snprintf (path, MAX_PATH, fmt, (const char *)req->ptr1);
1730 dirp = FindFirstFile (path, &entp);
1731 free (path);
1732
1733 if (dirp == INVALID_HANDLE_VALUE)
1734 {
1735 /* should steal _dosmaperr */
1736 switch (GetLastError ())
1737 {
1738 case ERROR_FILE_NOT_FOUND:
1739 req->result = 0;
1740 break;
1741
1742 case ERROR_INVALID_NAME:
1743 case ERROR_PATH_NOT_FOUND:
1744 case ERROR_NO_MORE_FILES:
1745 errno = ENOENT;
1746 break;
1747
1748 case ERROR_NOT_ENOUGH_MEMORY:
1749 errno = ENOMEM;
1750 break;
1751
1752 default:
1753 errno = EINVAL;
1754 break;
1755 }
1756
1757 return;
1758 }
1759 }
1760#else
1215 self->dirp = dirp = opendir (req->ptr1); 1761 dirp = opendir (req->ptr1);
1762
1763 if (!dirp)
1764 return;
1765#endif
1766
1767 if (req->flags & EIO_FLAG_PTR1_FREE)
1768 free (req->ptr1);
1769
1216 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1770 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1217 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; 1771 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1218 req->ptr2 = names = malloc (namesalloc); 1772 req->ptr2 = names = malloc (namesalloc);
1219 X_UNLOCK (wrklock);
1220 1773
1221 if (dirp && names && (!flags || dents)) 1774 if (!names || (flags && !dents))
1775 return;
1776
1222 for (;;) 1777 for (;;)
1223 { 1778 {
1779 int done;
1780
1781#ifdef _WIN32
1782 done = !dirp;
1783#else
1224 errno = 0; 1784 errno = 0;
1225 entp = readdir (dirp); 1785 entp = readdir (dirp);
1786 done = !entp;
1787#endif
1226 1788
1227 if (!entp) 1789 if (done)
1228 { 1790 {
1791#ifndef _WIN32
1792 int old_errno = errno;
1793 closedir (dirp);
1794 errno = old_errno;
1795
1229 if (errno) 1796 if (errno)
1230 break; 1797 break;
1798#endif
1231 1799
1232 /* sort etc. */ 1800 /* sort etc. */
1233 req->int1 = flags; 1801 req->int1 = flags;
1234 req->result = dentoffs; 1802 req->result = dentoffs;
1235 1803
1236 if (flags & EIO_READDIR_STAT_ORDER) 1804 if (flags & EIO_READDIR_STAT_ORDER)
1237 eio_dent_sort (dents, dentoffs, 0, inode_bits); /* sort by inode exclusively */ 1805 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1238 else if (flags & EIO_READDIR_DIRS_FIRST) 1806 else if (flags & EIO_READDIR_DIRS_FIRST)
1239 if (flags & EIO_READDIR_FOUND_UNKNOWN) 1807 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1240 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */ 1808 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1241 else 1809 else
1810 {
1811 /* in this case, all is known, and we just put dirs first and sort them */
1812 eio_dirent *oth = dents + dentoffs;
1813 eio_dirent *dir = dents;
1814
1815 /* now partition dirs to the front, and non-dirs to the back */
1816 /* by walking from both sides and swapping if necessary */
1817 while (oth > dir)
1818 {
1819 if (dir->type == EIO_DT_DIR)
1820 ++dir;
1821 else if ((--oth)->type == EIO_DT_DIR)
1822 {
1823 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1824
1825 ++dir;
1826 }
1827 }
1828
1829 /* now sort the dirs only (dirs all have the same score) */
1830 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1831 }
1832
1833 break;
1834 }
1835
1836 /* now add the entry to our list(s) */
1837 name = D_NAME (entp);
1838
1839 /* skip . and .. entries */
1840 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1841 {
1842 int len = D_NAMLEN (entp) + 1;
1843
1844 while (ecb_expect_false (namesoffs + len > namesalloc))
1845 {
1846 namesalloc *= 2;
1847 req->ptr2 = names = realloc (names, namesalloc);
1848
1849 if (!names)
1850 break;
1851 }
1852
1853 memcpy (names + namesoffs, name, len);
1854
1855 if (dents)
1856 {
1857 struct eio_dirent *ent;
1858
1859 if (ecb_expect_false (dentoffs == dentalloc))
1242 { 1860 {
1243 /* in this case, all is known, and we just put dirs first and sort them */ 1861 dentalloc *= 2;
1862 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1863
1864 if (!dents)
1865 break;
1866 }
1867
1244 eio_dirent *oth = dents + dentoffs; 1868 ent = dents + dentoffs;
1245 eio_dirent *dir = dents;
1246 1869
1247 /* now partition dirs to the front, and non-dirs to the back */ 1870 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1248 /* by walking from both sides and swapping if necessary */ 1871 ent->namelen = len - 1;
