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

Comparing libeio/eio.c (file contents):
Revision 1.30 by root, Wed Jun 3 12:24:49 2009 UTC vs.
Revision 1.106 by root, Mon Sep 26 20:19:08 2011 UTC

1/* 1/*
2 * libeio implementation 2 * libeio implementation
3 * 3 *
4 * Copyright (c) 2007,2008 Marc Alexander Lehmann <libeio@schmorp.de> 4 * Copyright (c) 2007,2008,2009,2010,2011 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#if _POSIX_VERSION >= 200809L
64# define HAVE_AT 1
65#else
66# define HAVE_AT 0
67#endif
68
69/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
70/* intptr_t only comes from stdint.h, says idiot openbsd coder */
71#if HAVE_STDINT_H
72# include <stdint.h>
73#endif
74
75#ifndef ECANCELED
76# define ECANCELED EDOM
77#endif
78#ifndef ELOOP
79# define ELOOP EDOM
80#endif
81
82#if !defined(ENOTSOCK) && defined(WSAENOTSOCK)
83# define ENOTSOCK WSAENOTSOCK
84#endif
85
86static void eio_destroy (eio_req *req);
87
58#ifndef EIO_FINISH 88#ifndef EIO_FINISH
59# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0 89# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
60#endif 90#endif
61 91
62#ifndef EIO_DESTROY 92#ifndef EIO_DESTROY
… …
65 95
66#ifndef EIO_FEED 96#ifndef EIO_FEED
67# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0) 97# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)
68#endif 98#endif
69 99
100#ifndef EIO_FD_TO_WIN32_HANDLE
101# define EIO_FD_TO_WIN32_HANDLE(fd) _get_osfhandle (fd)
102#endif
103#ifndef EIO_WIN32_HANDLE_TO_FD
104# define EIO_WIN32_HANDLE_TO_FD(handle) _open_osfhandle (handle, 0)
105#endif
106
107#define EIO_ERRNO(errval,retval) ((errno = errval), retval)
108
109#define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1)
110
70#ifdef _WIN32 111#ifdef _WIN32
71 112
72 /*doh*/ 113 #undef PAGESIZE
114 #define PAGESIZE 4096 /* GetSystemInfo? */
115
116 /* TODO: look at how perl does stat (non-sloppy), unlink (ro-files), utime, link */
117
118 #ifdef EIO_STRUCT_STATI64
119 /* look at perl's non-sloppy stat */
120 #define stat(path,buf) _stati64 (path,buf)
121 #define fstat(fd,buf) _fstati64 (fd,buf)
122 #endif
123 #define lstat(path,buf) stat (path,buf)
124 #define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
125 #define mkdir(path,mode) _mkdir (path)
126 #define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
127
128 #define chmod(path,mode) _chmod (path, mode)
129 #define dup(fd) _dup (fd)
130 #define dup2(fd1,fd2) _dup2 (fd1, fd2)
131
132 #define fchmod(fd,mode) EIO_ENOSYS ()
133 #define chown(path,uid,gid) EIO_ENOSYS ()
134 #define fchown(fd,uid,gid) EIO_ENOSYS ()
135 #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
136 #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
137 #define mknod(path,mode,dev) EIO_ENOSYS ()
138 #define sync() EIO_ENOSYS ()
139 #define readlink(path,buf,s) EIO_ENOSYS ()
140 #define statvfs(path,buf) EIO_ENOSYS ()
141 #define fstatvfs(fd,buf) EIO_ENOSYS ()
142
143 /* rename() uses MoveFile, which fails to overwrite */
144 #define rename(old,neu) eio__rename (old, neu)
145
146 static int
147 eio__rename (const char *old, const char *neu)
148 {
149 if (MoveFileEx (old, neu, MOVEFILE_REPLACE_EXISTING))
150 return 0;
151
152 /* should steal _dosmaperr */
153 switch (GetLastError ())
154 {
155 case ERROR_FILE_NOT_FOUND:
156 case ERROR_PATH_NOT_FOUND:
157 case ERROR_INVALID_DRIVE:
158 case ERROR_NO_MORE_FILES:
159 case ERROR_BAD_NETPATH:
160 case ERROR_BAD_NET_NAME:
161 case ERROR_BAD_PATHNAME:
162 case ERROR_FILENAME_EXCED_RANGE:
163 errno = ENOENT;
164 break;
165
166 default:
167 errno = EACCES;
168 break;
169 }
170
171 return -1;
172 }
173
174 /* we could even stat and see if it exists */
175 static int
176 symlink (const char *old, const char *neu)
177 {
178 #if WINVER >= 0x0600
179 if (CreateSymbolicLink (neu, old, 1))
180 return 0;
181
182 if (CreateSymbolicLink (neu, old, 0))
183 return 0;
184 #endif
185
186 return EIO_ERRNO (ENOENT, -1);
187 }
188
189 /* POSIX API only */
190 #define CreateHardLink(neu,old,flags) 0
191 #define CreateSymbolicLink(neu,old,flags) 0
192
193 struct statvfs
194 {
195 int dummy;
196 };
197
198 #define DT_DIR EIO_DT_DIR
199 #define DT_REG EIO_DT_REG
200 #define D_NAME(entp) entp.cFileName
201 #define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
73 202
74#else 203#else
75 204
76# include "config.h"
77# include <sys/time.h> 205 #include <sys/time.h>
78# include <sys/select.h> 206 #include <sys/select.h>
207 #include <sys/statvfs.h>
79# include <unistd.h> 208 #include <unistd.h>
209 #include <signal.h>
210 #include <dirent.h>
211
212 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
213 #include <sys/mman.h>
214 #endif
215
216 #define D_NAME(entp) entp->d_name
217
218 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
219 #if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
220 #define _DIRENT_HAVE_D_TYPE /* sigh */
221 #define D_INO(de) (de)->d_fileno
222 #define D_NAMLEN(de) (de)->d_namlen
223 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
224 #define D_INO(de) (de)->d_ino
225 #endif
226
227 #ifdef _D_EXACT_NAMLEN
228 #undef D_NAMLEN
229 #define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
230 #endif
231
232 #ifdef _DIRENT_HAVE_D_TYPE
233 #define D_TYPE(de) (de)->d_type
234 #endif
235
236 #ifndef EIO_STRUCT_DIRENT
237 #define EIO_STRUCT_DIRENT struct dirent
238 #endif
239
240#endif
241
242#if HAVE_UTIMES
80# include <utime.h> 243# include <utime.h>
81# include <signal.h>
82# include <dirent.h>
83
84# ifndef EIO_STRUCT_DIRENT
85# define EIO_STRUCT_DIRENT struct dirent
86# endif 244#endif
87 245
246#if HAVE_SYS_SYSCALL_H
247# include <sys/syscall.h>
248#endif
249
250#if HAVE_SYS_PRCTL_H
251# include <sys/prctl.h>
88#endif 252#endif
89 253
90#if HAVE_SENDFILE 254#if HAVE_SENDFILE
91# if __linux 255# if __linux
92# include <sys/sendfile.h> 256# include <sys/sendfile.h>
93# elif __freebsd 257# elif __FreeBSD__ || defined __APPLE__
94# include <sys/socket.h> 258# include <sys/socket.h>
95# include <sys/uio.h> 259# include <sys/uio.h>
96# elif __hpux 260# elif __hpux
97# include <sys/socket.h> 261# include <sys/socket.h>
98# elif __solaris /* not yet */ 262# elif __solaris
99# include <sys/sendfile.h> 263# include <sys/sendfile.h>
100# else 264# else
101# error sendfile support requested but not available 265# error sendfile support requested but not available
102# endif 266# endif
103#endif 267#endif
104 268
105/* number of seconds after which an idle threads exit */ 269#ifndef D_TYPE
106#define IDLE_TIMEOUT 10 270# define D_TYPE(de) 0
271#endif
272#ifndef D_INO
273# define D_INO(de) 0
274#endif
275#ifndef D_NAMLEN
276# define D_NAMLEN(entp) strlen (D_NAME (entp))
277#endif
107 278
108/* used for struct dirent, AIX doesn't provide it */ 279/* used for struct dirent, AIX doesn't provide it */
109#ifndef NAME_MAX 280#ifndef NAME_MAX
110# define NAME_MAX 4096 281# define NAME_MAX 4096
111#endif 282#endif
112 283
284/* used for readlink etc. */
285#ifndef PATH_MAX
286# define PATH_MAX 4096
287#endif
288
113/* buffer size for various temporary buffers */ 289/* buffer size for various temporary buffers */
114#define EIO_BUFSIZE 65536 290#define EIO_BUFSIZE 65536
115 291
116#define dBUF \ 292#define dBUF \
117 char *eio_buf; \
118 ETP_WORKER_LOCK (self); \
119 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 293 char *eio_buf = malloc (EIO_BUFSIZE); \
120 ETP_WORKER_UNLOCK (self); \
121 errno = ENOMEM; \ 294 errno = ENOMEM; \
122 if (!eio_buf) \ 295 if (!eio_buf) \
123 return -1; 296 return -1
297
298#define FUBd \
299 free (eio_buf)
124 300
125#define EIO_TICKS ((1000000 + 1023) >> 10) 301#define EIO_TICKS ((1000000 + 1023) >> 10)
126 302
127/*****************************************************************************/ 303/*****************************************************************************/
128 304
129#if __GNUC__ >= 3 305struct tmpbuf
130# define expect(expr,value) __builtin_expect ((expr),(value)) 306{
131#else 307 void *ptr;
132# define expect(expr,value) (expr) 308 int len;
133#endif 309};
134 310
135#define expect_false(expr) expect ((expr) != 0, 0) 311static void *
136#define expect_true(expr) expect ((expr) != 0, 1) 312tmpbuf_get (struct tmpbuf *buf, int len)
313{
314 if (buf->len < len)
315 {
316 free (buf->ptr);
317 buf->ptr = malloc (buf->len = len);
318 }
319
320 return buf->ptr;
321}
137 322
138/*****************************************************************************/ 323/*****************************************************************************/
139 324
140#define ETP_PRI_MIN EIO_PRI_MIN 325#define ETP_PRI_MIN EIO_PRI_MIN
141#define ETP_PRI_MAX EIO_PRI_MAX 326#define ETP_PRI_MAX EIO_PRI_MAX
… …
147static int eio_finish (eio_req *req); 332static int eio_finish (eio_req *req);
148#define ETP_FINISH(req) eio_finish (req) 333#define ETP_FINISH(req) eio_finish (req)
149static void eio_execute (struct etp_worker *self, eio_req *req); 334static void eio_execute (struct etp_worker *self, eio_req *req);
150#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) 335#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
151 336
152#define ETP_WORKER_CLEAR(req) \
153 if (wrk->dbuf) \
154 { \
155 free (wrk->dbuf); \
156 wrk->dbuf = 0; \
157 } \
158 \
159 if (wrk->dirp) \
160 { \
161 closedir (wrk->dirp); \
162 wrk->dirp = 0; \
163 }
164#define ETP_WORKER_COMMON \
165 void *dbuf; \
166 DIR *dirp;
167
168/*****************************************************************************/ 337/*****************************************************************************/
169 338
170#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) 339#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
171 340
172/* calculcate time difference in ~1/EIO_TICKS of a second */ 341/* calculate time difference in ~1/EIO_TICKS of a second */
342ecb_inline int
173static int tvdiff (struct timeval *tv1, struct timeval *tv2) 343tvdiff (struct timeval *tv1, struct timeval *tv2)
174{ 344{
175 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 345 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
176 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 346 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
177} 347}
178 348
… …
182static void (*done_poll_cb) (void); 352static void (*done_poll_cb) (void);
183 353
184static unsigned int max_poll_time; /* reslock */ 354static unsigned int max_poll_time; /* reslock */
185static unsigned int max_poll_reqs; /* reslock */ 355static unsigned int max_poll_reqs; /* reslock */
186 356
187static volatile unsigned int nreqs; /* reqlock */ 357static unsigned int nreqs; /* reqlock */
188static volatile unsigned int nready; /* reqlock */ 358static unsigned int nready; /* reqlock */
189static volatile unsigned int npending; /* reqlock */ 359static unsigned int npending; /* reqlock */
190static volatile unsigned int max_idle = 4; 360static unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
361static unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
191 362
192static mutex_t wrklock = X_MUTEX_INIT; 363static xmutex_t wrklock;
193static mutex_t reslock = X_MUTEX_INIT; 364static xmutex_t reslock;
194static mutex_t reqlock = X_MUTEX_INIT; 365static xmutex_t reqlock;
195static cond_t reqwait = X_COND_INIT; 366static xcond_t reqwait;
196 367
197#if !HAVE_PREADWRITE 368#if !HAVE_PREADWRITE
198/* 369/*
199 * make our pread/pwrite emulation safe against themselves, but not against 370 * make our pread/pwrite emulation safe against themselves, but not against
200 * normal read/write by using a mutex. slows down execution a lot, 371 * normal read/write by using a mutex. slows down execution a lot,
201 * but that's your problem, not mine. 372 * but that's your problem, not mine.
