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

Comparing libeio/eio.c (file contents):
Revision 1.38 by root, Fri Jun 12 20:01:42 2009 UTC vs.
Revision 1.99 by root, Tue Jul 26 11:07:08 2011 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 XTHREAD_STACKSIZE EIO_STACKSIZE
44#endif 49#endif
45#include "xthread.h" 50#include "xthread.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 #ifdef EIO_STRUCT_STATI64
111 #define stat(path,buf) _stati64 (path,buf)
112 #define fstat(fd,buf) _fstati64 (fd,buf)
113 #endif
114 #define lstat(path,buf) stat (path,buf)
115 #define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
116 #define mkdir(path,mode) _mkdir (path)
117 #define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
118
119 #define chmod(path,mode) _chmod (path, mode)
120 #define dup(fd) _dup (fd)
121 #define dup2(fd1,fd2) _dup2 (fd1, fd2)
122
123 #define fchmod(fd,mode) EIO_ENOSYS ()
124 #define chown(path,uid,gid) EIO_ENOSYS ()
125 #define fchown(fd,uid,gid) EIO_ENOSYS ()
126 #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
127 #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
128 #define mknod(path,mode,dev) EIO_ENOSYS ()
129 #define sync() EIO_ENOSYS ()
130 #define readlink(path,buf,s) EIO_ENOSYS ()
131 #define statvfs(path,buf) EIO_ENOSYS ()
132 #define fstatvfs(fd,buf) EIO_ENOSYS ()
133
134 /* rename() uses MoveFile, which fails to overwrite */
135 #define rename(old,neu) eio__rename (old, neu)
136
137 static int
138 eio__rename (const char *old, const char *neu)
139 {
140 if (MoveFileEx (old, neu, MOVEFILE_REPLACE_EXISTING))
141 return 0;
142
143 /* should steal _dosmaperr */
144 switch (GetLastError ())
145 {
146 case ERROR_FILE_NOT_FOUND:
147 case ERROR_PATH_NOT_FOUND:
148 case ERROR_INVALID_DRIVE:
149 case ERROR_NO_MORE_FILES:
150 case ERROR_BAD_NETPATH:
151 case ERROR_BAD_NET_NAME:
152 case ERROR_BAD_PATHNAME:
153 case ERROR_FILENAME_EXCED_RANGE:
154 errno = ENOENT;
155 break;
156
157 default:
158 errno = EACCES;
159 break;
160 }
161
162 return -1;
163 }
164
165 /* we could even stat and see if it exists */
166 static int
167 symlink (const char *old, const char *neu)
168 {
169 #if WINVER >= 0x0600
170 if (CreateSymbolicLink (neu, old, 1))
171 return 0;
172
173 if (CreateSymbolicLink (neu, old, 0))
174 return 0;
175 #endif
176
177 return EIO_ERRNO (ENOENT, -1);
178 }
179
180 /* POSIX API only */
181 #define CreateHardLink(neu,old,flags) 0
182 #define CreateSymbolicLink(neu,old,flags) 0
183
184 struct statvfs
185 {
186 int dummy;
187 };
188
189 #define DT_DIR EIO_DT_DIR
190 #define DT_REG EIO_DT_REG
191 #define D_NAME(entp) entp.cFileName
192 #define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
193
73#else 194#else
74 195
75# include "config.h"
76# include <sys/time.h> 196 #include <sys/time.h>
77# include <sys/select.h> 197 #include <sys/select.h>
78# include <sys/mman.h> 198 #include <sys/statvfs.h>
79# include <unistd.h> 199 #include <unistd.h>
200 #include <signal.h>
201 #include <dirent.h>
202
203 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
204 #include <sys/mman.h>
205 #endif
206
207 #define D_NAME(entp) entp->d_name
208
209 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
210 #if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
211 #define _DIRENT_HAVE_D_TYPE /* sigh */
212 #define D_INO(de) (de)->d_fileno
213 #define D_NAMLEN(de) (de)->d_namlen
214 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
215 #define D_INO(de) (de)->d_ino
216 #endif
217
218 #ifdef _D_EXACT_NAMLEN
219 #undef D_NAMLEN
220 #define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
221 #endif
222
223 #ifdef _DIRENT_HAVE_D_TYPE
224 #define D_TYPE(de) (de)->d_type
225 #endif
226
227 #ifndef EIO_STRUCT_DIRENT
228 #define EIO_STRUCT_DIRENT struct dirent
229 #endif
230
231#endif
232
233#if HAVE_UTIMES
80# include <utime.h> 234# 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 235#endif
92 236
93#ifdef _D_EXACT_NAMLEN 237#if HAVE_SYS_SYSCALL_H
94# undef D_NAMLEN 238# 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 239#endif
101 240
102# ifndef EIO_STRUCT_DIRENT 241#if HAVE_SYS_PRCTL_H
103# define EIO_STRUCT_DIRENT struct dirent 242# include <sys/prctl.h>
104# endif
105
106#endif 243#endif
107 244
108#if HAVE_SENDFILE 245#if HAVE_SENDFILE
109# if __linux 246# if __linux
110# include <sys/sendfile.h> 247# include <sys/sendfile.h>
111# elif __freebsd 248# elif __FreeBSD__ || defined __APPLE__
112# include <sys/socket.h> 249# include <sys/socket.h>
113# include <sys/uio.h> 250# include <sys/uio.h>
114# elif __hpux 251# elif __hpux
115# include <sys/socket.h> 252# include <sys/socket.h>
116# elif __solaris /* not yet */ 253# elif __solaris
117# include <sys/sendfile.h> 254# include <sys/sendfile.h>
118# else 255# else
119# error sendfile support requested but not available 256# error sendfile support requested but not available
120# endif 257# endif
121#endif 258#endif
125#endif 262#endif
126#ifndef D_INO 263#ifndef D_INO
127# define D_INO(de) 0 264# define D_INO(de) 0
128#endif 265#endif
129#ifndef D_NAMLEN 266#ifndef D_NAMLEN
130# define D_NAMLEN(de) strlen ((de)->d_name) 267# define D_NAMLEN(entp) strlen (D_NAME (entp))
131#endif 268#endif
132
133/* number of seconds after which an idle threads exit */
134#define IDLE_TIMEOUT 10
135 269
136/* used for struct dirent, AIX doesn't provide it */ 270/* used for struct dirent, AIX doesn't provide it */
137#ifndef NAME_MAX 271#ifndef NAME_MAX
138# define NAME_MAX 4096 272# define NAME_MAX 4096
139#endif 273#endif
140 274
275/* used for readlink etc. */
276#ifndef PATH_MAX
277# define PATH_MAX 4096
278#endif
279
141/* buffer size for various temporary buffers */ 280/* buffer size for various temporary buffers */
142#define EIO_BUFSIZE 65536 281#define EIO_BUFSIZE 65536
143 282
144#define dBUF \ 283#define dBUF \
145 char *eio_buf; \
146 ETP_WORKER_LOCK (self); \
147 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 284 char *eio_buf = malloc (EIO_BUFSIZE); \
148 ETP_WORKER_UNLOCK (self); \
149 errno = ENOMEM; \ 285 errno = ENOMEM; \
150 if (!eio_buf) \ 286 if (!eio_buf) \
151 return -1; 287 return -1
288
289#define FUBd \
290 free (eio_buf)
152 291
153#define EIO_TICKS ((1000000 + 1023) >> 10) 292#define EIO_TICKS ((1000000 + 1023) >> 10)
154
155/*****************************************************************************/
156
157#if __GNUC__ >= 3
158# define expect(expr,value) __builtin_expect ((expr),(value))
159#else
160# define expect(expr,value) (expr)
161#endif
162
163#define expect_false(expr) expect ((expr) != 0, 0)
164#define expect_true(expr) expect ((expr) != 0, 1)
165
166/*****************************************************************************/
167 293
168#define ETP_PRI_MIN EIO_PRI_MIN 294#define ETP_PRI_MIN EIO_PRI_MIN
169#define ETP_PRI_MAX EIO_PRI_MAX 295#define ETP_PRI_MAX EIO_PRI_MAX
170 296
171struct etp_worker; 297struct etp_worker;
175static int eio_finish (eio_req *req); 301static int eio_finish (eio_req *req);
176#define ETP_FINISH(req) eio_finish (req) 302#define ETP_FINISH(req) eio_finish (req)
177static void eio_execute (struct etp_worker *self, eio_req *req); 303static void eio_execute (struct etp_worker *self, eio_req *req);
178#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) 304#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
179 305
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/*****************************************************************************/ 306/*****************************************************************************/
198 307
199#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) 308#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
200 309
201/* calculcate time difference in ~1/EIO_TICKS of a second */ 310/* calculate time difference in ~1/EIO_TICKS of a second */
311ecb_inline int
202static int tvdiff (struct timeval *tv1, struct timeval *tv2) 312tvdiff (struct timeval *tv1, struct timeval *tv2)
203{ 313{
204 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 314 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
205 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 315 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
206} 316}
207 317
214static unsigned int max_poll_reqs; /* reslock */ 324static unsigned int max_poll_reqs; /* reslock */
215 325
216static volatile unsigned int nreqs; /* reqlock */ 326static volatile unsigned int nreqs; /* reqlock */
217static volatile unsigned int nready; /* reqlock */ 327static volatile unsigned int nready; /* reqlock */
218static volatile unsigned int npending; /* reqlock */ 328static volatile unsigned int npending; /* reqlock */
219static volatile unsigned int max_idle = 4; 329static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
330static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
220 331
221static mutex_t wrklock = X_MUTEX_INIT; 332static xmutex_t wrklock;
222static mutex_t reslock = X_MUTEX_INIT; 333static xmutex_t reslock;
223static mutex_t reqlock = X_MUTEX_INIT; 334static xmutex_t reqlock;
224static cond_t reqwait = X_COND_INIT; 335static xcond_t reqwait;
225 336
226#if !HAVE_PREADWRITE 337#if !HAVE_PREADWRITE
227/* 338/*
228 * make our pread/pwrite emulation safe against themselves, but not against 339 * make our pread/pwrite emulation safe against themselves, but not against
229 * normal read/write by using a mutex. slows down execution a lot, 340 * normal read/write by using a mutex. slows down execution a lot,
230 * but that's your problem, not mine. 341 * but that's your problem, not mine.