1249 /* also clear score, so it doesn't influence sorting */ 1872 ent->inode = D_INO (entp);
1250 while (oth > dir) 1873
1874 inode_bits |= ent->inode;
1875
1876 switch (D_TYPE (entp))
1877 {
1878 default:
1879 ent->type = EIO_DT_UNKNOWN;
1880 flags |= EIO_READDIR_FOUND_UNKNOWN;
1881 break;
1882
1883 #ifdef DT_FIFO
1884 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1885 #endif
1886 #ifdef DT_CHR
1887 case DT_CHR: ent->type = EIO_DT_CHR; break;
1888 #endif
1889 #ifdef DT_MPC
1890 case DT_MPC: ent->type = EIO_DT_MPC; break;
1891 #endif
1892 #ifdef DT_DIR
1893 case DT_DIR: ent->type = EIO_DT_DIR; break;
1894 #endif
1895 #ifdef DT_NAM
1896 case DT_NAM: ent->type = EIO_DT_NAM; break;
1897 #endif
1898 #ifdef DT_BLK
1899 case DT_BLK: ent->type = EIO_DT_BLK; break;
1900 #endif
1901 #ifdef DT_MPB
1902 case DT_MPB: ent->type = EIO_DT_MPB; break;
1903 #endif
1904 #ifdef DT_REG
1905 case DT_REG: ent->type = EIO_DT_REG; break;
1906 #endif
1907 #ifdef DT_NWK
1908 case DT_NWK: ent->type = EIO_DT_NWK; break;
1909 #endif
1910 #ifdef DT_CMP
1911 case DT_CMP: ent->type = EIO_DT_CMP; break;
1912 #endif
1913 #ifdef DT_LNK
1914 case DT_LNK: ent->type = EIO_DT_LNK; break;
1915 #endif
1916 #ifdef DT_SOCK
1917 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1918 #endif
1919 #ifdef DT_DOOR
1920 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1921 #endif
1922 #ifdef DT_WHT
1923 case DT_WHT: ent->type = EIO_DT_WHT; break;
1924 #endif
1925 }
1926
1927 ent->score = 7;
1928
1929 if (flags & EIO_READDIR_DIRS_FIRST)
1930 {
1931 if (ent->type == EIO_DT_UNKNOWN)
1251 { 1932 {
1252 if (dir->type == EIO_DT_DIR) 1933 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1253 ++dir; 1934 ent->score = 1;
1254 else if ((--oth)->type == EIO_DT_DIR) 1935 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1255 { 1936 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1256 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1257
1258 ++dir;
1259 }
1260 } 1937 }
1261 1938 else if (ent->type == EIO_DT_DIR)
1262 /* now sort the dirs only */ 1939 ent->score = 0;
1263 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1264 } 1940 }
1265
1266 break;
1267 }
1268
1269 /* now add the entry to our list(s) */
1270 name = entp->d_name;
1271
1272 /* skip . and .. entries */
1273 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1274 {
1275 int len = D_NAMLEN (entp) + 1;
1276
1277 while (expect_false (namesoffs + len > namesalloc))
1278 {
1279 namesalloc *= 2;
1280 X_LOCK (wrklock);
1281 req->ptr2 = names = realloc (names, namesalloc);
1282 X_UNLOCK (wrklock);
1283
1284 if (!names)
1285 break;
1286 } 1941 }
1287 1942
1288 memcpy (names + namesoffs, name, len);
1289
1290 if (dents)
1291 {
1292 struct eio_dirent *ent;
1293
1294 if (expect_false (dentoffs == dentalloc))
1295 {
1296 dentalloc *= 2;
1297 X_LOCK (wrklock);
1298 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1299 X_UNLOCK (wrklock);
1300
1301 if (!dents)
1302 break;
1303 }
1304
1305 ent = dents + dentoffs;
1306
1307 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1308 ent->namelen = len - 1;
1309 ent->inode = D_INO (entp);
1310
1311 inode_bits |= ent->inode;
1312
1313 switch (D_TYPE (entp))
1314 {
1315 default:
1316 ent->type = EIO_DT_UNKNOWN;
1317 flags |= EIO_READDIR_FOUND_UNKNOWN;
1318 break;
1319
1320 #ifdef DT_FIFO
1321 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1322 #endif
1323 #ifdef DT_CHR
1324 case DT_CHR: ent->type = EIO_DT_CHR; break;
1325 #endif
1326 #ifdef DT_MPC
1327 case DT_MPC: ent->type = EIO_DT_MPC; break;
1328 #endif
1329 #ifdef DT_DIR
1330 case DT_DIR: ent->type = EIO_DT_DIR; break;
1331 #endif
1332 #ifdef DT_NAM
1333 case DT_NAM: ent->type = EIO_DT_NAM; break;
1334 #endif
1335 #ifdef DT_BLK
1336 case DT_BLK: ent->type = EIO_DT_BLK; break;
1337 #endif
1338 #ifdef DT_MPB
1339 case DT_MPB: ent->type = EIO_DT_MPB; break;
1340 #endif
1341 #ifdef DT_REG
1342 case DT_REG: ent->type = EIO_DT_REG; break;
1343 #endif
1344 #ifdef DT_NWK
1345 case DT_NWK: ent->type = EIO_DT_NWK; break;
1346 #endif
1347 #ifdef DT_CMP
1348 case DT_CMP: ent->type = EIO_DT_CMP; break;
1349 #endif
1350 #ifdef DT_LNK
1351 case DT_LNK: ent->type = EIO_DT_LNK; break;
1352 #endif
1353 #ifdef DT_SOCK