202 */ 373 */
203static mutex_t preadwritelock = X_MUTEX_INIT; 374static xmutex_t preadwritelock;
204#endif 375#endif
205 376
206typedef struct etp_worker 377typedef struct etp_worker
207{ 378{
379 struct tmpbuf tmpbuf;
380
208 /* locked by wrklock */ 381 /* locked by wrklock */
209 struct etp_worker *prev, *next; 382 struct etp_worker *prev, *next;
210 383
211 thread_t tid; 384 xthread_t tid;
212 385
213 /* locked by reslock, reqlock or wrklock */ 386#ifdef ETP_WORKER_COMMON
214 ETP_REQ *req; /* currently processed request */
215
216 ETP_WORKER_COMMON 387 ETP_WORKER_COMMON
388#endif
217} etp_worker; 389} etp_worker;
218 390
219static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */ 391static etp_worker wrk_first; /* NOT etp */
220 392
221#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock) 393#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
222#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock) 394#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
223 395
224/* worker threads management */ 396/* worker threads management */
225 397
398static void ecb_cold
226static void etp_worker_clear (etp_worker *wrk) 399etp_worker_clear (etp_worker *wrk)
227{ 400{
228 ETP_WORKER_CLEAR (wrk);
229} 401}
230 402
403static void ecb_cold
231static void etp_worker_free (etp_worker *wrk) 404etp_worker_free (etp_worker *wrk)
232{ 405{
406 free (wrk->tmpbuf.ptr);
407
233 wrk->next->prev = wrk->prev; 408 wrk->next->prev = wrk->prev;
234 wrk->prev->next = wrk->next; 409 wrk->prev->next = wrk->next;
235 410
236 free (wrk); 411 free (wrk);
237} 412}
238 413
239static unsigned int etp_nreqs (void) 414static unsigned int
415etp_nreqs (void)
240{ 416{
241 int retval; 417 int retval;
242 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 418 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
243 retval = nreqs; 419 retval = nreqs;
244 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 420 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
245 return retval; 421 return retval;
246} 422}
247 423
248static unsigned int etp_nready (void) 424static unsigned int
425etp_nready (void)
249{ 426{
250 unsigned int retval; 427 unsigned int retval;
251 428
252 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 429 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
253 retval = nready; 430 retval = nready;
254 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 431 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
255 432
256 return retval; 433 return retval;
257} 434}
258 435
259static unsigned int etp_npending (void) 436static unsigned int
437etp_npending (void)
260{ 438{
261 unsigned int retval; 439 unsigned int retval;
262 440
263 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 441 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
264 retval = npending; 442 retval = npending;
265 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 443 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
266 444
267 return retval; 445 return retval;
268} 446}
269 447
270static unsigned int etp_nthreads (void) 448static unsigned int
449etp_nthreads (void)
271{ 450{
272 unsigned int retval; 451 unsigned int retval;
273 452
274 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 453 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
275 retval = started; 454 retval = started;
… …
289} etp_reqq; 468} etp_reqq;
290 469
291static etp_reqq req_queue; 470static etp_reqq req_queue;
292static etp_reqq res_queue; 471static etp_reqq res_queue;
293 472
473static void ecb_noinline ecb_cold
474reqq_init (etp_reqq *q)
475{
476 int pri;
477
478 for (pri = 0; pri < ETP_NUM_PRI; ++pri)
479 q->qs[pri] = q->qe[pri] = 0;
480
481 q->size = 0;
482}
483
484static int ecb_noinline
294static int reqq_push (etp_reqq *q, ETP_REQ *req) 485reqq_push (etp_reqq *q, ETP_REQ *req)
295{ 486{
296 int pri = req->pri; 487 int pri = req->pri;
297 req->next = 0; 488 req->next = 0;
298 489
299 if (q->qe[pri]) 490 if (q->qe[pri])
… …
305 q->qe[pri] = q->qs[pri] = req; 496 q->qe[pri] = q->qs[pri] = req;
306 497
307 return q->size++; 498 return q->size++;
308} 499}
309 500
501static ETP_REQ * ecb_noinline
310static ETP_REQ *reqq_shift (etp_reqq *q) 502reqq_shift (etp_reqq *q)
311{ 503{
312 int pri; 504 int pri;
313 505
314 if (!q->size) 506 if (!q->size)
315 return 0; 507 return 0;
… …
330 } 522 }
331 523
332 abort (); 524 abort ();
333} 525}
334 526
335static void etp_atfork_prepare (void) 527static int ecb_cold
528etp_init (void (*want_poll)(void), void (*done_poll)(void))
336{ 529{
337 X_LOCK (wrklock); 530 X_MUTEX_CREATE (wrklock);
338 X_LOCK (reqlock); 531 X_MUTEX_CREATE (reslock);
339 X_LOCK (reslock); 532 X_MUTEX_CREATE (reqlock);
340#if !HAVE_PREADWRITE 533 X_COND_CREATE (reqwait);
341 X_LOCK (preadwritelock);
342#endif
343}
344 534
345static void etp_atfork_parent (void) 535 reqq_init (&req_queue);
346{ 536 reqq_init (&res_queue);
347#if !HAVE_PREADWRITE
348 X_UNLOCK (preadwritelock);
349#endif
350 X_UNLOCK (reslock);
351 X_UNLOCK (reqlock);
352 X_UNLOCK (wrklock);
353}
354 537
355static void etp_atfork_child (void) 538 wrk_first.next =
356{ 539 wrk_first.prev = &wrk_first;
357 ETP_REQ *prv;
358
359 while ((prv = reqq_shift (&req_queue)))
360 ETP_DESTROY (prv);
361
362 while ((prv = reqq_shift (&res_queue)))
363 ETP_DESTROY (prv);
364
365 while (wrk_first.next != &wrk_first)
366 {
367 etp_worker *wrk = wrk_first.next;
368
369 if (wrk->req)
370 ETP_DESTROY (wrk->req);
371
372 etp_worker_clear (wrk);
373 etp_worker_free (wrk);
374 }
375 540
376 started = 0; 541 started = 0;
377 idle = 0; 542 idle = 0;
378 nreqs = 0; 543 nreqs = 0;
379 nready = 0; 544 nready = 0;
380 npending = 0; 545 npending = 0;
381 546
382 etp_atfork_parent ();
383}
384
385static void
386etp_once_init (void)
387{
388 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
389}
390
391static int
392etp_init (void (*want_poll)(void), void (*done_poll)(void))
393{
394 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
395
396 pthread_once (&doinit, etp_once_init);
397
398 want_poll_cb = want_poll; 547 want_poll_cb = want_poll;
399 done_poll_cb = done_poll; 548 done_poll_cb = done_poll;
400 549
401 return 0; 550 return 0;
402} 551}
403 552
404X_THREAD_PROC (etp_proc); 553X_THREAD_PROC (etp_proc);
405 554
555static void ecb_cold
406static void etp_start_thread (void) 556etp_start_thread (void)
407{ 557{
408 etp_worker *wrk = calloc (1, sizeof (etp_worker)); 558 etp_worker *wrk = calloc (1, sizeof (etp_worker));
409 559
410 /*TODO*/ 560 /*TODO*/
411 assert (("unable to allocate worker thread data", wrk)); 561 assert (("unable to allocate worker thread data", wrk));
… …
424 free (wrk); 574 free (wrk);
425 575
426 X_UNLOCK (wrklock); 576 X_UNLOCK (wrklock);
427} 577}
428 578
579static void
429static void etp_maybe_start_thread (void) 580etp_maybe_start_thread (void)
430{ 581{
431 if (expect_true (etp_nthreads () >= wanted)) 582 if (ecb_expect_true (etp_nthreads () >= wanted))
432 return; 583 return;
433 584
434 /* todo: maybe use idle here, but might be less exact */ 585 /* todo: maybe use idle here, but might be less exact */
435 if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ())) 586 if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
436 return; 587 return;
437 588
438 etp_start_thread (); 589 etp_start_thread ();
439} 590}
440 591
592static void ecb_cold
441static void etp_end_thread (void) 593etp_end_thread (void)
442{ 594{
443 eio_req *req = calloc (1, sizeof (eio_req)); 595 eio_req *req = calloc (1, sizeof (eio_req)); /* will be freed by worker */
444 596
445 req->type = -1; 597 req->type = -1;
446 req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 598 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
447 599
448 X_LOCK (reqlock); 600 X_LOCK (reqlock);
… …
453 X_LOCK (wrklock); 605 X_LOCK (wrklock);
454 --started; 606 --started;
455 X_UNLOCK (wrklock); 607 X_UNLOCK (wrklock);
456} 608}
457 609
458static int etp_poll (void) 610static int
611etp_poll (void)
459{ 612{
460 unsigned int maxreqs; 613 unsigned int maxreqs;
461 unsigned int maxtime; 614 unsigned int maxtime;
462 struct timeval tv_start, tv_now; 615 struct timeval tv_start, tv_now;
463 616
… …
493 646
494 X_LOCK (reqlock); 647 X_LOCK (reqlock);
495 --nreqs; 648 --nreqs;
496 X_UNLOCK (reqlock); 649 X_UNLOCK (reqlock);
497 650
498 if (expect_false (req->type == EIO_GROUP && req->size)) 651 if (ecb_expect_false (req->type == EIO_GROUP && req->size))
499 { 652 {
500 req->int1 = 1; /* mark request as delayed */ 653 req->int1 = 1; /* mark request as delayed */
501 continue; 654 continue;
502 } 655 }
503 else 656 else
504 { 657 {
505 int res = ETP_FINISH (req); 658 int res = ETP_FINISH (req);
506 if (expect_false (res)) 659 if (ecb_expect_false (res))
507 return res; 660 return res;
508 } 661 }
509 662
510 if (expect_false (maxreqs && !--maxreqs)) 663 if (ecb_expect_false (maxreqs && !--maxreqs))
511 break; 664 break;
512 665
513 if (maxtime) 666 if (maxtime)
514 { 667 {
515 gettimeofday (&tv_now, 0); 668 gettimeofday (&tv_now, 0);
… …
521 674
522 errno = EAGAIN; 675 errno = EAGAIN;
523 return -1; 676 return -1;
524} 677}
525 678
679static void
526static void etp_cancel (ETP_REQ *req) 680etp_cancel (ETP_REQ *req)
527{ 681{
528 X_LOCK (wrklock); 682 req->cancelled = 1;
529 req->flags |= EIO_FLAG_CANCELLED;
530 X_UNLOCK (wrklock);
531 683
532 eio_grp_cancel (req); 684 eio_grp_cancel (req);
533} 685}
534 686
687static void
535static void etp_submit (ETP_REQ *req) 688etp_submit (ETP_REQ *req)
536{ 689{
537 req->pri -= ETP_PRI_MIN; 690 req->pri -= ETP_PRI_MIN;