231 */ 342 */
232static mutex_t preadwritelock = X_MUTEX_INIT; 343static xmutex_t preadwritelock;
233#endif 344#endif
234 345
235typedef struct etp_worker 346typedef struct etp_worker
236{ 347{
237 /* locked by wrklock */ 348 /* locked by wrklock */
238 struct etp_worker *prev, *next; 349 struct etp_worker *prev, *next;
239 350
240 thread_t tid; 351 xthread_t tid;
241 352
242 /* locked by reslock, reqlock or wrklock */ 353 /* locked by reslock, reqlock or wrklock */
243 ETP_REQ *req; /* currently processed request */ 354 ETP_REQ *req; /* currently processed request */
244 355
356#ifdef ETP_WORKER_COMMON
245 ETP_WORKER_COMMON 357 ETP_WORKER_COMMON
358#endif
246} etp_worker; 359} etp_worker;
247 360
248static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */ 361static etp_worker wrk_first; /* NOT etp */
249 362
250#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock) 363#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
251#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock) 364#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
252 365
253/* worker threads management */ 366/* worker threads management */
254 367
368static void ecb_cold
255static void etp_worker_clear (etp_worker *wrk) 369etp_worker_clear (etp_worker *wrk)
256{ 370{
257 ETP_WORKER_CLEAR (wrk);
258} 371}
259 372
373static void ecb_cold
260static void etp_worker_free (etp_worker *wrk) 374etp_worker_free (etp_worker *wrk)
261{ 375{
262 wrk->next->prev = wrk->prev; 376 wrk->next->prev = wrk->prev;
263 wrk->prev->next = wrk->next; 377 wrk->prev->next = wrk->next;
264 378
265 free (wrk); 379 free (wrk);
266} 380}
267 381
268static unsigned int etp_nreqs (void) 382static unsigned int
383etp_nreqs (void)
269{ 384{
270 int retval; 385 int retval;
271 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 386 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
272 retval = nreqs; 387 retval = nreqs;
273 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 388 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
274 return retval; 389 return retval;
275} 390}
276 391
277static unsigned int etp_nready (void) 392static unsigned int
393etp_nready (void)
278{ 394{
279 unsigned int retval; 395 unsigned int retval;
280 396
281 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 397 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
282 retval = nready; 398 retval = nready;
283 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 399 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
284 400
285 return retval; 401 return retval;
286} 402}
287 403
288static unsigned int etp_npending (void) 404static unsigned int
405etp_npending (void)
289{ 406{
290 unsigned int retval; 407 unsigned int retval;
291 408
292 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 409 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
293 retval = npending; 410 retval = npending;
294 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 411 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
295 412
296 return retval; 413 return retval;
297} 414}
298 415
299static unsigned int etp_nthreads (void) 416static unsigned int
417etp_nthreads (void)
300{ 418{
301 unsigned int retval; 419 unsigned int retval;
302 420
303 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 421 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
304 retval = started; 422 retval = started;
318} etp_reqq; 436} etp_reqq;
319 437
320static etp_reqq req_queue; 438static etp_reqq req_queue;
321static etp_reqq res_queue; 439static etp_reqq res_queue;
322 440
441static void ecb_noinline ecb_cold
442reqq_init (etp_reqq *q)
443{
444 int pri;
445
446 for (pri = 0; pri < ETP_NUM_PRI; ++pri)
447 q->qs[pri] = q->qe[pri] = 0;
448
449 q->size = 0;
450}
451
452static int ecb_noinline
323static int reqq_push (etp_reqq *q, ETP_REQ *req) 453reqq_push (etp_reqq *q, ETP_REQ *req)
324{ 454{
325 int pri = req->pri; 455 int pri = req->pri;
326 req->next = 0; 456 req->next = 0;
327 457
328 if (q->qe[pri]) 458 if (q->qe[pri])
334 q->qe[pri] = q->qs[pri] = req; 464 q->qe[pri] = q->qs[pri] = req;
335 465
336 return q->size++; 466 return q->size++;
337} 467}
338 468
469static ETP_REQ * ecb_noinline
339static ETP_REQ *reqq_shift (etp_reqq *q) 470reqq_shift (etp_reqq *q)
340{ 471{
341 int pri; 472 int pri;
342 473
343 if (!q->size) 474 if (!q->size)
344 return 0; 475 return 0;
359 } 490 }
360 491
361 abort (); 492 abort ();
362} 493}
363 494
364static void etp_atfork_prepare (void) 495static int ecb_cold
496etp_init (void (*want_poll)(void), void (*done_poll)(void))
365{ 497{
366 X_LOCK (wrklock); 498 X_MUTEX_CREATE (wrklock);
367 X_LOCK (reqlock); 499 X_MUTEX_CREATE (reslock);
368 X_LOCK (reslock); 500 X_MUTEX_CREATE (reqlock);
369#if !HAVE_PREADWRITE 501 X_COND_CREATE (reqwait);
370 X_LOCK (preadwritelock);
371#endif
372}
373 502
374static void etp_atfork_parent (void) 503 reqq_init (&req_queue);
375{ 504 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}
383 505
384static void etp_atfork_child (void) 506 wrk_first.next =
385{ 507 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 }
404 508
405 started = 0; 509 started = 0;
406 idle = 0; 510 idle = 0;
407 nreqs = 0; 511 nreqs = 0;
408 nready = 0; 512 nready = 0;
409 npending = 0; 513 npending = 0;
410 514
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; 515 want_poll_cb = want_poll;
428 done_poll_cb = done_poll; 516 done_poll_cb = done_poll;
429 517
430 return 0; 518 return 0;
431} 519}
432 520
433X_THREAD_PROC (etp_proc); 521X_THREAD_PROC (etp_proc);
434 522
523static void ecb_cold
435static void etp_start_thread (void) 524etp_start_thread (void)
436{ 525{
437 etp_worker *wrk = calloc (1, sizeof (etp_worker)); 526 etp_worker *wrk = calloc (1, sizeof (etp_worker));
438 527
439 /*TODO*/ 528 /*TODO*/
440 assert (("unable to allocate worker thread data", wrk)); 529 assert (("unable to allocate worker thread data", wrk));
453 free (wrk); 542 free (wrk);
454 543
455 X_UNLOCK (wrklock); 544 X_UNLOCK (wrklock);
456} 545}
457 546
547static void
458static void etp_maybe_start_thread (void) 548etp_maybe_start_thread (void)
459{ 549{
460 if (expect_true (etp_nthreads () >= wanted)) 550 if (ecb_expect_true (etp_nthreads () >= wanted))
461 return; 551 return;
462 552
463 /* todo: maybe use idle here, but might be less exact */ 553 /* todo: maybe use idle here, but might be less exact */
464 if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ())) 554 if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
465 return; 555 return;
466 556
467 etp_start_thread (); 557 etp_start_thread ();
468} 558}
469 559
560static void ecb_cold
470static void etp_end_thread (void) 561etp_end_thread (void)
471{ 562{
472 eio_req *req = calloc (1, sizeof (eio_req)); 563 eio_req *req = calloc (1, sizeof (eio_req));
473 564
474 req->type = -1; 565 req->type = -1;
475 req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 566 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
482 X_LOCK (wrklock); 573 X_LOCK (wrklock);
483 --started; 574 --started;
484 X_UNLOCK (wrklock); 575 X_UNLOCK (wrklock);
485} 576}
486 577
487static int etp_poll (void) 578static int
579etp_poll (void)
488{ 580{
489 unsigned int maxreqs; 581 unsigned int maxreqs;
490 unsigned int maxtime; 582 unsigned int maxtime;
491 struct timeval tv_start, tv_now; 583 struct timeval tv_start, tv_now;
492 584
522 614
523 X_LOCK (reqlock); 615 X_LOCK (reqlock);
524 --nreqs; 616 --nreqs;
525 X_UNLOCK (reqlock); 617 X_UNLOCK (reqlock);
526 618
527 if (expect_false (req->type == EIO_GROUP && req->size)) 619 if (ecb_expect_false (req->type == EIO_GROUP && req->size))
528 { 620 {
529 req->int1 = 1; /* mark request as delayed */ 621 req->int1 = 1; /* mark request as delayed */
530 continue; 622 continue;
531 } 623 }
532 else 624 else
533 { 625 {
534 int res = ETP_FINISH (req); 626 int res = ETP_FINISH (req);
535 if (expect_false (res)) 627 if (ecb_expect_false (res))
536 return res; 628 return res;
537 } 629 }
538 630
539 if (expect_false (maxreqs && !--maxreqs)) 631 if (ecb_expect_false (maxreqs && !--maxreqs))
540 break; 632 break;
541 633
542 if (maxtime) 634 if (maxtime)
543 { 635 {
544 gettimeofday (&tv_now, 0); 636 gettimeofday (&tv_now, 0);
550 642
551 errno = EAGAIN; 643 errno = EAGAIN;
552 return -1; 644 return -1;
553} 645}
554 646
647static void
555static void etp_cancel (ETP_REQ *req) 648etp_cancel (ETP_REQ *req)
556{ 649{
557 X_LOCK (wrklock); 650 req->cancelled = 1;
558 req->flags |= EIO_FLAG_CANCELLED;
559 X_UNLOCK (wrklock);
560 651
561 eio_grp_cancel (req); 652 eio_grp_cancel (req);
562} 653}
563 654
655static void
564static void etp_submit (ETP_REQ *req) 656etp_submit (ETP_REQ *req)
565{ 657{
566 req->pri -= ETP_PRI_MIN; 658 req->pri -= ETP_PRI_MIN;
567 659
568 if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN; 660 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; 661 if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
570 662
571 if (expect_false (req->type == EIO_GROUP)) 663 if (ecb_expect_false (req->type == EIO_GROUP))