1354 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1355 #endif
1356 #ifdef DT_DOOR
1357 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1358 #endif
1359 #ifdef DT_WHT
1360 case DT_WHT: ent->type = EIO_DT_WHT; break;
1361 #endif
1362 }
1363
1364 ent->score = 7;
1365
1366 if (flags & EIO_READDIR_DIRS_FIRST)
1367 {
1368 if (ent->type == EIO_DT_UNKNOWN)
1369 {
1370 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1371 ent->score = 1;
1372 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1373 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1374 }
1375 else if (ent->type == EIO_DT_DIR)
1376 ent->score = 0;
1377 }
1378 }
1379
1380 namesoffs += len; 1943 namesoffs += len;
1381 ++dentoffs; 1944 ++dentoffs;
1382 } 1945 }
1383 1946
1384 if (EIO_CANCELLED (req)) 1947 if (EIO_CANCELLED (req))
1385 { 1948 {
1386 errno = ECANCELED; 1949 errno = ECANCELED;
1387 break; 1950 break;
1388 } 1951 }
1952
1953#ifdef _WIN32
1954 if (!FindNextFile (dirp, &entp))
1955 {
1956 FindClose (dirp);
1957 dirp = 0;
1389 } 1958 }
1959#endif
1960 }
1390} 1961}
1391 1962
1392#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) 1963/*****************************************************************************/
1393# define eio__msync(a,b,c) ((errno = ENOSYS), -1) 1964/* working directory stuff */
1965
1966#if HAVE_AT
1967
1968#define WD2FD(wd) ((wd) ? ((int)(long)(wd)) - 1 : AT_FDCWD)
1969
1970#ifndef O_SEARCH
1971# define O_SEARCH O_RDONLY
1972#endif
1973
1974eio_wd
1975eio_wd_open_sync (eio_wd wd, const char *path)
1976{
1977 int fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1978
1979 return fd >= 0 ? (eio_wd)(long)(fd + 1) : EIO_INVALID_WD;
1980}
1981
1982static eio_wd
1983eio__wd_open_sync (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
1984{
1985 return eio_wd_open_sync (wd, path);
1986}
1987
1988void
1989eio_wd_close_sync (eio_wd wd)
1990{
1991 int fd = WD2FD (wd);
1992
1993 if (fd >= 0)
1994 close (fd);
1995}
1996
1997static int
1998eio__truncateat (int dirfd, const char *path, off_t length)
1999{
2000 int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
2001 int res;
2002
2003 if (fd < 0)
2004 return fd;
2005
2006 res = ftruncate (fd, length);
2007 close (fd);
2008 return res;
2009}
2010
2011static int
2012eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
2013{
2014 int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
2015 int res;
2016
2017 if (fd < 0)
2018 return fd;
2019
2020 res = fstatvfs (fd, buf);
2021 close (fd);
2022 return res;
2023
2024}
2025
1394#else 2026#else
1395 2027
1396int 2028/* on legacy systems, we represent the working directories simply by their path strings */
1397eio__msync (void *mem, size_t len, int flags) 2029
2030static const char *
2031wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
1398{ 2032{
1399 if (EIO_MS_ASYNC != MS_SYNC 2033 if (!wd || *path == '/')
1400 || EIO_MS_INVALIDATE != MS_INVALIDATE 2034 return path;
1401 || EIO_MS_SYNC != MS_SYNC) 2035
2036 {
2037 int l1 = strlen ((const char *)wd);
2038 int l2 = strlen (path);
2039
2040 char *res = tmpbuf_get (tmpbuf, l1 + l2 + 2);
2041
2042 memcpy (res, wd, l1);
2043 res [l1] = '/';
2044 memcpy (res + l1 + 1, path, l2 + 1);
2045
2046 return res;
2047 }
2048}
2049
2050static eio_wd
2051eio__wd_open_sync (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
2052{
2053 if (*path == '/') /* absolute paths ignore wd */
2054 path = strdup (path);
2055 else if (path [0] == '.' && !path [1]) /* special case '.', as it is common */
2056 return wd;
2057 else
1402 { 2058 {
1403 flags = 0 2059 int len = eio__realpath (tmpbuf, wd, path);
1404 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1405 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1406 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1407 }
1408 2060
1409 return msync (mem, len, flags); 2061 path = EIO_INVALID_WD;
1410}
1411 2062
1412#endif 2063 if (len >= 0)
2064 {
2065 ((char *)tmpbuf->ptr)[len] = 0;
2066 path = strdup (tmpbuf->ptr);
2067 }
2068 }
1413 2069
1414int
1415eio__mtouch (void *mem, size_t len, int flags)
1416{
1417 intptr_t addr = (intptr_t)mem;
1418 intptr_t end = addr + len;
1419#ifdef PAGESIZE
1420 const intptr_t page = PAGESIZE;
1421#else
1422 static intptr_t page;