538 691
539 if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN; 692 if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
540 if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 693 if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
541 694
542 if (expect_false (req->type == EIO_GROUP)) 695 if (ecb_expect_false (req->type == EIO_GROUP))
543 { 696 {
544 /* I hope this is worth it :/ */ 697 /* I hope this is worth it :/ */
545 X_LOCK (reqlock); 698 X_LOCK (reqlock);
546 ++nreqs; 699 ++nreqs;
547 X_UNLOCK (reqlock); 700 X_UNLOCK (reqlock);
… …
566 719
567 etp_maybe_start_thread (); 720 etp_maybe_start_thread ();
568 } 721 }
569} 722}
570 723
724static void ecb_cold
571static void etp_set_max_poll_time (double nseconds) 725etp_set_max_poll_time (double nseconds)
572{ 726{
573 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 727 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
574 max_poll_time = nseconds; 728 max_poll_time = nseconds * EIO_TICKS;
575 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 729 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
576} 730}
577 731
732static void ecb_cold
578static void etp_set_max_poll_reqs (unsigned int maxreqs) 733etp_set_max_poll_reqs (unsigned int maxreqs)
579{ 734{
580 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 735 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
581 max_poll_reqs = maxreqs; 736 max_poll_reqs = maxreqs;
582 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 737 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
583} 738}
584 739
740static void ecb_cold
585static void etp_set_max_idle (unsigned int nthreads) 741etp_set_max_idle (unsigned int nthreads)
586{ 742{
587 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 743 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
588 max_idle = nthreads <= 0 ? 1 : nthreads; 744 max_idle = nthreads;
589 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 745 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
590} 746}
591 747
748static void ecb_cold
749etp_set_idle_timeout (unsigned int seconds)
750{
751 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
752 idle_timeout = seconds;
753 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
754}
755
756static void ecb_cold
592static void etp_set_min_parallel (unsigned int nthreads) 757etp_set_min_parallel (unsigned int nthreads)
593{ 758{
594 if (wanted < nthreads) 759 if (wanted < nthreads)
595 wanted = nthreads; 760 wanted = nthreads;
596} 761}
597 762
763static void ecb_cold
598static void etp_set_max_parallel (unsigned int nthreads) 764etp_set_max_parallel (unsigned int nthreads)
599{ 765{
600 if (wanted > nthreads) 766 if (wanted > nthreads)
601 wanted = nthreads; 767 wanted = nthreads;
602 768
603 while (started > wanted) 769 while (started > wanted)
604 etp_end_thread (); 770 etp_end_thread ();
605} 771}
606 772
607/*****************************************************************************/ 773/*****************************************************************************/
608 774
775static void
609static void grp_try_feed (eio_req *grp) 776grp_try_feed (eio_req *grp)
610{ 777{
611 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 778 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
612 { 779 {
613 grp->flags &= ~EIO_FLAG_GROUPADD; 780 grp->flags &= ~EIO_FLAG_GROUPADD;
614 781
… …
621 break; 788 break;
622 } 789 }
623 } 790 }
624} 791}
625 792
793static int
626static int grp_dec (eio_req *grp) 794grp_dec (eio_req *grp)
627{ 795{
628 --grp->size; 796 --grp->size;
629 797
630 /* call feeder, if applicable */ 798 /* call feeder, if applicable */
631 grp_try_feed (grp); 799 grp_try_feed (grp);
… …
635 return eio_finish (grp); 803 return eio_finish (grp);
636 else 804 else
637 return 0; 805 return 0;
638} 806}
639 807
808static void
640void eio_destroy (eio_req *req) 809eio_destroy (eio_req *req)
641{ 810{
642 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); 811 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
643 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); 812 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
644 813
645 EIO_DESTROY (req); 814 EIO_DESTROY (req);
646} 815}
647 816
817static int
648static int eio_finish (eio_req *req) 818eio_finish (eio_req *req)
649{ 819{
650 int res = EIO_FINISH (req); 820 int res = EIO_FINISH (req);
651 821
652 if (req->grp) 822 if (req->grp)
653 { 823 {
… …
661 if (grp->grp_first == req) 831 if (grp->grp_first == req)
662 grp->grp_first = req->grp_next; 832 grp->grp_first = req->grp_next;
663 833
664 res2 = grp_dec (grp); 834 res2 = grp_dec (grp);
665 835
666 if (!res && res2) 836 if (!res)
667 res = res2; 837 res = res2;
668 } 838 }
669 839
670 eio_destroy (req); 840 eio_destroy (req);
671 841
672 return res; 842 return res;
673} 843}
674 844
845void
675void eio_grp_cancel (eio_req *grp) 846eio_grp_cancel (eio_req *grp)
676{ 847{
677 for (grp = grp->grp_first; grp; grp = grp->grp_next) 848 for (grp = grp->grp_first; grp; grp = grp->grp_next)
678 eio_cancel (grp); 849 eio_cancel (grp);
679} 850}
680 851
852void
681void eio_cancel (eio_req *req) 853eio_cancel (eio_req *req)
682{ 854{
683 etp_cancel (req); 855 etp_cancel (req);
684} 856}
685 857
858void
686void eio_submit (eio_req *req) 859eio_submit (eio_req *req)
687{ 860{
688 etp_submit (req); 861 etp_submit (req);
689} 862}
690 863
691unsigned int eio_nreqs (void) 864unsigned int
865eio_nreqs (void)
692{ 866{
693 return etp_nreqs (); 867 return etp_nreqs ();
694} 868}
695 869
696unsigned int eio_nready (void) 870unsigned int
871eio_nready (void)
697{ 872{
698 return etp_nready (); 873 return etp_nready ();
699} 874}
700 875
701unsigned int eio_npending (void) 876unsigned int
877eio_npending (void)
702{ 878{
703 return etp_npending (); 879 return etp_npending ();
704} 880}
705 881
706unsigned int eio_nthreads (void) 882unsigned int ecb_cold
883eio_nthreads (void)
707{ 884{
708 return etp_nthreads (); 885 return etp_nthreads ();
709} 886}
710 887
888void ecb_cold
711void eio_set_max_poll_time (double nseconds) 889eio_set_max_poll_time (double nseconds)
712{ 890{
713 etp_set_max_poll_time (nseconds); 891 etp_set_max_poll_time (nseconds);
714} 892}
715 893
894void ecb_cold
716void eio_set_max_poll_reqs (unsigned int maxreqs) 895eio_set_max_poll_reqs (unsigned int maxreqs)
717{ 896{
718 etp_set_max_poll_reqs (maxreqs); 897 etp_set_max_poll_reqs (maxreqs);
719} 898}
720 899
900void ecb_cold
721void eio_set_max_idle (unsigned int nthreads) 901eio_set_max_idle (unsigned int nthreads)
722{ 902{
723 etp_set_max_idle (nthreads); 903 etp_set_max_idle (nthreads);
724} 904}
725 905
906void ecb_cold
907eio_set_idle_timeout (unsigned int seconds)
908{
909 etp_set_idle_timeout (seconds);
910}
911
912void ecb_cold
726void eio_set_min_parallel (unsigned int nthreads) 913eio_set_min_parallel (unsigned int nthreads)
727{ 914{
728 etp_set_min_parallel (nthreads); 915 etp_set_min_parallel (nthreads);
729} 916}
730 917
918void ecb_cold
731void eio_set_max_parallel (unsigned int nthreads) 919eio_set_max_parallel (unsigned int nthreads)
732{ 920{
733 etp_set_max_parallel (nthreads); 921 etp_set_max_parallel (nthreads);
734} 922}
735 923
736int eio_poll (void) 924int eio_poll (void)
… …
745# undef pread 933# undef pread
746# undef pwrite 934# undef pwrite
747# define pread eio__pread 935# define pread eio__pread
748# define pwrite eio__pwrite 936# define pwrite eio__pwrite
749 937
750static ssize_t 938static eio_ssize_t
751eio__pread (int fd, void *buf, size_t count, off_t offset) 939eio__pread (int fd, void *buf, size_t count, off_t offset)
752{ 940{
753 ssize_t res; 941 eio_ssize_t res;
754 off_t ooffset; 942 off_t ooffset;
755 943
756 X_LOCK (preadwritelock); 944 X_LOCK (preadwritelock);
757 ooffset = lseek (fd, 0, SEEK_CUR); 945 ooffset = lseek (fd, 0, SEEK_CUR);
758 lseek (fd, offset, SEEK_SET); 946 lseek (fd, offset, SEEK_SET);
… …
761 X_UNLOCK (preadwritelock); 949 X_UNLOCK (preadwritelock);
762 950
763 return res; 951 return res;
764} 952}
765 953
766static ssize_t 954static eio_ssize_t
767eio__pwrite (int fd, void *buf, size_t count, off_t offset) 955eio__pwrite (int fd, void *buf, size_t count, off_t offset)
768{ 956{
769 ssize_t res; 957 eio_ssize_t res;
770 off_t ooffset; 958 off_t ooffset;
771 959
772 X_LOCK (preadwritelock); 960 X_LOCK (preadwritelock);
773 ooffset = lseek (fd, 0, SEEK_CUR); 961 ooffset = lseek (fd, 0, SEEK_CUR);
774 lseek (fd, offset, SEEK_SET); 962 lseek (fd, offset, SEEK_SET);
… …
778 966
779 return res; 967 return res;
780} 968}
781#endif 969#endif
782 970
783#ifndef HAVE_FUTIMES 971#ifndef HAVE_UTIMES
784 972
785# undef utimes 973# undef utimes
786# undef futimes
787# define utimes(path,times) eio__utimes (path, times) 974# define utimes(path,times) eio__utimes (path, times)
788# define futimes(fd,times) eio__futimes (fd, times)
789 975
790static int 976static int
791eio__utimes (const char *filename, const struct timeval times[2]) 977eio__utimes (const char *filename, const struct timeval times[2])
792{ 978{
793 if (times) 979 if (times)
… …
801 } 987 }
802 else 988 else
803 return utime (filename, 0); 989 return utime (filename, 0);
804} 990}
805 991
992#endif
993
994#ifndef HAVE_FUTIMES
995
996# undef futimes
997# define futimes(fd,times) eio__futimes (fd, times)
998
999static int
806static int eio__futimes (int fd, const struct timeval tv[2]) 1000eio__futimes (int fd, const struct timeval tv[2])
807{ 1001{
808 errno = ENOSYS; 1002 errno = ENOSYS;
809 return -1; 1003 return -1;
810} 1004}
811 1005
… …
814#if !HAVE_FDATASYNC 1008#if !HAVE_FDATASYNC
815# undef fdatasync 1009# undef fdatasync
816# define fdatasync(fd) fsync (fd) 1010# define fdatasync(fd) fsync (fd)
817#endif 1011#endif
818 1012
1013static int
1014eio__syncfs (int fd)
1015{
1016 int res;
1017
1018#if HAVE_SYS_SYNCFS
1019 res = (int)syscall (__NR_syncfs, (int)(fd));
1020#else
1021 res = -1;
1022 errno = ENOSYS;
1023#endif
1024
1025 if (res < 0 && errno == ENOSYS && fd >= 0)
1026 sync ();
1027
1028 return res;
1029}
1030
819/* sync_file_range always needs emulation */ 1031/* sync_file_range always needs emulation */
820int 1032static int
821eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags) 1033eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
822{ 1034{
823#if HAVE_SYNC_FILE_RANGE 1035#if HAVE_SYNC_FILE_RANGE
824 int res; 1036 int res;
825 1037
… …
838 if (!res || errno != ENOSYS) 1050 if (!res || errno != ENOSYS)
839 return res; 1051 return res;
840#endif 1052#endif
841 1053
842 /* even though we could play tricks with the flags, it's better to always 1054 /* even though we could play tricks with the flags, it's better to always
843 * call fdatasync, as thta matches the expectation of it's users best */ 1055 * call fdatasync, as that matches the expectation of its users best */
844 return fdatasync (fd); 1056 return fdatasync (fd);
1057}
1058
1059static int
1060eio__fallocate (int fd, int mode, off_t offset, size_t len)
1061{
1062#if HAVE_FALLOCATE
1063 return fallocate (fd, mode, offset, len);
1064#else
1065 errno = ENOSYS;
1066 return -1;
1067#endif
845} 1068}
846 1069
847#if !HAVE_READAHEAD 1070#if !HAVE_READAHEAD
848# undef readahead 1071# undef readahead
849# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 1072# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
850 1073
851static ssize_t 1074static eio_ssize_t
852eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) 1075eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
853{ 1076{
854 size_t todo = count; 1077 size_t todo = count;
855 dBUF; 1078 dBUF;
856 1079
… …
861 pread (fd, eio_buf, len, offset); 1084 pread (fd, eio_buf, len, offset);
862 offset += len; 1085 offset += len;
863 todo -= len; 1086 todo -= len;
864 } 1087 }
865 1088
1089 FUBd;
1090
866 errno = 0; 1091 errno = 0;
867 return count; 1092 return count;
868} 1093}
869 1094
870#endif 1095#endif
871 1096
872/* sendfile always needs emulation */ 1097/* sendfile always needs emulation */
873static ssize_t 1098static eio_ssize_t
874eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) 1099eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
875{ 1100{
1101 eio_ssize_t written = 0;
876 ssize_t res; 1102 eio_ssize_t res;
877 1103
878 if (!count) 1104 if (!count)
879 return 0; 1105 return 0;
880 1106
1107 for (;;)
1108 {
1109#ifdef __APPLE__
1110# undef HAVE_SENDFILE /* broken, as everything on os x */
1111#endif
881#if HAVE_SENDFILE 1112#if HAVE_SENDFILE
882# if __linux 1113# if __linux
1114 off_t soffset = offset;
883 res = sendfile (ofd, ifd, &offset, count); 1115 res = sendfile (ofd, ifd, &soffset, count);
884 1116
885# elif __freebsd 1117# elif __FreeBSD__
886 /* 1118 /*
887 * Of course, the freebsd sendfile is a dire hack with no thoughts 1119 * Of course, the freebsd sendfile is a dire hack with no thoughts
888 * wasted on making it similar to other I/O functions. 1120 * wasted on making it similar to other I/O functions.
889 */ 1121 */
890 {
891 off_t sbytes; 1122 off_t sbytes;
892 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 1123 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
893 1124
894 if (res < 0 && sbytes) 1125 #if 0 /* according to the manpage, this is correct, but broken behaviour */
895 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 1126 /* freebsd' sendfile will return 0 on success */
1127 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
1128 /* not on e.g. EIO or EPIPE - sounds broken */
1129 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
896 res = sbytes; 1130 res = sbytes;
897 } 1131 #endif
1132
1133 /* according to source inspection, this is correct, and useful behaviour */
1134 if (sbytes)
1135 res = sbytes;
1136
1137# elif defined (__APPLE__)
1138 off_t sbytes = count;
1139 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
1140
1141 /* according to the manpage, sbytes is always valid */
1142 if (sbytes)
1143 res = sbytes;
898 1144
899# elif __hpux 1145# elif __hpux
900 res = sendfile (ofd, ifd, offset, count, 0, 0); 1146 res = sendfile (ofd, ifd, offset, count, 0, 0);
901 1147
902# elif __solaris 1148# elif __solaris
903 {
904 struct sendfilevec vec; 1149 struct sendfilevec vec;
905 size_t sbytes; 1150 size_t sbytes;
906 1151
907 vec.sfv_fd = ifd; 1152 vec.sfv_fd = ifd;
908 vec.sfv_flag = 0; 1153 vec.sfv_flag = 0;
909 vec.sfv_off = offset; 1154 vec.sfv_off = offset;
910 vec.sfv_len = count; 1155 vec.sfv_len = count;
911 1156
912 res = sendfilev (ofd, &vec, 1, &sbytes); 1157 res = sendfilev (ofd, &vec, 1, &sbytes);
913 1158
914 if (res < 0 && sbytes) 1159 if (res < 0 && sbytes)
915 res = sbytes; 1160 res = sbytes;
916 }
917 1161
918# endif 1162# endif
1163
1164#elif defined (_WIN32) && 0
1165 /* does not work, just for documentation of what would need to be done */
1166 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
1167 /* libeio guarantees that the file offset does not change, and windows */
1168 /* has no way to get an independent handle to the same file description */
1169 HANDLE h = TO_SOCKET (ifd);
1170 SetFilePointer (h, offset, 0, FILE_BEGIN);
1171 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1172
919#else 1173#else
920 res = -1; 1174 res = -1;
921 errno = ENOSYS; 1175 errno = ENOSYS;
922#endif 1176#endif
923 1177
1178 /* we assume sendfile can copy at least 128mb in one go */
1179 if (res <= 128 * 1024 * 1024)
1180 {
1181 if (res > 0)
1182 written += res;
1183
1184 if (written)
1185 return written;
1186
1187 break;
1188 }
1189 else
1190 {
1191 /* if we requested more, then probably the kernel was lazy */
1192 written += res;
1193 offset += res;
1194 count -= res;
1195
1196 if (!count)
1197 return written;
1198 }
1199 }
1200
924 if (res < 0 1201 if (res < 0
925 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1202 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1203 /* BSDs */
1204#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1205 || errno == ENOTSUP
1206#endif
1207#ifdef EOPNOTSUPP /* windows */
1208 || errno == EOPNOTSUPP /* BSDs */
1209#endif
926#if __solaris 1210#if __solaris
927 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1211 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
928#endif 1212#endif
929 ) 1213 )
930 ) 1214 )
… …
934 1218
935 res = 0; 1219 res = 0;
936 1220
937 while (count) 1221 while (count)
938 { 1222 {
939 ssize_t cnt; 1223 eio_ssize_t cnt;
940 1224
941 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 1225 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
942 1226
943 if (cnt <= 0) 1227 if (cnt <= 0)
944 { 1228 {
… …
956 1240
957 offset += cnt; 1241 offset += cnt;
958 res += cnt; 1242 res += cnt;
959 count -= cnt; 1243 count -= cnt;
960 } 1244 }
1245
1246 FUBd;
961 } 1247 }
962 1248
963 return res; 1249 return res;
1250}
1251
1252#ifdef PAGESIZE
1253# define eio_pagesize() PAGESIZE
1254#else
1255static intptr_t
1256eio_pagesize (void)
1257{
1258 static intptr_t page;
1259
1260 if (!page)
1261 page = sysconf (_SC_PAGESIZE);
1262
1263 return page;
1264}
1265#endif
1266
1267static void
1268eio_page_align (void **addr, size_t *length)
1269{
1270 intptr_t mask = eio_pagesize () - 1;
1271
1272 /* round down addr */
1273 intptr_t adj = mask & (intptr_t)*addr;
1274
1275 *addr = (void *)((intptr_t)*addr - adj);
1276 *length += adj;
1277
1278 /* round up length */
1279 *length = (*length + mask) & ~mask;
1280}
1281
1282#if !_POSIX_MEMLOCK
1283# define eio__mlockall(a) EIO_ENOSYS ()
1284#else
1285
1286static int
1287eio__mlockall (int flags)
1288{
1289 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1290 extern int mallopt (int, int);
1291 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1292 #endif
1293
1294 if (EIO_MCL_CURRENT != MCL_CURRENT
1295 || EIO_MCL_FUTURE != MCL_FUTURE)
1296 {
1297 flags = 0
1298 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1299 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1300 }
1301
1302 return mlockall (flags);
1303}
1304#endif
1305
1306#if !_POSIX_MEMLOCK_RANGE
1307# define eio__mlock(a,b) EIO_ENOSYS ()
1308#else
1309
1310static int
1311eio__mlock (void *addr, size_t length)
1312{
1313 eio_page_align (&addr, &length);
1314
1315 return mlock (addr, length);
1316}
1317
1318#endif
1319
1320#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1321# define eio__msync(a,b,c) EIO_ENOSYS ()
1322#else
1323
1324static int
1325eio__msync (void *mem, size_t len, int flags)
1326{
1327 eio_page_align (&mem, &len);
1328
1329 if (EIO_MS_ASYNC != MS_SYNC
1330 || EIO_MS_INVALIDATE != MS_INVALIDATE
1331 || EIO_MS_SYNC != MS_SYNC)
1332 {
1333 flags = 0
1334 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1335 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1336 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1337 }
1338
1339 return msync (mem, len, flags);
1340}
1341
1342#endif
1343
1344static int
1345eio__mtouch (eio_req *req)
1346{
1347 void *mem = req->ptr2;
1348 size_t len = req->size;
1349 int flags = req->int1;
1350
1351 eio_page_align (&mem, &len);
1352
1353 {
1354 intptr_t addr = (intptr_t)mem;
1355 intptr_t end = addr + len;
1356 intptr_t page = eio_pagesize ();