572 { 664 {
573 /* I hope this is worth it :/ */ 665 /* I hope this is worth it :/ */
574 X_LOCK (reqlock); 666 X_LOCK (reqlock);
575 ++nreqs; 667 ++nreqs;
576 X_UNLOCK (reqlock); 668 X_UNLOCK (reqlock);
595 687
596 etp_maybe_start_thread (); 688 etp_maybe_start_thread ();
597 } 689 }
598} 690}
599 691
692static void ecb_cold
600static void etp_set_max_poll_time (double nseconds) 693etp_set_max_poll_time (double nseconds)
601{ 694{
602 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 695 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
603 max_poll_time = nseconds; 696 max_poll_time = nseconds * EIO_TICKS;
604 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 697 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
605} 698}
606 699
700static void ecb_cold
607static void etp_set_max_poll_reqs (unsigned int maxreqs) 701etp_set_max_poll_reqs (unsigned int maxreqs)
608{ 702{
609 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 703 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
610 max_poll_reqs = maxreqs; 704 max_poll_reqs = maxreqs;
611 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 705 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
612} 706}
613 707
708static void ecb_cold
614static void etp_set_max_idle (unsigned int nthreads) 709etp_set_max_idle (unsigned int nthreads)
615{ 710{
616 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 711 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
617 max_idle = nthreads <= 0 ? 1 : nthreads; 712 max_idle = nthreads;
618 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 713 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
619} 714}
620 715
716static void ecb_cold
717etp_set_idle_timeout (unsigned int seconds)
718{
719 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
720 idle_timeout = seconds;
721 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
722}
723
724static void ecb_cold
621static void etp_set_min_parallel (unsigned int nthreads) 725etp_set_min_parallel (unsigned int nthreads)
622{ 726{
623 if (wanted < nthreads) 727 if (wanted < nthreads)
624 wanted = nthreads; 728 wanted = nthreads;
625} 729}
626 730
731static void ecb_cold
627static void etp_set_max_parallel (unsigned int nthreads) 732etp_set_max_parallel (unsigned int nthreads)
628{ 733{
629 if (wanted > nthreads) 734 if (wanted > nthreads)
630 wanted = nthreads; 735 wanted = nthreads;
631 736
632 while (started > wanted) 737 while (started > wanted)
633 etp_end_thread (); 738 etp_end_thread ();
634} 739}
635 740
636/*****************************************************************************/ 741/*****************************************************************************/
637 742
743static void
638static void grp_try_feed (eio_req *grp) 744grp_try_feed (eio_req *grp)
639{ 745{
640 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 746 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
641 { 747 {
642 grp->flags &= ~EIO_FLAG_GROUPADD; 748 grp->flags &= ~EIO_FLAG_GROUPADD;
643 749
650 break; 756 break;
651 } 757 }
652 } 758 }
653} 759}
654 760
761static int
655static int grp_dec (eio_req *grp) 762grp_dec (eio_req *grp)
656{ 763{
657 --grp->size; 764 --grp->size;
658 765
659 /* call feeder, if applicable */ 766 /* call feeder, if applicable */
660 grp_try_feed (grp); 767 grp_try_feed (grp);
664 return eio_finish (grp); 771 return eio_finish (grp);
665 else 772 else
666 return 0; 773 return 0;
667} 774}
668 775
776static void
669void eio_destroy (eio_req *req) 777eio_destroy (eio_req *req)
670{ 778{
671 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); 779 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
672 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); 780 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
673 781
674 EIO_DESTROY (req); 782 EIO_DESTROY (req);
675} 783}
676 784
785static int
677static int eio_finish (eio_req *req) 786eio_finish (eio_req *req)
678{ 787{
679 int res = EIO_FINISH (req); 788 int res = EIO_FINISH (req);
680 789
681 if (req->grp) 790 if (req->grp)
682 { 791 {
690 if (grp->grp_first == req) 799 if (grp->grp_first == req)
691 grp->grp_first = req->grp_next; 800 grp->grp_first = req->grp_next;
692 801
693 res2 = grp_dec (grp); 802 res2 = grp_dec (grp);
694 803
695 if (!res && res2) 804 if (!res)
696 res = res2; 805 res = res2;
697 } 806 }
698 807
699 eio_destroy (req); 808 eio_destroy (req);
700 809
701 return res; 810 return res;
702} 811}
703 812
813void
704void eio_grp_cancel (eio_req *grp) 814eio_grp_cancel (eio_req *grp)
705{ 815{
706 for (grp = grp->grp_first; grp; grp = grp->grp_next) 816 for (grp = grp->grp_first; grp; grp = grp->grp_next)
707 eio_cancel (grp); 817 eio_cancel (grp);
708} 818}
709 819
820void
710void eio_cancel (eio_req *req) 821eio_cancel (eio_req *req)
711{ 822{
712 etp_cancel (req); 823 etp_cancel (req);
713} 824}
714 825
826void
715void eio_submit (eio_req *req) 827eio_submit (eio_req *req)
716{ 828{
717 etp_submit (req); 829 etp_submit (req);
718} 830}
719 831
720unsigned int eio_nreqs (void) 832unsigned int
833eio_nreqs (void)
721{ 834{
722 return etp_nreqs (); 835 return etp_nreqs ();
723} 836}
724 837
725unsigned int eio_nready (void) 838unsigned int
839eio_nready (void)
726{ 840{
727 return etp_nready (); 841 return etp_nready ();
728} 842}
729 843
730unsigned int eio_npending (void) 844unsigned int
845eio_npending (void)
731{ 846{
732 return etp_npending (); 847 return etp_npending ();
733} 848}
734 849
735unsigned int eio_nthreads (void) 850unsigned int ecb_cold
851eio_nthreads (void)
736{ 852{
737 return etp_nthreads (); 853 return etp_nthreads ();
738} 854}
739 855
856void ecb_cold
740void eio_set_max_poll_time (double nseconds) 857eio_set_max_poll_time (double nseconds)
741{ 858{
742 etp_set_max_poll_time (nseconds); 859 etp_set_max_poll_time (nseconds);
743} 860}
744 861
862void ecb_cold
745void eio_set_max_poll_reqs (unsigned int maxreqs) 863eio_set_max_poll_reqs (unsigned int maxreqs)
746{ 864{
747 etp_set_max_poll_reqs (maxreqs); 865 etp_set_max_poll_reqs (maxreqs);
748} 866}
749 867
868void ecb_cold
750void eio_set_max_idle (unsigned int nthreads) 869eio_set_max_idle (unsigned int nthreads)
751{ 870{
752 etp_set_max_idle (nthreads); 871 etp_set_max_idle (nthreads);
753} 872}
754 873
874void ecb_cold
875eio_set_idle_timeout (unsigned int seconds)
876{
877 etp_set_idle_timeout (seconds);
878}
879
880void ecb_cold
755void eio_set_min_parallel (unsigned int nthreads) 881eio_set_min_parallel (unsigned int nthreads)
756{ 882{
757 etp_set_min_parallel (nthreads); 883 etp_set_min_parallel (nthreads);
758} 884}
759 885
886void ecb_cold
760void eio_set_max_parallel (unsigned int nthreads) 887eio_set_max_parallel (unsigned int nthreads)
761{ 888{
762 etp_set_max_parallel (nthreads); 889 etp_set_max_parallel (nthreads);
763} 890}
764 891
765int eio_poll (void) 892int eio_poll (void)
774# undef pread 901# undef pread
775# undef pwrite 902# undef pwrite
776# define pread eio__pread 903# define pread eio__pread
777# define pwrite eio__pwrite 904# define pwrite eio__pwrite
778 905
779static ssize_t 906static eio_ssize_t
780eio__pread (int fd, void *buf, size_t count, off_t offset) 907eio__pread (int fd, void *buf, size_t count, off_t offset)
781{ 908{
782 ssize_t res; 909 eio_ssize_t res;
783 off_t ooffset; 910 off_t ooffset;
784 911
785 X_LOCK (preadwritelock); 912 X_LOCK (preadwritelock);
786 ooffset = lseek (fd, 0, SEEK_CUR); 913 ooffset = lseek (fd, 0, SEEK_CUR);
787 lseek (fd, offset, SEEK_SET); 914 lseek (fd, offset, SEEK_SET);
790 X_UNLOCK (preadwritelock); 917 X_UNLOCK (preadwritelock);
791 918
792 return res; 919 return res;
793} 920}
794 921
795static ssize_t 922static eio_ssize_t
796eio__pwrite (int fd, void *buf, size_t count, off_t offset) 923eio__pwrite (int fd, void *buf, size_t count, off_t offset)
797{ 924{
798 ssize_t res; 925 eio_ssize_t res;
799 off_t ooffset; 926 off_t ooffset;
800 927
801 X_LOCK (preadwritelock); 928 X_LOCK (preadwritelock);
802 ooffset = lseek (fd, 0, SEEK_CUR); 929 ooffset = lseek (fd, 0, SEEK_CUR);
803 lseek (fd, offset, SEEK_SET); 930 lseek (fd, offset, SEEK_SET);
807 934
808 return res; 935 return res;
809} 936}
810#endif 937#endif
811 938
812#ifndef HAVE_FUTIMES 939#ifndef HAVE_UTIMES
813 940
814# undef utimes 941# undef utimes
815# undef futimes
816# define utimes(path,times) eio__utimes (path, times) 942# define utimes(path,times) eio__utimes (path, times)
817# define futimes(fd,times) eio__futimes (fd, times)
818 943
819static int 944static int
820eio__utimes (const char *filename, const struct timeval times[2]) 945eio__utimes (const char *filename, const struct timeval times[2])
821{ 946{
822 if (times) 947 if (times)
830 } 955 }
831 else 956 else
832 return utime (filename, 0); 957 return utime (filename, 0);
833} 958}
834 959
960#endif
961
962#ifndef HAVE_FUTIMES
963
964# undef futimes
965# define futimes(fd,times) eio__futimes (fd, times)
966
967static int