1423
1424 if (!page) 2070 if (!path)
1425 page = sysconf (_SC_PAGESIZE); 2071 path = EIO_INVALID_WD;
1426#endif
1427 2072
1428 /* round down to start of page, although this is probably useless */ 2073 return (eio_wd)path;
1429 addr &= ~(page - 1); /* assume page size is always a power of two */ 2074}
1430 2075
1431 if (addr < end) 2076eio_wd
1432 if (flags & EIO_MT_MODIFY) /* modify */ 2077eio_wd_open_sync (eio_wd wd, const char *path)
1433 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len); 2078{
1434 else 2079 struct tmpbuf tmpbuf = { 0 };
1435 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len); 2080 wd = eio__wd_open_sync (&tmpbuf, wd, path);
2081 free (tmpbuf.ptr);
1436 2082
1437 return 0; 2083 return wd;
1438} 2084}
2085
2086void
2087eio_wd_close_sync (eio_wd wd)
2088{
2089 if (wd != EIO_INVALID_WD)
2090 free (wd);
2091}
2092
2093#endif
1439 2094
1440/*****************************************************************************/ 2095/*****************************************************************************/
1441 2096
1442#define ALLOC(len) \ 2097#define ALLOC(len) \
1443 if (!req->ptr2) \ 2098 if (!req->ptr2) \
1458{ 2113{
1459 ETP_REQ *req; 2114 ETP_REQ *req;
1460 struct timespec ts; 2115 struct timespec ts;
1461 etp_worker *self = (etp_worker *)thr_arg; 2116 etp_worker *self = (etp_worker *)thr_arg;
1462 2117
2118#if HAVE_PRCTL_SET_NAME
2119 prctl (PR_SET_NAME, (unsigned long)"eio_thread", 0, 0, 0);
2120#endif
2121
1463 /* try to distribute timeouts somewhat randomly */ 2122 /* try to distribute timeouts somewhat evenly */
1464 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 2123 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1465 2124
1466 for (;;) 2125 for (;;)
1467 { 2126 {
2127 ts.tv_sec = 0;
2128
1468 X_LOCK (reqlock); 2129 X_LOCK (reqlock);
1469 2130
1470 for (;;) 2131 for (;;)
1471 { 2132 {
1472 self->req = req = reqq_shift (&req_queue); 2133 req = reqq_shift (&req_queue);
1473 2134
1474 if (req) 2135 if (req)
1475 break; 2136 break;
1476 2137
2138 if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
2139 {
2140 X_UNLOCK (reqlock);
2141 X_LOCK (wrklock);
2142 --started;
2143 X_UNLOCK (wrklock);
2144 goto quit;
2145 }
2146
1477 ++idle; 2147 ++idle;
1478 2148
1479 ts.tv_sec = time (0) + IDLE_TIMEOUT; 2149 if (idle <= max_idle)
1480 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 2150 /* we are allowed to idle, so do so without any timeout */
2151 X_COND_WAIT (reqwait, reqlock);
2152 else
1481 { 2153 {
1482 if (idle > max_idle) 2154 /* initialise timeout once */
1483 { 2155 if (!ts.tv_sec)
1484 --idle; 2156 ts.tv_sec = time (0) + idle_timeout;
1485 X_UNLOCK (reqlock);
1486 X_LOCK (wrklock);
1487 --started;
1488 X_UNLOCK (wrklock);
1489 goto quit;
1490 }
1491 2157
1492 /* we are allowed to idle, so do so without any timeout */
1493 X_COND_WAIT (reqwait, reqlock); 2158 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
2159 ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1494 } 2160 }
1495 2161
1496 --idle; 2162 --idle;
1497 } 2163 }
1498 2164
1501 X_UNLOCK (reqlock); 2167 X_UNLOCK (reqlock);
1502 2168
1503 if (req->type < 0) 2169 if (req->type < 0)
1504 goto quit; 2170 goto quit;
1505 2171
1506 if (!EIO_CANCELLED (req))
1507 ETP_EXECUTE (self, req); 2172 ETP_EXECUTE (self, req);
1508 2173
1509 X_LOCK (reslock); 2174 X_LOCK (reslock);
1510 2175
1511 ++npending; 2176 ++npending;
1512 2177
1513 if (!reqq_push (&res_queue, req) && want_poll_cb) 2178 if (!reqq_push (&res_queue, req) && want_poll_cb)
1514 want_poll_cb (); 2179 want_poll_cb ();
1515 2180
1516 self->req = 0;
1517 etp_worker_clear (self); 2181 etp_worker_clear (self);
1518 2182
1519 X_UNLOCK (reslock); 2183 X_UNLOCK (reslock);
1520 } 2184 }
1521 2185
1522quit: 2186quit:
2187 free (req);
2188
1523 X_LOCK (wrklock); 2189 X_LOCK (wrklock);
1524 etp_worker_free (self); 2190 etp_worker_free (self);
1525 X_UNLOCK (wrklock); 2191 X_UNLOCK (wrklock);
1526 2192
1527 return 0; 2193 return 0;
1528} 2194}
1529 2195
1530/*****************************************************************************/ 2196/*****************************************************************************/
1531 2197
2198int ecb_cold