1357
1358 if (addr < end)
1359 if (flags & EIO_MT_MODIFY) /* modify */
1360 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1361 else
1362 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1363 }
1364
1365 return 0;
1366}
1367
1368/*****************************************************************************/
1369/* requests implemented outside eio_execute, because they are so large */
1370
1371/* copies some absolute path to tmpbuf */
1372static char *
1373eio__getwd (struct tmpbuf *tmpbuf, eio_wd wd)
1374{
1375 if (wd == EIO_CWD)
1376 return getcwd (tmpbuf->ptr, PATH_MAX);
1377
1378#if HAVE_AT
1379 abort (); /*TODO*/
1380#else
1381 strcpy (tmpbuf->ptr, wd);
1382#endif
1383 return tmpbuf->ptr;
1384}
1385
1386/* result will always end up in tmpbuf, there is always space for adding a 0-byte */
1387static int
1388eio__realpath (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
1389{
1390 const char *rel = path;
1391 char *res;
1392 char *tmp1, *tmp2;
1393#if SYMLOOP_MAX > 32
1394 int symlinks = SYMLOOP_MAX;
1395#else
1396 int symlinks = 32;
1397#endif
1398
1399 /*D*/ /*TODO: wd ignored */
1400
1401 errno = EINVAL;
1402 if (!rel)
1403 return -1;
1404
1405 errno = ENOENT;
1406 if (!*rel)
1407 return -1;
1408
1409 res = tmpbuf_get (tmpbuf, PATH_MAX * 3);
1410 tmp1 = res + PATH_MAX;
1411 tmp2 = tmp1 + PATH_MAX;
1412
1413#if 0 /* disabled, the musl way to do things is just too racy */
1414#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1415 /* on linux we may be able to ask the kernel */
1416 {
1417 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1418
1419 if (fd >= 0)
1420 {
1421 sprintf (tmp1, "/proc/self/fd/%d", fd);
1422 req->result = readlink (tmp1, res, PATH_MAX);
1423 close (fd);
1424
1425 /* here we should probably stat the open file and the disk file, to make sure they still match */
1426
1427 if (req->result > 0)
1428 goto done;
1429 }
1430 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1431 return;
1432 }
1433#endif
1434#endif
1435
1436 if (*rel != '/')
1437 {
1438 if (!eio__getwd (tmpbuf, wd))
1439 return -1;
1440
1441 if (res [1]) /* only use if not / */
1442 res += strlen (res);
1443 }
1444
1445 while (*rel)
1446 {
1447 eio_ssize_t len, linklen;
1448 const char *beg = rel;
1449
1450 while (*rel && *rel != '/')
1451 ++rel;
1452
1453 len = rel - beg;
1454
1455 if (!len) /* skip slashes */
1456 {
1457 ++rel;
1458 continue;
1459 }
1460
1461 if (beg [0] == '.')
1462 {
1463 if (len == 1)
1464 continue; /* . - nop */
1465
1466 if (beg [1] == '.' && len == 2)
1467 {
1468 /* .. - back up one component, if possible */
1469
1470 while (res != tmpbuf->ptr)
1471 if (*--res == '/')
1472 break;
1473
1474 continue;
1475 }
1476 }
1477
1478 errno = ENAMETOOLONG;
1479 if (res + 1 + len + 1 >= tmp1)
1480 return;
1481
1482 /* copy one component */
1483 *res = '/';
1484 memcpy (res + 1, beg, len);
1485
1486 /* zero-terminate, for readlink */
1487 res [len + 1] = 0;
1488
1489 /* now check if it's a symlink */
1490 linklen = readlink (tmpbuf->ptr, tmp1, PATH_MAX);
1491
1492 if (linklen < 0)
1493 {
1494 if (errno != EINVAL)
1495 return -1;
1496
1497 /* it's a normal directory. hopefully */
1498 res += len + 1;
1499 }
1500 else
1501 {
1502 /* yay, it was a symlink - build new path in tmp2 */
1503 int rellen = strlen (rel);
1504
1505 errno = ENAMETOOLONG;
1506 if (linklen + 1 + rellen >= PATH_MAX)
1507 return -1;
1508
1509 errno = ELOOP;
1510 if (!--symlinks)
1511 return -1;
1512
1513 if (*tmp1 == '/')
1514 res = tmpbuf->ptr; /* symlink resolves to an absolute path */
1515
1516 /* we need to be careful, as rel might point into tmp2 already */
1517 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1518 tmp2 [linklen] = '/';
1519 memcpy (tmp2, tmp1, linklen);
1520
1521 rel = tmp2;
1522 }
1523 }
1524
1525 /* special case for the lone root path */
1526 if (res == tmpbuf->ptr)
1527 *res++ = '/';
1528
1529 return res - (char *)tmpbuf->ptr;
1530}
1531
1532static signed char
1533eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1534{
1535 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1536 : a->inode < b->inode ? -1
1537 : a->inode > b->inode ? 1
1538 : 0;
1539}
1540
1541#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1542
1543#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1544#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1545
1546static void
1547eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1548{
1549 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1550 unsigned char *bit = bits;
1551
1552 assert (CHAR_BIT == 8);
1553 assert (sizeof (eio_dirent) * 8 < 256);
1554 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1555 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1556
1557 if (size <= EIO_SORT_FAST)
1558 return;
1559
1560 /* first prepare an array of bits to test in our radix sort */
1561 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1562 /* inode_bits must contain all inodes ORed together */
1563 /* which is used to skip bits that are 0 everywhere, which is very common */
1564 {
1565 eio_ino_t endianness;
1566 int i, j;
1567
1568 /* we store the byte offset of byte n into byte n of "endianness" */
1569 for (i = 0; i < sizeof (eio_ino_t); ++i)
1570 ((unsigned char *)&endianness)[i] = i;
1571
1572 *bit++ = 0;
1573
1574 for (i = 0; i < sizeof (eio_ino_t); ++i)
1575 {
1576 /* shifting off the byte offsets out of "endianness" */
1577 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1578 endianness >>= 8;
1579
1580 for (j = 0; j < 8; ++j)
1581 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1582 *bit++ = offs + j;
1583 }
1584
1585 for (j = 0; j < 8; ++j)
1586 if (score_bits & (1 << j))
1587 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1588 }
1589
1590 /* now actually do the sorting (a variant of MSD radix sort) */
1591 {
1592 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1593 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1594 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1595 int stk_idx = 0;
1596
1597 base_stk [stk_idx] = dents;
1598 end_stk [stk_idx] = dents + size;
1599 bit_stk [stk_idx] = bit - 1;
1600
1601 do
1602 {
1603 base = base_stk [stk_idx];
1604 end = end_stk [stk_idx];
1605 bit = bit_stk [stk_idx];
1606
1607 for (;;)
1608 {
1609 unsigned char O = *bit >> 3;
1610 unsigned char M = 1 << (*bit & 7);
1611
1612 eio_dirent *a = base;
1613 eio_dirent *b = end;
1614
1615 if (b - a < EIO_SORT_CUTOFF)
1616 break;
1617
1618 /* now bit-partition the array on the bit */
1619 /* this ugly asymmetric loop seems to perform much better than typical */
1620 /* partition algos found in the literature */
1621 do
1622 if (!(((unsigned char *)a)[O] & M))
1623 ++a;
1624 else if (!(((unsigned char *)--b)[O] & M))
1625 {
1626 eio_dirent tmp = *a; *a = *b; *b = tmp;
1627 ++a;
1628 }
1629 while (b > a);
1630
1631 /* next bit, or stop, if no bits left in this path */
1632 if (!*--bit)
1633 break;
1634
1635 base_stk [stk_idx] = a;
1636 end_stk [stk_idx] = end;
1637 bit_stk [stk_idx] = bit;
1638 ++stk_idx;
1639
1640 end = a;
1641 }
1642 }
1643 while (stk_idx--);
1644 }
1645}
1646
1647static void
1648eio_dent_insertion_sort (eio_dirent *dents, int size)
1649{
1650 /* first move the smallest element to the front, to act as a sentinel */
1651 {
1652 int i;
1653 eio_dirent *min = dents;
1654
1655 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1656 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1657 if (EIO_DENT_CMP (dents [i], <, *min))
1658 min = &dents [i];
1659
1660 /* swap elements 0 and j (minimum) */
1661 {
1662 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1663 }
1664 }
1665
1666 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1667 {
1668 eio_dirent *i, *j;
1669
1670 for (i = dents + 1; i < dents + size; ++i)
1671 {
1672 eio_dirent value = *i;
1673
1674 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1675 j [1] = j [0];
1676
1677 j [1] = value;
1678 }
1679 }
1680}
1681
1682static void
1683eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1684{
1685 if (size <= 1)
1686 return; /* our insertion sort relies on size > 0 */
1687
1688 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1689 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1690 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1691
1692 /* use an insertion sort at the end, or for small arrays, */
1693 /* as insertion sort is more efficient for small partitions */
1694 eio_dent_insertion_sort (dents, size);
964} 1695}
965 1696
966/* read a full directory */ 1697/* read a full directory */
967static void 1698static void
968eio__scandir (eio_req *req, etp_worker *self) 1699eio__scandir (eio_req *req)
969{ 1700{
1701 char *name, *names;
1702 int namesalloc = 4096 - sizeof (void *) * 4;
1703 int namesoffs = 0;
1704 int flags = req->int1;
1705 eio_dirent *dents = 0;
1706 int dentalloc = 128;
1707 int dentoffs = 0;
1708 eio_ino_t inode_bits = 0;
1709#ifdef _WIN32
1710 HANDLE dirp;
1711 WIN32_FIND_DATA entp;
1712#else
970 DIR *dirp; 1713 DIR *dirp;
971 EIO_STRUCT_DIRENT *entp; 1714 EIO_STRUCT_DIRENT *entp;
972 char *name, *names; 1715#endif
973 int memlen = 4096;
974 int memofs = 0;
975 int res = 0;
976 1716
977 X_LOCK (wrklock); 1717 req->result = -1;
978 /* the corresponding closedir is in ETP_WORKER_CLEAR */
979 self->dirp = dirp = opendir (req->ptr1);
980 req->flags |= EIO_FLAG_PTR2_FREE;
981 req->ptr2 = names = malloc (memlen);
982 X_UNLOCK (wrklock);
983 1718
984 if (dirp && names) 1719 if (!(flags & EIO_READDIR_DENTS))
985 for (;;) 1720 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1721