835static int eio__futimes (int fd, const struct timeval tv[2]) 968eio__futimes (int fd, const struct timeval tv[2])
836{ 969{
837 errno = ENOSYS; 970 errno = ENOSYS;
838 return -1; 971 return -1;
839} 972}
840 973
843#if !HAVE_FDATASYNC 976#if !HAVE_FDATASYNC
844# undef fdatasync 977# undef fdatasync
845# define fdatasync(fd) fsync (fd) 978# define fdatasync(fd) fsync (fd)
846#endif 979#endif
847 980
981static int
982eio__syncfs (int fd)
983{
984 int res;
985
986#if HAVE_SYS_SYNCFS
987 res = (int)syscall (__NR_syncfs, (int)(fd));
988#else
989 res = -1;
990 errno = ENOSYS;
991#endif
992
993 if (res < 0 && errno == ENOSYS && fd >= 0)
994 sync ();
995
996 return res;
997}
998
848/* sync_file_range always needs emulation */ 999/* sync_file_range always needs emulation */
849int 1000static int
850eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags) 1001eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
851{ 1002{
852#if HAVE_SYNC_FILE_RANGE 1003#if HAVE_SYNC_FILE_RANGE
853 int res; 1004 int res;
854 1005
867 if (!res || errno != ENOSYS) 1018 if (!res || errno != ENOSYS)
868 return res; 1019 return res;
869#endif 1020#endif
870 1021
871 /* even though we could play tricks with the flags, it's better to always 1022 /* 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 */ 1023 * call fdatasync, as that matches the expectation of its users best */
873 return fdatasync (fd); 1024 return fdatasync (fd);
1025}
1026
1027static int
1028eio__fallocate (int fd, int mode, off_t offset, size_t len)
1029{
1030#if HAVE_FALLOCATE
1031 return fallocate (fd, mode, offset, len);
1032#else
1033 errno = ENOSYS;
1034 return -1;
1035#endif
874} 1036}
875 1037
876#if !HAVE_READAHEAD 1038#if !HAVE_READAHEAD
877# undef readahead 1039# undef readahead
878# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 1040# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
879 1041
880static ssize_t 1042static eio_ssize_t
881eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) 1043eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
882{ 1044{
883 size_t todo = count; 1045 size_t todo = count;
884 dBUF; 1046 dBUF;
885 1047
890 pread (fd, eio_buf, len, offset); 1052 pread (fd, eio_buf, len, offset);
891 offset += len; 1053 offset += len;
892 todo -= len; 1054 todo -= len;
893 } 1055 }
894 1056
1057 FUBd;
1058
895 errno = 0; 1059 errno = 0;
896 return count; 1060 return count;
897} 1061}
898 1062
899#endif 1063#endif
900 1064
901/* sendfile always needs emulation */ 1065/* sendfile always needs emulation */
902static ssize_t 1066static eio_ssize_t
903eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) 1067eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
904{ 1068{
1069 eio_ssize_t written = 0;
905 ssize_t res; 1070 eio_ssize_t res;
906 1071
907 if (!count) 1072 if (!count)
908 return 0; 1073 return 0;
909 1074
1075 for (;;)
1076 {
1077#ifdef __APPLE__
1078# undef HAVE_SENDFILE /* broken, as everything on os x */
1079#endif
910#if HAVE_SENDFILE 1080#if HAVE_SENDFILE
911# if __linux 1081# if __linux
1082 off_t soffset = offset;
912 res = sendfile (ofd, ifd, &offset, count); 1083 res = sendfile (ofd, ifd, &soffset, count);
913 1084
914# elif __freebsd 1085# elif __FreeBSD__
915 /* 1086 /*
916 * Of course, the freebsd sendfile is a dire hack with no thoughts 1087 * Of course, the freebsd sendfile is a dire hack with no thoughts
917 * wasted on making it similar to other I/O functions. 1088 * wasted on making it similar to other I/O functions.
918 */ 1089 */
919 {
920 off_t sbytes; 1090 off_t sbytes;
921 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 1091 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
922 1092
923 if (res < 0 && sbytes) 1093 #if 0 /* according to the manpage, this is correct, but broken behaviour */
924 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 1094 /* freebsd' sendfile will return 0 on success */
1095 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
1096 /* not on e.g. EIO or EPIPE - sounds broken */
1097 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
925 res = sbytes; 1098 res = sbytes;
926 } 1099 #endif
1100
1101 /* according to source inspection, this is correct, and useful behaviour */
1102 if (sbytes)
1103 res = sbytes;
1104
1105# elif defined (__APPLE__)
1106 off_t sbytes = count;
1107 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
1108
1109 /* according to the manpage, sbytes is always valid */
1110 if (sbytes)
1111 res = sbytes;
927 1112
928# elif __hpux 1113# elif __hpux
929 res = sendfile (ofd, ifd, offset, count, 0, 0); 1114 res = sendfile (ofd, ifd, offset, count, 0, 0);
930 1115
931# elif __solaris 1116# elif __solaris
932 {
933 struct sendfilevec vec; 1117 struct sendfilevec vec;
934 size_t sbytes; 1118 size_t sbytes;
935 1119
936 vec.sfv_fd = ifd; 1120 vec.sfv_fd = ifd;
937 vec.sfv_flag = 0; 1121 vec.sfv_flag = 0;
938 vec.sfv_off = offset; 1122 vec.sfv_off = offset;
939 vec.sfv_len = count; 1123 vec.sfv_len = count;
940 1124
941 res = sendfilev (ofd, &vec, 1, &sbytes); 1125 res = sendfilev (ofd, &vec, 1, &sbytes);
942 1126
943 if (res < 0 && sbytes) 1127 if (res < 0 && sbytes)
944 res = sbytes; 1128 res = sbytes;
945 }
946 1129
947# endif 1130# endif
1131
1132#elif defined (_WIN32) && 0
1133 /* does not work, just for documentation of what would need to be done */
1134 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
1135 /* libeio guarantees that the file offset does not change, and windows */
1136 /* has no way to get an independent handle to the same file description */
1137 HANDLE h = TO_SOCKET (ifd);
1138 SetFilePointer (h, offset, 0, FILE_BEGIN);
1139 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1140
948#else 1141#else
949 res = -1; 1142 res = -1;
950 errno = ENOSYS; 1143 errno = ENOSYS;
951#endif 1144#endif
952 1145
1146 /* we assume sendfile can copy at least 128mb in one go */
1147 if (res <= 128 * 1024 * 1024)
1148 {
1149 if (res > 0)
1150 written += res;
1151
1152 if (written)
1153 return written;
1154
1155 break;
1156 }
1157 else
1158 {
1159 /* if we requested more, then probably the kernel was lazy */
1160 written += res;
1161 offset += res;
1162 count -= res;
1163
1164 if (!count)
1165 return written;
1166 }
1167 }
1168
953 if (res < 0 1169 if (res < 0
954 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1170 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1171 /* BSDs */
1172#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1173 || errno == ENOTSUP
1174#endif
1175#ifdef EOPNOTSUPP /* windows */
1176 || errno == EOPNOTSUPP /* BSDs */
1177#endif
955#if __solaris 1178#if __solaris
956 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1179 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
957#endif 1180#endif
958 ) 1181 )
959 ) 1182 )
963 1186
964 res = 0; 1187 res = 0;
965 1188
966 while (count) 1189 while (count)
967 { 1190 {
968 ssize_t cnt; 1191 eio_ssize_t cnt;
969 1192
970 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 1193 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
971 1194
972 if (cnt <= 0) 1195 if (cnt <= 0)
973 { 1196 {
985 1208
986 offset += cnt; 1209 offset += cnt;
987 res += cnt; 1210 res += cnt;
988 count -= cnt; 1211 count -= cnt;
989 } 1212 }
1213
1214 FUBd;
990 } 1215 }
991 1216
992 return res; 1217 return res;
1218}
1219
1220#ifdef PAGESIZE
1221# define eio_pagesize() PAGESIZE
1222#else
1223static intptr_t
1224eio_pagesize (void)
1225{
1226 static intptr_t page;
1227
1228 if (!page)
1229 page = sysconf (_SC_PAGESIZE);
1230
1231 return page;
1232}
1233#endif
1234
1235static void
1236eio_page_align (void **addr, size_t *length)
1237{
1238 intptr_t mask = eio_pagesize () - 1;
1239
1240 /* round down addr */
1241 intptr_t adj = mask & (intptr_t)*addr;
1242
1243 *addr = (void *)((intptr_t)*addr - adj);
1244 *length += adj;
1245
1246 /* round up length */
1247 *length = (*length + mask) & ~mask;
1248}
1249
1250#if !_POSIX_MEMLOCK
1251# define eio__mlockall(a) EIO_ENOSYS ()
1252#else
1253
1254static int
1255eio__mlockall (int flags)
1256{
1257 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1258 extern int mallopt (int, int);
1259 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1260 #endif
1261
1262 if (EIO_MCL_CURRENT != MCL_CURRENT
1263 || EIO_MCL_FUTURE != MCL_FUTURE)
1264 {
1265 flags = 0
1266 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1267 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1268 }
1269
1270 return mlockall (flags);
1271}
1272#endif
1273
1274#if !_POSIX_MEMLOCK_RANGE
1275# define eio__mlock(a,b) EIO_ENOSYS ()
1276#else
1277
1278static int
1279eio__mlock (void *addr, size_t length)
1280{
1281 eio_page_align (&addr, &length);
1282
1283 return mlock (addr, length);
1284}
1285
1286#endif
1287
1288#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1289# define eio__msync(a,b,c) EIO_ENOSYS ()
1290#else
1291
1292static int
1293eio__msync (void *mem, size_t len, int flags)
1294{
1295 eio_page_align (&mem, &len);
1296
1297 if (EIO_MS_ASYNC != MS_SYNC
1298 || EIO_MS_INVALIDATE != MS_INVALIDATE
1299 || EIO_MS_SYNC != MS_SYNC)
1300 {
1301 flags = 0
1302 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1303 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1304 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1305 }
1306
1307 return msync (mem, len, flags);
1308}
1309
1310#endif
1311
1312static int
1313eio__mtouch (eio_req *req)
1314{
1315 void *mem = req->ptr2;
1316 size_t len = req->size;
1317 int flags = req->int1;
1318
1319 eio_page_align (&mem, &len);
1320
1321 {
1322 intptr_t addr = (intptr_t)mem;
1323 intptr_t end = addr + len;
1324 intptr_t page = eio_pagesize ();
1325
1326 if (addr < end)
1327 if (flags & EIO_MT_MODIFY) /* modify */
1328 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1329 else
1330 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1331 }
1332
1333 return 0;
1334}
1335
1336/*****************************************************************************/
1337/* requests implemented outside eio_execute, because they are so large */
1338
1339static void
1340eio__realpath (eio_req *req, etp_worker *self)
1341{
1342 char *rel = req->ptr1;
1343 char *res;
1344 char *tmp1, *tmp2;
1345#if SYMLOOP_MAX > 32
1346 int symlinks = SYMLOOP_MAX;
1347#else
1348 int symlinks = 32;
1349#endif
1350
1351 req->result = -1;
1352
1353 errno = EINVAL;
1354 if (!rel)
1355 return;
1356
1357 errno = ENOENT;
1358 if (!*rel)
1359 return;
1360
1361 if (!req->ptr2)
1362 {
1363 X_LOCK (wrklock);
1364 req->flags |= EIO_FLAG_PTR2_FREE;
1365 X_UNLOCK (wrklock);
1366 req->ptr2 = malloc (PATH_MAX * 3);
1367
1368 errno = ENOMEM;
1369 if (!req->ptr2)
1370 return;
1371 }
1372
1373 res = req->ptr2;
1374 tmp1 = res + PATH_MAX;
1375 tmp2 = tmp1 + PATH_MAX;
1376
1377#if 0 /* disabled, the musl way to do things is just too racy */
1378#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1379 /* on linux we may be able to ask the kernel */
1380 {
1381 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1382
1383 if (fd >= 0)
1384 {
1385 sprintf (tmp1, "/proc/self/fd/%d", fd);
1386 req->result = readlink (tmp1, res, PATH_MAX);
1387 close (fd);
1388
1389 /* here we should probably stat the open file and the disk file, to make sure they still match */
1390
1391 if (req->result > 0)
1392 goto done;
1393 }
1394 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1395 return;
1396 }
1397#endif
1398#endif
1399
1400 if (*rel != '/')
1401 {
1402 if (!getcwd (res, PATH_MAX))
1403 return;
1404
1405 if (res [1]) /* only use if not / */
1406 res += strlen (res);
1407 }
1408
1409 while (*rel)
1410 {
1411 eio_ssize_t len, linklen;
1412 char *beg = rel;
1413
1414 while (*rel && *rel != '/')
1415 ++rel;
1416
1417 len = rel - beg;
1418
1419 if (!len) /* skip slashes */
1420 {
1421 ++rel;
1422 continue;
1423 }
1424
1425 if (beg [0] == '.')
1426 {
1427 if (len == 1)
1428 continue; /* . - nop */
1429
1430 if (beg [1] == '.' && len == 2)
1431 {
1432 /* .. - back up one component, if possible */
1433
1434 while (res != req->ptr2)
1435 if (*--res == '/')
1436 break;
1437
1438 continue;
1439 }
1440 }
1441
1442 errno = ENAMETOOLONG;
1443 if (res + 1 + len + 1 >= tmp1)
1444 return;
1445
1446 /* copy one component */
1447 *res = '/';
1448 memcpy (res + 1, beg, len);
1449
1450 /* zero-terminate, for readlink */
1451 res [len + 1] = 0;
1452
1453 /* now check if it's a symlink */
1454 linklen = readlink (req->ptr2, tmp1, PATH_MAX);
1455
1456 if (linklen < 0)
1457 {
1458 if (errno != EINVAL)
1459 return;
1460
1461 /* it's a normal directory. hopefully */
1462 res += len + 1;
1463 }
1464 else
1465 {
1466 /* yay, it was a symlink - build new path in tmp2 */
1467 int rellen = strlen (rel);
1468
1469 errno = ENAMETOOLONG;
1470 if (linklen + 1 + rellen >= PATH_MAX)
1471 return;
1472
1473 errno = ELOOP;
1474 if (!--symlinks)
1475 return;
1476
1477 if (*tmp1 == '/')
1478 res = req->ptr2; /* symlink resolves to an absolute path */
1479
1480 /* we need to be careful, as rel might point into tmp2 already */
1481 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1482 tmp2 [linklen] = '/';
1483 memcpy (tmp2, tmp1, linklen);
1484
1485 rel = tmp2;
1486 }
1487 }
1488
1489 /* special case for the lone root path */
1490 if (res == req->ptr2)
1491 *res++ = '/';
1492
1493 req->result = res - (char *)req->ptr2;
1494
1495done:
1496 req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
993} 1497}
994 1498
995static signed char 1499static signed char
996eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) 1500eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
997{ 1501{
998 return b->score - a->score ? b->score - a->score /* works because our signed char is always 0..100 */ 1502 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; 1503 : a->inode < b->inode ? -1
1504 : a->inode > b->inode ? 1
1505 : 0;
1000} 1506}
1001 1507
1508#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1509
1002#define EIO_QSORT_CUTOFF 20 /* quite high, but performs well on many filesystems */ 1510#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1511#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1003 1512
1004static void 1513static void
1005eio_dent_sort (eio_dirent *dents, int size) 1514eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1006{ 1515{
1516 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1517 unsigned char *bit = bits;
1518
1519 assert (CHAR_BIT == 8);
1520 assert (sizeof (eio_dirent) * 8 < 256);
1521 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1522 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1523
1524 if (size <= EIO_SORT_FAST)
1525 return;
1526
1527 /* first prepare an array of bits to test in our radix sort */
1528 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1529 /* inode_bits must contain all inodes ORed together */
1530 /* which is used to skip bits that are 0 everywhere, which is very common */
1531 {
1532 eio_ino_t endianness;
1007 int i, j; 1533 int i, j;
1008 1534
1009 if (size > EIO_QSORT_CUTOFF * 3) /* skip quicksort for small directories */ 1535 /* we store the byte offset of byte n into byte n of "endianness" */
1536 for (i = 0; i < sizeof (eio_ino_t); ++i)
1537 ((unsigned char *)&endianness)[i] = i;
1538
1539 *bit++ = 0;
1540
1541 for (i = 0; i < sizeof (eio_ino_t); ++i)
1542 {
1543 /* shifting off the byte offsets out of "endianness" */
1544 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1545 endianness >>= 8;
1546
1547 for (j = 0; j < 8; ++j)
1548 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1549 *bit++ = offs + j;
1550 }
1551
1552 for (j = 0; j < 8; ++j)
1553 if (score_bits & (1 << j))
1554 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1555 }
1556
1557 /* now actually do the sorting (a variant of MSD radix sort) */
1558 {
1559 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1560 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1561 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1562 int stk_idx = 0;
1563
1564 base_stk [stk_idx] = dents;
1565 end_stk [stk_idx] = dents + size;
1566 bit_stk [stk_idx] = bit - 1;
1567
1568 do
1569 {
1570 base = base_stk [stk_idx];
1571 end = end_stk [stk_idx];
1572 bit = bit_stk [stk_idx];
1573
1574 for (;;)
1575 {
1576 unsigned char O = *bit >> 3;
1577 unsigned char M = 1 << (*bit & 7);
1578
1579 eio_dirent *a = base;
1580 eio_dirent *b = end;
1581
1582 if (b - a < EIO_SORT_CUTOFF)
1583 break;
1584
1585 /* now bit-partition the array on the bit */
1586 /* this ugly asymmetric loop seems to perform much better than typical */
1587 /* partition algos found in the literature */
1588 do
1589 if (!(((unsigned char *)a)[O] & M))
1590 ++a;
1591 else if (!(((unsigned char *)--b)[O] & M))
1592 {
1593 eio_dirent tmp = *a; *a = *b; *b = tmp;
1594 ++a;
1595 }
1596 while (b > a);
1597
1598 /* next bit, or stop, if no bits left in this path */
1599 if (!*--bit)
1600 break;
1601
1602 base_stk [stk_idx] = a;
1603 end_stk [stk_idx] = end;
1604 bit_stk [stk_idx] = bit;
1605 ++stk_idx;
1606
1607 end = a;
1608 }
1609 }
1610 while (stk_idx--);
1611 }
1612}
1613
1614static void
1615eio_dent_insertion_sort (eio_dirent *dents, int size)
1616{
1617 /* first move the smallest element to the front, to act as a sentinel */
1618 {
1619 int i;
1620 eio_dirent *min = dents;
1621
1622 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1623 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1624 if (EIO_DENT_CMP (dents [i], <, *min))
1625 min = &dents [i];
1626
1627 /* swap elements 0 and j (minimum) */
1010 { 1628 {
1011 /* first, use quicksort */ 1629 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 } 1630 }
1631 }
1061 1632
1062 /* use a simple insertion sort at the end */ 1633 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1634 {
1635 eio_dirent *i, *j;
1636
1063 for (i = 1; i < size; ++i) 1637 for (i = dents + 1; i < dents + size; ++i)
1064 { 1638 {
1065 eio_dirent value = dents [i]; 1639 eio_dirent value = *i;
1066 1640