1532int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 2199eio_init (void (*want_poll)(void), void (*done_poll)(void))
1533{ 2200{
2201#if !HAVE_PREADWRITE
2202 X_MUTEX_CREATE (preadwritelock);
2203#endif
2204
1534 return etp_init (want_poll, done_poll); 2205 return etp_init (want_poll, done_poll);
1535} 2206}
1536 2207
2208ecb_inline void
1537static void eio_api_destroy (eio_req *req) 2209eio_api_destroy (eio_req *req)
1538{ 2210{
1539 free (req); 2211 free (req);
1540} 2212}
1541 2213
1542#define REQ(rtype) \ 2214#define REQ(rtype) \
1561 { \ 2233 { \
1562 eio_api_destroy (req); \ 2234 eio_api_destroy (req); \
1563 return 0; \ 2235 return 0; \
1564 } 2236 }
1565 2237
2238static void
1566static void eio_execute (etp_worker *self, eio_req *req) 2239eio_execute (etp_worker *self, eio_req *req)
1567{ 2240{
1568 errno = 0; 2241#if HAVE_AT
2242 int dirfd;
2243#else
2244 const char *path;
2245#endif
2246
2247 if (ecb_expect_false (EIO_CANCELLED (req)))
2248 {
2249 req->result = -1;
2250 req->errorno = ECANCELED;
2251 return;
2252 }
2253
2254 if (ecb_expect_false (req->wd == EIO_INVALID_WD))
2255 {
2256 req->result = -1;
2257 req->errorno = ENOENT;
2258 return;
2259 }
2260
2261 if (req->type >= EIO_OPEN)
2262 {
2263 #if HAVE_AT
2264 dirfd = WD2FD (req->wd);
2265 #else
2266 path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
2267 #endif
2268 }
1569 2269
1570 switch (req->type) 2270 switch (req->type)
1571 { 2271 {
2272 case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
2273 req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
2274 break;
2275 case EIO_WD_CLOSE: eio_wd_close_sync (req->wd); break;
2276
1572 case EIO_READ: ALLOC (req->size); 2277 case EIO_READ: ALLOC (req->size);
1573 req->result = req->offs >= 0 2278 req->result = req->offs >= 0
1574 ? pread (req->int1, req->ptr2, req->size, req->offs) 2279 ? pread (req->int1, req->ptr2, req->size, req->offs)
1575 : read (req->int1, req->ptr2, req->size); break; 2280 : read (req->int1, req->ptr2, req->size); break;
1576 case EIO_WRITE: req->result = req->offs >= 0 2281 case EIO_WRITE: req->result = req->offs >= 0
1577 ? pwrite (req->int1, req->ptr2, req->size, req->offs) 2282 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1578 : write (req->int1, req->ptr2, req->size); break; 2283 : write (req->int1, req->ptr2, req->size); break;
1579 2284
1580 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break; 2285 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1581 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break; 2286 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
1582 2287
2288#if HAVE_AT
1583 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2289 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1584 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2290 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break;
1585 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2291 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2292 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, AT_SYMLINK_NOFOLLOW); break;
2293 case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break;
2294 case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break;
2295 case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break;
2296 case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break;
2297
2298 case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
2299 case EIO_RMDIR: req->result = unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break;
2300 case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); break;
2301 case EIO_RENAME: req->result = renameat (dirfd, req->ptr1, WD2FD (req->int3), req->ptr2); break;
2302 case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD (req->int3), req->ptr2, 0); break;
2303 case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
2304 case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
2305 case EIO_READLINK: ALLOC (PATH_MAX);
2306 req->result = readlinkat (dirfd, req->ptr1, req->ptr2, PATH_MAX); break;
2307 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2308 req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2309#else
2310 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2311 req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
