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

Comparing cvsroot/libeio/eio.c (file contents):
Revision 1.38 by root, Fri Jun 12 20:01:42 2009 UTC vs.
Revision 1.103 by root, Thu Sep 15 13:20:54 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 X_STACKSIZE EIO_STACKSIZE
44#endif 49#endif
45#include "xthread.h" 50#include "xthread.h"
46 51
47#include <errno.h> 52#include <errno.h>
48#include <stddef.h> 53#include <stddef.h>
… …
53#include <sys/stat.h> 58#include <sys/stat.h>
54#include <limits.h> 59#include <limits.h>
55#include <fcntl.h> 60#include <fcntl.h>
56#include <assert.h> 61#include <assert.h>
57 62
63/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
64/* intptr_t only comes from stdint.h, says idiot openbsd coder */
65#if HAVE_STDINT_H
66# include <stdint.h>
67#endif
68
69#ifndef ECANCELED
70# define ECANCELED EDOM
71#endif
72#ifndef ELOOP
73# define ELOOP EDOM
74#endif
75
76#if !defined(ENOTSOCK) && defined(WSAENOTSOCK)
77# define ENOTSOCK WSAENOTSOCK
78#endif
79
80static void eio_destroy (eio_req *req);
81
58#ifndef EIO_FINISH 82#ifndef EIO_FINISH
59# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0 83# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
60#endif 84#endif
61 85
62#ifndef EIO_DESTROY 86#ifndef EIO_DESTROY
… …
65 89
66#ifndef EIO_FEED 90#ifndef EIO_FEED
67# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0) 91# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)
68#endif 92#endif
69 93
94#ifndef EIO_FD_TO_WIN32_HANDLE
95# define EIO_FD_TO_WIN32_HANDLE(fd) _get_osfhandle (fd)
96#endif
97#ifndef EIO_WIN32_HANDLE_TO_FD
98# define EIO_WIN32_HANDLE_TO_FD(handle) _open_osfhandle (handle, 0)
99#endif
100
101#define EIO_ERRNO(errval,retval) ((errno = errval), retval)
102
103#define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1)
104
70#ifdef _WIN32 105#ifdef _WIN32
71 106
72 /*doh*/ 107 #undef PAGESIZE
108 #define PAGESIZE 4096 /* GetSystemInfo? */
109
110 /* TODO: look at how perl does stat (non-sloppy), unlink (ro-files), utime, link */
111
112 #ifdef EIO_STRUCT_STATI64
113 /* look at perl's non-sloppy stat */
114 #define stat(path,buf) _stati64 (path,buf)
115 #define fstat(fd,buf) _fstati64 (fd,buf)
116 #endif
117 #define lstat(path,buf) stat (path,buf)
118 #define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
119 #define mkdir(path,mode) _mkdir (path)
120 #define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
121
122 #define chmod(path,mode) _chmod (path, mode)
123 #define dup(fd) _dup (fd)
124 #define dup2(fd1,fd2) _dup2 (fd1, fd2)
125
126 #define fchmod(fd,mode) EIO_ENOSYS ()
127 #define chown(path,uid,gid) EIO_ENOSYS ()
128 #define fchown(fd,uid,gid) EIO_ENOSYS ()
129 #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
130 #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
131 #define mknod(path,mode,dev) EIO_ENOSYS ()
132 #define sync() EIO_ENOSYS ()
133 #define readlink(path,buf,s) EIO_ENOSYS ()
134 #define statvfs(path,buf) EIO_ENOSYS ()
135 #define fstatvfs(fd,buf) EIO_ENOSYS ()
136
137 /* rename() uses MoveFile, which fails to overwrite */
138 #define rename(old,neu) eio__rename (old, neu)
139
140 static int
141 eio__rename (const char *old, const char *neu)
142 {
143 if (MoveFileEx (old, neu, MOVEFILE_REPLACE_EXISTING))
144 return 0;
145
146 /* should steal _dosmaperr */
147 switch (GetLastError ())
148 {
149 case ERROR_FILE_NOT_FOUND:
150 case ERROR_PATH_NOT_FOUND:
151 case ERROR_INVALID_DRIVE:
152 case ERROR_NO_MORE_FILES:
153 case ERROR_BAD_NETPATH:
154 case ERROR_BAD_NET_NAME:
155 case ERROR_BAD_PATHNAME:
156 case ERROR_FILENAME_EXCED_RANGE:
157 errno = ENOENT;
158 break;
159
160 default:
161 errno = EACCES;
162 break;
163 }
164
165 return -1;
166 }
167
168 /* we could even stat and see if it exists */
169 static int
170 symlink (const char *old, const char *neu)
171 {
172 #if WINVER >= 0x0600
173 if (CreateSymbolicLink (neu, old, 1))
174 return 0;
175
176 if (CreateSymbolicLink (neu, old, 0))
177 return 0;
178 #endif
179
180 return EIO_ERRNO (ENOENT, -1);
181 }
182
183 /* POSIX API only */
184 #define CreateHardLink(neu,old,flags) 0
185 #define CreateSymbolicLink(neu,old,flags) 0
186
187 struct statvfs
188 {
189 int dummy;
190 };
191
192 #define DT_DIR EIO_DT_DIR
193 #define DT_REG EIO_DT_REG
194 #define D_NAME(entp) entp.cFileName
195 #define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
196
73#else 197#else
74 198
75# include "config.h"
76# include <sys/time.h> 199 #include <sys/time.h>
77# include <sys/select.h> 200 #include <sys/select.h>
78# include <sys/mman.h> 201 #include <sys/statvfs.h>
79# include <unistd.h> 202 #include <unistd.h>
203 #include <signal.h>
204 #include <dirent.h>
205
206 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
207 #include <sys/mman.h>
208 #endif
209
210 #define D_NAME(entp) entp->d_name
211
212 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
213 #if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
214 #define _DIRENT_HAVE_D_TYPE /* sigh */
215 #define D_INO(de) (de)->d_fileno
216 #define D_NAMLEN(de) (de)->d_namlen
217 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
218 #define D_INO(de) (de)->d_ino
219 #endif
220
221 #ifdef _D_EXACT_NAMLEN
222 #undef D_NAMLEN
223 #define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
224 #endif
225
226 #ifdef _DIRENT_HAVE_D_TYPE
227 #define D_TYPE(de) (de)->d_type
228 #endif
229
230 #ifndef EIO_STRUCT_DIRENT
231 #define EIO_STRUCT_DIRENT struct dirent
232 #endif
233
234#endif
235
236#if HAVE_UTIMES
80# include <utime.h> 237# include <utime.h>
81# include <signal.h>
82# include <dirent.h>
83
84/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
85# if defined(__FreeBSD__) || defined(__NetBSD__) || defined(__OpenBSD__)
86# define _DIRENT_HAVE_D_TYPE /* sigh */
87# define D_INO(de) (de)->d_fileno
88# define D_NAMLEN(de) (de)->d_namlen
89# elif defined(__linux) || defined(d_ino) || _XOPEN_SOURCE >= 600
90# define D_INO(de) (de)->d_ino
91# endif 238#endif
92 239
93#ifdef _D_EXACT_NAMLEN 240#if HAVE_SYS_SYSCALL_H
94# undef D_NAMLEN 241# include <sys/syscall.h>
95# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
96#endif
97
98# ifdef _DIRENT_HAVE_D_TYPE
99# define D_TYPE(de) (de)->d_type
100# endif 242#endif
101 243
102# ifndef EIO_STRUCT_DIRENT 244#if HAVE_SYS_PRCTL_H
103# define EIO_STRUCT_DIRENT struct dirent 245# include <sys/prctl.h>
104# endif
105
106#endif 246#endif
107 247
108#if HAVE_SENDFILE 248#if HAVE_SENDFILE
109# if __linux 249# if __linux
110# include <sys/sendfile.h> 250# include <sys/sendfile.h>
111# elif __freebsd 251# elif __FreeBSD__ || defined __APPLE__
112# include <sys/socket.h> 252# include <sys/socket.h>
113# include <sys/uio.h> 253# include <sys/uio.h>
114# elif __hpux 254# elif __hpux
115# include <sys/socket.h> 255# include <sys/socket.h>
116# elif __solaris /* not yet */ 256# elif __solaris
117# include <sys/sendfile.h> 257# include <sys/sendfile.h>
118# else 258# else
119# error sendfile support requested but not available 259# error sendfile support requested but not available
120# endif 260# endif
121#endif 261#endif
… …
125#endif 265#endif
126#ifndef D_INO 266#ifndef D_INO
127# define D_INO(de) 0 267# define D_INO(de) 0
128#endif 268#endif
129#ifndef D_NAMLEN 269#ifndef D_NAMLEN
130# define D_NAMLEN(de) strlen ((de)->d_name) 270# define D_NAMLEN(entp) strlen (D_NAME (entp))
131#endif 271#endif
132
133/* number of seconds after which an idle threads exit */
134#define IDLE_TIMEOUT 10
135 272
136/* used for struct dirent, AIX doesn't provide it */ 273/* used for struct dirent, AIX doesn't provide it */
137#ifndef NAME_MAX 274#ifndef NAME_MAX
138# define NAME_MAX 4096 275# define NAME_MAX 4096
139#endif 276#endif
140 277
278/* used for readlink etc. */
279#ifndef PATH_MAX
280# define PATH_MAX 4096
281#endif
282
141/* buffer size for various temporary buffers */ 283/* buffer size for various temporary buffers */
142#define EIO_BUFSIZE 65536 284#define EIO_BUFSIZE 65536
143 285
144#define dBUF \ 286#define dBUF \
145 char *eio_buf; \
146 ETP_WORKER_LOCK (self); \
147 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 287 char *eio_buf = malloc (EIO_BUFSIZE); \
148 ETP_WORKER_UNLOCK (self); \
149 errno = ENOMEM; \ 288 errno = ENOMEM; \
150 if (!eio_buf) \ 289 if (!eio_buf) \
151 return -1; 290 return -1
291
292#define FUBd \
293 free (eio_buf)
152 294
153#define EIO_TICKS ((1000000 + 1023) >> 10) 295#define EIO_TICKS ((1000000 + 1023) >> 10)
154
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/*****************************************************************************/
167 296
168#define ETP_PRI_MIN EIO_PRI_MIN 297#define ETP_PRI_MIN EIO_PRI_MIN
169#define ETP_PRI_MAX EIO_PRI_MAX 298#define ETP_PRI_MAX EIO_PRI_MAX
170 299
171struct etp_worker; 300struct etp_worker;
… …
175static int eio_finish (eio_req *req); 304static int eio_finish (eio_req *req);
176#define ETP_FINISH(req) eio_finish (req) 305#define ETP_FINISH(req) eio_finish (req)
177static void eio_execute (struct etp_worker *self, eio_req *req); 306static void eio_execute (struct etp_worker *self, eio_req *req);
178#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) 307#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
179 308
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/*****************************************************************************/ 309/*****************************************************************************/
198 310
199#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) 311#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
200 312
201/* calculcate time difference in ~1/EIO_TICKS of a second */ 313/* calculate time difference in ~1/EIO_TICKS of a second */
314ecb_inline int
202static int tvdiff (struct timeval *tv1, struct timeval *tv2) 315tvdiff (struct timeval *tv1, struct timeval *tv2)
203{ 316{
204 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 317 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
205 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 318 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
206} 319}
207 320
… …
211static void (*done_poll_cb) (void); 324static void (*done_poll_cb) (void);
212 325
213static unsigned int max_poll_time; /* reslock */ 326static unsigned int max_poll_time; /* reslock */
214static unsigned int max_poll_reqs; /* reslock */ 327static unsigned int max_poll_reqs; /* reslock */
215 328
216static volatile unsigned int nreqs; /* reqlock */ 329static unsigned int nreqs; /* reqlock */
217static volatile unsigned int nready; /* reqlock */ 330static unsigned int nready; /* reqlock */
218static volatile unsigned int npending; /* reqlock */ 331static unsigned int npending; /* reqlock */
219static volatile unsigned int max_idle = 4; 332static unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
333static unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
220 334
221static mutex_t wrklock = X_MUTEX_INIT; 335static xmutex_t wrklock;
222static mutex_t reslock = X_MUTEX_INIT; 336static xmutex_t reslock;
223static mutex_t reqlock = X_MUTEX_INIT; 337static xmutex_t reqlock;
224static cond_t reqwait = X_COND_INIT; 338static xcond_t reqwait;
225 339
226#if !HAVE_PREADWRITE 340#if !HAVE_PREADWRITE
227/* 341/*
228 * make our pread/pwrite emulation safe against themselves, but not against 342 * make our pread/pwrite emulation safe against themselves, but not against
229 * normal read/write by using a mutex. slows down execution a lot, 343 * normal read/write by using a mutex. slows down execution a lot,
230 * but that's your problem, not mine. 344 * but that's your problem, not mine.
231 */ 345 */
232static mutex_t preadwritelock = X_MUTEX_INIT; 346static xmutex_t preadwritelock;
233#endif 347#endif
234 348
235typedef struct etp_worker 349typedef struct etp_worker
236{ 350{
237 /* locked by wrklock */ 351 /* locked by wrklock */
238 struct etp_worker *prev, *next; 352 struct etp_worker *prev, *next;
239 353
240 thread_t tid; 354 xthread_t tid;
241 355
242 /* locked by reslock, reqlock or wrklock */ 356 /* locked by reslock, reqlock or wrklock */
243 ETP_REQ *req; /* currently processed request */ 357 ETP_REQ *req; /* currently processed request */
244 358
359#ifdef ETP_WORKER_COMMON
245 ETP_WORKER_COMMON 360 ETP_WORKER_COMMON
361#endif
246} etp_worker; 362} etp_worker;
247 363
248static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */ 364static etp_worker wrk_first; /* NOT etp */
249 365
250#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock) 366#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
251#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock) 367#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
252 368
253/* worker threads management */ 369/* worker threads management */
254 370
371static void ecb_cold
255static void etp_worker_clear (etp_worker *wrk) 372etp_worker_clear (etp_worker *wrk)
256{ 373{
257 ETP_WORKER_CLEAR (wrk);
258} 374}
259 375
376static void ecb_cold
260static void etp_worker_free (etp_worker *wrk) 377etp_worker_free (etp_worker *wrk)
261{ 378{
262 wrk->next->prev = wrk->prev; 379 wrk->next->prev = wrk->prev;
263 wrk->prev->next = wrk->next; 380 wrk->prev->next = wrk->next;
264 381
265 free (wrk); 382 free (wrk);
266} 383}
267 384
268static unsigned int etp_nreqs (void) 385static unsigned int
386etp_nreqs (void)
269{ 387{
270 int retval; 388 int retval;
271 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 389 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
272 retval = nreqs; 390 retval = nreqs;
273 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 391 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
274 return retval; 392 return retval;
275} 393}
276 394
277static unsigned int etp_nready (void) 395static unsigned int
396etp_nready (void)
278{ 397{
279 unsigned int retval; 398 unsigned int retval;
280 399
281 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 400 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
282 retval = nready; 401 retval = nready;
283 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 402 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
284 403
285 return retval; 404 return retval;
286} 405}
287 406
288static unsigned int etp_npending (void) 407static unsigned int
408etp_npending (void)
289{ 409{
290 unsigned int retval; 410 unsigned int retval;
291 411
292 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 412 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
293 retval = npending; 413 retval = npending;
294 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 414 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
295 415
296 return retval; 416 return retval;
297} 417}
298 418
299static unsigned int etp_nthreads (void) 419static unsigned int
420etp_nthreads (void)
300{ 421{
301 unsigned int retval; 422 unsigned int retval;
302 423
303 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 424 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
304 retval = started; 425 retval = started;
… …
318} etp_reqq; 439} etp_reqq;
319 440
320static etp_reqq req_queue; 441static etp_reqq req_queue;
321static etp_reqq res_queue; 442static etp_reqq res_queue;
322 443
444static void ecb_noinline ecb_cold
445reqq_init (etp_reqq *q)
446{
447 int pri;
448
449 for (pri = 0; pri < ETP_NUM_PRI; ++pri)
450 q->qs[pri] = q->qe[pri] = 0;
451
452 q->size = 0;
453}
454
455static int ecb_noinline
323static int reqq_push (etp_reqq *q, ETP_REQ *req) 456reqq_push (etp_reqq *q, ETP_REQ *req)
324{ 457{
325 int pri = req->pri; 458 int pri = req->pri;
326 req->next = 0; 459 req->next = 0;
327 460
328 if (q->qe[pri]) 461 if (q->qe[pri])
… …
334 q->qe[pri] = q->qs[pri] = req; 467 q->qe[pri] = q->qs[pri] = req;
335 468
336 return q->size++; 469 return q->size++;
337} 470}
338 471
472static ETP_REQ * ecb_noinline
339static ETP_REQ *reqq_shift (etp_reqq *q) 473reqq_shift (etp_reqq *q)
340{ 474{
341 int pri; 475 int pri;
342 476
343 if (!q->size) 477 if (!q->size)
344 return 0; 478 return 0;
… …
359 } 493 }
360 494
361 abort (); 495 abort ();
362} 496}
363 497
364static void etp_atfork_prepare (void) 498static int ecb_cold
499etp_init (void (*want_poll)(void), void (*done_poll)(void))
365{ 500{
366 X_LOCK (wrklock); 501 X_MUTEX_CREATE (wrklock);
367 X_LOCK (reqlock); 502 X_MUTEX_CREATE (reslock);
368 X_LOCK (reslock); 503 X_MUTEX_CREATE (reqlock);
369#if !HAVE_PREADWRITE 504 X_COND_CREATE (reqwait);
370 X_LOCK (preadwritelock);
371#endif
372}
373 505
374static void etp_atfork_parent (void) 506 reqq_init (&req_queue);
375{ 507 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}
383 508
384static void etp_atfork_child (void) 509 wrk_first.next =
385{ 510 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 }
404 511
405 started = 0; 512 started = 0;
406 idle = 0; 513 idle = 0;
407 nreqs = 0; 514 nreqs = 0;
408 nready = 0; 515 nready = 0;
409 npending = 0; 516 npending = 0;
410 517
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; 518 want_poll_cb = want_poll;
428 done_poll_cb = done_poll; 519 done_poll_cb = done_poll;
429 520
430 return 0; 521 return 0;
431} 522}
432 523
433X_THREAD_PROC (etp_proc); 524X_THREAD_PROC (etp_proc);
434 525
526static void ecb_cold
435static void etp_start_thread (void) 527etp_start_thread (void)
436{ 528{
437 etp_worker *wrk = calloc (1, sizeof (etp_worker)); 529 etp_worker *wrk = calloc (1, sizeof (etp_worker));
438 530
439 /*TODO*/ 531 /*TODO*/
440 assert (("unable to allocate worker thread data", wrk)); 532 assert (("unable to allocate worker thread data", wrk));
… …
453 free (wrk); 545 free (wrk);
454 546
455 X_UNLOCK (wrklock); 547 X_UNLOCK (wrklock);
456} 548}
457 549
550static void
458static void etp_maybe_start_thread (void) 551etp_maybe_start_thread (void)
459{ 552{
460 if (expect_true (etp_nthreads () >= wanted)) 553 if (ecb_expect_true (etp_nthreads () >= wanted))
461 return; 554 return;
462 555
463 /* todo: maybe use idle here, but might be less exact */ 556 /* todo: maybe use idle here, but might be less exact */
464 if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ())) 557 if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
465 return; 558 return;
466 559
467 etp_start_thread (); 560 etp_start_thread ();
468} 561}
469 562
563static void ecb_cold
470static void etp_end_thread (void) 564etp_end_thread (void)
471{ 565{
472 eio_req *req = calloc (1, sizeof (eio_req)); 566 eio_req *req = calloc (1, sizeof (eio_req));
473 567
474 req->type = -1; 568 req->type = -1;
475 req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 569 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
… …
482 X_LOCK (wrklock); 576 X_LOCK (wrklock);
483 --started; 577 --started;
484 X_UNLOCK (wrklock); 578 X_UNLOCK (wrklock);
485} 579}
486 580
487static int etp_poll (void) 581static int
582etp_poll (void)
488{ 583{
489 unsigned int maxreqs; 584 unsigned int maxreqs;
490 unsigned int maxtime; 585 unsigned int maxtime;
491 struct timeval tv_start, tv_now; 586 struct timeval tv_start, tv_now;
492 587
… …
522 617
523 X_LOCK (reqlock); 618 X_LOCK (reqlock);
524 --nreqs; 619 --nreqs;
525 X_UNLOCK (reqlock); 620 X_UNLOCK (reqlock);
526 621
527 if (expect_false (req->type == EIO_GROUP && req->size)) 622 if (ecb_expect_false (req->type == EIO_GROUP && req->size))
528 { 623 {
529 req->int1 = 1; /* mark request as delayed */ 624 req->int1 = 1; /* mark request as delayed */
530 continue; 625 continue;
531 } 626 }
532 else 627 else
533 { 628 {
534 int res = ETP_FINISH (req); 629 int res = ETP_FINISH (req);
535 if (expect_false (res)) 630 if (ecb_expect_false (res))
536 return res; 631 return res;
537 } 632 }
538 633
539 if (expect_false (maxreqs && !--maxreqs)) 634 if (ecb_expect_false (maxreqs && !--maxreqs))
540 break; 635 break;
541 636
542 if (maxtime) 637 if (maxtime)
543 { 638 {
544 gettimeofday (&tv_now, 0); 639 gettimeofday (&tv_now, 0);
… …
550 645
551 errno = EAGAIN; 646 errno = EAGAIN;
552 return -1; 647 return -1;
553} 648}
554 649
650static void
555static void etp_cancel (ETP_REQ *req) 651etp_cancel (ETP_REQ *req)
556{ 652{
557 X_LOCK (wrklock); 653 req->cancelled = 1;
558 req->flags |= EIO_FLAG_CANCELLED;
559 X_UNLOCK (wrklock);
560 654
561 eio_grp_cancel (req); 655 eio_grp_cancel (req);
562} 656}
563 657
658static void
564static void etp_submit (ETP_REQ *req) 659etp_submit (ETP_REQ *req)
565{ 660{
566 req->pri -= ETP_PRI_MIN; 661 req->pri -= ETP_PRI_MIN;
567 662
568 if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN; 663 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; 664 if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
570 665
571 if (expect_false (req->type == EIO_GROUP)) 666 if (ecb_expect_false (req->type == EIO_GROUP))
572 { 667 {
573 /* I hope this is worth it :/ */ 668 /* I hope this is worth it :/ */
574 X_LOCK (reqlock); 669 X_LOCK (reqlock);
575 ++nreqs; 670 ++nreqs;
576 X_UNLOCK (reqlock); 671 X_UNLOCK (reqlock);
… …
595 690
596 etp_maybe_start_thread (); 691 etp_maybe_start_thread ();
597 } 692 }
598} 693}
599 694
695static void ecb_cold
600static void etp_set_max_poll_time (double nseconds) 696etp_set_max_poll_time (double nseconds)
601{ 697{
602 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 698 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
603 max_poll_time = nseconds; 699 max_poll_time = nseconds * EIO_TICKS;
604 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 700 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
605} 701}
606 702
703static void ecb_cold
607static void etp_set_max_poll_reqs (unsigned int maxreqs) 704etp_set_max_poll_reqs (unsigned int maxreqs)
608{ 705{
609 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 706 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
610 max_poll_reqs = maxreqs; 707 max_poll_reqs = maxreqs;
611 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 708 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
612} 709}
613 710
711static void ecb_cold
614static void etp_set_max_idle (unsigned int nthreads) 712etp_set_max_idle (unsigned int nthreads)
615{ 713{
616 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 714 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
617 max_idle = nthreads <= 0 ? 1 : nthreads; 715 max_idle = nthreads;
618 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 716 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
619} 717}
620 718
719static void ecb_cold
720etp_set_idle_timeout (unsigned int seconds)
721{
722 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
723 idle_timeout = seconds;
724 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
725}
726
727static void ecb_cold
621static void etp_set_min_parallel (unsigned int nthreads) 728etp_set_min_parallel (unsigned int nthreads)
622{ 729{
623 if (wanted < nthreads) 730 if (wanted < nthreads)
624 wanted = nthreads; 731 wanted = nthreads;
625} 732}
626 733
734static void ecb_cold
627static void etp_set_max_parallel (unsigned int nthreads) 735etp_set_max_parallel (unsigned int nthreads)
628{ 736{
629 if (wanted > nthreads) 737 if (wanted > nthreads)
630 wanted = nthreads; 738 wanted = nthreads;
631 739
632 while (started > wanted) 740 while (started > wanted)
633 etp_end_thread (); 741 etp_end_thread ();
634} 742}
635 743
636/*****************************************************************************/ 744/*****************************************************************************/
637 745
746static void
638static void grp_try_feed (eio_req *grp) 747grp_try_feed (eio_req *grp)
639{ 748{
640 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 749 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
641 { 750 {
642 grp->flags &= ~EIO_FLAG_GROUPADD; 751 grp->flags &= ~EIO_FLAG_GROUPADD;
643 752
… …
650 break; 759 break;
651 } 760 }
652 } 761 }
653} 762}
654 763
764static int
655static int grp_dec (eio_req *grp) 765grp_dec (eio_req *grp)
656{ 766{
657 --grp->size; 767 --grp->size;
658 768
659 /* call feeder, if applicable */ 769 /* call feeder, if applicable */
660 grp_try_feed (grp); 770 grp_try_feed (grp);
… …
664 return eio_finish (grp); 774 return eio_finish (grp);
665 else 775 else
666 return 0; 776 return 0;
667} 777}
668 778
779static void
669void eio_destroy (eio_req *req) 780eio_destroy (eio_req *req)
670{ 781{
671 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); 782 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
672 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); 783 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
673 784
674 EIO_DESTROY (req); 785 EIO_DESTROY (req);
675} 786}
676 787
788static int
677static int eio_finish (eio_req *req) 789eio_finish (eio_req *req)
678{ 790{
679 int res = EIO_FINISH (req); 791 int res = EIO_FINISH (req);
680 792
681 if (req->grp) 793 if (req->grp)
682 { 794 {
… …
690 if (grp->grp_first == req) 802 if (grp->grp_first == req)
691 grp->grp_first = req->grp_next; 803 grp->grp_first = req->grp_next;
692 804
693 res2 = grp_dec (grp); 805 res2 = grp_dec (grp);
694 806
695 if (!res && res2) 807 if (!res)
696 res = res2; 808 res = res2;
697 } 809 }
698 810
699 eio_destroy (req); 811 eio_destroy (req);
700 812
701 return res; 813 return res;
702} 814}
703 815
816void
704void eio_grp_cancel (eio_req *grp) 817eio_grp_cancel (eio_req *grp)
705{ 818{
706 for (grp = grp->grp_first; grp; grp = grp->grp_next) 819 for (grp = grp->grp_first; grp; grp = grp->grp_next)
707 eio_cancel (grp); 820 eio_cancel (grp);
708} 821}
709 822
823void
710void eio_cancel (eio_req *req) 824eio_cancel (eio_req *req)
711{ 825{
712 etp_cancel (req); 826 etp_cancel (req);
713} 827}
714 828
829void
715void eio_submit (eio_req *req) 830eio_submit (eio_req *req)
716{ 831{
717 etp_submit (req); 832 etp_submit (req);
718} 833}
719 834
720unsigned int eio_nreqs (void) 835unsigned int
836eio_nreqs (void)
721{ 837{
722 return etp_nreqs (); 838 return etp_nreqs ();
723} 839}
724 840
725unsigned int eio_nready (void) 841unsigned int
842eio_nready (void)
726{ 843{
727 return etp_nready (); 844 return etp_nready ();
728} 845}
729 846
730unsigned int eio_npending (void) 847unsigned int
848eio_npending (void)
731{ 849{
732 return etp_npending (); 850 return etp_npending ();
733} 851}
734 852
735unsigned int eio_nthreads (void) 853unsigned int ecb_cold
854eio_nthreads (void)
736{ 855{
737 return etp_nthreads (); 856 return etp_nthreads ();
738} 857}
739 858
859void ecb_cold
740void eio_set_max_poll_time (double nseconds) 860eio_set_max_poll_time (double nseconds)
741{ 861{
742 etp_set_max_poll_time (nseconds); 862 etp_set_max_poll_time (nseconds);
743} 863}
744 864
865void ecb_cold
745void eio_set_max_poll_reqs (unsigned int maxreqs) 866eio_set_max_poll_reqs (unsigned int maxreqs)
746{ 867{
747 etp_set_max_poll_reqs (maxreqs); 868 etp_set_max_poll_reqs (maxreqs);
748} 869}
749 870
871void ecb_cold
750void eio_set_max_idle (unsigned int nthreads) 872eio_set_max_idle (unsigned int nthreads)
751{ 873{
752 etp_set_max_idle (nthreads); 874 etp_set_max_idle (nthreads);
753} 875}
754 876
877void ecb_cold
878eio_set_idle_timeout (unsigned int seconds)
879{
880 etp_set_idle_timeout (seconds);
881}
882
883void ecb_cold
755void eio_set_min_parallel (unsigned int nthreads) 884eio_set_min_parallel (unsigned int nthreads)
756{ 885{
757 etp_set_min_parallel (nthreads); 886 etp_set_min_parallel (nthreads);
758} 887}
759 888
889void ecb_cold
760void eio_set_max_parallel (unsigned int nthreads) 890eio_set_max_parallel (unsigned int nthreads)
761{ 891{
762 etp_set_max_parallel (nthreads); 892 etp_set_max_parallel (nthreads);
763} 893}
764 894
765int eio_poll (void) 895int eio_poll (void)
… …
774# undef pread 904# undef pread
775# undef pwrite 905# undef pwrite
776# define pread eio__pread 906# define pread eio__pread
777# define pwrite eio__pwrite 907# define pwrite eio__pwrite
778 908
779static ssize_t 909static eio_ssize_t
780eio__pread (int fd, void *buf, size_t count, off_t offset) 910eio__pread (int fd, void *buf, size_t count, off_t offset)
781{ 911{
782 ssize_t res; 912 eio_ssize_t res;
783 off_t ooffset; 913 off_t ooffset;
784 914
785 X_LOCK (preadwritelock); 915 X_LOCK (preadwritelock);
786 ooffset = lseek (fd, 0, SEEK_CUR); 916 ooffset = lseek (fd, 0, SEEK_CUR);
787 lseek (fd, offset, SEEK_SET); 917 lseek (fd, offset, SEEK_SET);
… …
790 X_UNLOCK (preadwritelock); 920 X_UNLOCK (preadwritelock);
791 921
792 return res; 922 return res;
793} 923}
794 924
795static ssize_t 925static eio_ssize_t
796eio__pwrite (int fd, void *buf, size_t count, off_t offset) 926eio__pwrite (int fd, void *buf, size_t count, off_t offset)
797{ 927{
798 ssize_t res; 928 eio_ssize_t res;
799 off_t ooffset; 929 off_t ooffset;
800 930
801 X_LOCK (preadwritelock); 931 X_LOCK (preadwritelock);
802 ooffset = lseek (fd, 0, SEEK_CUR); 932 ooffset = lseek (fd, 0, SEEK_CUR);
803 lseek (fd, offset, SEEK_SET); 933 lseek (fd, offset, SEEK_SET);
… …
807 937
808 return res; 938 return res;
809} 939}
810#endif 940#endif
811 941
812#ifndef HAVE_FUTIMES 942#ifndef HAVE_UTIMES
813 943
814# undef utimes 944# undef utimes
815# undef futimes
816# define utimes(path,times) eio__utimes (path, times) 945# define utimes(path,times) eio__utimes (path, times)
817# define futimes(fd,times) eio__futimes (fd, times)
818 946
819static int 947static int
820eio__utimes (const char *filename, const struct timeval times[2]) 948eio__utimes (const char *filename, const struct timeval times[2])
821{ 949{
822 if (times) 950 if (times)
… …
830 } 958 }
831 else 959 else
832 return utime (filename, 0); 960 return utime (filename, 0);
833} 961}
834 962
963#endif
964
965#ifndef HAVE_FUTIMES
966
967# undef futimes
968# define futimes(fd,times) eio__futimes (fd, times)
969
970static int
835static int eio__futimes (int fd, const struct timeval tv[2]) 971eio__futimes (int fd, const struct timeval tv[2])
836{ 972{
837 errno = ENOSYS; 973 errno = ENOSYS;
838 return -1; 974 return -1;
839} 975}
840 976
… …
843#if !HAVE_FDATASYNC 979#if !HAVE_FDATASYNC
844# undef fdatasync 980# undef fdatasync
845# define fdatasync(fd) fsync (fd) 981# define fdatasync(fd) fsync (fd)
846#endif 982#endif
847 983
984static int
985eio__syncfs (int fd)
986{
987 int res;
988
989#if HAVE_SYS_SYNCFS
990 res = (int)syscall (__NR_syncfs, (int)(fd));
991#else
992 res = -1;
993 errno = ENOSYS;
994#endif
995
996 if (res < 0 && errno == ENOSYS && fd >= 0)
997 sync ();
998
999 return res;
1000}
1001
848/* sync_file_range always needs emulation */ 1002/* sync_file_range always needs emulation */
849int 1003static int
850eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags) 1004eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
851{ 1005{
852#if HAVE_SYNC_FILE_RANGE 1006#if HAVE_SYNC_FILE_RANGE
853 int res; 1007 int res;
854 1008
… …
867 if (!res || errno != ENOSYS) 1021 if (!res || errno != ENOSYS)
868 return res; 1022 return res;
869#endif 1023#endif
870 1024
871 /* even though we could play tricks with the flags, it's better to always 1025 /* 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 */ 1026 * call fdatasync, as that matches the expectation of its users best */
873 return fdatasync (fd); 1027 return fdatasync (fd);
1028}
1029
1030static int
1031eio__fallocate (int fd, int mode, off_t offset, size_t len)
1032{
1033#if HAVE_FALLOCATE
1034 return fallocate (fd, mode, offset, len);
1035#else
1036 errno = ENOSYS;
1037 return -1;
1038#endif
874} 1039}
875 1040
876#if !HAVE_READAHEAD 1041#if !HAVE_READAHEAD
877# undef readahead 1042# undef readahead
878# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 1043# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
879 1044
880static ssize_t 1045static eio_ssize_t
881eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) 1046eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
882{ 1047{
883 size_t todo = count; 1048 size_t todo = count;
884 dBUF; 1049 dBUF;
885 1050
… …
890 pread (fd, eio_buf, len, offset); 1055 pread (fd, eio_buf, len, offset);
891 offset += len; 1056 offset += len;
892 todo -= len; 1057 todo -= len;
893 } 1058 }
894 1059
1060 FUBd;
1061
895 errno = 0; 1062 errno = 0;
896 return count; 1063 return count;
897} 1064}
898 1065
899#endif 1066#endif
900 1067
901/* sendfile always needs emulation */ 1068/* sendfile always needs emulation */
902static ssize_t 1069static eio_ssize_t
903eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) 1070eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
904{ 1071{
1072 eio_ssize_t written = 0;
905 ssize_t res; 1073 eio_ssize_t res;
906 1074
907 if (!count) 1075 if (!count)
908 return 0; 1076 return 0;
909 1077
1078 for (;;)
1079 {
1080#ifdef __APPLE__
1081# undef HAVE_SENDFILE /* broken, as everything on os x */
1082#endif
910#if HAVE_SENDFILE 1083#if HAVE_SENDFILE
911# if __linux 1084# if __linux
1085 off_t soffset = offset;
912 res = sendfile (ofd, ifd, &offset, count); 1086 res = sendfile (ofd, ifd, &soffset, count);
913 1087
914# elif __freebsd 1088# elif __FreeBSD__
915 /* 1089 /*
916 * Of course, the freebsd sendfile is a dire hack with no thoughts 1090 * Of course, the freebsd sendfile is a dire hack with no thoughts
917 * wasted on making it similar to other I/O functions. 1091 * wasted on making it similar to other I/O functions.
918 */ 1092 */
919 {
920 off_t sbytes; 1093 off_t sbytes;
921 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 1094 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
922 1095
923 if (res < 0 && sbytes) 1096 #if 0 /* according to the manpage, this is correct, but broken behaviour */
924 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 1097 /* freebsd' sendfile will return 0 on success */
1098 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
1099 /* not on e.g. EIO or EPIPE - sounds broken */
1100 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
925 res = sbytes; 1101 res = sbytes;
926 } 1102 #endif
1103
1104 /* according to source inspection, this is correct, and useful behaviour */
1105 if (sbytes)
1106 res = sbytes;
1107
1108# elif defined (__APPLE__)
1109 off_t sbytes = count;
1110 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
1111
1112 /* according to the manpage, sbytes is always valid */
1113 if (sbytes)
1114 res = sbytes;
927 1115
928# elif __hpux 1116# elif __hpux
929 res = sendfile (ofd, ifd, offset, count, 0, 0); 1117 res = sendfile (ofd, ifd, offset, count, 0, 0);
930 1118
931# elif __solaris 1119# elif __solaris
932 {
933 struct sendfilevec vec; 1120 struct sendfilevec vec;
934 size_t sbytes; 1121 size_t sbytes;
935 1122
936 vec.sfv_fd = ifd; 1123 vec.sfv_fd = ifd;
937 vec.sfv_flag = 0; 1124 vec.sfv_flag = 0;
938 vec.sfv_off = offset; 1125 vec.sfv_off = offset;
939 vec.sfv_len = count; 1126 vec.sfv_len = count;
940 1127
941 res = sendfilev (ofd, &vec, 1, &sbytes); 1128 res = sendfilev (ofd, &vec, 1, &sbytes);
942 1129
943 if (res < 0 && sbytes) 1130 if (res < 0 && sbytes)
944 res = sbytes; 1131 res = sbytes;
945 }
946 1132
947# endif 1133# endif
1134
1135#elif defined (_WIN32) && 0
1136 /* does not work, just for documentation of what would need to be done */
1137 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
1138 /* libeio guarantees that the file offset does not change, and windows */
1139 /* has no way to get an independent handle to the same file description */
1140 HANDLE h = TO_SOCKET (ifd);
1141 SetFilePointer (h, offset, 0, FILE_BEGIN);
1142 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1143
948#else 1144#else
949 res = -1; 1145 res = -1;
950 errno = ENOSYS; 1146 errno = ENOSYS;
951#endif 1147#endif
952 1148
1149 /* we assume sendfile can copy at least 128mb in one go */
1150 if (res <= 128 * 1024 * 1024)
1151 {
1152 if (res > 0)
1153 written += res;
1154
1155 if (written)
1156 return written;
1157
1158 break;
1159 }
1160 else
1161 {
1162 /* if we requested more, then probably the kernel was lazy */
1163 written += res;
1164 offset += res;
1165 count -= res;
1166
1167 if (!count)
1168 return written;
1169 }
1170 }
1171
953 if (res < 0 1172 if (res < 0
954 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1173 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1174 /* BSDs */
1175#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1176 || errno == ENOTSUP
1177#endif
1178#ifdef EOPNOTSUPP /* windows */
1179 || errno == EOPNOTSUPP /* BSDs */
1180#endif
955#if __solaris 1181#if __solaris
956 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1182 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
957#endif 1183#endif
958 ) 1184 )
959 ) 1185 )
… …
963 1189
964 res = 0; 1190 res = 0;
965 1191
966 while (count) 1192 while (count)
967 { 1193 {
968 ssize_t cnt; 1194 eio_ssize_t cnt;
969 1195
970 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 1196 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
971 1197
972 if (cnt <= 0) 1198 if (cnt <= 0)
973 { 1199 {
… …
985 1211
986 offset += cnt; 1212 offset += cnt;
987 res += cnt; 1213 res += cnt;
988 count -= cnt; 1214 count -= cnt;
989 } 1215 }
1216
1217 FUBd;
990 } 1218 }
991 1219
992 return res; 1220 return res;
1221}
1222
1223#ifdef PAGESIZE
1224# define eio_pagesize() PAGESIZE
1225#else
1226static intptr_t
1227eio_pagesize (void)
1228{
1229 static intptr_t page;
1230
1231 if (!page)
1232 page = sysconf (_SC_PAGESIZE);
1233
1234 return page;
1235}
1236#endif
1237
1238static void
1239eio_page_align (void **addr, size_t *length)
1240{
1241 intptr_t mask = eio_pagesize () - 1;
1242
1243 /* round down addr */
1244 intptr_t adj = mask & (intptr_t)*addr;
1245
1246 *addr = (void *)((intptr_t)*addr - adj);
1247 *length += adj;
1248
1249 /* round up length */
1250 *length = (*length + mask) & ~mask;
1251}
1252
1253#if !_POSIX_MEMLOCK
1254# define eio__mlockall(a) EIO_ENOSYS ()
1255#else
1256
1257static int
1258eio__mlockall (int flags)
1259{
1260 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1261 extern int mallopt (int, int);
1262 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1263 #endif
1264
1265 if (EIO_MCL_CURRENT != MCL_CURRENT
1266 || EIO_MCL_FUTURE != MCL_FUTURE)
1267 {
1268 flags = 0
1269 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1270 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1271 }
1272
1273 return mlockall (flags);
1274}
1275#endif
1276
1277#if !_POSIX_MEMLOCK_RANGE
1278# define eio__mlock(a,b) EIO_ENOSYS ()
1279#else
1280
1281static int
1282eio__mlock (void *addr, size_t length)
1283{
1284 eio_page_align (&addr, &length);
1285
1286 return mlock (addr, length);
1287}
1288
1289#endif
1290
1291#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1292# define eio__msync(a,b,c) EIO_ENOSYS ()
1293#else
1294
1295static int
1296eio__msync (void *mem, size_t len, int flags)
1297{
1298 eio_page_align (&mem, &len);
1299
1300 if (EIO_MS_ASYNC != MS_SYNC
1301 || EIO_MS_INVALIDATE != MS_INVALIDATE
1302 || EIO_MS_SYNC != MS_SYNC)
1303 {
1304 flags = 0
1305 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1306 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1307 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1308 }
1309
1310 return msync (mem, len, flags);
1311}
1312
1313#endif
1314
1315static int
1316eio__mtouch (eio_req *req)
1317{
1318 void *mem = req->ptr2;
1319 size_t len = req->size;
1320 int flags = req->int1;
1321
1322 eio_page_align (&mem, &len);
1323
1324 {
1325 intptr_t addr = (intptr_t)mem;
1326 intptr_t end = addr + len;
1327 intptr_t page = eio_pagesize ();
1328
1329 if (addr < end)
1330 if (flags & EIO_MT_MODIFY) /* modify */
1331 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1332 else
1333 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1334 }
1335
1336 return 0;
1337}
1338
1339/*****************************************************************************/
1340/* requests implemented outside eio_execute, because they are so large */
1341
1342static void
1343eio__realpath (eio_req *req, etp_worker *self)
1344{
1345 char *rel = req->ptr1;
1346 char *res;
1347 char *tmp1, *tmp2;
1348#if SYMLOOP_MAX > 32
1349 int symlinks = SYMLOOP_MAX;
1350#else
1351 int symlinks = 32;
1352#endif
1353
1354 req->result = -1;
1355
1356 errno = EINVAL;
1357 if (!rel)
1358 return;
1359
1360 errno = ENOENT;
1361 if (!*rel)
1362 return;
1363
1364 if (!req->ptr2)
1365 {
1366 X_LOCK (wrklock);
1367 req->flags |= EIO_FLAG_PTR2_FREE;
1368 X_UNLOCK (wrklock);
1369 req->ptr2 = malloc (PATH_MAX * 3);
1370
1371 errno = ENOMEM;
1372 if (!req->ptr2)
1373 return;
1374 }
1375
1376 res = req->ptr2;
1377 tmp1 = res + PATH_MAX;
1378 tmp2 = tmp1 + PATH_MAX;
1379
1380#if 0 /* disabled, the musl way to do things is just too racy */
1381#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1382 /* on linux we may be able to ask the kernel */
1383 {
1384 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1385
1386 if (fd >= 0)
1387 {
1388 sprintf (tmp1, "/proc/self/fd/%d", fd);
1389 req->result = readlink (tmp1, res, PATH_MAX);
1390 close (fd);
1391
1392 /* here we should probably stat the open file and the disk file, to make sure they still match */
1393
1394 if (req->result > 0)
1395 goto done;
1396 }
1397 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1398 return;
1399 }
1400#endif
1401#endif
1402
1403 if (*rel != '/')
1404 {
1405 if (!getcwd (res, PATH_MAX))
1406 return;
1407
1408 if (res [1]) /* only use if not / */
1409 res += strlen (res);
1410 }
1411
1412 while (*rel)
1413 {
1414 eio_ssize_t len, linklen;
1415 char *beg = rel;
1416
1417 while (*rel && *rel != '/')
1418 ++rel;
1419
1420 len = rel - beg;
1421
1422 if (!len) /* skip slashes */
1423 {
1424 ++rel;
1425 continue;
1426 }
1427
1428 if (beg [0] == '.')
1429 {
1430 if (len == 1)
1431 continue; /* . - nop */
1432
1433 if (beg [1] == '.' && len == 2)
1434 {
1435 /* .. - back up one component, if possible */
1436
1437 while (res != req->ptr2)
1438 if (*--res == '/')
1439 break;
1440
1441 continue;
1442 }
1443 }
1444
1445 errno = ENAMETOOLONG;
1446 if (res + 1 + len + 1 >= tmp1)
1447 return;
1448
1449 /* copy one component */
1450 *res = '/';
1451 memcpy (res + 1, beg, len);
1452
1453 /* zero-terminate, for readlink */
1454 res [len + 1] = 0;
1455
1456 /* now check if it's a symlink */
1457 linklen = readlink (req->ptr2, tmp1, PATH_MAX);
1458
1459 if (linklen < 0)
1460 {
1461 if (errno != EINVAL)
1462 return;
1463
1464 /* it's a normal directory. hopefully */
1465 res += len + 1;
1466 }
1467 else
1468 {
1469 /* yay, it was a symlink - build new path in tmp2 */
1470 int rellen = strlen (rel);
1471
1472 errno = ENAMETOOLONG;
1473 if (linklen + 1 + rellen >= PATH_MAX)
1474 return;
1475
1476 errno = ELOOP;
1477 if (!--symlinks)
1478 return;
1479
1480 if (*tmp1 == '/')
1481 res = req->ptr2; /* symlink resolves to an absolute path */
1482
1483 /* we need to be careful, as rel might point into tmp2 already */
1484 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1485 tmp2 [linklen] = '/';
1486 memcpy (tmp2, tmp1, linklen);
1487
1488 rel = tmp2;
1489 }
1490 }
1491
1492 /* special case for the lone root path */
1493 if (res == req->ptr2)
1494 *res++ = '/';
1495
1496 req->result = res - (char *)req->ptr2;
1497
1498done:
1499 req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
993} 1500}
994 1501
995static signed char 1502static signed char
996eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) 1503eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
997{ 1504{
998 return b->score - a->score ? b->score - a->score /* works because our signed char is always 0..100 */ 1505 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; 1506 : a->inode < b->inode ? -1
1507 : a->inode > b->inode ? 1
1508 : 0;
1000} 1509}
1001 1510
1511#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1512
1002#define EIO_QSORT_CUTOFF 20 /* quite high, but performs well on many filesystems */ 1513#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1514#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1003 1515
1004static void 1516static void
1005eio_dent_sort (eio_dirent *dents, int size) 1517eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1006{ 1518{
1519 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1520 unsigned char *bit = bits;
1521
1522 assert (CHAR_BIT == 8);
1523 assert (sizeof (eio_dirent) * 8 < 256);
1524 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1525 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1526
1527 if (size <= EIO_SORT_FAST)
1528 return;
1529
1530 /* first prepare an array of bits to test in our radix sort */
1531 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1532 /* inode_bits must contain all inodes ORed together */
1533 /* which is used to skip bits that are 0 everywhere, which is very common */
1534 {
1535 eio_ino_t endianness;
1007 int i, j; 1536 int i, j;
1008 1537
1009 if (size > EIO_QSORT_CUTOFF * 3) /* skip quicksort for small directories */ 1538 /* we store the byte offset of byte n into byte n of "endianness" */
1539 for (i = 0; i < sizeof (eio_ino_t); ++i)
1540 ((unsigned char *)&endianness)[i] = i;
1541
1542 *bit++ = 0;
1543
1544 for (i = 0; i < sizeof (eio_ino_t); ++i)
1545 {
1546 /* shifting off the byte offsets out of "endianness" */
1547 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1548 endianness >>= 8;
1549
1550 for (j = 0; j < 8; ++j)
1551 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1552 *bit++ = offs + j;
1553 }
1554
1555 for (j = 0; j < 8; ++j)
1556 if (score_bits & (1 << j))
1557 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1558 }
1559
1560 /* now actually do the sorting (a variant of MSD radix sort) */
1561 {
1562 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1563 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1564 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1565 int stk_idx = 0;
1566
1567 base_stk [stk_idx] = dents;
1568 end_stk [stk_idx] = dents + size;
1569 bit_stk [stk_idx] = bit - 1;
1570
1571 do
1572 {
1573 base = base_stk [stk_idx];
1574 end = end_stk [stk_idx];
1575 bit = bit_stk [stk_idx];
1576
1577 for (;;)
1578 {
1579 unsigned char O = *bit >> 3;
1580 unsigned char M = 1 << (*bit & 7);
1581
1582 eio_dirent *a = base;
1583 eio_dirent *b = end;
1584
1585 if (b - a < EIO_SORT_CUTOFF)
1586 break;
1587
1588 /* now bit-partition the array on the bit */
1589 /* this ugly asymmetric loop seems to perform much better than typical */
1590 /* partition algos found in the literature */
1591 do
1592 if (!(((unsigned char *)a)[O] & M))
1593 ++a;
1594 else if (!(((unsigned char *)--b)[O] & M))
1595 {
1596 eio_dirent tmp = *a; *a = *b; *b = tmp;
1597 ++a;
1598 }
1599 while (b > a);
1600
1601 /* next bit, or stop, if no bits left in this path */
1602 if (!*--bit)
1603 break;
1604
1605 base_stk [stk_idx] = a;
1606 end_stk [stk_idx] = end;
1607 bit_stk [stk_idx] = bit;
1608 ++stk_idx;
1609
1610 end = a;
1611 }
1612 }
1613 while (stk_idx--);
1614 }
1615}
1616
1617static void
1618eio_dent_insertion_sort (eio_dirent *dents, int size)
1619{
1620 /* first move the smallest element to the front, to act as a sentinel */
1621 {
1622 int i;
1623 eio_dirent *min = dents;
1624
1625 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1626 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1627 if (EIO_DENT_CMP (dents [i], <, *min))
1628 min = &dents [i];
1629
1630 /* swap elements 0 and j (minimum) */
1010 { 1631 {
1011 /* first, use quicksort */ 1632 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 } 1633 }
1634 }
1061 1635
1062 /* use a simple insertion sort at the end */ 1636 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1637 {
1638 eio_dirent *i, *j;
1639
1063 for (i = 1; i < size; ++i) 1640 for (i = dents + 1; i < dents + size; ++i)
1064 { 1641 {
1065 eio_dirent value = dents [i]; 1642 eio_dirent value = *i;
1066 1643
1067 for (j = i - 1; j >= 0 && eio_dent_cmp (&dents [j], &value) > 0; --j) 1644 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1068 dents [j + 1] = dents [j]; 1645 j [1] = j [0];
1069 1646
1070 dents [j + 1] = value; 1647 j [1] = value;
1071 } 1648 }
1649 }
1650}
1651
1652static void
1653eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1654{
1655 if (size <= 1)
1656 return; /* our insertion sort relies on size > 0 */
1657
1658 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1659 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1660 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1661
1662 /* use an insertion sort at the end, or for small arrays, */
1663 /* as insertion sort is more efficient for small partitions */
1664 eio_dent_insertion_sort (dents, size);
1072} 1665}
1073 1666
1074/* read a full directory */ 1667/* read a full directory */
1075static void 1668static void
1076eio__scandir (eio_req *req, etp_worker *self) 1669eio__scandir (eio_req *req, etp_worker *self)
1077{ 1670{
1078 DIR *dirp;
1079 EIO_STRUCT_DIRENT *entp;
1080 char *name, *names; 1671 char *name, *names;
1081 int namesalloc = 4096; 1672 int namesalloc = 4096 - sizeof (void *) * 4;
1082 int namesoffs = 0; 1673 int namesoffs = 0;
1083 int flags = req->int1; 1674 int flags = req->int1;
1084 eio_dirent *dents = 0; 1675 eio_dirent *dents = 0;
1085 int dentalloc = 128; 1676 int dentalloc = 128;
1086 int dentoffs = 0; 1677 int dentoffs = 0;
1678 eio_ino_t inode_bits = 0;
1679#ifdef _WIN32
1680 HANDLE dirp;
1681 WIN32_FIND_DATA entp;
1682#else
1683 DIR *dirp;
1684 EIO_STRUCT_DIRENT *entp;
1685#endif
1087 1686
1088 req->result = -1; 1687 req->result = -1;
1089 1688
1090 if (!(flags & EIO_READDIR_DENTS)) 1689 if (!(flags & EIO_READDIR_DENTS))
1091 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); 1690 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1092 1691
1093 X_LOCK (wrklock); 1692#ifdef _WIN32
1094 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 1693 {
1694 int len = strlen ((const char *)req->ptr1);
1695 char *path = malloc (MAX_PATH);
1696 const char *fmt;
1697
1698 if (!len)
1699 fmt = "./*";
1700 else if (((const char *)req->ptr1)[len - 1] == '/' || ((const char *)req->ptr1)[len - 1] == '\\')
1701 fmt = "%s*";
1702 else
1703 fmt = "%s/*";
1704
1705 _snprintf (path, MAX_PATH, fmt, (const char *)req->ptr1);
1706 dirp = FindFirstFile (path, &entp);
1707 free (path);
1708
1709 if (dirp == INVALID_HANDLE_VALUE)
1710 {
1711 /* should steal _dosmaperr */
1712 switch (GetLastError ())
1713 {
1714 case ERROR_FILE_NOT_FOUND:
1715 req->result = 0;
1716 break;
1717
1718 case ERROR_INVALID_NAME:
1719 case ERROR_PATH_NOT_FOUND:
1720 case ERROR_NO_MORE_FILES:
1721 errno = ENOENT;
1722 break;
1723
1724 case ERROR_NOT_ENOUGH_MEMORY:
1725 errno = ENOMEM;
1726 break;
1727
1728 default:
1729 errno = EINVAL;
1730 break;
1731 }
1732
1733 return;
1734 }
1735 }
1736#else
1095 self->dirp = dirp = opendir (req->ptr1); 1737 dirp = opendir (req->ptr1);
1738
1739 if (!dirp)
1740 return;
1741#endif
1742
1743 if (req->flags & EIO_FLAG_PTR1_FREE)
1744 free (req->ptr1);
1745
1096 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1746 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; 1747 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1099 X_UNLOCK (wrklock); 1748 req->ptr2 = names = malloc (namesalloc);
1100 1749
1101 if (dirp && names && (!flags || dents)) 1750 if (!names || (flags && !dents))
1751 return;
1752
1102 for (;;) 1753 for (;;)
1103 { 1754 {
1755 int done;
1756
1757#ifdef _WIN32
1758 done = !dirp;
1759#else
1104 errno = 0; 1760 errno = 0;
1105 entp = readdir (dirp); 1761 entp = readdir (dirp);
1762 done = !entp;
1763#endif
1106 1764
1107 if (!entp) 1765 if (done)
1108 { 1766 {
1767#ifndef _WIN32
1768 int old_errno = errno;
1769 closedir (dirp);
1770 errno = old_errno;
1771
1109 if (errno) 1772 if (errno)
1110 break; 1773 break;
1774#endif
1111 1775
1112 /* sort etc. */ 1776 /* sort etc. */
1113 req->int1 = flags; 1777 req->int1 = flags;
1114 req->result = dentoffs; 1778 req->result = dentoffs;
1115 1779
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 1780 if (flags & EIO_READDIR_STAT_ORDER)
1125 || !(~flags & (EIO_READDIR_DIRS_FIRST | EIO_READDIR_FOUND_UNKNOWN))) 1781 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) 1782 else if (flags & EIO_READDIR_DIRS_FIRST)
1783 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1784 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1785 else
1128 { 1786 {
1129 /* in this case, all is known, and we just put dirs first and sort them */ 1787 /* in this case, all is known, and we just put dirs first and sort them */
1130 eio_dirent *ent = dents + dentoffs; 1788 eio_dirent *oth = dents + dentoffs;
1131 eio_dirent *dir = dents; 1789 eio_dirent *dir = dents;
1132 1790
1133 /* now move dirs to the front, and non-dirs to the back */ 1791 /* now partition dirs to the front, and non-dirs to the back */
1134 /* by walking from both sides and swapping if necessary */ 1792 /* by walking from both sides and swapping if necessary */
1135 while (ent > dir) 1793 while (oth > dir)
1136 { 1794 {
1137 if (dir->type == DT_DIR) 1795 if (dir->type == EIO_DT_DIR)
1138 ++dir; 1796 ++dir;
1139 else 1797 else if ((--oth)->type == EIO_DT_DIR)
1140 { 1798 {
1141 --ent;
1142
1143 if (ent->type == DT_DIR)
1144 {
1145 eio_dirent tmp = *dir; 1799 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1146 *dir = *ent;
1147 *ent = tmp;
1148 1800
1149 ++dir; 1801 ++dir;
1150 }
1151 } 1802 }
1152 } 1803 }
1153 1804
1154 /* now sort the dirs only */ 1805 /* now sort the dirs only (dirs all have the same score) */
1155 eio_dent_sort (dents, dir - dents); 1806 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1156 } 1807 }
1157 1808
1158 /* only provide the names array unless DENTS is specified */
1159 if (!(flags & EIO_READDIR_DENTS))
1160 {
1161 X_LOCK (wrklock);
1162 assert (!dents);
1163 req->ptr1 = 0;
1164 req->ptr2 = names;
1165 X_UNLOCK (wrklock);
1166 }
1167
1168 break; 1809 break;
1169 } 1810 }
1170 1811
1171 /* now add the entry to our list(s) */ 1812 /* now add the entry to our list(s) */
1172 name = entp->d_name; 1813 name = D_NAME (entp);
1173 1814
1174 /* skip . and .. entries */ 1815 /* skip . and .. entries */
1175 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1816 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1176 { 1817 {
1177 int len = D_NAMLEN (entp) + 1; 1818 int len = D_NAMLEN (entp) + 1;
1178 1819
1179 while (expect_false (namesoffs + len > namesalloc)) 1820 while (ecb_expect_false (namesoffs + len > namesalloc))
1180 { 1821 {
1181 namesalloc *= 2; 1822 namesalloc *= 2;
1182 X_LOCK (wrklock);
1183 req->ptr1 = names = realloc (names, namesalloc); 1823 req->ptr2 = names = realloc (names, namesalloc);
1184 X_UNLOCK (wrklock);
1185 1824
1186 if (!names) 1825 if (!names)
1187 break; 1826 break;
1188 } 1827 }
1189 1828
1190 memcpy (names + namesoffs, name, len); 1829 memcpy (names + namesoffs, name, len);
1191 1830
1192 if (dents) 1831 if (dents)
1193 { 1832 {
1194 struct eio_dirent *ent; 1833 struct eio_dirent *ent;
1195 1834
1196 if (expect_false (dentoffs == dentalloc)) 1835 if (ecb_expect_false (dentoffs == dentalloc))
1197 { 1836 {
1198 dentalloc *= 2; 1837 dentalloc *= 2;
1199 X_LOCK (wrklock);
1200 req->ptr2 = dents = realloc (dents, dentalloc * sizeof (eio_dirent)); 1838 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1201 X_UNLOCK (wrklock);
1202 1839
1203 if (!dents) 1840 if (!dents)
1204 break; 1841 break;
1205 } 1842 }
1206 1843
1207 ent = dents + dentoffs; 1844 ent = dents + dentoffs;
1208 1845
1209 ent->name = (char *)(size_t)namesoffs; /* rather dirtily we store the offset in the pointer */ 1846 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1210 ent->namelen = len - 1; 1847 ent->namelen = len - 1;
1211 ent->inode = D_INO (entp); 1848 ent->inode = D_INO (entp);
1212 1849
1850 inode_bits |= ent->inode;
1851
1213 switch (D_TYPE (entp)) 1852 switch (D_TYPE (entp))
1214 { 1853 {
1215 default: 1854 default:
1216 ent->type = EIO_DT_UNKNOWN; 1855 ent->type = EIO_DT_UNKNOWN;
1217 flags |= EIO_READDIR_FOUND_UNKNOWN; 1856 flags |= EIO_READDIR_FOUND_UNKNOWN;
1218 break; 1857 break;
1219 1858
1220 #ifdef DT_FIFO 1859 #ifdef DT_FIFO
1221 case DT_FIFO: ent->type = EIO_DT_FIFO; break; 1860 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1222 #endif 1861 #endif
1223 #ifdef DT_CHR 1862 #ifdef DT_CHR
1224 case DT_CHR: ent->type = EIO_DT_CHR; break; 1863 case DT_CHR: ent->type = EIO_DT_CHR; break;
1225 #endif 1864 #endif
1226 #ifdef DT_MPC 1865 #ifdef DT_MPC
1227 case DT_MPC: ent->type = EIO_DT_MPC; break; 1866 case DT_MPC: ent->type = EIO_DT_MPC; break;
1228 #endif 1867 #endif
1229 #ifdef DT_DIR 1868 #ifdef DT_DIR
1230 case DT_DIR: ent->type = EIO_DT_DIR; break; 1869 case DT_DIR: ent->type = EIO_DT_DIR; break;
1231 #endif 1870 #endif
1232 #ifdef DT_NAM 1871 #ifdef DT_NAM
1233 case DT_NAM: ent->type = EIO_DT_NAM; break; 1872 case DT_NAM: ent->type = EIO_DT_NAM; break;
1234 #endif 1873 #endif
1235 #ifdef DT_BLK 1874 #ifdef DT_BLK
1236 case DT_BLK: ent->type = EIO_DT_BLK; break; 1875 case DT_BLK: ent->type = EIO_DT_BLK; break;
1237 #endif 1876 #endif
1238 #ifdef DT_MPB 1877 #ifdef DT_MPB
1239 case DT_MPB: ent->type = EIO_DT_MPB; break; 1878 case DT_MPB: ent->type = EIO_DT_MPB; break;
1240 #endif 1879 #endif
1241 #ifdef DT_REG 1880 #ifdef DT_REG
1242 case DT_REG: ent->type = EIO_DT_REG; break; 1881 case DT_REG: ent->type = EIO_DT_REG; break;
1243 #endif 1882 #endif
1244 #ifdef DT_NWK 1883 #ifdef DT_NWK
1245 case DT_NWK: ent->type = EIO_DT_NWK; break; 1884 case DT_NWK: ent->type = EIO_DT_NWK; break;
1246 #endif 1885 #endif
1247 #ifdef DT_CMP 1886 #ifdef DT_CMP
1248 case DT_CMP: ent->type = EIO_DT_CMP; break; 1887 case DT_CMP: ent->type = EIO_DT_CMP; break;
1249 #endif 1888 #endif
1250 #ifdef DT_LNK 1889 #ifdef DT_LNK
1251 case DT_LNK: ent->type = EIO_DT_LNK; break; 1890 case DT_LNK: ent->type = EIO_DT_LNK; break;
1252 #endif 1891 #endif
1253 #ifdef DT_SOCK 1892 #ifdef DT_SOCK
1254 case DT_SOCK: ent->type = EIO_DT_SOCK; break; 1893 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1255 #endif 1894 #endif
1256 #ifdef DT_DOOR 1895 #ifdef DT_DOOR
1257 case DT_DOOR: ent->type = EIO_DT_DOOR; break; 1896 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1258 #endif 1897 #endif
1259 #ifdef DT_WHT 1898 #ifdef DT_WHT
1260 case DT_WHT: ent->type = EIO_DT_WHT; break; 1899 case DT_WHT: ent->type = EIO_DT_WHT; break;
1261 #endif 1900 #endif
1262 } 1901 }
1263 1902
1264 ent->score = 0; 1903 ent->score = 7;
1265 1904
1266 if (flags & EIO_READDIR_DIRS_FIRST) 1905 if (flags & EIO_READDIR_DIRS_FIRST)
1267 { 1906 {
1268 if (ent->type == EIO_DT_UNKNOWN) 1907 if (ent->type == EIO_DT_UNKNOWN)
1269 { 1908 {
1270 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */ 1909 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1271 ent->score = 98; 1910 ent->score = 1;
1272 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */ 1911 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 */ 1912 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1274 } 1913 }
1275 else if (ent->type == EIO_DT_DIR) 1914 else if (ent->type == EIO_DT_DIR)
1276 ent->score = 100; 1915 ent->score = 0;
1277 } 1916 }
1278 } 1917 }
1279 1918
1280 namesoffs += len; 1919 namesoffs += len;
1281 ++dentoffs; 1920 ++dentoffs;
1282 } 1921 }
1283 1922
1284 if (EIO_CANCELLED (req)) 1923 if (EIO_CANCELLED (req))
1285 { 1924 {
1286 errno = ECANCELED; 1925 errno = ECANCELED;
1287 break; 1926 break;
1288 } 1927 }
1928
1929#ifdef _WIN32
1930 if (!FindNextFile (dirp, &entp))
1931 {
1932 FindClose (dirp);
1933 dirp = 0;
1289 } 1934 }
1290}
1291
1292#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1293# undef msync
1294# define msync(a,b,c) ((errno = ENOSYS), -1)
1295#endif 1935#endif
1296 1936 }
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} 1937}
1321 1938
1322/*****************************************************************************/ 1939/*****************************************************************************/
1323 1940
1324#define ALLOC(len) \ 1941#define ALLOC(len) \
… …
1340{ 1957{
1341 ETP_REQ *req; 1958 ETP_REQ *req;
1342 struct timespec ts; 1959 struct timespec ts;
1343 etp_worker *self = (etp_worker *)thr_arg; 1960 etp_worker *self = (etp_worker *)thr_arg;
1344 1961
1962#if HAVE_PRCTL_SET_NAME
1963 prctl (PR_SET_NAME, (unsigned long)"eio_thread", 0, 0, 0);
1964#endif
1965
1345 /* try to distribute timeouts somewhat randomly */ 1966 /* try to distribute timeouts somewhat evenly */
1346 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 1967 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1347 1968
1348 for (;;) 1969 for (;;)
1349 { 1970 {
1971 ts.tv_sec = 0;
1972
1350 X_LOCK (reqlock); 1973 X_LOCK (reqlock);
1351 1974
1352 for (;;) 1975 for (;;)
1353 { 1976 {
1354 self->req = req = reqq_shift (&req_queue); 1977 self->req = req = reqq_shift (&req_queue);
1355 1978
1356 if (req) 1979 if (req)
1357 break; 1980 break;
1358 1981
1982 if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
1983 {
1984 X_UNLOCK (reqlock);
1985 X_LOCK (wrklock);
1986 --started;
1987 X_UNLOCK (wrklock);
1988 goto quit;
1989 }
1990
1359 ++idle; 1991 ++idle;
1360 1992
1361 ts.tv_sec = time (0) + IDLE_TIMEOUT; 1993 if (idle <= max_idle)
1362 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 1994 /* we are allowed to idle, so do so without any timeout */
1995 X_COND_WAIT (reqwait, reqlock);
1996 else
1363 { 1997 {
1364 if (idle > max_idle) 1998 /* initialise timeout once */
1365 { 1999 if (!ts.tv_sec)
1366 --idle; 2000 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 2001
1374 /* we are allowed to idle, so do so without any timeout */
1375 X_COND_WAIT (reqwait, reqlock); 2002 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
2003 ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1376 } 2004 }
1377 2005
1378 --idle; 2006 --idle;
1379 } 2007 }
1380 2008
… …
1383 X_UNLOCK (reqlock); 2011 X_UNLOCK (reqlock);
1384 2012
1385 if (req->type < 0) 2013 if (req->type < 0)
1386 goto quit; 2014 goto quit;
1387 2015
1388 if (!EIO_CANCELLED (req))
1389 ETP_EXECUTE (self, req); 2016 ETP_EXECUTE (self, req);
1390 2017
1391 X_LOCK (reslock); 2018 X_LOCK (reslock);
1392 2019
1393 ++npending; 2020 ++npending;
1394 2021
… …
1409 return 0; 2036 return 0;
1410} 2037}
1411 2038
1412/*****************************************************************************/ 2039/*****************************************************************************/
1413 2040
2041int ecb_cold
1414int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 2042eio_init (void (*want_poll)(void), void (*done_poll)(void))
1415{ 2043{
2044#if !HAVE_PREADWRITE
2045 X_MUTEX_CREATE (preadwritelock);
2046#endif
2047
1416 return etp_init (want_poll, done_poll); 2048 return etp_init (want_poll, done_poll);
1417} 2049}
1418 2050
2051ecb_inline void
1419static void eio_api_destroy (eio_req *req) 2052eio_api_destroy (eio_req *req)
1420{ 2053{
1421 free (req); 2054 free (req);
1422} 2055}
1423 2056
1424#define REQ(rtype) \ 2057#define REQ(rtype) \
… …
1443 { \ 2076 { \
1444 eio_api_destroy (req); \ 2077 eio_api_destroy (req); \
1445 return 0; \ 2078 return 0; \
1446 } 2079 }
1447 2080
2081static void
1448static void eio_execute (etp_worker *self, eio_req *req) 2082eio_execute (etp_worker *self, eio_req *req)
1449{ 2083{
1450 errno = 0; 2084 if (ecb_expect_false (EIO_CANCELLED (req)))
2085 {
2086 req->result = -1;
2087 req->errorno = ECANCELED;
2088 return;
2089 }
1451 2090
1452 switch (req->type) 2091 switch (req->type)
1453 { 2092 {
1454 case EIO_READ: ALLOC (req->size); 2093 case EIO_READ: ALLOC (req->size);
1455 req->result = req->offs >= 0 2094 req->result = req->offs >= 0
… …
1458 case EIO_WRITE: req->result = req->offs >= 0 2097 case EIO_WRITE: req->result = req->offs >= 0
1459 ? pwrite (req->int1, req->ptr2, req->size, req->offs) 2098 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1460 : write (req->int1, req->ptr2, req->size); break; 2099 : write (req->int1, req->ptr2, req->size); break;
1461 2100
1462 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break; 2101 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; 2102 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
1464 2103
1465 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2104 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1466 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2105 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1467 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2106 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1468 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2107 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1469 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2108 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1470 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; 2109 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
2110
2111 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2112 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2113 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2114 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1471 2115
1472 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; 2116 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; 2117 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; 2118 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; 2119 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
… …
1483 case EIO_RMDIR: req->result = rmdir (req->ptr1); break; 2127 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1484 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; 2128 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; 2129 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1486 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; 2130 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1487 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; 2131 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; 2132 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1489 2133
2134 case EIO_REALPATH: eio__realpath (req, self); break;
2135
1490 case EIO_READLINK: ALLOC (NAME_MAX); 2136 case EIO_READLINK: ALLOC (PATH_MAX);
1491 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; 2137 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1492 2138
1493 case EIO_SYNC: req->result = 0; sync (); break; 2139 case EIO_SYNC: req->result = 0; sync (); break;
1494 case EIO_FSYNC: req->result = fsync (req->int1); break; 2140 case EIO_FSYNC: req->result = fsync (req->int1); break;
1495 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 2141 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1496 case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break; 2142 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; 2143 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
2144 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
2145 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
2146 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
2147 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
2148 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
1499 2149
1500 case EIO_READDIR: eio__scandir (req, self); break; 2150 case EIO_READDIR: eio__scandir (req, self); break;
1501 2151
1502 case EIO_BUSY: 2152 case EIO_BUSY:
1503#ifdef _WIN32 2153#ifdef _WIN32
1504 Sleep (req->nv1 * 1000.); 2154 Sleep (req->nv1 * 1e3);
1505#else 2155#else
1506 { 2156 {
1507 struct timeval tv; 2157 struct timeval tv;
1508 2158
1509 tv.tv_sec = req->nv1; 2159 tv.tv_sec = req->nv1;
1510 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; 2160 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1511 2161
1512 req->result = select (0, 0, 0, 0, &tv); 2162 req->result = select (0, 0, 0, 0, &tv);
1513 } 2163 }
1514#endif 2164#endif
1515 break; 2165 break;
… …
1530 times = tv; 2180 times = tv;
1531 } 2181 }
1532 else 2182 else
1533 times = 0; 2183 times = 0;
1534 2184
1535
1536 req->result = req->type == EIO_FUTIME 2185 req->result = req->type == EIO_FUTIME
1537 ? futimes (req->int1, times) 2186 ? futimes (req->int1, times)
1538 : utimes (req->ptr1, times); 2187 : utimes (req->ptr1, times);
1539 } 2188 }
1540 break; 2189 break;
… …
1545 case EIO_NOP: 2194 case EIO_NOP:
1546 req->result = 0; 2195 req->result = 0;
1547 break; 2196 break;
1548 2197
1549 case EIO_CUSTOM: 2198 case EIO_CUSTOM:
1550 ((void (*)(eio_req *))req->feed) (req); 2199 req->feed (req);
1551 break; 2200 break;
1552 2201
1553 default: 2202 default:
2203 errno = ENOSYS;
1554 req->result = -1; 2204 req->result = -1;
1555 break; 2205 break;
1556 } 2206 }
1557 2207
1558 req->errorno = errno; 2208 req->errorno = errno;
… …
1583eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2233eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1584{ 2234{
1585 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2235 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1586} 2236}
1587 2237
2238eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2239{
2240 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2241}
2242
2243eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
2244{
2245 REQ (EIO_SYNCFS); req->int1 = fd; SEND;
2246}
2247
2248eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2249{
2250 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2251}
2252
1588eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2253eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1589{ 2254{
1590 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2255 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1591} 2256}
1592 2257
1593eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) 2258eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
1594{ 2259{
1595 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; 2260 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
1596} 2261}
1597 2262
1598eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) 2263eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
1599{ 2264{
1600 REQ (EIO_FDATASYNC); req->int1 = fd; SEND; 2265 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2266}
2267
2268eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2269{
2270 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
1601} 2271}
1602 2272
1603eio_req *eio_close (int fd, int pri, eio_cb cb, void *data) 2273eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
1604{ 2274{
1605 REQ (EIO_CLOSE); req->int1 = fd; SEND; 2275 REQ (EIO_CLOSE); req->int1 = fd; SEND;
… …
1623eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) 2293eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1624{ 2294{
1625 REQ (EIO_FSTAT); req->int1 = fd; SEND; 2295 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1626} 2296}
1627 2297
2298eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
2299{
2300 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
2301}
2302
1628eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) 2303eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1629{ 2304{
1630 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; 2305 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1631} 2306}
1632 2307
… …
1638eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) 2313eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1639{ 2314{
1640 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2315 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1641} 2316}
1642 2317
1643eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2318eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1644{ 2319{
1645 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2320 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1646} 2321}
1647 2322
1648eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) 2323eio_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) 2343eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1669{ 2344{
1670 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2345 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1671} 2346}
1672 2347
1673eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2348eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1674{ 2349{
1675 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2350 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1676} 2351}
1677 2352
1678eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) 2353eio_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) 2369eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1695{ 2370{
1696 return eio__1path (EIO_READLINK, path, pri, cb, data); 2371 return eio__1path (EIO_READLINK, path, pri, cb, data);
1697} 2372}
1698 2373
2374eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2375{
2376 return eio__1path (EIO_REALPATH, path, pri, cb, data);
2377}
2378
1699eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) 2379eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1700{ 2380{
1701 return eio__1path (EIO_STAT, path, pri, cb, data); 2381 return eio__1path (EIO_STAT, path, pri, cb, data);
1702} 2382}
1703 2383
1704eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) 2384eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1705{ 2385{
1706 return eio__1path (EIO_LSTAT, path, pri, cb, data); 2386 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1707} 2387}
1708 2388
2389eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
2390{
2391 return eio__1path (EIO_STATVFS, path, pri, cb, data);
2392}
2393
1709eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) 2394eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1710{ 2395{
1711 return eio__1path (EIO_UNLINK, path, pri, cb, data); 2396 return eio__1path (EIO_UNLINK, path, pri, cb, data);
1712} 2397}
1713 2398
… …
1721 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND; 2406 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1722} 2407}
1723 2408
1724eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) 2409eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1725{ 2410{
1726 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; 2411 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1727} 2412}
1728 2413
1729static eio_req * 2414static eio_req *
1730eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2415eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1731{ 2416{
… …
1755eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2440eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1756{ 2441{
1757 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2442 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1758} 2443}
1759 2444
1760eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) 2445eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
1761{ 2446{
1762 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND; 2447 REQ (EIO_CUSTOM); req->feed = execute; SEND;
1763} 2448}
1764 2449
1765#endif 2450#endif
1766 2451
1767eio_req *eio_grp (eio_cb cb, void *data) 2452eio_req *eio_grp (eio_cb cb, void *data)
… …
1776#undef SEND 2461#undef SEND
1777 2462
1778/*****************************************************************************/ 2463/*****************************************************************************/
1779/* grp functions */ 2464/* grp functions */
1780 2465
2466void
1781void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2467eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1782{ 2468{
1783 grp->int2 = limit; 2469 grp->int2 = limit;
1784 grp->feed = feed; 2470 grp->feed = feed;
1785 2471
1786 grp_try_feed (grp); 2472 grp_try_feed (grp);
1787} 2473}
1788 2474
2475void
1789void eio_grp_limit (eio_req *grp, int limit) 2476eio_grp_limit (eio_req *grp, int limit)
1790{ 2477{
1791 grp->int2 = limit; 2478 grp->int2 = limit;
1792 2479
1793 grp_try_feed (grp); 2480 grp_try_feed (grp);
1794} 2481}
1795 2482
2483void
1796void eio_grp_add (eio_req *grp, eio_req *req) 2484eio_grp_add (eio_req *grp, eio_req *req)
1797{ 2485{
1798 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2486 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1799 2487
1800 grp->flags |= EIO_FLAG_GROUPADD; 2488 grp->flags |= EIO_FLAG_GROUPADD;
1801 2489
… …
1812} 2500}
1813 2501
1814/*****************************************************************************/ 2502/*****************************************************************************/
1815/* misc garbage */ 2503/* misc garbage */
1816 2504
2505eio_ssize_t
1817ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2506eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1818{ 2507{
1819 etp_worker wrk;
1820 ssize_t ret;
1821
1822 wrk.dbuf = 0;
1823
1824 ret = eio__sendfile (ofd, ifd, offset, count, &wrk); 2508 return eio__sendfile (ofd, ifd, offset, count);
1825
1826 if (wrk.dbuf)
1827 free (wrk.dbuf);
1828
1829 return ret;
1830} 2509}
1831 2510

Diff Legend

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