1067 for (j = i - 1; j >= 0 && eio_dent_cmp (&dents [j], &value) > 0; --j) 1641 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1068 dents [j + 1] = dents [j]; 1642 j [1] = j [0];
1069 1643
1070 dents [j + 1] = value; 1644 j [1] = value;
1071 } 1645 }
1646 }
1647}
1648
1649static void
1650eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1651{
1652 if (size <= 1)
1653 return; /* our insertion sort relies on size > 0 */
1654
1655 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1656 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1657 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1658
1659 /* use an insertion sort at the end, or for small arrays, */
1660 /* as insertion sort is more efficient for small partitions */
1661 eio_dent_insertion_sort (dents, size);
1072} 1662}
1073 1663
1074/* read a full directory */ 1664/* read a full directory */
1075static void 1665static void
1076eio__scandir (eio_req *req, etp_worker *self) 1666eio__scandir (eio_req *req, etp_worker *self)
1077{ 1667{
1078 DIR *dirp;
1079 EIO_STRUCT_DIRENT *entp;
1080 char *name, *names; 1668 char *name, *names;
1081 int namesalloc = 4096; 1669 int namesalloc = 4096 - sizeof (void *) * 4;
1082 int namesoffs = 0; 1670 int namesoffs = 0;
1083 int flags = req->int1; 1671 int flags = req->int1;
1084 eio_dirent *dents = 0; 1672 eio_dirent *dents = 0;
1085 int dentalloc = 128; 1673 int dentalloc = 128;
1086 int dentoffs = 0; 1674 int dentoffs = 0;
1675 eio_ino_t inode_bits = 0;
1676#ifdef _WIN32
1677 HANDLE dirp;
1678 WIN32_FIND_DATA entp;
1679#else
1680 DIR *dirp;
1681 EIO_STRUCT_DIRENT *entp;
1682#endif
1087 1683
1088 req->result = -1; 1684 req->result = -1;
1089 1685
1090 if (!(flags & EIO_READDIR_DENTS)) 1686 if (!(flags & EIO_READDIR_DENTS))
1091 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); 1687 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1092 1688
1093 X_LOCK (wrklock); 1689#ifdef _WIN32
1094 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 1690 {
1691 int len = strlen ((const char *)req->ptr1);
1692 char *path = malloc (MAX_PATH);
1693 const char *fmt;
1694
1695 if (!len)
1696 fmt = "./*";
1697 else if (((const char *)req->ptr1)[len - 1] == '/' || ((const char *)req->ptr1)[len - 1] == '\\')
1698 fmt = "%s*";
1699 else
1700 fmt = "%s/*";
1701
1702 _snprintf (path, MAX_PATH, fmt, (const char *)req->ptr1);
1703 dirp = FindFirstFile (path, &entp);
1704 free (path);
1705
1706 if (dirp == INVALID_HANDLE_VALUE)
1707 {
1708 dirp = 0;
1709
1710 /* should steal _dosmaperr */
1711 switch (GetLastError ())
1712 {
1713 case ERROR_FILE_NOT_FOUND:
1714 req->result = 0;
1715 break;
1716
1717 case ERROR_INVALID_NAME:
1718 case ERROR_PATH_NOT_FOUND:
1719 case ERROR_NO_MORE_FILES:
1720 errno = ENOENT;
1721 break;
1722
1723 case ERROR_NOT_ENOUGH_MEMORY:
1724 errno = ENOMEM;
1725 break;
1726
1727 default:
1728 errno = EINVAL;
1729 break;
1730 }
1731 }
1732 }
1733#else
1095 self->dirp = dirp = opendir (req->ptr1); 1734 dirp = opendir (req->ptr1);
1735#endif
1736
1737 if (req->flags & EIO_FLAG_PTR1_FREE)
1738 free (req->ptr1);
1739
1096 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1740 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; 1741 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1099 X_UNLOCK (wrklock); 1742 req->ptr2 = names = malloc (namesalloc);
1100 1743
1101 if (dirp && names && (!flags || dents)) 1744 if (dirp && names && (!flags || dents))
1102 for (;;) 1745 for (;;)
1103 { 1746 {
1747 int done;
1748
1749#ifdef _WIN32
1750 done = !dirp;
1751#else
1104 errno = 0; 1752 errno = 0;
1105 entp = readdir (dirp); 1753 entp = readdir (dirp);
1754 done = !entp;
1755#endif
1106 1756
1107 if (!entp) 1757 if (done)
1108 { 1758 {
1759#ifndef _WIN32
1760 int old_errno = errno;
1761 closedir (dirp);
1762 errno = old_errno;
1763
1109 if (errno) 1764 if (errno)
1110 break; 1765 break;
1766#endif
1111 1767
1112 /* sort etc. */ 1768 /* sort etc. */
1113 req->int1 = flags; 1769 req->int1 = flags;
1114 req->result = dentoffs; 1770 req->result = dentoffs;
1115 1771
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 1772 if (flags & EIO_READDIR_STAT_ORDER)
1125 || !(~flags & (EIO_READDIR_DIRS_FIRST | EIO_READDIR_FOUND_UNKNOWN))) 1773 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) 1774 else if (flags & EIO_READDIR_DIRS_FIRST)
1775 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1776 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1777 else
1128 { 1778 {
1129 /* in this case, all is known, and we just put dirs first and sort them */ 1779 /* in this case, all is known, and we just put dirs first and sort them */
1130 eio_dirent *ent = dents + dentoffs; 1780 eio_dirent *oth = dents + dentoffs;
1131 eio_dirent *dir = dents; 1781 eio_dirent *dir = dents;
1132 1782
1133 /* now move dirs to the front, and non-dirs to the back */ 1783 /* now partition dirs to the front, and non-dirs to the back */
1134 /* by walking from both sides and swapping if necessary */ 1784 /* by walking from both sides and swapping if necessary */
1135 while (ent > dir) 1785 while (oth > dir)
1136 { 1786 {
1137 if (dir->type == DT_DIR) 1787 if (dir->type == EIO_DT_DIR)
1138 ++dir; 1788 ++dir;
1139 else
1140 {
1141 --ent;
1142
1143 if (ent->type == DT_DIR) 1789 else if ((--oth)->type == EIO_DT_DIR)
1144 { 1790 {
1145 eio_dirent tmp = *dir; 1791 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1146 *dir = *ent;
1147 *ent = tmp;
1148 1792
1149 ++dir; 1793 ++dir;
1150 } 1794 }
1151 } 1795 }
1796
1797 /* now sort the dirs only (dirs all have the same score) */
1798 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1152 } 1799 }
1153
1154 /* now sort the dirs only */
1155 eio_dent_sort (dents, dir - dents);
1156 }
1157
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 1800
1168 break; 1801 break;
1169 } 1802 }
1170 1803
1171 /* now add the entry to our list(s) */ 1804 /* now add the entry to our list(s) */
1172 name = entp->d_name; 1805 name = D_NAME (entp);
1173 1806
1174 /* skip . and .. entries */ 1807 /* skip . and .. entries */
1175 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1808 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1176 { 1809 {
1177 int len = D_NAMLEN (entp) + 1; 1810 int len = D_NAMLEN (entp) + 1;
1178 1811
1179 while (expect_false (namesoffs + len > namesalloc)) 1812 while (ecb_expect_false (namesoffs + len > namesalloc))
1180 { 1813 {
1181 namesalloc *= 2; 1814 namesalloc *= 2;
1182 X_LOCK (wrklock);
1183 req->ptr1 = names = realloc (names, namesalloc); 1815 req->ptr2 = names = realloc (names, namesalloc);
1184 X_UNLOCK (wrklock);
1185 1816
1186 if (!names) 1817 if (!names)
1187 break; 1818 break;
1188 } 1819 }
1189 1820
1191 1822
1192 if (dents) 1823 if (dents)
1193 { 1824 {
1194 struct eio_dirent *ent; 1825 struct eio_dirent *ent;
1195 1826
1196 if (expect_false (dentoffs == dentalloc)) 1827 if (ecb_expect_false (dentoffs == dentalloc))
1197 { 1828 {
1198 dentalloc *= 2; 1829 dentalloc *= 2;
1199 X_LOCK (wrklock);
1200 req->ptr2 = dents = realloc (dents, dentalloc * sizeof (eio_dirent)); 1830 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1201 X_UNLOCK (wrklock);
1202 1831
1203 if (!dents) 1832 if (!dents)
1204 break; 1833 break;
1205 } 1834 }
1206 1835
1207 ent = dents + dentoffs; 1836 ent = dents + dentoffs;
1208 1837
1209 ent->name = (char *)(size_t)namesoffs; /* rather dirtily we store the offset in the pointer */ 1838 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1210 ent->namelen = len - 1; 1839 ent->namelen = len - 1;
1211 ent->inode = D_INO (entp); 1840 ent->inode = D_INO (entp);
1841
1842 inode_bits |= ent->inode;
1212 1843
1213 switch (D_TYPE (entp)) 1844 switch (D_TYPE (entp))
1214 { 1845 {
1215 default: 1846 default:
1216 ent->type = EIO_DT_UNKNOWN; 1847 ent->type = EIO_DT_UNKNOWN;
1259 #ifdef DT_WHT 1890 #ifdef DT_WHT
1260 case DT_WHT: ent->type = EIO_DT_WHT; break; 1891 case DT_WHT: ent->type = EIO_DT_WHT; break;
1261 #endif 1892 #endif
1262 } 1893 }
1263 1894
1264 ent->score = 0; 1895 ent->score = 7;
1265 1896
1266 if (flags & EIO_READDIR_DIRS_FIRST) 1897 if (flags & EIO_READDIR_DIRS_FIRST)
1267 { 1898 {
1268 if (ent->type == EIO_DT_UNKNOWN) 1899 if (ent->type == EIO_DT_UNKNOWN)
1269 { 1900 {
1270 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */ 1901 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1271 ent->score = 98; 1902 ent->score = 1;
1272 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */ 1903 else if (!strchr (name, '.')) /* absense 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 */ 1904 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1274 } 1905 }
1275 else if (ent->type == EIO_DT_DIR) 1906 else if (ent->type == EIO_DT_DIR)
1276 ent->score = 100; 1907 ent->score = 0;
1277 } 1908 }
1278 } 1909 }
1279 1910
1280 namesoffs += len; 1911 namesoffs += len;
1281 ++dentoffs; 1912 ++dentoffs;
1284 if (EIO_CANCELLED (req)) 1915 if (EIO_CANCELLED (req))
1285 { 1916 {
1286 errno = ECANCELED; 1917 errno = ECANCELED;
1287 break; 1918 break;
1288 } 1919 }
1920
1921#ifdef _WIN32
1922 if (!FindNextFile (dirp, &entp))
1923 {