2312 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1586 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2313 req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
2314 case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
2315 case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
2316 case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
2317 case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
2318
2319 case EIO_UNLINK: req->result = unlink (path ); break;
2320 case EIO_RMDIR: req->result = rmdir (path ); break;
2321 case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
2322 case EIO_RENAME: req->result = rename (path , req->ptr2); break;
2323 case EIO_LINK: req->result = link (path , req->ptr2); break;
2324 case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
2325 case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
2326 case EIO_READLINK: ALLOC (PATH_MAX);
2327 req->result = readlink (path, req->ptr2, PATH_MAX); break;
2328 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2329 req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
2330#endif
2331
2332 case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
2333 {
2334 ALLOC (req->result);
2335 memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
2336 }
2337 break;
2338
1587 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2339 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1588 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; 2340 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1589 2341
1590 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1591 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1592 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS)); 2342 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1593 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break; 2343 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1594 2344
1595 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1596 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; 2345 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1597 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1598 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break; 2346 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1599 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1600 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break; 2347 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1601 2348
1602 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
1603 case EIO_CLOSE: req->result = close (req->int1); break; 2349 case EIO_CLOSE: req->result = close (req->int1); break;
1604 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break; 2350 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1605 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
1606 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1607 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1608 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1609 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1610 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1611 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break;
1612
1613 case EIO_READLINK: ALLOC (NAME_MAX);
1614 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
1615
1616 case EIO_SYNC: req->result = 0; sync (); break; 2351 case EIO_SYNC: req->result = 0; sync (); break;
1617 case EIO_FSYNC: req->result = fsync (req->int1); break; 2352 case EIO_FSYNC: req->result = fsync (req->int1); break;
1618 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 2353 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
2354 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
2355 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1619 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break; 2356 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
2357 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1620 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break; 2358 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1621 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break; 2359 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