1722#ifdef _WIN32
1723 {
1724 int len = strlen ((const char *)req->ptr1);
1725 char *path = malloc (MAX_PATH);
1726 const char *fmt;
1727
1728 if (!len)
1729 fmt = "./*";
1730 else if (((const char *)req->ptr1)[len - 1] == '/' || ((const char *)req->ptr1)[len - 1] == '\\')
1731 fmt = "%s*";
1732 else
1733 fmt = "%s/*";
1734
1735 _snprintf (path, MAX_PATH, fmt, (const char *)req->ptr1);
1736 dirp = FindFirstFile (path, &entp);
1737 free (path);
1738
1739 if (dirp == INVALID_HANDLE_VALUE)
986 { 1740 {
1741 /* should steal _dosmaperr */
1742 switch (GetLastError ())
1743 {
1744 case ERROR_FILE_NOT_FOUND:
1745 req->result = 0;
1746 break;
1747
1748 case ERROR_INVALID_NAME:
1749 case ERROR_PATH_NOT_FOUND:
1750 case ERROR_NO_MORE_FILES:
1751 errno = ENOENT;
1752 break;
1753
1754 case ERROR_NOT_ENOUGH_MEMORY:
1755 errno = ENOMEM;
1756 break;
1757
1758 default:
1759 errno = EINVAL;
1760 break;
1761 }
1762
1763 return;
1764 }
1765 }
1766#else
1767 dirp = opendir (req->ptr1);
1768
1769 if (!dirp)
1770 return;
1771#endif
1772
1773 if (req->flags & EIO_FLAG_PTR1_FREE)
1774 free (req->ptr1);
1775
1776 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1777 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1778 req->ptr2 = names = malloc (namesalloc);
1779
1780 if (!names || (flags && !dents))
1781 return;
1782
1783 for (;;)
1784 {
1785 int done;
1786
1787#ifdef _WIN32
1788 done = !dirp;
1789#else
987 errno = 0; 1790 errno = 0;
988 entp = readdir (dirp); 1791 entp = readdir (dirp);
1792 done = !entp;
1793#endif
989 1794
1795 if (done)
1796 {
1797#ifndef _WIN32
1798 int old_errno = errno;
1799 closedir (dirp);
1800 errno = old_errno;
1801
990 if (!entp) 1802 if (errno)
1803 break;
1804#endif
1805
1806 /* sort etc. */
1807 req->int1 = flags;
1808 req->result = dentoffs;
1809
1810 if (flags & EIO_READDIR_STAT_ORDER)
1811 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1812 else if (flags & EIO_READDIR_DIRS_FIRST)
1813 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1814 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1815 else
1816 {
1817 /* in this case, all is known, and we just put dirs first and sort them */
1818 eio_dirent *oth = dents + dentoffs;
1819 eio_dirent *dir = dents;
1820
1821 /* now partition dirs to the front, and non-dirs to the back */
1822 /* by walking from both sides and swapping if necessary */
1823 while (oth > dir)
1824 {
1825 if (dir->type == EIO_DT_DIR)
1826 ++dir;
1827 else if ((--oth)->type == EIO_DT_DIR)
1828 {
1829 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1830
1831 ++dir;
1832 }
1833 }
1834
1835 /* now sort the dirs only (dirs all have the same score) */
1836 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1837 }
1838
991 break; 1839 break;
1840 }
992 1841
993 name = entp->d_name; 1842 /* now add the entry to our list(s) */
1843 name = D_NAME (entp);
994 1844
1845 /* skip . and .. entries */
995 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1846 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
996 { 1847 {
997 int len = strlen (name) + 1; 1848 int len = D_NAMLEN (entp) + 1;
998 1849
999 res++; 1850 while (ecb_expect_false (namesoffs + len > namesalloc))
1000
1001 while (memofs + len > memlen)
1002 { 1851 {
1003 memlen *= 2; 1852 namesalloc *= 2;
1004 X_LOCK (wrklock);
1005 req->ptr2 = names = realloc (names, memlen); 1853 req->ptr2 = names = realloc (names, namesalloc);
1006 X_UNLOCK (wrklock);
1007 1854
1008 if (!names) 1855 if (!names)
1856 break;
1857 }
1858
1859 memcpy (names + namesoffs, name, len);
1860
1861 if (dents)
1862 {
1863 struct eio_dirent *ent;
1864
1865 if (ecb_expect_false (dentoffs == dentalloc))
1866 {
1867 dentalloc *= 2;
1868 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1869
1870 if (!dents)
1009 break; 1871 break;
1010 } 1872 }
1011 1873
1012 memcpy (names + memofs, name, len); 1874 ent = dents + dentoffs;
1013 memofs += len; 1875
1876 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1877 ent->namelen = len - 1;
1878 ent->inode = D_INO (entp);
1879
1880 inode_bits |= ent->inode;
1881
1882 switch (D_TYPE (entp))
1883 {
1884 default:
1885 ent->type = EIO_DT_UNKNOWN;
1886 flags |= EIO_READDIR_FOUND_UNKNOWN;
1887 break;
1888
1889 #ifdef DT_FIFO
1890 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1891 #endif
1892 #ifdef DT_CHR
1893 case DT_CHR: ent->type = EIO_DT_CHR; break;
1894 #endif
1895 #ifdef DT_MPC
1896 case DT_MPC: ent->type = EIO_DT_MPC; break;
1897 #endif
1898 #ifdef DT_DIR
1899 case DT_DIR: ent->type = EIO_DT_DIR; break;
1900 #endif
1901 #ifdef DT_NAM
1902 case DT_NAM: ent->type = EIO_DT_NAM; break;
1903 #endif
1904 #ifdef DT_BLK
1905 case DT_BLK: ent->type = EIO_DT_BLK; break;
1906 #endif
1907 #ifdef DT_MPB
1908 case DT_MPB: ent->type = EIO_DT_MPB; break;
1909 #endif
1910 #ifdef DT_REG
1911 case DT_REG: ent->type = EIO_DT_REG; break;
1912 #endif
1913 #ifdef DT_NWK
1914 case DT_NWK: ent->type = EIO_DT_NWK; break;
1915 #endif
1916 #ifdef DT_CMP
1917 case DT_CMP: ent->type = EIO_DT_CMP; break;
1918 #endif
1919 #ifdef DT_LNK
1920 case DT_LNK: ent->type = EIO_DT_LNK; break;
1921 #endif
1922 #ifdef DT_SOCK
1923 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1924 #endif
1925 #ifdef DT_DOOR
1926 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1927 #endif
1928 #ifdef DT_WHT
1929 case DT_WHT: ent->type = EIO_DT_WHT; break;
1930 #endif
1931 }
1932
1933 ent->score = 7;
1934
1935 if (flags & EIO_READDIR_DIRS_FIRST)
1936 {
1937 if (ent->type == EIO_DT_UNKNOWN)
1938 {
1939 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1940 ent->score = 1;
1941 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1942 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1943 }
1944 else if (ent->type == EIO_DT_DIR)
1945 ent->score = 0;
1946 }
1014 } 1947 }
1948
1949 namesoffs += len;
1950 ++dentoffs;
1015 } 1951 }
1016 1952
1017 if (errno) 1953 if (EIO_CANCELLED (req))
1018 res = -1; 1954 {
1019 1955 errno = ECANCELED;
1020 req->result = res; 1956 break;
1021} 1957 }
1022 1958
1023#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) 1959#ifdef _WIN32
1024# undef msync 1960 if (!FindNextFile (dirp, &entp))
1025# define msync(a,b,c) ((errno = ENOSYS), -1) 1961 {
1962 FindClose (dirp);
1963 dirp = 0;
1964 }
1026#endif 1965#endif
1966 }
1967}
1027 1968
1028int 1969/*****************************************************************************/
1029eio__mtouch (void *mem, size_t len, int flags) 1970/* working directory stuff */
1971
1972#if HAVE_AT
1973
1974#define WD2FD(wd) ((wd) ? ((int)(long)(wd)) - 1 : AT_FDCWD)
1975
1976#ifndef O_SEARCH
1977# define O_SEARCH O_RDONLY
1978#endif
1979
1980eio_wd
1981eio_wd_open_sync (eio_wd wd, const char *path)
1030{ 1982{
1031 intptr_t addr = (intptr_t)mem; 1983 int fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1032 intptr_t end = addr + len; 1984
1033#ifdef PAGESIZE 1985 return fd >= 0 ? (eio_wd)(long)(fd + 1) : EIO_INVALID_WD;
1034 const intptr_t page = PAGESIZE; 1986}
1987
1988static eio_wd
1989eio__wd_open_sync (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
1990{
1991 return eio_wd_open_sync (wd, path);
1992}
1993
1994void
1995eio_wd_close_sync (eio_wd wd)
1996{
1997 int fd = WD2FD (wd);
1998
1999 if (fd >= 0)
2000 close (fd);
2001}
2002
2003static int
2004eio__truncateat (int dirfd, const char *path, off_t length)
2005{
2006 int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
2007 int res;
2008
2009 if (fd < 0)
2010 return fd;
2011
2012 res = ftruncate (fd, length);
2013 close (fd);
2014 return res;
2015}
2016
2017static int
2018eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
2019{
2020 int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
2021 int res;
2022
2023 if (fd < 0)
2024 return fd;
2025
2026 res = fstatvfs (fd, buf);
2027 close (fd);
2028 return res;
2029
2030}
2031
1035#else 2032#else
1036 static intptr_t page;
1037 2033
1038 if (!page) 2034/* on legacy systems, we represent the working directories simply by their path strings */
1039 page = sysconf (_SC_PAGESIZE);
1040#endif
1041 2035
1042 addr &= ~(page - 1); /* assume page size is always a power of two */ 2036static const char *
2037wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
2038{
2039 if (!wd || *path == '/')
2040 return path;
1043 2041
1044 if (addr < end) 2042 {
1045 if (flags) /* modify */ 2043 int l1 = strlen ((const char *)wd);
1046 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len); 2044 int l2 = strlen (path);
2045
2046 char *res = tmpbuf_get (tmpbuf, l1 + l2 + 2);
2047
2048 memcpy (res, wd, l1);
2049 res [l1] = '/';
2050 memcpy (res + l1 + 1, path, l2 + 1);
2051
2052 return res;
2053 }
2054}
2055
2056static eio_wd
2057eio__wd_open_sync (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
2058{
2059 if (*path == '/') /* absolute paths ignore wd */
2060 path = strdup (path);
2061 else if (path [0] == '.' && !path [1]) /* special case '.', as it is common */
2062 return wd;
1047 else 2063 else
1048 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len); 2064 {
2065 int len = eio__realpath (tmpbuf, wd, path);
1049 2066
2067 path = EIO_INVALID_WD;
2068
2069 if (len >= 0)
2070 {
2071 ((char *)tmpbuf->ptr)[len] = 0;
2072 path = strdup (tmpbuf->ptr);
2073 }
2074 }
2075
2076 if (!path)
2077 path = EIO_INVALID_WD;
2078
2079 return (eio_wd)path;
2080}
2081
2082eio_wd
2083eio_wd_open_sync (eio_wd wd, const char *path)
2084{
2085 struct tmpbuf tmpbuf = { 0 };
2086 wd = eio__wd_open_sync (&tmpbuf, wd, path);
2087 free (tmpbuf.ptr);
2088
1050 return 0; 2089 return wd;
1051} 2090}
2091
2092void
2093eio_wd_close_sync (eio_wd wd)
2094{
2095 if (wd != EIO_INVALID_WD)
2096 free (wd);
2097}
2098
2099#endif
1052 2100
1053/*****************************************************************************/ 2101/*****************************************************************************/
1054 2102
1055#define ALLOC(len) \ 2103#define ALLOC(len) \
1056 if (!req->ptr2) \ 2104 if (!req->ptr2) \
… …
1071{ 2119{
1072 ETP_REQ *req; 2120 ETP_REQ *req;
1073 struct timespec ts; 2121 struct timespec ts;
1074 etp_worker *self = (etp_worker *)thr_arg; 2122 etp_worker *self = (etp_worker *)thr_arg;
1075 2123
2124#if HAVE_PRCTL_SET_NAME
2125 prctl (PR_SET_NAME, (unsigned long)"eio_thread", 0, 0, 0);
2126#endif
2127
1076 /* try to distribute timeouts somewhat randomly */ 2128 /* try to distribute timeouts somewhat evenly */
1077 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 2129 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1078 2130
1079 for (;;) 2131 for (;;)
1080 { 2132 {
2133 ts.tv_sec = 0;
2134
1081 X_LOCK (reqlock); 2135 X_LOCK (reqlock);
1082 2136
1083 for (;;) 2137 for (;;)
1084 { 2138 {
1085 self->req = req = reqq_shift (&req_queue); 2139 req = reqq_shift (&req_queue);
1086 2140
1087 if (req) 2141 if (req)
1088 break; 2142 break;
1089 2143
2144 if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
2145 {
2146 X_UNLOCK (reqlock);
2147 X_LOCK (wrklock);
2148 --started;
2149 X_UNLOCK (wrklock);
2150 goto quit;
2151 }
2152
1090 ++idle; 2153 ++idle;
1091 2154
1092 ts.tv_sec = time (0) + IDLE_TIMEOUT; 2155 if (idle <= max_idle)
1093 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 2156 /* we are allowed to idle, so do so without any timeout */
2157 X_COND_WAIT (reqwait, reqlock);
2158 else
1094 { 2159 {
1095 if (idle > max_idle) 2160 /* initialise timeout once */
1096 { 2161 if (!ts.tv_sec)
1097 --idle; 2162 ts.tv_sec = time (0) + idle_timeout;
1098 X_UNLOCK (reqlock);
1099 X_LOCK (wrklock);
1100 --started;
1101 X_UNLOCK (wrklock);
1102 goto quit;
1103 }
1104 2163
1105 /* we are allowed to idle, so do so without any timeout */
1106 X_COND_WAIT (reqwait, reqlock); 2164 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
2165 ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1107 } 2166 }
1108 2167
1109 --idle; 2168 --idle;
1110 } 2169 }
1111 2170
… …
1114 X_UNLOCK (reqlock); 2173 X_UNLOCK (reqlock);
1115 2174
1116 if (req->type < 0) 2175 if (req->type < 0)
1117 goto quit; 2176 goto quit;
1118 2177
1119 if (!EIO_CANCELLED (req))
1120 ETP_EXECUTE (self, req); 2178 ETP_EXECUTE (self, req);
1121 2179
1122 X_LOCK (reslock); 2180 X_LOCK (reslock);
1123 2181
1124 ++npending; 2182 ++npending;
1125 2183
1126 if (!reqq_push (&res_queue, req) && want_poll_cb) 2184 if (!reqq_push (&res_queue, req) && want_poll_cb)
1127 want_poll_cb (); 2185 want_poll_cb ();
1128 2186
1129 self->req = 0;
1130 etp_worker_clear (self); 2187 etp_worker_clear (self);
1131 2188
1132 X_UNLOCK (reslock); 2189 X_UNLOCK (reslock);
1133 } 2190 }
1134 2191
1135quit: 2192quit:
2193 free (req);
2194
1136 X_LOCK (wrklock); 2195 X_LOCK (wrklock);
1137 etp_worker_free (self); 2196 etp_worker_free (self);
1138 X_UNLOCK (wrklock); 2197 X_UNLOCK (wrklock);
1139 2198
1140 return 0; 2199 return 0;
1141} 2200}
1142 2201
1143/*****************************************************************************/ 2202/*****************************************************************************/
1144 2203
2204int ecb_cold
1145int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 2205eio_init (void (*want_poll)(void), void (*done_poll)(void))
1146{ 2206{
2207#if !HAVE_PREADWRITE
2208 X_MUTEX_CREATE (preadwritelock);
2209#endif
2210
1147 return etp_init (want_poll, done_poll); 2211 return etp_init (want_poll, done_poll);
1148} 2212}
1149 2213
2214ecb_inline void
1150static void eio_api_destroy (eio_req *req) 2215eio_api_destroy (eio_req *req)
1151{ 2216{
1152 free (req); 2217 free (req);
1153} 2218}
1154 2219
1155#define REQ(rtype) \ 2220#define REQ(rtype) \
… …
1174 { \ 2239 { \
1175 eio_api_destroy (req); \ 2240 eio_api_destroy (req); \
1176 return 0; \ 2241 return 0; \
1177 } 2242 }
1178 2243
2244static void
1179static void eio_execute (etp_worker *self, eio_req *req) 2245eio_execute (etp_worker *self, eio_req *req)
1180{ 2246{
1181 errno = 0; 2247#if HAVE_AT
2248 int dirfd;
2249#else
2250 const char *path;
2251#endif
2252
2253 if (ecb_expect_false (EIO_CANCELLED (req)))
2254 {
2255 req->result = -1;
2256 req->errorno = ECANCELED;
2257 return;
2258 }
2259
2260 if (ecb_expect_false (req->wd == EIO_INVALID_WD))
2261 {
2262 req->result = -1;
2263 req->errorno = ENOENT;
2264 return;
2265 }
2266
2267 if (req->type >= EIO_OPEN)
2268 {
2269 #if HAVE_AT
2270 dirfd = WD2FD (req->wd);
2271 #else
2272 path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
2273 #endif
2274 }
1182 2275
1183 switch (req->type) 2276 switch (req->type)
1184 { 2277 {
2278 case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
2279 req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
2280 break;
2281 case EIO_WD_CLOSE: req->result = 0;
2282 eio_wd_close_sync (req->wd); break;
2283
1185 case EIO_READ: ALLOC (req->size); 2284 case EIO_READ: ALLOC (req->size);
1186 req->result = req->offs >= 0 2285 req->result = req->offs >= 0
1187 ? pread (req->int1, req->ptr2, req->size, req->offs) 2286 ? pread (req->int1, req->ptr2, req->size, req->offs)
1188 : read (req->int1, req->ptr2, req->size); break; 2287 : read (req->int1, req->ptr2, req->size); break;
1189 case EIO_WRITE: req->result = req->offs >= 0 2288 case EIO_WRITE: req->result = req->offs >= 0
1190 ? pwrite (req->int1, req->ptr2, req->size, req->offs) 2289 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1191 : write (req->int1, req->ptr2, req->size); break; 2290 : write (req->int1, req->ptr2, req->size); break;
1192 2291
1193 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break; 2292 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1194 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break; 2293 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
2294
2295#if HAVE_AT
1195 2296
1196 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2297 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1197 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2298 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break;
1198 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2299 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1199 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2300 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, AT_SYMLINK_NOFOLLOW); break;
1200 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1201 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1202
1203 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; 2301 case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break;
1204 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1205 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break; 2302 case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break;
1206 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1207 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break; 2303 case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break;
1208 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1209
1210 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break; 2304 case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break;
1211 case EIO_CLOSE: req->result = close (req->int1); break; 2305
1212 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1213 case EIO_UNLINK: req->result = unlink (req->ptr1); break; 2306 case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
1214 case EIO_RMDIR: req->result = rmdir (req->ptr1); break; 2307 case EIO_RMDIR: req->result = unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break;
1215 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; 2308 case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); break;
1216 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; 2309 case EIO_RENAME: req->result = renameat (dirfd, req->ptr1, WD2FD (req->int3), req->ptr2); break;
1217 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; 2310 case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD (req->int3), req->ptr2, 0); break;
1218 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; 2311 case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
1219 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break; 2312 case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1220
1221 case EIO_READLINK: ALLOC (NAME_MAX); 2313 case EIO_READLINK: ALLOC (PATH_MAX);
1222 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; 2314 req->result = readlinkat (dirfd, req->ptr1, req->ptr2, PATH_MAX); break;
1223 2315 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1224 case EIO_SYNC: req->result = 0; sync (); break; 2316 req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1225 case EIO_FSYNC: req->result = fsync (req->int1); break;
1226 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1227 case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break;
1228 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;
1229 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1230
1231 case EIO_READDIR: eio__scandir (req, self); break;
1232
1233 case EIO_BUSY: 2317 case EIO_UTIME:
1234#ifdef _WIN32 2318 case EIO_FUTIME:
1235 Sleep (req->nv1 * 1000.); 2319 {
2320 struct timespec ts[2];
2321 struct timespec *times;
2322
2323 if (req->nv1 != -1. || req->nv2 != -1.)
2324 {
2325 ts[0].tv_sec = req->nv1;
2326 ts[0].tv_nsec = (req->nv1 - ts[0].tv_sec) * 1e9;
2327 ts[1].tv_sec = req->nv2;
2328 ts[1].tv_nsec = (req->nv2 - ts[1].tv_sec) * 1e9;
2329
2330 times = ts;
2331 }
2332 else
2333 times = 0;
2334
2335 req->result = req->type == EIO_FUTIME
2336 ? futimens (req->int1, times)
2337 : utimensat (dirfd, req->ptr1, times, 0);
2338 }
2339 break;
2340
1236#else 2341#else
1237 {
1238 struct timeval tv;
1239 2342
1240 tv.tv_sec = req->nv1; 2343 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1241 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; 2344 req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
2345 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2346 req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
2347 case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
2348 case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
2349 case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
2350 case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
1242 2351
1243 req->result = select (0, 0, 0, 0, &tv); 2352 case EIO_UNLINK: req->result = unlink (path ); break;
1244 } 2353 case EIO_RMDIR: req->result = rmdir (path ); break;
1245#endif 2354 case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
1246 break; 2355 case EIO_RENAME: req->result = rename (path , req->ptr2); break;
2356 case EIO_LINK: req->result = link (path , req->ptr2); break;
2357 case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
2358 case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
2359 case EIO_READLINK: ALLOC (PATH_MAX);
2360 req->result = readlink (path, req->ptr2, PATH_MAX); break;
2361 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2362 req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
1247 2363
1248 case EIO_UTIME: 2364 case EIO_UTIME:
1249 case EIO_FUTIME: 2365 case EIO_FUTIME:
1250 { 2366 {
1251 struct timeval tv[2]; 2367 struct timeval tv[2];
1252 struct timeval *times; 2368 struct timeval *times;
1253 2369
1254 if (req->nv1 != -1. || req->nv2 != -1.) 2370 if (req->nv1 != -1. || req->nv2 != -1.)
1255 { 2371 {
1256 tv[0].tv_sec = req->nv1; 2372 tv[0].tv_sec = req->nv1;
1257 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.; 2373 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
1258 tv[1].tv_sec = req->nv2; 2374 tv[1].tv_sec = req->nv2;
1259 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.; 2375 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
1260 2376
1261 times = tv; 2377 times = tv;
1262 } 2378 }
1263 else 2379 else
1264 times = 0; 2380 times = 0;
1265
1266 2381
1267 req->result = req->type == EIO_FUTIME 2382 req->result = req->type == EIO_FUTIME
1268 ? futimes (req->int1, times) 2383 ? futimes (req->int1, times)
1269 : utimes (req->ptr1, times); 2384 : utimes (req->ptr1, times);
1270 } 2385 }
1271 break; 2386 break;
1272 2387
2388#endif
2389
2390 case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
2391 {
2392 ALLOC (req->result);
2393 memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
2394 }
2395 break;
2396
2397 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2398 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
2399
2400 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2401 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2402
2403 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
2404 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
2405 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
2406
2407 case EIO_CLOSE: req->result = close (req->int1); break;
2408 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
2409 case EIO_SYNC: req->result = 0; sync (); break;
2410 case EIO_FSYNC: req->result = fsync (req->int1); break;
2411 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
2412 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
2413 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
2414 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
2415 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
2416 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
2417 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
2418 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
2419
2420 case EIO_READDIR: eio__scandir (req); break;
2421
2422 case EIO_BUSY:
2423#ifdef _WIN32
2424 Sleep (req->nv1 * 1e3);
2425#else
2426 {
2427 struct timeval tv;
2428
2429 tv.tv_sec = req->nv1;
2430 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
2431
2432 req->result = select (0, 0, 0, 0, &tv);
2433 }
2434#endif
2435 break;
2436
1273 case EIO_GROUP: 2437 case EIO_GROUP:
1274 abort (); /* handled in eio_request */ 2438 abort (); /* handled in eio_request */
1275 2439
1276 case EIO_NOP: 2440 case EIO_NOP:
1277 req->result = 0; 2441 req->result = 0;
1278 break; 2442 break;
1279 2443
1280 case EIO_CUSTOM: 2444 case EIO_CUSTOM:
1281 ((void (*)(eio_req *))req->feed) (req); 2445 req->feed (req);
1282 break; 2446 break;
1283 2447
1284 default: 2448 default:
2449 errno = ENOSYS;
1285 req->result = -1; 2450 req->result = -1;
1286 break; 2451 break;
1287 } 2452 }
1288 2453
1289 req->errorno = errno; 2454 req->errorno = errno;
… …
1314eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2479eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1315{ 2480{
1316 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2481 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1317} 2482}
1318 2483
2484eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2485{
2486 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2487}
2488
2489eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
2490{
2491 REQ (EIO_SYNCFS); req->int1 = fd; SEND;
2492}
2493
2494eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2495{
2496 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2497}
2498
1319eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2499eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1320{ 2500{
1321 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2501 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1322} 2502}
1323 2503
1324eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) 2504eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
1325{ 2505{
1326 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; 2506 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
1327} 2507}
1328 2508
1329eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) 2509eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
1330{ 2510{
1331 REQ (EIO_FDATASYNC); req->int1 = fd; SEND; 2511 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2512}
2513
2514eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2515{
2516 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
1332} 2517}
1333 2518
1334eio_req *eio_close (int fd, int pri, eio_cb cb, void *data) 2519eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
1335{ 2520{
1336 REQ (EIO_CLOSE); req->int1 = fd; SEND; 2521 REQ (EIO_CLOSE); req->int1 = fd; SEND;
… …
1354eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) 2539eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1355{ 2540{
1356 REQ (EIO_FSTAT); req->int1 = fd; SEND; 2541 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1357} 2542}
1358 2543
2544eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
2545{
2546 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
2547}
2548
1359eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) 2549eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1360{ 2550{
1361 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; 2551 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1362} 2552}
1363 2553
… …
1369eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) 2559eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1370{ 2560{
1371 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2561 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1372} 2562}
1373 2563
1374eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2564eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1375{ 2565{
1376 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2566 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1377} 2567}
1378 2568
1379eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) 2569eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
… …
1399eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data) 2589eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1400{ 2590{
1401 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2591 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1402} 2592}
1403 2593
1404eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2594eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1405{ 2595{
1406 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2596 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1407} 2597}
1408 2598
1409eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) 2599eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
… …
1425eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data) 2615eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1426{ 2616{
1427 return eio__1path (EIO_READLINK, path, pri, cb, data); 2617 return eio__1path (EIO_READLINK, path, pri, cb, data);
1428} 2618}
1429 2619
2620eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2621{
2622 return eio__1path (EIO_REALPATH, path, pri, cb, data);
2623}
2624
1430eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) 2625eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1431{ 2626{
1432 return eio__1path (EIO_STAT, path, pri, cb, data); 2627 return eio__1path (EIO_STAT, path, pri, cb, data);
1433} 2628}
1434 2629
1435eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) 2630eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1436{ 2631{
1437 return eio__1path (EIO_LSTAT, path, pri, cb, data); 2632 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1438} 2633}
1439 2634
2635eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
2636{
2637 return eio__1path (EIO_STATVFS, path, pri, cb, data);
2638}
2639
1440eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) 2640eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1441{ 2641{
1442 return eio__1path (EIO_UNLINK, path, pri, cb, data); 2642 return eio__1path (EIO_UNLINK, path, pri, cb, data);
1443} 2643}
1444 2644
1445eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data) 2645eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1446{ 2646{
1447 return eio__1path (EIO_RMDIR, path, pri, cb, data); 2647 return eio__1path (EIO_RMDIR, path, pri, cb, data);
1448} 2648}
1449 2649
1450eio_req *eio_readdir (const char *path, int pri, eio_cb cb, void *data) 2650eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1451{ 2651{
1452 return eio__1path (EIO_READDIR, path, pri, cb, data); 2652 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1453} 2653}
1454 2654
1455eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) 2655eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1456{ 2656{
1457 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; 2657 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1458} 2658}
1459 2659
1460static eio_req * 2660static eio_req *
1461eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2661eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1462{ 2662{
… …
1486eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2686eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1487{ 2687{
1488 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2688 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1489} 2689}
1490 2690
1491eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) 2691eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
1492{ 2692{
1493 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND; 2693 REQ (EIO_CUSTOM); req->feed = execute; SEND;
1494} 2694}
1495 2695
1496#endif 2696#endif
1497 2697
1498eio_req *eio_grp (eio_cb cb, void *data) 2698eio_req *eio_grp (eio_cb cb, void *data)
… …
1507#undef SEND 2707#undef SEND
1508 2708
1509/*****************************************************************************/ 2709/*****************************************************************************/
1510/* grp functions */ 2710/* grp functions */
1511 2711
2712void
1512void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2713eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1513{ 2714{
1514 grp->int2 = limit; 2715 grp->int2 = limit;
1515 grp->feed = feed; 2716 grp->feed = feed;
1516 2717
1517 grp_try_feed (grp); 2718 grp_try_feed (grp);
1518} 2719}
1519 2720
2721void
1520void eio_grp_limit (eio_req *grp, int limit) 2722eio_grp_limit (eio_req *grp, int limit)
1521{ 2723{
1522 grp->int2 = limit; 2724 grp->int2 = limit;
1523 2725
1524 grp_try_feed (grp); 2726 grp_try_feed (grp);
1525} 2727}
1526 2728
2729void
1527void eio_grp_add (eio_req *grp, eio_req *req) 2730eio_grp_add (eio_req *grp, eio_req *req)
1528{ 2731{
1529 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2732 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1530 2733
1531 grp->flags |= EIO_FLAG_GROUPADD; 2734 grp->flags |= EIO_FLAG_GROUPADD;
1532 2735
… …
1543} 2746}
1544 2747
1545/*****************************************************************************/ 2748/*****************************************************************************/
1546/* misc garbage */ 2749/* misc garbage */
1547 2750
2751eio_ssize_t
1548ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2752eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1549{ 2753{
1550 etp_worker wrk;
1551
1552 wrk.dbuf = 0;
1553
1554 eio__sendfile (ofd, ifd, offset, count, &wrk); 2754 return eio__sendfile (ofd, ifd, offset, count);
1555
1556 if (wrk.dbuf)
1557 free (wrk.dbuf);
1558} 2755}
1559 2756

Diff Legend

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