1924 FindClose (dirp);
1925 dirp = 0;
1926 }
1927#endif
1289 } 1928 }
1290}
1291
1292#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1293# undef msync
1294# define msync(a,b,c) ((errno = ENOSYS), -1)
1295#endif
1296
1297int
1298eio__mtouch (void *mem, size_t len, int flags)
1299{
1300 intptr_t addr = (intptr_t)mem;
1301 intptr_t end = addr + len;
1302#ifdef PAGESIZE
1303 const intptr_t page = PAGESIZE;
1304#else
1305 static intptr_t page;
1306
1307 if (!page)
1308 page = sysconf (_SC_PAGESIZE);
1309#endif
1310
1311 addr &= ~(page - 1); /* assume page size is always a power of two */
1312
1313 if (addr < end)
1314 if (flags) /* modify */
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
1319 return 0;
1320} 1929}
1321 1930
1322/*****************************************************************************/ 1931/*****************************************************************************/
1323 1932
1324#define ALLOC(len) \ 1933#define ALLOC(len) \
1340{ 1949{
1341 ETP_REQ *req; 1950 ETP_REQ *req;
1342 struct timespec ts; 1951 struct timespec ts;
1343 etp_worker *self = (etp_worker *)thr_arg; 1952 etp_worker *self = (etp_worker *)thr_arg;
1344 1953
1954#if HAVE_PRCTL_SET_NAME
1955 prctl (PR_SET_NAME, (unsigned long)"eio_thread", 0, 0, 0);
1956#endif
1957
1345 /* try to distribute timeouts somewhat randomly */ 1958 /* try to distribute timeouts somewhat evenly */
1346 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 1959 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1347 1960
1348 for (;;) 1961 for (;;)
1349 { 1962 {
1963 ts.tv_sec = 0;
1964
1350 X_LOCK (reqlock); 1965 X_LOCK (reqlock);
1351 1966
1352 for (;;) 1967 for (;;)
1353 { 1968 {
1354 self->req = req = reqq_shift (&req_queue); 1969 self->req = req = reqq_shift (&req_queue);
1355 1970
1356 if (req) 1971 if (req)
1357 break; 1972 break;
1358 1973
1974 if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
1975 {
1976 X_UNLOCK (reqlock);
1977 X_LOCK (wrklock);
1978 --started;
1979 X_UNLOCK (wrklock);
1980 goto quit;
1981 }
1982
1359 ++idle; 1983 ++idle;
1360 1984
1361 ts.tv_sec = time (0) + IDLE_TIMEOUT; 1985 if (idle <= max_idle)
1362 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 1986 /* we are allowed to idle, so do so without any timeout */
1987 X_COND_WAIT (reqwait, reqlock);
1988 else
1363 { 1989 {
1364 if (idle > max_idle) 1990 /* initialise timeout once */
1365 { 1991 if (!ts.tv_sec)
1366 --idle; 1992 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 1993
1374 /* we are allowed to idle, so do so without any timeout */
1375 X_COND_WAIT (reqwait, reqlock); 1994 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
1995 ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1376 } 1996 }
1377 1997
1378 --idle; 1998 --idle;
1379 } 1999 }
1380 2000
1383 X_UNLOCK (reqlock); 2003 X_UNLOCK (reqlock);
1384 2004
1385 if (req->type < 0) 2005 if (req->type < 0)
1386 goto quit; 2006 goto quit;
1387 2007
1388 if (!EIO_CANCELLED (req))
1389 ETP_EXECUTE (self, req); 2008 ETP_EXECUTE (self, req);
1390 2009
1391 X_LOCK (reslock); 2010 X_LOCK (reslock);
1392 2011
1393 ++npending; 2012 ++npending;
1394 2013
1409 return 0; 2028 return 0;
1410} 2029}
1411 2030
1412/*****************************************************************************/ 2031/*****************************************************************************/
1413 2032
2033int ecb_cold
1414int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 2034eio_init (void (*want_poll)(void), void (*done_poll)(void))
1415{ 2035{
2036#if !HAVE_PREADWRITE
2037 X_MUTEX_CREATE (preadwritelock);
2038#endif
2039
1416 return etp_init (want_poll, done_poll); 2040 return etp_init (want_poll, done_poll);
1417} 2041}
1418 2042
2043ecb_inline void
1419static void eio_api_destroy (eio_req *req) 2044eio_api_destroy (eio_req *req)
1420{ 2045{
1421 free (req); 2046 free (req);
1422} 2047}
1423 2048
1424#define REQ(rtype) \ 2049#define REQ(rtype) \
1443 { \ 2068 { \
1444 eio_api_destroy (req); \ 2069 eio_api_destroy (req); \
1445 return 0; \ 2070 return 0; \
1446 } 2071 }
1447 2072
2073static void
1448static void eio_execute (etp_worker *self, eio_req *req) 2074eio_execute (etp_worker *self, eio_req *req)
1449{ 2075{
1450 errno = 0; 2076 if (ecb_expect_false (EIO_CANCELLED (req)))
2077 {
2078 req->result = -1;
2079 req->errorno = ECANCELED;
2080 return;
2081 }
1451 2082
1452 switch (req->type) 2083 switch (req->type)
1453 { 2084 {
1454 case EIO_READ: ALLOC (req->size); 2085 case EIO_READ: ALLOC (req->size);
1455 req->result = req->offs >= 0 2086 req->result = req->offs >= 0
1458 case EIO_WRITE: req->result = req->offs >= 0 2089 case EIO_WRITE: req->result = req->offs >= 0
1459 ? pwrite (req->int1, req->ptr2, req->size, req->offs) 2090 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1460 : write (req->int1, req->ptr2, req->size); break; 2091 : write (req->int1, req->ptr2, req->size); break;
1461 2092
1462 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break; 2093 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; 2094 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
1464 2095
1465 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2096 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1466 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2097 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1467 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2098 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1468 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2099 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1469 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2100 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1470 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; 2101 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
2102
2103 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2104 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2105 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2106 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1471 2107
1472 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; 2108 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1473 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; 2109 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; 2110 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1475 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break; 2111 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1483 case EIO_RMDIR: req->result = rmdir (req->ptr1); break; 2119 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1484 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; 2120 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1485 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; 2121 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1486 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; 2122 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1487 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; 2123 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1488 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break; 2124 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1489 2125
2126 case EIO_REALPATH: eio__realpath (req, self); break;
2127
1490 case EIO_READLINK: ALLOC (NAME_MAX); 2128 case EIO_READLINK: ALLOC (PATH_MAX);
1491 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; 2129 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1492 2130
1493 case EIO_SYNC: req->result = 0; sync (); break; 2131 case EIO_SYNC: req->result = 0; sync (); break;
1494 case EIO_FSYNC: req->result = fsync (req->int1); break; 2132 case EIO_FSYNC: req->result = fsync (req->int1); break;
1495 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 2133 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1496 case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break; 2134 case EIO_SYNCFS: req->result = eio__syncfs (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; 2135 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
2136 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
2137 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
2138 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
2139 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
2140 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
1499 2141
1500 case EIO_READDIR: eio__scandir (req, self); break; 2142 case EIO_READDIR: eio__scandir (req, self); break;
1501 2143
1502 case EIO_BUSY: 2144 case EIO_BUSY:
1503#ifdef _WIN32 2145#ifdef _WIN32
1504 Sleep (req->nv1 * 1000.); 2146 Sleep (req->nv1 * 1e3);
1505#else 2147#else
1506 { 2148 {
1507 struct timeval tv; 2149 struct timeval tv;
1508 2150
1509 tv.tv_sec = req->nv1; 2151 tv.tv_sec = req->nv1;
1510 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; 2152 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1511 2153
1512 req->result = select (0, 0, 0, 0, &tv); 2154 req->result = select (0, 0, 0, 0, &tv);
1513 } 2155 }