2360 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
1622 2361
1623 case EIO_READDIR: eio__scandir (req, self); break; 2362 case EIO_READDIR: eio__scandir (req); break;
1624 2363
1625 case EIO_BUSY: 2364 case EIO_BUSY:
1626#ifdef _WIN32 2365#ifdef _WIN32
1627 Sleep (req->nv1 * 1000.); 2366 Sleep (req->nv1 * 1e3);
1628#else 2367#else
1629 { 2368 {
1630 struct timeval tv; 2369 struct timeval tv;
1631 2370
1632 tv.tv_sec = req->nv1; 2371 tv.tv_sec = req->nv1;
1633 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; 2372 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1634 2373
1635 req->result = select (0, 0, 0, 0, &tv); 2374 req->result = select (0, 0, 0, 0, &tv);
1636 } 2375 }
1637#endif 2376#endif
1638 break; 2377 break;
1653 times = tv; 2392 times = tv;
1654 } 2393 }
1655 else 2394 else
1656 times = 0; 2395 times = 0;
1657 2396
1658
1659 req->result = req->type == EIO_FUTIME 2397 req->result = req->type == EIO_FUTIME
1660 ? futimes (req->int1, times) 2398 ? futimes (req->int1, times)
1661 : utimes (req->ptr1, times); 2399 : utimes (req->ptr1, times);
1662 } 2400 }
1663 break; 2401 break;
1668 case EIO_NOP: 2406 case EIO_NOP:
1669 req->result = 0; 2407 req->result = 0;
1670 break; 2408 break;
1671 2409
1672 case EIO_CUSTOM: 2410 case EIO_CUSTOM:
1673 ((void (*)(eio_req *))req->feed) (req); 2411 req->feed (req);
1674 break; 2412 break;
1675 2413
1676 default: 2414 default:
2415 errno = ENOSYS;
1677 req->result = -1; 2416 req->result = -1;
1678 break; 2417 break;
1679 } 2418 }
1680 2419
1681 req->errorno = errno; 2420 req->errorno = errno;
1706eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2445eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1707{ 2446{
1708 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2447 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1709} 2448}
1710 2449
2450eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2451{
2452 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2453}
2454
2455eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
2456{
2457 REQ (EIO_SYNCFS); req->int1 = fd; SEND;
2458}
2459
2460eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2461{
2462 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2463}
2464
1711eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2465eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1712{ 2466{
1713 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2467 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1714} 2468}
1715 2469
1716eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) 2470eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
1717{ 2471{
1718 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; 2472 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
1719} 2473}
1720 2474
1721eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) 2475eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
1722{ 2476{
1723 REQ (EIO_FDATASYNC); req->int1 = fd; SEND; 2477 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2478}
2479
2480eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2481{
2482 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
1724} 2483}
1725 2484
1726eio_req *eio_close (int fd, int pri, eio_cb cb, void *data) 2485eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
1727{ 2486{
1728 REQ (EIO_CLOSE); req->int1 = fd; SEND; 2487 REQ (EIO_CLOSE); req->int1 = fd; SEND;
1766eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) 2525eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1767{ 2526{
1768 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2527 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1769} 2528}
1770 2529
1771eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2530eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1772{ 2531{
1773 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2532 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1774} 2533}
1775 2534
1776eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) 2535eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
1796eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data) 2555eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1797{ 2556{
1798 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2557 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1799} 2558}
1800 2559
1801eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2560eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1802{ 2561{
1803 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2562 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1804} 2563}
1805 2564
1806eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) 2565eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
1822eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data) 2581eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1823{ 2582{
1824 return eio__1path (EIO_READLINK, path, pri, cb, data); 2583 return eio__1path (EIO_READLINK, path, pri, cb, data);
1825} 2584}
1826 2585
2586eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2587{
2588 return eio__1path (EIO_REALPATH, path, pri, cb, data);
2589}
2590
1827eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) 2591eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1828{ 2592{
1829 return eio__1path (EIO_STAT, path, pri, cb, data); 2593 return eio__1path (EIO_STAT, path, pri, cb, data);
1830} 2594}
1831 2595
1854 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND; 2618 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1855} 2619}
1856 2620
1857eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) 2621eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1858{ 2622{
1859 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; 2623 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1860} 2624}
1861 2625
1862static eio_req * 2626static eio_req *
1863eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2627eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1864{ 2628{
1888eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2652eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1889{ 2653{
1890 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2654 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1891} 2655}
1892 2656
1893eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) 2657eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
1894{ 2658{
1895 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND; 2659 REQ (EIO_CUSTOM); req->feed = execute; SEND;
1896} 2660}
1897 2661
1898#endif 2662#endif
1899 2663
1900eio_req *eio_grp (eio_cb cb, void *data) 2664eio_req *eio_grp (eio_cb cb, void *data)
1909#undef SEND 2673#undef SEND
1910 2674
1911/*****************************************************************************/ 2675/*****************************************************************************/
1912/* grp functions */ 2676/* grp functions */
1913 2677
2678void
1914void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2679eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1915{ 2680{
1916 grp->int2 = limit; 2681 grp->int2 = limit;
1917 grp->feed = feed; 2682 grp->feed = feed;
1918 2683
1919 grp_try_feed (grp); 2684 grp_try_feed (grp);
1920} 2685}
1921 2686
2687void
1922void eio_grp_limit (eio_req *grp, int limit) 2688eio_grp_limit (eio_req *grp, int limit)
1923{ 2689{
1924 grp->int2 = limit; 2690 grp->int2 = limit;
1925 2691
1926 grp_try_feed (grp); 2692 grp_try_feed (grp);
1927} 2693}
1928 2694
2695void
1929void eio_grp_add (eio_req *grp, eio_req *req) 2696eio_grp_add (eio_req *grp, eio_req *req)
1930{ 2697{
1931 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2698 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1932 2699
1933 grp->flags |= EIO_FLAG_GROUPADD; 2700 grp->flags |= EIO_FLAG_GROUPADD;
1934 2701
1945} 2712}
1946 2713
1947/*****************************************************************************/ 2714/*****************************************************************************/
1948/* misc garbage */ 2715/* misc garbage */
1949 2716
2717eio_ssize_t
1950ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2718eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1951{ 2719{
1952 etp_worker wrk;
1953 ssize_t ret;
1954
1955 wrk.dbuf = 0;
1956
1957 ret = eio__sendfile (ofd, ifd, offset, count, &wrk); 2720 return eio__sendfile (ofd, ifd, offset, count);
1958
1959 if (wrk.dbuf)
1960 free (wrk.dbuf);
1961
1962 return ret;
1963} 2721}
1964 2722

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines