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

Comparing libeio/eio.c (file contents):
Revision 1.55 by root, Wed Mar 31 00:46:18 2010 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) 2007,2008,2009,2010 Marc Alexander Lehmann <libeio@schmorp.de> 4 * Copyright (c) 2007,2008,2009,2010,2011 Marc Alexander Lehmann <libeio@schmorp.de>
5 * All rights reserved. 5 * All rights reserved.
6 * 6 *
7 * Redistribution and use in source and binary forms, with or without modifica- 7 * Redistribution and use in source and binary forms, with or without modifica-
8 * tion, are permitted provided that the following conditions are met: 8 * tion, are permitted provided that the following conditions are met:
9 * 9 *
… …
35 * and other provisions required by the GPL. If you do not delete the 35 * and other provisions required by the GPL. If you do not delete the
36 * provisions above, a recipient may use your version of this file under 36 * provisions above, a recipient may use your version of this file under
37 * either the BSD or the GPL. 37 * either the BSD or the GPL.
38 */ 38 */
39 39
40#ifndef _WIN32
41# include "config.h"
42#endif
43
40#include "eio.h" 44#include "eio.h"
45#include "ecb.h"
41 46
42#ifdef EIO_STACKSIZE 47#ifdef EIO_STACKSIZE
43# define XTHREAD_STACKSIZE EIO_STACKSIZE 48# define X_STACKSIZE EIO_STACKSIZE
44#endif 49#endif
45#include "xthread.h" 50#include "xthread.h"
46 51
47#include <errno.h> 52#include <errno.h>
48#include <stddef.h> 53#include <stddef.h>
49#include <stdlib.h> 54#include <stdlib.h>
50#include <string.h> 55#include <string.h>
51#include <errno.h> 56#include <errno.h>
52#include <sys/types.h> 57#include <sys/types.h>
53#include <sys/stat.h> 58#include <sys/stat.h>
54#include <sys/statvfs.h>
55#include <limits.h> 59#include <limits.h>
56#include <fcntl.h> 60#include <fcntl.h>
57#include <assert.h> 61#include <assert.h>
58 62
63#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
59#ifndef EIO_FINISH 88#ifndef EIO_FINISH
60# 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
61#endif 90#endif
62 91
63#ifndef EIO_DESTROY 92#ifndef EIO_DESTROY
… …
66 95
67#ifndef EIO_FEED 96#ifndef EIO_FEED
68# 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)
69#endif 98#endif
70 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
71#ifdef _WIN32 111#ifdef _WIN32
72 112
73 /*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)
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>
79# include <sys/mman.h> 207 #include <sys/statvfs.h>
80# 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
81# include <utime.h> 243# include <utime.h>
82# include <signal.h>
83# include <dirent.h>
84
85/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
86# if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
87# define _DIRENT_HAVE_D_TYPE /* sigh */
88# define D_INO(de) (de)->d_fileno
89# define D_NAMLEN(de) (de)->d_namlen
90# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
91# define D_INO(de) (de)->d_ino
92# endif 244#endif
93 245
94#ifdef _D_EXACT_NAMLEN 246#if HAVE_SYS_SYSCALL_H
95# undef D_NAMLEN 247# include <sys/syscall.h>
96# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
97#endif
98
99# ifdef _DIRENT_HAVE_D_TYPE
100# define D_TYPE(de) (de)->d_type
101# endif 248#endif
102 249
103# ifndef EIO_STRUCT_DIRENT 250#if HAVE_SYS_PRCTL_H
104# define EIO_STRUCT_DIRENT struct dirent 251# include <sys/prctl.h>
105# endif
106
107#endif 252#endif
108 253
109#if HAVE_SENDFILE 254#if HAVE_SENDFILE
110# if __linux 255# if __linux
111# include <sys/sendfile.h> 256# include <sys/sendfile.h>
… …
126#endif 271#endif
127#ifndef D_INO 272#ifndef D_INO
128# define D_INO(de) 0 273# define D_INO(de) 0
129#endif 274#endif
130#ifndef D_NAMLEN 275#ifndef D_NAMLEN
131# define D_NAMLEN(de) strlen ((de)->d_name) 276# define D_NAMLEN(entp) strlen (D_NAME (entp))
132#endif 277#endif
133
134/* number of seconds after which an idle threads exit */
135#define IDLE_TIMEOUT 10
136 278
137/* used for struct dirent, AIX doesn't provide it */ 279/* used for struct dirent, AIX doesn't provide it */
138#ifndef NAME_MAX 280#ifndef NAME_MAX
139# define NAME_MAX 4096 281# define NAME_MAX 4096
140#endif 282#endif
… …
146 288
147/* buffer size for various temporary buffers */ 289/* buffer size for various temporary buffers */
148#define EIO_BUFSIZE 65536 290#define EIO_BUFSIZE 65536
149 291
150#define dBUF \ 292#define dBUF \
151 char *eio_buf; \
152 ETP_WORKER_LOCK (self); \
153 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 293 char *eio_buf = malloc (EIO_BUFSIZE); \
154 ETP_WORKER_UNLOCK (self); \
155 errno = ENOMEM; \ 294 errno = ENOMEM; \
156 if (!eio_buf) \ 295 if (!eio_buf) \
157 return -1; 296 return -1
297
298#define FUBd \
299 free (eio_buf)
158 300
159#define EIO_TICKS ((1000000 + 1023) >> 10) 301#define EIO_TICKS ((1000000 + 1023) >> 10)
160 302
161/*****************************************************************************/ 303/*****************************************************************************/
162 304
163#if __GNUC__ >= 3 305struct tmpbuf
164# define expect(expr,value) __builtin_expect ((expr),(value)) 306{
165#else 307 void *ptr;
166# define expect(expr,value) (expr) 308 int len;
167#endif 309};
168 310
169#define expect_false(expr) expect ((expr) != 0, 0) 311static void *
170#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}
171 322
172/*****************************************************************************/ 323/*****************************************************************************/
173 324
174#define ETP_PRI_MIN EIO_PRI_MIN 325#define ETP_PRI_MIN EIO_PRI_MIN
175#define ETP_PRI_MAX EIO_PRI_MAX 326#define ETP_PRI_MAX EIO_PRI_MAX
… …
181static int eio_finish (eio_req *req); 332static int eio_finish (eio_req *req);
182#define ETP_FINISH(req) eio_finish (req) 333#define ETP_FINISH(req) eio_finish (req)
183static void eio_execute (struct etp_worker *self, eio_req *req); 334static void eio_execute (struct etp_worker *self, eio_req *req);
184#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) 335#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
185 336
186#define ETP_WORKER_CLEAR(req) \
187 if (wrk->dbuf) \
188 { \
189 free (wrk->dbuf); \
190 wrk->dbuf = 0; \
191 } \
192 \
193 if (wrk->dirp) \
194 { \
195 closedir (wrk->dirp); \
196 wrk->dirp = 0; \
197 }
198
199#define ETP_WORKER_COMMON \
200 void *dbuf; \
201 DIR *dirp;
202
203/*****************************************************************************/ 337/*****************************************************************************/
204 338
205#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) 339#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
206 340
207/* calculate time difference in ~1/EIO_TICKS of a second */ 341/* calculate time difference in ~1/EIO_TICKS of a second */
342ecb_inline int
208static int tvdiff (struct timeval *tv1, struct timeval *tv2) 343tvdiff (struct timeval *tv1, struct timeval *tv2)
209{ 344{
210 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 345 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
211 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 346 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
212} 347}
213 348
… …
217static void (*done_poll_cb) (void); 352static void (*done_poll_cb) (void);
218 353
219static unsigned int max_poll_time; /* reslock */ 354static unsigned int max_poll_time; /* reslock */
220static unsigned int max_poll_reqs; /* reslock */ 355static unsigned int max_poll_reqs; /* reslock */
221 356
222static volatile unsigned int nreqs; /* reqlock */ 357static unsigned int nreqs; /* reqlock */
223static volatile unsigned int nready; /* reqlock */ 358static unsigned int nready; /* reqlock */
224static volatile unsigned int npending; /* reqlock */ 359static unsigned int npending; /* reqlock */
225static 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 */
226 362
227static xmutex_t wrklock = X_MUTEX_INIT; 363static xmutex_t wrklock;
228static xmutex_t reslock = X_MUTEX_INIT; 364static xmutex_t reslock;
229static xmutex_t reqlock = X_MUTEX_INIT; 365static xmutex_t reqlock;
230static xcond_t reqwait = X_COND_INIT; 366static xcond_t reqwait;
231 367
232#if !HAVE_PREADWRITE 368#if !HAVE_PREADWRITE
233/* 369/*
234 * make our pread/pwrite emulation safe against themselves, but not against 370 * make our pread/pwrite emulation safe against themselves, but not against
235 * 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,
236 * but that's your problem, not mine. 372 * but that's your problem, not mine.
237 */ 373 */
238static xmutex_t preadwritelock = X_MUTEX_INIT; 374static xmutex_t preadwritelock;
239#endif 375#endif
240 376
241typedef struct etp_worker 377typedef struct etp_worker
242{ 378{
379 struct tmpbuf tmpbuf;
380
243 /* locked by wrklock */ 381 /* locked by wrklock */
244 struct etp_worker *prev, *next; 382 struct etp_worker *prev, *next;
245 383
246 xthread_t tid; 384 xthread_t tid;
247 385
248 /* locked by reslock, reqlock or wrklock */ 386#ifdef ETP_WORKER_COMMON
249 ETP_REQ *req; /* currently processed request */
250
251 ETP_WORKER_COMMON 387 ETP_WORKER_COMMON
388#endif
252} etp_worker; 389} etp_worker;
253 390
254static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */ 391static etp_worker wrk_first; /* NOT etp */
255 392
256#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock) 393#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
257#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock) 394#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
258 395
259/* worker threads management */ 396/* worker threads management */
260 397
398static void ecb_cold
261static void etp_worker_clear (etp_worker *wrk) 399etp_worker_clear (etp_worker *wrk)
262{ 400{
263 ETP_WORKER_CLEAR (wrk);
264} 401}
265 402
403static void ecb_cold
266static void etp_worker_free (etp_worker *wrk) 404etp_worker_free (etp_worker *wrk)
267{ 405{
406 free (wrk->tmpbuf.ptr);
407
268 wrk->next->prev = wrk->prev; 408 wrk->next->prev = wrk->prev;
269 wrk->prev->next = wrk->next; 409 wrk->prev->next = wrk->next;
270 410
271 free (wrk); 411 free (wrk);
272} 412}
273 413
274static unsigned int etp_nreqs (void) 414static unsigned int
415etp_nreqs (void)
275{ 416{
276 int retval; 417 int retval;
277 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 418 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
278 retval = nreqs; 419 retval = nreqs;
279 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 420 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
280 return retval; 421 return retval;
281} 422}
282 423
283static unsigned int etp_nready (void) 424static unsigned int
425etp_nready (void)
284{ 426{
285 unsigned int retval; 427 unsigned int retval;
286 428
287 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 429 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
288 retval = nready; 430 retval = nready;
289 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 431 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
290 432
291 return retval; 433 return retval;
292} 434}
293 435
294static unsigned int etp_npending (void) 436static unsigned int
437etp_npending (void)
295{ 438{
296 unsigned int retval; 439 unsigned int retval;
297 440
298 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 441 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
299 retval = npending; 442 retval = npending;
300 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 443 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
301 444
302 return retval; 445 return retval;
303} 446}
304 447
305static unsigned int etp_nthreads (void) 448static unsigned int
449etp_nthreads (void)
306{ 450{
307 unsigned int retval; 451 unsigned int retval;
308 452
309 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 453 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
310 retval = started; 454 retval = started;
… …
324} etp_reqq; 468} etp_reqq;
325 469
326static etp_reqq req_queue; 470static etp_reqq req_queue;
327static etp_reqq res_queue; 471static etp_reqq res_queue;
328 472
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
329static int reqq_push (etp_reqq *q, ETP_REQ *req) 485reqq_push (etp_reqq *q, ETP_REQ *req)
330{ 486{
331 int pri = req->pri; 487 int pri = req->pri;
332 req->next = 0; 488 req->next = 0;
333 489
334 if (q->qe[pri]) 490 if (q->qe[pri])
… …
340 q->qe[pri] = q->qs[pri] = req; 496 q->qe[pri] = q->qs[pri] = req;
341 497
342 return q->size++; 498 return q->size++;
343} 499}
344 500
501static ETP_REQ * ecb_noinline
345static ETP_REQ *reqq_shift (etp_reqq *q) 502reqq_shift (etp_reqq *q)
346{ 503{
347 int pri; 504 int pri;
348 505
349 if (!q->size) 506 if (!q->size)
350 return 0; 507 return 0;
… …
365 } 522 }
366 523
367 abort (); 524 abort ();
368} 525}
369 526
370static void etp_atfork_prepare (void) 527static int ecb_cold
528etp_init (void (*want_poll)(void), void (*done_poll)(void))
371{ 529{
372 X_LOCK (wrklock); 530 X_MUTEX_CREATE (wrklock);
373 X_LOCK (reqlock); 531 X_MUTEX_CREATE (reslock);
374 X_LOCK (reslock); 532 X_MUTEX_CREATE (reqlock);
375#if !HAVE_PREADWRITE 533 X_COND_CREATE (reqwait);
376 X_LOCK (preadwritelock);
377#endif
378}
379 534
380static void etp_atfork_parent (void) 535 reqq_init (&req_queue);
381{ 536 reqq_init (&res_queue);
382#if !HAVE_PREADWRITE
383 X_UNLOCK (preadwritelock);
384#endif
385 X_UNLOCK (reslock);
386 X_UNLOCK (reqlock);
387 X_UNLOCK (wrklock);
388}
389 537
390static void etp_atfork_child (void) 538 wrk_first.next =
391{ 539 wrk_first.prev = &wrk_first;
392 ETP_REQ *prv;
393
394 while ((prv = reqq_shift (&req_queue)))
395 ETP_DESTROY (prv);
396
397 while ((prv = reqq_shift (&res_queue)))
398 ETP_DESTROY (prv);
399
400 while (wrk_first.next != &wrk_first)
401 {
402 etp_worker *wrk = wrk_first.next;
403
404 if (wrk->req)
405 ETP_DESTROY (wrk->req);
406
407 etp_worker_clear (wrk);
408 etp_worker_free (wrk);
409 }
410 540
411 started = 0; 541 started = 0;
412 idle = 0; 542 idle = 0;
413 nreqs = 0; 543 nreqs = 0;
414 nready = 0; 544 nready = 0;
415 npending = 0; 545 npending = 0;
416 546
417 etp_atfork_parent ();
418}
419
420static void
421etp_once_init (void)
422{
423 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
424}
425
426static int
427etp_init (void (*want_poll)(void), void (*done_poll)(void))
428{
429 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
430
431 pthread_once (&doinit, etp_once_init);
432
433 want_poll_cb = want_poll; 547 want_poll_cb = want_poll;
434 done_poll_cb = done_poll; 548 done_poll_cb = done_poll;
435 549
436 return 0; 550 return 0;
437} 551}
438 552
439X_THREAD_PROC (etp_proc); 553X_THREAD_PROC (etp_proc);
440 554
555static void ecb_cold
441static void etp_start_thread (void) 556etp_start_thread (void)
442{ 557{
443 etp_worker *wrk = calloc (1, sizeof (etp_worker)); 558 etp_worker *wrk = calloc (1, sizeof (etp_worker));
444 559
445 /*TODO*/ 560 /*TODO*/
446 assert (("unable to allocate worker thread data", wrk)); 561 assert (("unable to allocate worker thread data", wrk));
… …
459 free (wrk); 574 free (wrk);
460 575
461 X_UNLOCK (wrklock); 576 X_UNLOCK (wrklock);
462} 577}
463 578
579static void
464static void etp_maybe_start_thread (void) 580etp_maybe_start_thread (void)
465{ 581{
466 if (expect_true (etp_nthreads () >= wanted)) 582 if (ecb_expect_true (etp_nthreads () >= wanted))
467 return; 583 return;
468 584
469 /* todo: maybe use idle here, but might be less exact */ 585 /* todo: maybe use idle here, but might be less exact */
470 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 ()))
471 return; 587 return;
472 588
473 etp_start_thread (); 589 etp_start_thread ();
474} 590}
475 591
592static void ecb_cold
476static void etp_end_thread (void) 593etp_end_thread (void)
477{ 594{
478 eio_req *req = calloc (1, sizeof (eio_req)); 595 eio_req *req = calloc (1, sizeof (eio_req)); /* will be freed by worker */
479 596
480 req->type = -1; 597 req->type = -1;
481 req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 598 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
482 599
483 X_LOCK (reqlock); 600 X_LOCK (reqlock);
… …
488 X_LOCK (wrklock); 605 X_LOCK (wrklock);
489 --started; 606 --started;
490 X_UNLOCK (wrklock); 607 X_UNLOCK (wrklock);
491} 608}
492 609
493static int etp_poll (void) 610static int
611etp_poll (void)
494{ 612{
495 unsigned int maxreqs; 613 unsigned int maxreqs;
496 unsigned int maxtime; 614 unsigned int maxtime;
497 struct timeval tv_start, tv_now; 615 struct timeval tv_start, tv_now;
498 616
… …
528 646
529 X_LOCK (reqlock); 647 X_LOCK (reqlock);
530 --nreqs; 648 --nreqs;
531 X_UNLOCK (reqlock); 649 X_UNLOCK (reqlock);
532 650
533 if (expect_false (req->type == EIO_GROUP && req->size)) 651 if (ecb_expect_false (req->type == EIO_GROUP && req->size))
534 { 652 {
535 req->int1 = 1; /* mark request as delayed */ 653 req->int1 = 1; /* mark request as delayed */
536 continue; 654 continue;
537 } 655 }
538 else 656 else
539 { 657 {
540 int res = ETP_FINISH (req); 658 int res = ETP_FINISH (req);
541 if (expect_false (res)) 659 if (ecb_expect_false (res))
542 return res; 660 return res;
543 } 661 }
544 662
545 if (expect_false (maxreqs && !--maxreqs)) 663 if (ecb_expect_false (maxreqs && !--maxreqs))
546 break; 664 break;
547 665
548 if (maxtime) 666 if (maxtime)
549 { 667 {
550 gettimeofday (&tv_now, 0); 668 gettimeofday (&tv_now, 0);
… …
556 674
557 errno = EAGAIN; 675 errno = EAGAIN;
558 return -1; 676 return -1;
559} 677}
560 678
679static void
561static void etp_cancel (ETP_REQ *req) 680etp_cancel (ETP_REQ *req)
562{ 681{
563 X_LOCK (wrklock); 682 req->cancelled = 1;
564 req->flags |= EIO_FLAG_CANCELLED;
565 X_UNLOCK (wrklock);
566 683
567 eio_grp_cancel (req); 684 eio_grp_cancel (req);
568} 685}
569 686
687static void
570static void etp_submit (ETP_REQ *req) 688etp_submit (ETP_REQ *req)
571{ 689{
572 req->pri -= ETP_PRI_MIN; 690 req->pri -= ETP_PRI_MIN;
573 691
574 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;
575 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;
576 694
577 if (expect_false (req->type == EIO_GROUP)) 695 if (ecb_expect_false (req->type == EIO_GROUP))
578 { 696 {
579 /* I hope this is worth it :/ */ 697 /* I hope this is worth it :/ */
580 X_LOCK (reqlock); 698 X_LOCK (reqlock);
581 ++nreqs; 699 ++nreqs;
582 X_UNLOCK (reqlock); 700 X_UNLOCK (reqlock);
… …
601 719
602 etp_maybe_start_thread (); 720 etp_maybe_start_thread ();
603 } 721 }
604} 722}
605 723
724static void ecb_cold
606static void etp_set_max_poll_time (double nseconds) 725etp_set_max_poll_time (double nseconds)
607{ 726{
608 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 727 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
609 max_poll_time = nseconds * EIO_TICKS; 728 max_poll_time = nseconds * EIO_TICKS;
610 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 729 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
611} 730}
612 731
732static void ecb_cold
613static void etp_set_max_poll_reqs (unsigned int maxreqs) 733etp_set_max_poll_reqs (unsigned int maxreqs)
614{ 734{
615 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 735 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
616 max_poll_reqs = maxreqs; 736 max_poll_reqs = maxreqs;
617 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 737 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
618} 738}
619 739
740static void ecb_cold
620static void etp_set_max_idle (unsigned int nthreads) 741etp_set_max_idle (unsigned int nthreads)
621{ 742{
622 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 743 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
623 max_idle = nthreads <= 0 ? 1 : nthreads; 744 max_idle = nthreads;
624 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 745 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
625} 746}
626 747
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
627static void etp_set_min_parallel (unsigned int nthreads) 757etp_set_min_parallel (unsigned int nthreads)
628{ 758{
629 if (wanted < nthreads) 759 if (wanted < nthreads)
630 wanted = nthreads; 760 wanted = nthreads;
631} 761}
632 762
763static void ecb_cold
633static void etp_set_max_parallel (unsigned int nthreads) 764etp_set_max_parallel (unsigned int nthreads)
634{ 765{
635 if (wanted > nthreads) 766 if (wanted > nthreads)
636 wanted = nthreads; 767 wanted = nthreads;
637 768
638 while (started > wanted) 769 while (started > wanted)
639 etp_end_thread (); 770 etp_end_thread ();
640} 771}
641 772
642/*****************************************************************************/ 773/*****************************************************************************/
643 774
775static void
644static void grp_try_feed (eio_req *grp) 776grp_try_feed (eio_req *grp)
645{ 777{
646 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 778 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
647 { 779 {
648 grp->flags &= ~EIO_FLAG_GROUPADD; 780 grp->flags &= ~EIO_FLAG_GROUPADD;
649 781
… …
656 break; 788 break;
657 } 789 }
658 } 790 }
659} 791}
660 792
793static int
661static int grp_dec (eio_req *grp) 794grp_dec (eio_req *grp)
662{ 795{
663 --grp->size; 796 --grp->size;
664 797
665 /* call feeder, if applicable */ 798 /* call feeder, if applicable */
666 grp_try_feed (grp); 799 grp_try_feed (grp);
… …
670 return eio_finish (grp); 803 return eio_finish (grp);
671 else 804 else
672 return 0; 805 return 0;
673} 806}
674 807
808static void
675void eio_destroy (eio_req *req) 809eio_destroy (eio_req *req)
676{ 810{
677 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); 811 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
678 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); 812 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
679 813
680 EIO_DESTROY (req); 814 EIO_DESTROY (req);
681} 815}
682 816
817static int
683static int eio_finish (eio_req *req) 818eio_finish (eio_req *req)
684{ 819{
685 int res = EIO_FINISH (req); 820 int res = EIO_FINISH (req);
686 821
687 if (req->grp) 822 if (req->grp)
688 { 823 {
… …
696 if (grp->grp_first == req) 831 if (grp->grp_first == req)
697 grp->grp_first = req->grp_next; 832 grp->grp_first = req->grp_next;
698 833
699 res2 = grp_dec (grp); 834 res2 = grp_dec (grp);
700 835
701 if (!res && res2) 836 if (!res)
702 res = res2; 837 res = res2;
703 } 838 }
704 839
705 eio_destroy (req); 840 eio_destroy (req);
706 841
707 return res; 842 return res;
708} 843}
709 844
845void
710void eio_grp_cancel (eio_req *grp) 846eio_grp_cancel (eio_req *grp)
711{ 847{
712 for (grp = grp->grp_first; grp; grp = grp->grp_next) 848 for (grp = grp->grp_first; grp; grp = grp->grp_next)
713 eio_cancel (grp); 849 eio_cancel (grp);
714} 850}
715 851
852void
716void eio_cancel (eio_req *req) 853eio_cancel (eio_req *req)
717{ 854{
718 etp_cancel (req); 855 etp_cancel (req);
719} 856}
720 857
858void
721void eio_submit (eio_req *req) 859eio_submit (eio_req *req)
722{ 860{
723 etp_submit (req); 861 etp_submit (req);
724} 862}
725 863
726unsigned int eio_nreqs (void) 864unsigned int
865eio_nreqs (void)
727{ 866{
728 return etp_nreqs (); 867 return etp_nreqs ();
729} 868}
730 869
731unsigned int eio_nready (void) 870unsigned int
871eio_nready (void)
732{ 872{
733 return etp_nready (); 873 return etp_nready ();
734} 874}
735 875
736unsigned int eio_npending (void) 876unsigned int
877eio_npending (void)
737{ 878{
738 return etp_npending (); 879 return etp_npending ();
739} 880}
740 881
741unsigned int eio_nthreads (void) 882unsigned int ecb_cold
883eio_nthreads (void)
742{ 884{
743 return etp_nthreads (); 885 return etp_nthreads ();
744} 886}
745 887
888void ecb_cold
746void eio_set_max_poll_time (double nseconds) 889eio_set_max_poll_time (double nseconds)
747{ 890{
748 etp_set_max_poll_time (nseconds); 891 etp_set_max_poll_time (nseconds);
749} 892}
750 893
894void ecb_cold
751void eio_set_max_poll_reqs (unsigned int maxreqs) 895eio_set_max_poll_reqs (unsigned int maxreqs)
752{ 896{
753 etp_set_max_poll_reqs (maxreqs); 897 etp_set_max_poll_reqs (maxreqs);
754} 898}
755 899
900void ecb_cold
756void eio_set_max_idle (unsigned int nthreads) 901eio_set_max_idle (unsigned int nthreads)
757{ 902{
758 etp_set_max_idle (nthreads); 903 etp_set_max_idle (nthreads);
759} 904}
760 905
906void ecb_cold
907eio_set_idle_timeout (unsigned int seconds)
908{
909 etp_set_idle_timeout (seconds);
910}
911
912void ecb_cold
761void eio_set_min_parallel (unsigned int nthreads) 913eio_set_min_parallel (unsigned int nthreads)
762{ 914{
763 etp_set_min_parallel (nthreads); 915 etp_set_min_parallel (nthreads);
764} 916}
765 917
918void ecb_cold
766void eio_set_max_parallel (unsigned int nthreads) 919eio_set_max_parallel (unsigned int nthreads)
767{ 920{
768 etp_set_max_parallel (nthreads); 921 etp_set_max_parallel (nthreads);
769} 922}
770 923
771int eio_poll (void) 924int eio_poll (void)
… …
780# undef pread 933# undef pread
781# undef pwrite 934# undef pwrite
782# define pread eio__pread 935# define pread eio__pread
783# define pwrite eio__pwrite 936# define pwrite eio__pwrite
784 937
785static ssize_t 938static eio_ssize_t
786eio__pread (int fd, void *buf, size_t count, off_t offset) 939eio__pread (int fd, void *buf, size_t count, off_t offset)
787{ 940{
788 ssize_t res; 941 eio_ssize_t res;
789 off_t ooffset; 942 off_t ooffset;
790 943
791 X_LOCK (preadwritelock); 944 X_LOCK (preadwritelock);
792 ooffset = lseek (fd, 0, SEEK_CUR); 945 ooffset = lseek (fd, 0, SEEK_CUR);
793 lseek (fd, offset, SEEK_SET); 946 lseek (fd, offset, SEEK_SET);
… …
796 X_UNLOCK (preadwritelock); 949 X_UNLOCK (preadwritelock);
797 950
798 return res; 951 return res;
799} 952}
800 953
801static ssize_t 954static eio_ssize_t
802eio__pwrite (int fd, void *buf, size_t count, off_t offset) 955eio__pwrite (int fd, void *buf, size_t count, off_t offset)
803{ 956{
804 ssize_t res; 957 eio_ssize_t res;
805 off_t ooffset; 958 off_t ooffset;
806 959
807 X_LOCK (preadwritelock); 960 X_LOCK (preadwritelock);
808 ooffset = lseek (fd, 0, SEEK_CUR); 961 ooffset = lseek (fd, 0, SEEK_CUR);
809 lseek (fd, offset, SEEK_SET); 962 lseek (fd, offset, SEEK_SET);
… …
813 966
814 return res; 967 return res;
815} 968}
816#endif 969#endif
817 970
818#ifndef HAVE_FUTIMES 971#ifndef HAVE_UTIMES
819 972
820# undef utimes 973# undef utimes
821# undef futimes
822# define utimes(path,times) eio__utimes (path, times) 974# define utimes(path,times) eio__utimes (path, times)
823# define futimes(fd,times) eio__futimes (fd, times)
824 975
825static int 976static int
826eio__utimes (const char *filename, const struct timeval times[2]) 977eio__utimes (const char *filename, const struct timeval times[2])
827{ 978{
828 if (times) 979 if (times)
… …
836 } 987 }
837 else 988 else
838 return utime (filename, 0); 989 return utime (filename, 0);
839} 990}
840 991
992#endif
993
994#ifndef HAVE_FUTIMES
995
996# undef futimes
997# define futimes(fd,times) eio__futimes (fd, times)
998
999static int
841static int eio__futimes (int fd, const struct timeval tv[2]) 1000eio__futimes (int fd, const struct timeval tv[2])
842{ 1001{
843 errno = ENOSYS; 1002 errno = ENOSYS;
844 return -1; 1003 return -1;
845} 1004}
846 1005
… …
849#if !HAVE_FDATASYNC 1008#if !HAVE_FDATASYNC
850# undef fdatasync 1009# undef fdatasync
851# define fdatasync(fd) fsync (fd) 1010# define fdatasync(fd) fsync (fd)
852#endif 1011#endif
853 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
854/* sync_file_range always needs emulation */ 1031/* sync_file_range always needs emulation */
855int 1032static int
856eio__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)
857{ 1034{
858#if HAVE_SYNC_FILE_RANGE 1035#if HAVE_SYNC_FILE_RANGE
859 int res; 1036 int res;
860 1037
… …
877 /* 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
878 * call fdatasync, as that matches the expectation of its users best */ 1055 * call fdatasync, as that matches the expectation of its users best */
879 return fdatasync (fd); 1056 return fdatasync (fd);
880} 1057}
881 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
1068}
1069
882#if !HAVE_READAHEAD 1070#if !HAVE_READAHEAD
883# undef readahead 1071# undef readahead
884# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 1072# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
885 1073
886static ssize_t 1074static eio_ssize_t
887eio__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)
888{ 1076{
889 size_t todo = count; 1077 size_t todo = count;
890 dBUF; 1078 dBUF;
891 1079
… …
896 pread (fd, eio_buf, len, offset); 1084 pread (fd, eio_buf, len, offset);
897 offset += len; 1085 offset += len;
898 todo -= len; 1086 todo -= len;
899 } 1087 }
900 1088
1089 FUBd;
1090
901 errno = 0; 1091 errno = 0;
902 return count; 1092 return count;
903} 1093}
904 1094
905#endif 1095#endif
906 1096
907/* sendfile always needs emulation */ 1097/* sendfile always needs emulation */
908static ssize_t 1098static eio_ssize_t
909eio__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)
910{ 1100{
1101 eio_ssize_t written = 0;
911 ssize_t res; 1102 eio_ssize_t res;
912 1103
913 if (!count) 1104 if (!count)
914 return 0; 1105 return 0;
915 1106
1107 for (;;)
1108 {
1109#ifdef __APPLE__
1110# undef HAVE_SENDFILE /* broken, as everything on os x */
1111#endif
916#if HAVE_SENDFILE 1112#if HAVE_SENDFILE
917# if __linux 1113# if __linux
1114 off_t soffset = offset;
918 res = sendfile (ofd, ifd, &offset, count); 1115 res = sendfile (ofd, ifd, &soffset, count);
919 1116
920# elif __FreeBSD__ 1117# elif __FreeBSD__
921 /* 1118 /*
922 * 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
923 * wasted on making it similar to other I/O functions. 1120 * wasted on making it similar to other I/O functions.
924 */ 1121 */
925 {
926 off_t sbytes; 1122 off_t sbytes;
927 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 1123 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
928 1124
929 #if 0 /* according to the manpage, this is correct, but broken behaviour */ 1125 #if 0 /* according to the manpage, this is correct, but broken behaviour */
930 /* freebsd' sendfile will return 0 on success */ 1126 /* freebsd' sendfile will return 0 on success */
931 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */ 1127 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
932 /* not on e.g. EIO or EPIPE - sounds broken */ 1128 /* not on e.g. EIO or EPIPE - sounds broken */
933 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0) 1129 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
934 res = sbytes; 1130 res = sbytes;
935 #endif 1131 #endif
936 1132
937 /* according to source inspection, this is correct, and useful behaviour */ 1133 /* according to source inspection, this is correct, and useful behaviour */
938 if (sbytes) 1134 if (sbytes)
939 res = sbytes; 1135 res = sbytes;
940 }
941 1136
942# elif defined (__APPLE__) 1137# elif defined (__APPLE__)
943
944 {
945 off_t sbytes = count; 1138 off_t sbytes = count;
946 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0); 1139 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
947 1140
948 /* according to the manpage, sbytes is always valid */ 1141 /* according to the manpage, sbytes is always valid */
949 if (sbytes) 1142 if (sbytes)
950 res = sbytes; 1143 res = sbytes;
951 }
952 1144
953# elif __hpux 1145# elif __hpux
954 res = sendfile (ofd, ifd, offset, count, 0, 0); 1146 res = sendfile (ofd, ifd, offset, count, 0, 0);
955 1147
956# elif __solaris 1148# elif __solaris
957 {
958 struct sendfilevec vec; 1149 struct sendfilevec vec;
959 size_t sbytes; 1150 size_t sbytes;
960 1151
961 vec.sfv_fd = ifd; 1152 vec.sfv_fd = ifd;
962 vec.sfv_flag = 0; 1153 vec.sfv_flag = 0;
963 vec.sfv_off = offset; 1154 vec.sfv_off = offset;
964 vec.sfv_len = count; 1155 vec.sfv_len = count;
965 1156
966 res = sendfilev (ofd, &vec, 1, &sbytes); 1157 res = sendfilev (ofd, &vec, 1, &sbytes);
967 1158
968 if (res < 0 && sbytes) 1159 if (res < 0 && sbytes)
969 res = sbytes; 1160 res = sbytes;
970 }
971 1161
972# endif 1162# endif
973 1163
974#elif defined (_WIN32) 1164#elif defined (_WIN32) && 0
975
976 /* does not work, just for documentation of what would need to be done */ 1165 /* does not work, just for documentation of what would need to be done */
977 { 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 */
978 HANDLE h = TO_SOCKET (ifd); 1169 HANDLE h = TO_SOCKET (ifd);
979 SetFilePointer (h, offset, 0, FILE_BEGIN); 1170 SetFilePointer (h, offset, 0, FILE_BEGIN);
980 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0); 1171 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
981 }
982 1172
983#else 1173#else
984 res = -1; 1174 res = -1;
985 errno = ENOSYS; 1175 errno = ENOSYS;
986#endif 1176#endif
987 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
988 if (res < 0 1201 if (res < 0
989 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1202 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
990 /* BSDs */ 1203 /* BSDs */
991#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */ 1204#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
992 || errno == ENOTSUP 1205 || errno == ENOTSUP
993#endif 1206#endif
1207#ifdef EOPNOTSUPP /* windows */
994 || errno == EOPNOTSUPP /* BSDs */ 1208 || errno == EOPNOTSUPP /* BSDs */
1209#endif
995#if __solaris 1210#if __solaris
996 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1211 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
997#endif 1212#endif
998 ) 1213 )
999 ) 1214 )
… …
1003 1218
1004 res = 0; 1219 res = 0;
1005 1220
1006 while (count) 1221 while (count)
1007 { 1222 {
1008 ssize_t cnt; 1223 eio_ssize_t cnt;
1009 1224
1010 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);
1011 1226
1012 if (cnt <= 0) 1227 if (cnt <= 0)
1013 { 1228 {
… …
1025 1240
1026 offset += cnt; 1241 offset += cnt;
1027 res += cnt; 1242 res += cnt;
1028 count -= cnt; 1243 count -= cnt;
1029 } 1244 }
1245
1246 FUBd;
1030 } 1247 }
1031 1248
1032 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;
1033} 1530}
1034 1531
1035static signed char 1532static signed char
1036eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) 1533eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1037{ 1534{
1038 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */ 1535 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1039 : a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0; 1536 : a->inode < b->inode ? -1
1537 : a->inode > b->inode ? 1
1538 : 0;
1040} 1539}
1041 1540
1042#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0 1541#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1043 1542
1044#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */ 1543#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1045#define EIO_SORT_FAST 60 /* when to only use insertion sort */ 1544#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1046 1545
1047static void 1546static void
1048eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1547eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1049{ 1548{
1050 unsigned char bits [9 + sizeof (ino_t) * 8]; 1549 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1051 unsigned char *bit = bits; 1550 unsigned char *bit = bits;
1052 1551
1053 assert (CHAR_BIT == 8); 1552 assert (CHAR_BIT == 8);
1054 assert (sizeof (eio_dirent) * 8 < 256); 1553 assert (sizeof (eio_dirent) * 8 < 256);
1055 assert (offsetof (eio_dirent, inode)); /* we use 0 as sentinel */ 1554 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1056 assert (offsetof (eio_dirent, score)); /* we use 0 as sentinel */ 1555 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1057 1556
1058 if (size <= EIO_SORT_FAST) 1557 if (size <= EIO_SORT_FAST)
1059 return; 1558 return;
1060 1559
1061 /* first prepare an array of bits to test in our radix sort */ 1560 /* first prepare an array of bits to test in our radix sort */
1062 /* try to take endianness into account, as well as differences in ino_t sizes */ 1561 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1063 /* inode_bits must contain all inodes ORed together */ 1562 /* inode_bits must contain all inodes ORed together */
1064 /* which is used to skip bits that are 0 everywhere, which is very common */ 1563 /* which is used to skip bits that are 0 everywhere, which is very common */
1065 { 1564 {
1066 ino_t endianness; 1565 eio_ino_t endianness;
1067 int i, j; 1566 int i, j;
1068 1567
1069 /* we store the byte offset of byte n into byte n of "endianness" */ 1568 /* we store the byte offset of byte n into byte n of "endianness" */
1070 for (i = 0; i < sizeof (ino_t); ++i) 1569 for (i = 0; i < sizeof (eio_ino_t); ++i)
1071 ((unsigned char *)&endianness)[i] = i; 1570 ((unsigned char *)&endianness)[i] = i;
1072 1571
1073 *bit++ = 0; 1572 *bit++ = 0;
1074 1573
1075 for (i = 0; i < sizeof (ino_t); ++i) 1574 for (i = 0; i < sizeof (eio_ino_t); ++i)
1076 { 1575 {
1077 /* shifting off the byte offsets out of "endianness" */ 1576 /* shifting off the byte offsets out of "endianness" */
1078 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8; 1577 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1079 endianness >>= 8; 1578 endianness >>= 8;
1080 1579
1081 for (j = 0; j < 8; ++j) 1580 for (j = 0; j < 8; ++j)
1082 if (inode_bits & (((ino_t)1) << (i * 8 + j))) 1581 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1083 *bit++ = offs + j; 1582 *bit++ = offs + j;
1084 } 1583 }
1085 1584
1086 for (j = 0; j < 8; ++j) 1585 for (j = 0; j < 8; ++j)
1087 if (score_bits & (1 << j)) 1586 if (score_bits & (1 << j))
1088 *bit++ = offsetof (eio_dirent, score) * 8 + j; 1587 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1089 } 1588 }
1090 1589
1091 /* now actually do the sorting (a variant of MSD radix sort) */ 1590 /* now actually do the sorting (a variant of MSD radix sort) */
1092 { 1591 {
1093 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base; 1592 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1094 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end; 1593 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1095 unsigned char *bit_stk [9 + sizeof (ino_t) * 8]; 1594 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1096 int stk_idx = 0; 1595 int stk_idx = 0;
1097 1596
1098 base_stk [stk_idx] = dents; 1597 base_stk [stk_idx] = dents;
1099 end_stk [stk_idx] = dents + size; 1598 end_stk [stk_idx] = dents + size;
1100 bit_stk [stk_idx] = bit - 1; 1599 bit_stk [stk_idx] = bit - 1;
… …
1179 } 1678 }
1180 } 1679 }
1181} 1680}
1182 1681
1183static void 1682static void
1184eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1683eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1185{ 1684{
1186 if (size <= 1) 1685 if (size <= 1)
1187 return; /* our insertion sort relies on size > 0 */ 1686 return; /* our insertion sort relies on size > 0 */
1188 1687
1189 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */ 1688 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
… …
1195 eio_dent_insertion_sort (dents, size); 1694 eio_dent_insertion_sort (dents, size);
1196} 1695}
1197 1696
1198/* read a full directory */ 1697/* read a full directory */
1199static void 1698static void
1200eio__scandir (eio_req *req, etp_worker *self) 1699eio__scandir (eio_req *req)
1201{ 1700{
1202 DIR *dirp;
1203 EIO_STRUCT_DIRENT *entp;
1204 char *name, *names; 1701 char *name, *names;
1205 int namesalloc = 4096; 1702 int namesalloc = 4096 - sizeof (void *) * 4;
1206 int namesoffs = 0; 1703 int namesoffs = 0;
1207 int flags = req->int1; 1704 int flags = req->int1;
1208 eio_dirent *dents = 0; 1705 eio_dirent *dents = 0;
1209 int dentalloc = 128; 1706 int dentalloc = 128;
1210 int dentoffs = 0; 1707 int dentoffs = 0;
1211 ino_t inode_bits = 0; 1708 eio_ino_t inode_bits = 0;
1709#ifdef _WIN32
1710 HANDLE dirp;
1711 WIN32_FIND_DATA entp;
1712#else
1713 DIR *dirp;
1714 EIO_STRUCT_DIRENT *entp;
1715#endif
1212 1716
1213 req->result = -1; 1717 req->result = -1;
1214 1718
1215 if (!(flags & EIO_READDIR_DENTS)) 1719 if (!(flags & EIO_READDIR_DENTS))
1216 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); 1720 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1217 1721
1218 X_LOCK (wrklock); 1722#ifdef _WIN32
1219 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 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)
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
1220 self->dirp = dirp = opendir (req->ptr1); 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
1221 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1776 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1222 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; 1777 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1223 req->ptr2 = names = malloc (namesalloc); 1778 req->ptr2 = names = malloc (namesalloc);
1224 X_UNLOCK (wrklock);
1225 1779
1226 if (dirp && names && (!flags || dents)) 1780 if (!names || (flags && !dents))
1781 return;
1782
1227 for (;;) 1783 for (;;)
1228 { 1784 {
1785 int done;
1786
1787#ifdef _WIN32
1788 done = !dirp;
1789#else
1229 errno = 0; 1790 errno = 0;
1230 entp = readdir (dirp); 1791 entp = readdir (dirp);
1792 done = !entp;
1793#endif
1231 1794
1232 if (!entp) 1795 if (done)
1233 { 1796 {
1797#ifndef _WIN32
1798 int old_errno = errno;
1799 closedir (dirp);
1800 errno = old_errno;
1801
1234 if (errno) 1802 if (errno)
1235 break; 1803 break;
1804#endif
1236 1805
1237 /* sort etc. */ 1806 /* sort etc. */
1238 req->int1 = flags; 1807 req->int1 = flags;
1239 req->result = dentoffs; 1808 req->result = dentoffs;
1240 1809
1241 if (flags & EIO_READDIR_STAT_ORDER) 1810 if (flags & EIO_READDIR_STAT_ORDER)
1242 eio_dent_sort (dents, dentoffs, 0, inode_bits); /* sort by inode exclusively */ 1811 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1243 else if (flags & EIO_READDIR_DIRS_FIRST) 1812 else if (flags & EIO_READDIR_DIRS_FIRST)
1244 if (flags & EIO_READDIR_FOUND_UNKNOWN) 1813 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1245 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */ 1814 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1246 else 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
1839 break;
1840 }
1841
1842 /* now add the entry to our list(s) */
1843 name = D_NAME (entp);
1844
1845 /* skip . and .. entries */
1846 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1847 {
1848 int len = D_NAMLEN (entp) + 1;
1849
1850 while (ecb_expect_false (namesoffs + len > namesalloc))
1851 {
1852 namesalloc *= 2;
1853 req->ptr2 = names = realloc (names, namesalloc);
1854
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))
1247 { 1866 {
1248 /* in this case, all is known, and we just put dirs first and sort them */ 1867 dentalloc *= 2;
1868 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1869
1870 if (!dents)
1871 break;
1872 }
1873
1249 eio_dirent *oth = dents + dentoffs; 1874 ent = dents + dentoffs;
1250 eio_dirent *dir = dents;
1251 1875
1252 /* now partition dirs to the front, and non-dirs to the back */ 1876 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1253 /* by walking from both sides and swapping if necessary */ 1877 ent->namelen = len - 1;
1254 /* also clear score, so it doesn't influence sorting */ 1878 ent->inode = D_INO (entp);
1255 while (oth > dir) 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)
1256 { 1938 {
1257 if (dir->type == EIO_DT_DIR) 1939 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1258 ++dir; 1940 ent->score = 1;
1259 else if ((--oth)->type == EIO_DT_DIR) 1941 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1260 { 1942 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1261 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1262
1263 ++dir;
1264 }
1265 } 1943 }
1266 1944 else if (ent->type == EIO_DT_DIR)
1267 /* now sort the dirs only */ 1945 ent->score = 0;
1268 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1269 } 1946 }
1270
1271 break;
1272 }
1273
1274 /* now add the entry to our list(s) */
1275 name = entp->d_name;
1276
1277 /* skip . and .. entries */
1278 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1279 {
1280 int len = D_NAMLEN (entp) + 1;
1281
1282 while (expect_false (namesoffs + len > namesalloc))
1283 {
1284 namesalloc *= 2;
1285 X_LOCK (wrklock);
1286 req->ptr2 = names = realloc (names, namesalloc);
1287 X_UNLOCK (wrklock);
1288
1289 if (!names)
1290 break;
1291 } 1947 }
1292 1948
1293 memcpy (names + namesoffs, name, len);
1294
1295 if (dents)
1296 {
1297 struct eio_dirent *ent;
1298
1299 if (expect_false (dentoffs == dentalloc))
1300 {
1301 dentalloc *= 2;
1302 X_LOCK (wrklock);
1303 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1304 X_UNLOCK (wrklock);
1305
1306 if (!dents)
1307 break;
1308 }
1309
1310 ent = dents + dentoffs;
1311
1312 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1313 ent->namelen = len - 1;
1314 ent->inode = D_INO (entp);
1315
1316 inode_bits |= ent->inode;
1317
1318 switch (D_TYPE (entp))
1319 {
1320 default:
1321 ent->type = EIO_DT_UNKNOWN;
1322 flags |= EIO_READDIR_FOUND_UNKNOWN;
1323 break;
1324
1325 #ifdef DT_FIFO
1326 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1327 #endif
1328 #ifdef DT_CHR
1329 case DT_CHR: ent->type = EIO_DT_CHR; break;
1330 #endif
1331 #ifdef DT_MPC
1332 case DT_MPC: ent->type = EIO_DT_MPC; break;
1333 #endif
1334 #ifdef DT_DIR
1335 case DT_DIR: ent->type = EIO_DT_DIR; break;
1336 #endif
1337 #ifdef DT_NAM
1338 case DT_NAM: ent->type = EIO_DT_NAM; break;
1339 #endif
1340 #ifdef DT_BLK
1341 case DT_BLK: ent->type = EIO_DT_BLK; break;
1342 #endif
1343 #ifdef DT_MPB
1344 case DT_MPB: ent->type = EIO_DT_MPB; break;
1345 #endif
1346 #ifdef DT_REG
1347 case DT_REG: ent->type = EIO_DT_REG; break;
1348 #endif
1349 #ifdef DT_NWK
1350 case DT_NWK: ent->type = EIO_DT_NWK; break;
1351 #endif
1352 #ifdef DT_CMP
1353 case DT_CMP: ent->type = EIO_DT_CMP; break;
1354 #endif
1355 #ifdef DT_LNK
1356 case DT_LNK: ent->type = EIO_DT_LNK; break;
1357 #endif
1358 #ifdef DT_SOCK
1359 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1360 #endif
1361 #ifdef DT_DOOR
1362 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1363 #endif
1364 #ifdef DT_WHT
1365 case DT_WHT: ent->type = EIO_DT_WHT; break;
1366 #endif
1367 }
1368
1369 ent->score = 7;
1370
1371 if (flags & EIO_READDIR_DIRS_FIRST)
1372 {
1373 if (ent->type == EIO_DT_UNKNOWN)
1374 {
1375 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1376 ent->score = 1;
1377 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1378 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1379 }
1380 else if (ent->type == EIO_DT_DIR)
1381 ent->score = 0;
1382 }
1383 }
1384
1385 namesoffs += len; 1949 namesoffs += len;
1386 ++dentoffs; 1950 ++dentoffs;
1387 } 1951 }
1388 1952
1389 if (EIO_CANCELLED (req)) 1953 if (EIO_CANCELLED (req))
1390 { 1954 {
1391 errno = ECANCELED; 1955 errno = ECANCELED;
1392 break; 1956 break;
1393 } 1957 }
1958
1959#ifdef _WIN32
1960 if (!FindNextFile (dirp, &entp))
1961 {
1962 FindClose (dirp);
1963 dirp = 0;
1394 } 1964 }
1965#endif
1966 }
1395} 1967}
1396 1968
1397#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) 1969/*****************************************************************************/
1398# define eio__msync(a,b,c) ((errno = ENOSYS), -1) 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)
1982{
1983 int fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1984
1985 return fd >= 0 ? (eio_wd)(long)(fd + 1) : EIO_INVALID_WD;
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
1399#else 2032#else
1400 2033
1401int 2034/* on legacy systems, we represent the working directories simply by their path strings */
1402eio__msync (void *mem, size_t len, int flags) 2035
2036static const char *
2037wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
1403{ 2038{
1404 if (EIO_MS_ASYNC != MS_SYNC 2039 if (!wd || *path == '/')
1405 || EIO_MS_INVALIDATE != MS_INVALIDATE 2040 return path;
1406 || EIO_MS_SYNC != MS_SYNC) 2041
2042 {
2043 int l1 = strlen ((const char *)wd);
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;
2063 else
1407 { 2064 {
1408 flags = 0 2065 int len = eio__realpath (tmpbuf, wd, path);
1409 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1410 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1411 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1412 }
1413 2066
1414 return msync (mem, len, flags); 2067 path = EIO_INVALID_WD;
1415}
1416 2068
1417#endif 2069 if (len >= 0)
2070 {
2071 ((char *)tmpbuf->ptr)[len] = 0;
2072 path = strdup (tmpbuf->ptr);
2073 }
2074 }
1418 2075
1419int
1420eio__mtouch (void *mem, size_t len, int flags)
1421{
1422 intptr_t addr = (intptr_t)mem;
1423 intptr_t end = addr + len;
1424#ifdef PAGESIZE
1425 const intptr_t page = PAGESIZE;
1426#else
1427 static intptr_t page;
1428
1429 if (!page) 2076 if (!path)
1430 page = sysconf (_SC_PAGESIZE); 2077 path = EIO_INVALID_WD;
1431#endif
1432 2078
1433 /* round down to start of page, although this is probably useless */ 2079 return (eio_wd)path;
1434 addr &= ~(page - 1); /* assume page size is always a power of two */ 2080}
1435 2081
1436 if (addr < end) 2082eio_wd
1437 if (flags & EIO_MT_MODIFY) /* modify */ 2083eio_wd_open_sync (eio_wd wd, const char *path)
1438 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len); 2084{
1439 else 2085 struct tmpbuf tmpbuf = { 0 };
1440 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len); 2086 wd = eio__wd_open_sync (&tmpbuf, wd, path);
2087 free (tmpbuf.ptr);
1441 2088
1442 return 0; 2089 return wd;
1443} 2090}
2091
2092void
2093eio_wd_close_sync (eio_wd wd)
2094{
2095 if (wd != EIO_INVALID_WD)
2096 free (wd);
2097}
2098
2099#endif
1444 2100
1445/*****************************************************************************/ 2101/*****************************************************************************/
1446 2102
1447#define ALLOC(len) \ 2103#define ALLOC(len) \
1448 if (!req->ptr2) \ 2104 if (!req->ptr2) \
… …
1463{ 2119{
1464 ETP_REQ *req; 2120 ETP_REQ *req;
1465 struct timespec ts; 2121 struct timespec ts;
1466 etp_worker *self = (etp_worker *)thr_arg; 2122 etp_worker *self = (etp_worker *)thr_arg;
1467 2123
2124#if HAVE_PRCTL_SET_NAME
2125 prctl (PR_SET_NAME, (unsigned long)"eio_thread", 0, 0, 0);
2126#endif
2127
1468 /* try to distribute timeouts somewhat randomly */ 2128 /* try to distribute timeouts somewhat evenly */
1469 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 2129 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1470 2130
1471 for (;;) 2131 for (;;)
1472 { 2132 {
2133 ts.tv_sec = 0;
2134
1473 X_LOCK (reqlock); 2135 X_LOCK (reqlock);
1474 2136
1475 for (;;) 2137 for (;;)
1476 { 2138 {
1477 self->req = req = reqq_shift (&req_queue); 2139 req = reqq_shift (&req_queue);
1478 2140
1479 if (req) 2141 if (req)
1480 break; 2142 break;
1481 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
1482 ++idle; 2153 ++idle;
1483 2154
1484 ts.tv_sec = time (0) + IDLE_TIMEOUT; 2155 if (idle <= max_idle)
1485 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
1486 { 2159 {
1487 if (idle > max_idle) 2160 /* initialise timeout once */
1488 { 2161 if (!ts.tv_sec)
1489 --idle; 2162 ts.tv_sec = time (0) + idle_timeout;
1490 X_UNLOCK (reqlock);
1491 X_LOCK (wrklock);
1492 --started;
1493 X_UNLOCK (wrklock);
1494 goto quit;
1495 }
1496 2163
1497 /* we are allowed to idle, so do so without any timeout */
1498 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.,.. */
1499 } 2166 }
1500 2167
1501 --idle; 2168 --idle;
1502 } 2169 }
1503 2170
… …
1506 X_UNLOCK (reqlock); 2173 X_UNLOCK (reqlock);
1507 2174
1508 if (req->type < 0) 2175 if (req->type < 0)
1509 goto quit; 2176 goto quit;
1510 2177
1511 if (!EIO_CANCELLED (req))
1512 ETP_EXECUTE (self, req); 2178 ETP_EXECUTE (self, req);
1513 2179
1514 X_LOCK (reslock); 2180 X_LOCK (reslock);
1515 2181
1516 ++npending; 2182 ++npending;
1517 2183
1518 if (!reqq_push (&res_queue, req) && want_poll_cb) 2184 if (!reqq_push (&res_queue, req) && want_poll_cb)
1519 want_poll_cb (); 2185 want_poll_cb ();
1520 2186
1521 self->req = 0;
1522 etp_worker_clear (self); 2187 etp_worker_clear (self);
1523 2188
1524 X_UNLOCK (reslock); 2189 X_UNLOCK (reslock);
1525 } 2190 }
1526 2191
1527quit: 2192quit:
2193 free (req);
2194
1528 X_LOCK (wrklock); 2195 X_LOCK (wrklock);
1529 etp_worker_free (self); 2196 etp_worker_free (self);
1530 X_UNLOCK (wrklock); 2197 X_UNLOCK (wrklock);
1531 2198
1532 return 0; 2199 return 0;
1533} 2200}
1534 2201
1535/*****************************************************************************/ 2202/*****************************************************************************/
1536 2203
2204int ecb_cold
1537int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 2205eio_init (void (*want_poll)(void), void (*done_poll)(void))
1538{ 2206{
2207#if !HAVE_PREADWRITE
2208 X_MUTEX_CREATE (preadwritelock);
2209#endif
2210
1539 return etp_init (want_poll, done_poll); 2211 return etp_init (want_poll, done_poll);
1540} 2212}
1541 2213
2214ecb_inline void
1542static void eio_api_destroy (eio_req *req) 2215eio_api_destroy (eio_req *req)
1543{ 2216{
1544 free (req); 2217 free (req);
1545} 2218}
1546 2219
1547#define REQ(rtype) \ 2220#define REQ(rtype) \
… …
1566 { \ 2239 { \
1567 eio_api_destroy (req); \ 2240 eio_api_destroy (req); \
1568 return 0; \ 2241 return 0; \
1569 } 2242 }
1570 2243
2244static void
1571static void eio_execute (etp_worker *self, eio_req *req) 2245eio_execute (etp_worker *self, eio_req *req)
1572{ 2246{
1573 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 }
1574 2275
1575 switch (req->type) 2276 switch (req->type)
1576 { 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
1577 case EIO_READ: ALLOC (req->size); 2284 case EIO_READ: ALLOC (req->size);
1578 req->result = req->offs >= 0 2285 req->result = req->offs >= 0
1579 ? pread (req->int1, req->ptr2, req->size, req->offs) 2286 ? pread (req->int1, req->ptr2, req->size, req->offs)
1580 : read (req->int1, req->ptr2, req->size); break; 2287 : read (req->int1, req->ptr2, req->size); break;
1581 case EIO_WRITE: req->result = req->offs >= 0 2288 case EIO_WRITE: req->result = req->offs >= 0
1582 ? pwrite (req->int1, req->ptr2, req->size, req->offs) 2289 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1583 : write (req->int1, req->ptr2, req->size); break; 2290 : write (req->int1, req->ptr2, req->size); break;
1584 2291
1585 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;
1586 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
1587 2296
1588 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2297 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1589 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;
1590 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2299 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1591 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;
1592 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2301 case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break;
1593 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; 2302 case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break;
2303 case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break;
2304 case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break;
1594 2305
2306 case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
2307 case EIO_RMDIR: req->result = unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break;
2308 case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); break;
2309 case EIO_RENAME: req->result = renameat (dirfd, req->ptr1, WD2FD (req->int3), req->ptr2); break;
2310 case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD (req->int3), req->ptr2, 0); break;
2311 case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
2312 case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
2313 case EIO_READLINK: ALLOC (PATH_MAX);
2314 req->result = readlinkat (dirfd, req->ptr1, req->ptr2, PATH_MAX); break;
1595 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS)); 2315 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1596 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break; 2316 req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2317 case EIO_UTIME:
2318 case EIO_FUTIME:
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
2341#else
2342
1597 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS)); 2343 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1598 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break; 2344 req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
1599 2345 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2346 req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
1600 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; 2347 case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
1601 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1602 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break; 2348 case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
1603 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1604 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break; 2349 case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
1605 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1606
1607 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break; 2350 case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
1608 case EIO_CLOSE: req->result = close (req->int1); break; 2351
1609 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1610 case EIO_UNLINK: req->result = unlink (req->ptr1); break; 2352 case EIO_UNLINK: req->result = unlink (path ); break;
1611 case EIO_RMDIR: req->result = rmdir (req->ptr1); break; 2353 case EIO_RMDIR: req->result = rmdir (path ); break;
1612 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; 2354 case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
1613 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; 2355 case EIO_RENAME: req->result = rename (path , req->ptr2); break;
1614 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; 2356 case EIO_LINK: req->result = link (path , req->ptr2); break;
1615 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; 2357 case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
1616 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break; 2358 case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
1617
1618 case EIO_READLINK: ALLOC (PATH_MAX); 2359 case EIO_READLINK: ALLOC (PATH_MAX);
1619 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break; 2360 req->result = readlink (path, req->ptr2, PATH_MAX); break;
1620 2361 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1621 case EIO_SYNC: req->result = 0; sync (); break; 2362 req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
1622 case EIO_FSYNC: req->result = fsync (req->int1); break;
1623 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1624 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1625 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;
1626 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1627
1628 case EIO_READDIR: eio__scandir (req, self); break;
1629
1630 case EIO_BUSY:
1631#ifdef _WIN32
1632 Sleep (req->nv1 * 1000.);
1633#else
1634 {
1635 struct timeval tv;
1636
1637 tv.tv_sec = req->nv1;
1638 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;
1639
1640 req->result = select (0, 0, 0, 0, &tv);
1641 }
1642#endif
1643 break;
1644 2363
1645 case EIO_UTIME: 2364 case EIO_UTIME:
1646 case EIO_FUTIME: 2365 case EIO_FUTIME:
1647 { 2366 {
1648 struct timeval tv[2]; 2367 struct timeval tv[2];
1649 struct timeval *times; 2368 struct timeval *times;
1650 2369
1651 if (req->nv1 != -1. || req->nv2 != -1.) 2370 if (req->nv1 != -1. || req->nv2 != -1.)
1652 { 2371 {
1653 tv[0].tv_sec = req->nv1; 2372 tv[0].tv_sec = req->nv1;
1654 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.; 2373 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
1655 tv[1].tv_sec = req->nv2; 2374 tv[1].tv_sec = req->nv2;
1656 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.; 2375 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
1657 2376
1658 times = tv; 2377 times = tv;
1659 } 2378 }
1660 else 2379 else
1661 times = 0; 2380 times = 0;
1662
1663 2381
1664 req->result = req->type == EIO_FUTIME 2382 req->result = req->type == EIO_FUTIME
1665 ? futimes (req->int1, times) 2383 ? futimes (req->int1, times)
1666 : utimes (req->ptr1, times); 2384 : utimes (req->ptr1, times);
1667 } 2385 }
1668 break; 2386 break;
1669 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
1670 case EIO_GROUP: 2437 case EIO_GROUP:
1671 abort (); /* handled in eio_request */ 2438 abort (); /* handled in eio_request */
1672 2439
1673 case EIO_NOP: 2440 case EIO_NOP:
1674 req->result = 0; 2441 req->result = 0;
1675 break; 2442 break;
1676 2443
1677 case EIO_CUSTOM: 2444 case EIO_CUSTOM:
1678 ((void (*)(eio_req *))req->feed) (req); 2445 req->feed (req);
1679 break; 2446 break;
1680 2447
1681 default: 2448 default:
2449 errno = ENOSYS;
1682 req->result = -1; 2450 req->result = -1;
1683 break; 2451 break;
1684 } 2452 }
1685 2453
1686 req->errorno = errno; 2454 req->errorno = errno;
… …
1711eio_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)
1712{ 2480{
1713 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;
1714} 2482}
1715 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
1716eio_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)
1717{ 2500{
1718 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;
1719} 2502}
1720 2503
1721eio_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)
1722{ 2505{
1723 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;
1724} 2507}
1725 2508
1726eio_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)
1727{ 2510{
1728 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;
1729} 2517}
1730 2518
1731eio_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)
1732{ 2520{
1733 REQ (EIO_CLOSE); req->int1 = fd; SEND; 2521 REQ (EIO_CLOSE); req->int1 = fd; SEND;
… …
1771eio_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)
1772{ 2560{
1773 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2561 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1774} 2562}
1775 2563
1776eio_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)
1777{ 2565{
1778 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;
1779} 2567}
1780 2568
1781eio_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)
… …
1801eio_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)
1802{ 2590{
1803 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2591 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1804} 2592}
1805 2593
1806eio_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)
1807{ 2595{
1808 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;
1809} 2597}
1810 2598
1811eio_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)
… …
1827eio_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)
1828{ 2616{
1829 return eio__1path (EIO_READLINK, path, pri, cb, data); 2617 return eio__1path (EIO_READLINK, path, pri, cb, data);
1830} 2618}
1831 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
1832eio_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)
1833{ 2626{
1834 return eio__1path (EIO_STAT, path, pri, cb, data); 2627 return eio__1path (EIO_STAT, path, pri, cb, data);
1835} 2628}
1836 2629
… …
1859 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND; 2652 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1860} 2653}
1861 2654
1862eio_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)
1863{ 2656{
1864 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;
1865} 2658}
1866 2659
1867static eio_req * 2660static eio_req *
1868eio__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)
1869{ 2662{
… …
1893eio_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)
1894{ 2687{
1895 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2688 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1896} 2689}
1897 2690
1898eio_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)
1899{ 2692{
1900 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND; 2693 REQ (EIO_CUSTOM); req->feed = execute; SEND;
1901} 2694}
1902 2695
1903#endif 2696#endif
1904 2697
1905eio_req *eio_grp (eio_cb cb, void *data) 2698eio_req *eio_grp (eio_cb cb, void *data)
… …
1914#undef SEND 2707#undef SEND
1915 2708
1916/*****************************************************************************/ 2709/*****************************************************************************/
1917/* grp functions */ 2710/* grp functions */
1918 2711
2712void
1919void 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)
1920{ 2714{
1921 grp->int2 = limit; 2715 grp->int2 = limit;
1922 grp->feed = feed; 2716 grp->feed = feed;
1923 2717
1924 grp_try_feed (grp); 2718 grp_try_feed (grp);
1925} 2719}
1926 2720
2721void
1927void eio_grp_limit (eio_req *grp, int limit) 2722eio_grp_limit (eio_req *grp, int limit)
1928{ 2723{
1929 grp->int2 = limit; 2724 grp->int2 = limit;
1930 2725
1931 grp_try_feed (grp); 2726 grp_try_feed (grp);
1932} 2727}
1933 2728
2729void
1934void eio_grp_add (eio_req *grp, eio_req *req) 2730eio_grp_add (eio_req *grp, eio_req *req)
1935{ 2731{
1936 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));
1937 2733
1938 grp->flags |= EIO_FLAG_GROUPADD; 2734 grp->flags |= EIO_FLAG_GROUPADD;
1939 2735
… …
1950} 2746}
1951 2747
1952/*****************************************************************************/ 2748/*****************************************************************************/
1953/* misc garbage */ 2749/* misc garbage */
1954 2750
2751eio_ssize_t
1955ssize_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)
1956{ 2753{
1957 etp_worker wrk;
1958 ssize_t ret;
1959
1960 wrk.dbuf = 0;
1961
1962 ret = eio__sendfile (ofd, ifd, offset, count, &wrk); 2754 return eio__sendfile (ofd, ifd, offset, count);
1963
1964 if (wrk.dbuf)
1965 free (wrk.dbuf);
1966
1967 return ret;
1968} 2755}
1969 2756

Diff Legend

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