1514#endif 2156#endif
1515 break; 2157 break;
1530 times = tv; 2172 times = tv;
1531 } 2173 }
1532 else 2174 else
1533 times = 0; 2175 times = 0;
1534 2176
1535
1536 req->result = req->type == EIO_FUTIME 2177 req->result = req->type == EIO_FUTIME
1537 ? futimes (req->int1, times) 2178 ? futimes (req->int1, times)
1538 : utimes (req->ptr1, times); 2179 : utimes (req->ptr1, times);
1539 } 2180 }
1540 break; 2181 break;
1545 case EIO_NOP: 2186 case EIO_NOP:
1546 req->result = 0; 2187 req->result = 0;
1547 break; 2188 break;
1548 2189
1549 case EIO_CUSTOM: 2190 case EIO_CUSTOM:
1550 ((void (*)(eio_req *))req->feed) (req); 2191 req->feed (req);
1551 break; 2192 break;
1552 2193
1553 default: 2194 default:
2195 errno = ENOSYS;
1554 req->result = -1; 2196 req->result = -1;
1555 break; 2197 break;
1556 } 2198 }
1557 2199
1558 req->errorno = errno; 2200 req->errorno = errno;
1583eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2225eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1584{ 2226{
1585 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2227 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1586} 2228}
1587 2229
2230eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2231{
2232 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2233}
2234
2235eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
2236{
2237 REQ (EIO_SYNCFS); req->int1 = fd; SEND;
2238}
2239
2240eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2241{
2242 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2243}
2244
1588eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2245eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1589{ 2246{
1590 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2247 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1591} 2248}
1592 2249
1593eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) 2250eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
1594{ 2251{
1595 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; 2252 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
1596} 2253}
1597 2254
1598eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) 2255eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
1599{ 2256{
1600 REQ (EIO_FDATASYNC); req->int1 = fd; SEND; 2257 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2258}
2259
2260eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2261{
2262 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
1601} 2263}
1602 2264
1603eio_req *eio_close (int fd, int pri, eio_cb cb, void *data) 2265eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
1604{ 2266{
1605 REQ (EIO_CLOSE); req->int1 = fd; SEND; 2267 REQ (EIO_CLOSE); req->int1 = fd; SEND;
1623eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) 2285eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1624{ 2286{
1625 REQ (EIO_FSTAT); req->int1 = fd; SEND; 2287 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1626} 2288}
1627 2289
2290eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
2291{
2292 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
2293}
2294
1628eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) 2295eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1629{ 2296{
1630 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; 2297 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1631} 2298}
1632 2299
1638eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) 2305eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1639{ 2306{
1640 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2307 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1641} 2308}
1642 2309
1643eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2310eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1644{ 2311{
1645 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2312 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1646} 2313}
1647 2314
1648eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) 2315eio_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) 2335eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1669{ 2336{
1670 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2337 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1671} 2338}
1672 2339
1673eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2340eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1674{ 2341{
1675 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2342 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1676} 2343}
1677 2344
1678eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) 2345eio_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) 2361eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1695{ 2362{
1696 return eio__1path (EIO_READLINK, path, pri, cb, data); 2363 return eio__1path (EIO_READLINK, path, pri, cb, data);
1697} 2364}
1698 2365
2366eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2367{
2368 return eio__1path (EIO_REALPATH, path, pri, cb, data);
2369}
2370
1699eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) 2371eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1700{ 2372{
1701 return eio__1path (EIO_STAT, path, pri, cb, data); 2373 return eio__1path (EIO_STAT, path, pri, cb, data);
1702} 2374}
1703 2375
1704eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) 2376eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1705{ 2377{
1706 return eio__1path (EIO_LSTAT, path, pri, cb, data); 2378 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1707} 2379}
1708 2380
2381eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
2382{
2383 return eio__1path (EIO_STATVFS, path, pri, cb, data);
2384}
2385
1709eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) 2386eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1710{ 2387{
1711 return eio__1path (EIO_UNLINK, path, pri, cb, data); 2388 return eio__1path (EIO_UNLINK, path, pri, cb, data);
1712} 2389}
1713 2390
1721 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND; 2398 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1722} 2399}
1723 2400
1724eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) 2401eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1725{ 2402{
1726 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; 2403 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1727} 2404}
1728 2405
1729static eio_req * 2406static eio_req *
1730eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2407eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1731{ 2408{
1755eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2432eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1756{ 2433{
1757 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2434 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1758} 2435}
1759 2436
1760eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) 2437eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
1761{ 2438{
1762 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND; 2439 REQ (EIO_CUSTOM); req->feed = execute; SEND;
1763} 2440}
1764 2441
1765#endif 2442#endif
1766 2443
1767eio_req *eio_grp (eio_cb cb, void *data) 2444eio_req *eio_grp (eio_cb cb, void *data)
1776#undef SEND 2453#undef SEND
1777 2454
1778/*****************************************************************************/ 2455/*****************************************************************************/
1779/* grp functions */ 2456/* grp functions */
1780 2457
2458void
1781void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2459eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1782{ 2460{
1783 grp->int2 = limit; 2461 grp->int2 = limit;
1784 grp->feed = feed; 2462 grp->feed = feed;
1785 2463
1786 grp_try_feed (grp); 2464 grp_try_feed (grp);
1787} 2465}
1788 2466
2467void
1789void eio_grp_limit (eio_req *grp, int limit) 2468eio_grp_limit (eio_req *grp, int limit)
1790{ 2469{
1791 grp->int2 = limit; 2470 grp->int2 = limit;
1792 2471
1793 grp_try_feed (grp); 2472 grp_try_feed (grp);
1794} 2473}
1795 2474
2475void
1796void eio_grp_add (eio_req *grp, eio_req *req) 2476eio_grp_add (eio_req *grp, eio_req *req)
1797{ 2477{
1798 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2478 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1799 2479
1800 grp->flags |= EIO_FLAG_GROUPADD; 2480 grp->flags |= EIO_FLAG_GROUPADD;
1801 2481
1812} 2492}
1813 2493
1814/*****************************************************************************/ 2494/*****************************************************************************/
1815/* misc garbage */ 2495/* misc garbage */
1816 2496
2497eio_ssize_t
1817ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2498eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1818{ 2499{
1819 etp_worker wrk;
1820 ssize_t ret;
1821
1822 wrk.dbuf = 0;
1823
1824 ret = eio__sendfile (ofd, ifd, offset, count, &wrk); 2500 return eio__sendfile (ofd, ifd, offset, count);
1825
1826 if (wrk.dbuf)
1827 free (wrk.dbuf);
1828
1829 return ret;
1830} 2501}
1831 2502

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines