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

Comparing libeio/eio.c (file contents):
Revision 1.49 by root, Sat Jan 2 14:24:32 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>
53#include <sys/stat.h> 58#include <sys/stat.h>
54#include <limits.h> 59#include <limits.h>
55#include <fcntl.h> 60#include <fcntl.h>
56#include <assert.h> 61#include <assert.h>
57 62
63#if _POSIX_VERSION >= 200809L
64# define HAVE_AT 1
65#else
66# define HAVE_AT 0
67#endif
68
69/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
70/* intptr_t only comes from stdint.h, says idiot openbsd coder */
71#if HAVE_STDINT_H
72# include <stdint.h>
73#endif
74
75#ifndef ECANCELED
76# define ECANCELED EDOM
77#endif
78#ifndef ELOOP
79# define ELOOP EDOM
80#endif
81
82#if !defined(ENOTSOCK) && defined(WSAENOTSOCK)
83# define ENOTSOCK WSAENOTSOCK
84#endif
85
86static void eio_destroy (eio_req *req);
87
58#ifndef EIO_FINISH 88#ifndef EIO_FINISH
59# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0 89# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
60#endif 90#endif
61 91
62#ifndef EIO_DESTROY 92#ifndef EIO_DESTROY
65 95
66#ifndef EIO_FEED 96#ifndef EIO_FEED
67# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0) 97# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)
68#endif 98#endif
69 99
100#ifndef EIO_FD_TO_WIN32_HANDLE
101# define EIO_FD_TO_WIN32_HANDLE(fd) _get_osfhandle (fd)
102#endif
103#ifndef EIO_WIN32_HANDLE_TO_FD
104# define EIO_WIN32_HANDLE_TO_FD(handle) _open_osfhandle (handle, 0)
105#endif
106
107#define EIO_ERRNO(errval,retval) ((errno = errval), retval)
108
109#define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1)
110
70#ifdef _WIN32 111#ifdef _WIN32
71 112
72 /*doh*/ 113 #undef PAGESIZE
114 #define PAGESIZE 4096 /* GetSystemInfo? */
115
116 /* TODO: look at how perl does stat (non-sloppy), unlink (ro-files), utime, link */
117
118 #ifdef EIO_STRUCT_STATI64
119 /* look at perl's non-sloppy stat */
120 #define stat(path,buf) _stati64 (path,buf)
121 #define fstat(fd,buf) _fstati64 (fd,buf)
122 #endif
123 #define lstat(path,buf) stat (path,buf)
124 #define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
125 #define mkdir(path,mode) _mkdir (path)
126 #define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
127
128 #define chmod(path,mode) _chmod (path, mode)
129 #define dup(fd) _dup (fd)
130 #define dup2(fd1,fd2) _dup2 (fd1, fd2)
131
132 #define fchmod(fd,mode) EIO_ENOSYS ()
133 #define chown(path,uid,gid) EIO_ENOSYS ()
134 #define fchown(fd,uid,gid) EIO_ENOSYS ()
135 #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
136 #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
137 #define mknod(path,mode,dev) EIO_ENOSYS ()
138 #define sync() EIO_ENOSYS ()
139 #define readlink(path,buf,s) EIO_ENOSYS ()
140 #define statvfs(path,buf) EIO_ENOSYS ()
141 #define fstatvfs(fd,buf) EIO_ENOSYS ()
142
143 /* rename() uses MoveFile, which fails to overwrite */
144 #define rename(old,neu) eio__rename (old, neu)
145
146 static int
147 eio__rename (const char *old, const char *neu)
148 {
149 if (MoveFileEx (old, neu, MOVEFILE_REPLACE_EXISTING))
150 return 0;
151
152 /* should steal _dosmaperr */
153 switch (GetLastError ())
154 {
155 case ERROR_FILE_NOT_FOUND:
156 case ERROR_PATH_NOT_FOUND:
157 case ERROR_INVALID_DRIVE:
158 case ERROR_NO_MORE_FILES:
159 case ERROR_BAD_NETPATH:
160 case ERROR_BAD_NET_NAME:
161 case ERROR_BAD_PATHNAME:
162 case ERROR_FILENAME_EXCED_RANGE:
163 errno = ENOENT;
164 break;
165
166 default:
167 errno = EACCES;
168 break;
169 }
170
171 return -1;
172 }
173
174 /* we could even stat and see if it exists */
175 static int
176 symlink (const char *old, const char *neu)
177 {
178 #if WINVER >= 0x0600
179 if (CreateSymbolicLink (neu, old, 1))
180 return 0;
181
182 if (CreateSymbolicLink (neu, old, 0))
183 return 0;
184 #endif
185
186 return EIO_ERRNO (ENOENT, -1);
187 }
188
189 /* POSIX API only */
190 #define CreateHardLink(neu,old,flags) 0
191 #define CreateSymbolicLink(neu,old,flags) 0
192
193 struct statvfs
194 {
195 int dummy;
196 };
197
198 #define DT_DIR EIO_DT_DIR
199 #define DT_REG EIO_DT_REG
200 #define D_NAME(entp) entp.cFileName
201 #define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
202
73#else 203#else
74 204
75# include "config.h"
76# include <sys/time.h> 205 #include <sys/time.h>
77# include <sys/select.h> 206 #include <sys/select.h>
78# include <sys/mman.h> 207 #include <sys/statvfs.h>
79# include <unistd.h> 208 #include <unistd.h>
209 #include <signal.h>
210 #include <dirent.h>
211
212 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
213 #include <sys/mman.h>
214 #endif
215
216 #define D_NAME(entp) entp->d_name
217
218 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
219 #if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
220 #define _DIRENT_HAVE_D_TYPE /* sigh */
221 #define D_INO(de) (de)->d_fileno
222 #define D_NAMLEN(de) (de)->d_namlen
223 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
224 #define D_INO(de) (de)->d_ino
225 #endif
226
227 #ifdef _D_EXACT_NAMLEN
228 #undef D_NAMLEN
229 #define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
230 #endif
231
232 #ifdef _DIRENT_HAVE_D_TYPE
233 #define D_TYPE(de) (de)->d_type
234 #endif
235
236 #ifndef EIO_STRUCT_DIRENT
237 #define EIO_STRUCT_DIRENT struct dirent
238 #endif
239
240#endif
241
242#if HAVE_UTIMES
80# include <utime.h> 243# include <utime.h>
81# include <signal.h>
82# include <dirent.h>
83
84/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
85# if __freebsd || defined __NetBSD__ || defined __OpenBSD__
86# define _DIRENT_HAVE_D_TYPE /* sigh */
87# define D_INO(de) (de)->d_fileno
88# define D_NAMLEN(de) (de)->d_namlen
89# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
90# define D_INO(de) (de)->d_ino
91# endif 244#endif
92 245
93#ifdef _D_EXACT_NAMLEN 246#if HAVE_SYS_SYSCALL_H
94# undef D_NAMLEN 247# include <sys/syscall.h>
95# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
96#endif
97
98# ifdef _DIRENT_HAVE_D_TYPE
99# define D_TYPE(de) (de)->d_type
100# endif 248#endif
101 249
102# ifndef EIO_STRUCT_DIRENT 250#if HAVE_SYS_PRCTL_H
103# define EIO_STRUCT_DIRENT struct dirent 251# include <sys/prctl.h>
104# endif
105
106#endif 252#endif
107 253
108#if HAVE_SENDFILE 254#if HAVE_SENDFILE
109# if __linux 255# if __linux
110# include <sys/sendfile.h> 256# include <sys/sendfile.h>
111# elif __freebsd || defined __APPLE__ 257# elif __FreeBSD__ || defined __APPLE__
112# include <sys/socket.h> 258# include <sys/socket.h>
113# include <sys/uio.h> 259# include <sys/uio.h>
114# elif __hpux 260# elif __hpux
115# include <sys/socket.h> 261# include <sys/socket.h>
116# elif __solaris 262# elif __solaris
125#endif 271#endif
126#ifndef D_INO 272#ifndef D_INO
127# define D_INO(de) 0 273# define D_INO(de) 0
128#endif 274#endif
129#ifndef D_NAMLEN 275#ifndef D_NAMLEN
130# define D_NAMLEN(de) strlen ((de)->d_name) 276# define D_NAMLEN(entp) strlen (D_NAME (entp))
131#endif 277#endif
132
133/* number of seconds after which an idle threads exit */
134#define IDLE_TIMEOUT 10
135 278
136/* used for struct dirent, AIX doesn't provide it */ 279/* used for struct dirent, AIX doesn't provide it */
137#ifndef NAME_MAX 280#ifndef NAME_MAX
138# define NAME_MAX 4096 281# define NAME_MAX 4096
139#endif 282#endif
140 283
284/* used for readlink etc. */
285#ifndef PATH_MAX
286# define PATH_MAX 4096
287#endif
288
141/* buffer size for various temporary buffers */ 289/* buffer size for various temporary buffers */
142#define EIO_BUFSIZE 65536 290#define EIO_BUFSIZE 65536
143 291
144#define dBUF \ 292#define dBUF \
145 char *eio_buf; \
146 ETP_WORKER_LOCK (self); \
147 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 293 char *eio_buf = malloc (EIO_BUFSIZE); \
148 ETP_WORKER_UNLOCK (self); \
149 errno = ENOMEM; \ 294 errno = ENOMEM; \
150 if (!eio_buf) \ 295 if (!eio_buf) \
151 return -1; 296 return -1
297
298#define FUBd \
299 free (eio_buf)
152 300
153#define EIO_TICKS ((1000000 + 1023) >> 10) 301#define EIO_TICKS ((1000000 + 1023) >> 10)
154 302
155/*****************************************************************************/ 303/*****************************************************************************/
156 304
157#if __GNUC__ >= 3 305struct tmpbuf
158# define expect(expr,value) __builtin_expect ((expr),(value)) 306{
159#else 307 void *ptr;
160# define expect(expr,value) (expr) 308 int len;
161#endif 309};
162 310
163#define expect_false(expr) expect ((expr) != 0, 0) 311static void *
164#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}
165 322
166/*****************************************************************************/ 323/*****************************************************************************/
167 324
168#define ETP_PRI_MIN EIO_PRI_MIN 325#define ETP_PRI_MIN EIO_PRI_MIN
169#define ETP_PRI_MAX EIO_PRI_MAX 326#define ETP_PRI_MAX EIO_PRI_MAX
175static int eio_finish (eio_req *req); 332static int eio_finish (eio_req *req);
176#define ETP_FINISH(req) eio_finish (req) 333#define ETP_FINISH(req) eio_finish (req)
177static void eio_execute (struct etp_worker *self, eio_req *req); 334static void eio_execute (struct etp_worker *self, eio_req *req);
178#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) 335#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
179 336
180#define ETP_WORKER_CLEAR(req) \
181 if (wrk->dbuf) \
182 { \
183 free (wrk->dbuf); \
184 wrk->dbuf = 0; \
185 } \
186 \
187 if (wrk->dirp) \
188 { \
189 closedir (wrk->dirp); \
190 wrk->dirp = 0; \
191 }
192
193#define ETP_WORKER_COMMON \
194 void *dbuf; \
195 DIR *dirp;
196
197/*****************************************************************************/ 337/*****************************************************************************/
198 338
199#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) 339#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
200 340
201/* calculate time difference in ~1/EIO_TICKS of a second */ 341/* calculate time difference in ~1/EIO_TICKS of a second */
342ecb_inline int
202static int tvdiff (struct timeval *tv1, struct timeval *tv2) 343tvdiff (struct timeval *tv1, struct timeval *tv2)
203{ 344{
204 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 345 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
205 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 346 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
206} 347}
207 348
211static void (*done_poll_cb) (void); 352static void (*done_poll_cb) (void);
212 353
213static unsigned int max_poll_time; /* reslock */ 354static unsigned int max_poll_time; /* reslock */
214static unsigned int max_poll_reqs; /* reslock */ 355static unsigned int max_poll_reqs; /* reslock */
215 356
216static volatile unsigned int nreqs; /* reqlock */ 357static unsigned int nreqs; /* reqlock */
217static volatile unsigned int nready; /* reqlock */ 358static unsigned int nready; /* reqlock */
218static volatile unsigned int npending; /* reqlock */ 359static unsigned int npending; /* reqlock */
219static 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 */
220 362
221static mutex_t wrklock = X_MUTEX_INIT; 363static xmutex_t wrklock;
222static mutex_t reslock = X_MUTEX_INIT; 364static xmutex_t reslock;
223static mutex_t reqlock = X_MUTEX_INIT; 365static xmutex_t reqlock;
224static cond_t reqwait = X_COND_INIT; 366static xcond_t reqwait;
225 367
226#if !HAVE_PREADWRITE 368#if !HAVE_PREADWRITE
227/* 369/*
228 * make our pread/pwrite emulation safe against themselves, but not against 370 * make our pread/pwrite emulation safe against themselves, but not against
229 * 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,
230 * but that's your problem, not mine. 372 * but that's your problem, not mine.
231 */ 373 */
232static mutex_t preadwritelock = X_MUTEX_INIT; 374static xmutex_t preadwritelock;
233#endif 375#endif
234 376
235typedef struct etp_worker 377typedef struct etp_worker
236{ 378{
379 struct tmpbuf tmpbuf;
380
237 /* locked by wrklock */ 381 /* locked by wrklock */
238 struct etp_worker *prev, *next; 382 struct etp_worker *prev, *next;
239 383
240 thread_t tid; 384 xthread_t tid;
241 385
242 /* locked by reslock, reqlock or wrklock */ 386#ifdef ETP_WORKER_COMMON
243 ETP_REQ *req; /* currently processed request */
244
245 ETP_WORKER_COMMON 387 ETP_WORKER_COMMON
388#endif
246} etp_worker; 389} etp_worker;
247 390
248static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */ 391static etp_worker wrk_first; /* NOT etp */
249 392
250#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock) 393#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
251#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock) 394#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
252 395
253/* worker threads management */ 396/* worker threads management */
254 397
398static void ecb_cold
255static void etp_worker_clear (etp_worker *wrk) 399etp_worker_clear (etp_worker *wrk)
256{ 400{
257 ETP_WORKER_CLEAR (wrk);
258} 401}
259 402
403static void ecb_cold
260static void etp_worker_free (etp_worker *wrk) 404etp_worker_free (etp_worker *wrk)
261{ 405{
406 free (wrk->tmpbuf.ptr);
407
262 wrk->next->prev = wrk->prev; 408 wrk->next->prev = wrk->prev;
263 wrk->prev->next = wrk->next; 409 wrk->prev->next = wrk->next;
264 410
265 free (wrk); 411 free (wrk);
266} 412}
267 413
268static unsigned int etp_nreqs (void) 414static unsigned int
415etp_nreqs (void)
269{ 416{
270 int retval; 417 int retval;
271 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 418 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
272 retval = nreqs; 419 retval = nreqs;
273 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 420 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
274 return retval; 421 return retval;
275} 422}
276 423
277static unsigned int etp_nready (void) 424static unsigned int
425etp_nready (void)
278{ 426{
279 unsigned int retval; 427 unsigned int retval;
280 428
281 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 429 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
282 retval = nready; 430 retval = nready;
283 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 431 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
284 432
285 return retval; 433 return retval;
286} 434}
287 435
288static unsigned int etp_npending (void) 436static unsigned int
437etp_npending (void)
289{ 438{
290 unsigned int retval; 439 unsigned int retval;
291 440
292 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 441 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
293 retval = npending; 442 retval = npending;
294 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 443 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
295 444
296 return retval; 445 return retval;
297} 446}
298 447
299static unsigned int etp_nthreads (void) 448static unsigned int
449etp_nthreads (void)
300{ 450{
301 unsigned int retval; 451 unsigned int retval;
302 452
303 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 453 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
304 retval = started; 454 retval = started;
318} etp_reqq; 468} etp_reqq;
319 469
320static etp_reqq req_queue; 470static etp_reqq req_queue;
321static etp_reqq res_queue; 471static etp_reqq res_queue;
322 472
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
323static int reqq_push (etp_reqq *q, ETP_REQ *req) 485reqq_push (etp_reqq *q, ETP_REQ *req)
324{ 486{
325 int pri = req->pri; 487 int pri = req->pri;
326 req->next = 0; 488 req->next = 0;
327 489
328 if (q->qe[pri]) 490 if (q->qe[pri])
334 q->qe[pri] = q->qs[pri] = req; 496 q->qe[pri] = q->qs[pri] = req;
335 497
336 return q->size++; 498 return q->size++;
337} 499}
338 500
501static ETP_REQ * ecb_noinline
339static ETP_REQ *reqq_shift (etp_reqq *q) 502reqq_shift (etp_reqq *q)
340{ 503{
341 int pri; 504 int pri;
342 505
343 if (!q->size) 506 if (!q->size)
344 return 0; 507 return 0;
359 } 522 }
360 523
361 abort (); 524 abort ();
362} 525}
363 526
364static void etp_atfork_prepare (void) 527static int ecb_cold
528etp_init (void (*want_poll)(void), void (*done_poll)(void))
365{ 529{
366 X_LOCK (wrklock); 530 X_MUTEX_CREATE (wrklock);
367 X_LOCK (reqlock); 531 X_MUTEX_CREATE (reslock);
368 X_LOCK (reslock); 532 X_MUTEX_CREATE (reqlock);
369#if !HAVE_PREADWRITE 533 X_COND_CREATE (reqwait);
370 X_LOCK (preadwritelock);
371#endif
372}
373 534
374static void etp_atfork_parent (void) 535 reqq_init (&req_queue);
375{ 536 reqq_init (&res_queue);
376#if !HAVE_PREADWRITE
377 X_UNLOCK (preadwritelock);
378#endif
379 X_UNLOCK (reslock);
380 X_UNLOCK (reqlock);
381 X_UNLOCK (wrklock);
382}
383 537
384static void etp_atfork_child (void) 538 wrk_first.next =
385{ 539 wrk_first.prev = &wrk_first;
386 ETP_REQ *prv;
387
388 while ((prv = reqq_shift (&req_queue)))
389 ETP_DESTROY (prv);
390
391 while ((prv = reqq_shift (&res_queue)))
392 ETP_DESTROY (prv);
393
394 while (wrk_first.next != &wrk_first)
395 {
396 etp_worker *wrk = wrk_first.next;
397
398 if (wrk->req)
399 ETP_DESTROY (wrk->req);
400
401 etp_worker_clear (wrk);
402 etp_worker_free (wrk);
403 }
404 540
405 started = 0; 541 started = 0;
406 idle = 0; 542 idle = 0;
407 nreqs = 0; 543 nreqs = 0;
408 nready = 0; 544 nready = 0;
409 npending = 0; 545 npending = 0;
410 546
411 etp_atfork_parent ();
412}
413
414static void
415etp_once_init (void)
416{
417 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
418}
419
420static int
421etp_init (void (*want_poll)(void), void (*done_poll)(void))
422{
423 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
424
425 pthread_once (&doinit, etp_once_init);
426
427 want_poll_cb = want_poll; 547 want_poll_cb = want_poll;
428 done_poll_cb = done_poll; 548 done_poll_cb = done_poll;
429 549
430 return 0; 550 return 0;
431} 551}
432 552
433X_THREAD_PROC (etp_proc); 553X_THREAD_PROC (etp_proc);
434 554
555static void ecb_cold
435static void etp_start_thread (void) 556etp_start_thread (void)
436{ 557{
437 etp_worker *wrk = calloc (1, sizeof (etp_worker)); 558 etp_worker *wrk = calloc (1, sizeof (etp_worker));
438 559
439 /*TODO*/ 560 /*TODO*/
440 assert (("unable to allocate worker thread data", wrk)); 561 assert (("unable to allocate worker thread data", wrk));
453 free (wrk); 574 free (wrk);
454 575
455 X_UNLOCK (wrklock); 576 X_UNLOCK (wrklock);
456} 577}
457 578
579static void
458static void etp_maybe_start_thread (void) 580etp_maybe_start_thread (void)
459{ 581{
460 if (expect_true (etp_nthreads () >= wanted)) 582 if (ecb_expect_true (etp_nthreads () >= wanted))
461 return; 583 return;
462 584
463 /* todo: maybe use idle here, but might be less exact */ 585 /* todo: maybe use idle here, but might be less exact */
464 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 ()))
465 return; 587 return;
466 588
467 etp_start_thread (); 589 etp_start_thread ();
468} 590}
469 591
592static void ecb_cold
470static void etp_end_thread (void) 593etp_end_thread (void)
471{ 594{
472 eio_req *req = calloc (1, sizeof (eio_req)); 595 eio_req *req = calloc (1, sizeof (eio_req)); /* will be freed by worker */
473 596
474 req->type = -1; 597 req->type = -1;
475 req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 598 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
476 599
477 X_LOCK (reqlock); 600 X_LOCK (reqlock);
482 X_LOCK (wrklock); 605 X_LOCK (wrklock);
483 --started; 606 --started;
484 X_UNLOCK (wrklock); 607 X_UNLOCK (wrklock);
485} 608}
486 609
487static int etp_poll (void) 610static int
611etp_poll (void)
488{ 612{
489 unsigned int maxreqs; 613 unsigned int maxreqs;
490 unsigned int maxtime; 614 unsigned int maxtime;
491 struct timeval tv_start, tv_now; 615 struct timeval tv_start, tv_now;
492 616
522 646
523 X_LOCK (reqlock); 647 X_LOCK (reqlock);
524 --nreqs; 648 --nreqs;
525 X_UNLOCK (reqlock); 649 X_UNLOCK (reqlock);
526 650
527 if (expect_false (req->type == EIO_GROUP && req->size)) 651 if (ecb_expect_false (req->type == EIO_GROUP && req->size))
528 { 652 {
529 req->int1 = 1; /* mark request as delayed */ 653 req->int1 = 1; /* mark request as delayed */
530 continue; 654 continue;
531 } 655 }
532 else 656 else
533 { 657 {
534 int res = ETP_FINISH (req); 658 int res = ETP_FINISH (req);
535 if (expect_false (res)) 659 if (ecb_expect_false (res))
536 return res; 660 return res;
537 } 661 }
538 662
539 if (expect_false (maxreqs && !--maxreqs)) 663 if (ecb_expect_false (maxreqs && !--maxreqs))
540 break; 664 break;
541 665
542 if (maxtime) 666 if (maxtime)
543 { 667 {
544 gettimeofday (&tv_now, 0); 668 gettimeofday (&tv_now, 0);
550 674
551 errno = EAGAIN; 675 errno = EAGAIN;
552 return -1; 676 return -1;
553} 677}
554 678
679static void
555static void etp_cancel (ETP_REQ *req) 680etp_cancel (ETP_REQ *req)
556{ 681{
557 X_LOCK (wrklock); 682 req->cancelled = 1;
558 req->flags |= EIO_FLAG_CANCELLED;
559 X_UNLOCK (wrklock);
560 683
561 eio_grp_cancel (req); 684 eio_grp_cancel (req);
562} 685}
563 686
687static void
564static void etp_submit (ETP_REQ *req) 688etp_submit (ETP_REQ *req)
565{ 689{
566 req->pri -= ETP_PRI_MIN; 690 req->pri -= ETP_PRI_MIN;
567 691
568 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;
569 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;
570 694
571 if (expect_false (req->type == EIO_GROUP)) 695 if (ecb_expect_false (req->type == EIO_GROUP))
572 { 696 {
573 /* I hope this is worth it :/ */ 697 /* I hope this is worth it :/ */
574 X_LOCK (reqlock); 698 X_LOCK (reqlock);
575 ++nreqs; 699 ++nreqs;
576 X_UNLOCK (reqlock); 700 X_UNLOCK (reqlock);
595 719
596 etp_maybe_start_thread (); 720 etp_maybe_start_thread ();
597 } 721 }
598} 722}
599 723
724static void ecb_cold
600static void etp_set_max_poll_time (double nseconds) 725etp_set_max_poll_time (double nseconds)
601{ 726{
602 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 727 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
603 max_poll_time = nseconds * EIO_TICKS; 728 max_poll_time = nseconds * EIO_TICKS;
604 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 729 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
605} 730}
606 731
732static void ecb_cold
607static void etp_set_max_poll_reqs (unsigned int maxreqs) 733etp_set_max_poll_reqs (unsigned int maxreqs)
608{ 734{
609 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 735 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
610 max_poll_reqs = maxreqs; 736 max_poll_reqs = maxreqs;
611 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 737 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
612} 738}
613 739
740static void ecb_cold
614static void etp_set_max_idle (unsigned int nthreads) 741etp_set_max_idle (unsigned int nthreads)
615{ 742{
616 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 743 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
617 max_idle = nthreads <= 0 ? 1 : nthreads; 744 max_idle = nthreads;
618 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 745 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
619} 746}
620 747
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
621static void etp_set_min_parallel (unsigned int nthreads) 757etp_set_min_parallel (unsigned int nthreads)
622{ 758{
623 if (wanted < nthreads) 759 if (wanted < nthreads)
624 wanted = nthreads; 760 wanted = nthreads;
625} 761}
626 762
763static void ecb_cold
627static void etp_set_max_parallel (unsigned int nthreads) 764etp_set_max_parallel (unsigned int nthreads)
628{ 765{
629 if (wanted > nthreads) 766 if (wanted > nthreads)
630 wanted = nthreads; 767 wanted = nthreads;
631 768
632 while (started > wanted) 769 while (started > wanted)
633 etp_end_thread (); 770 etp_end_thread ();
634} 771}
635 772
636/*****************************************************************************/ 773/*****************************************************************************/
637 774
775static void
638static void grp_try_feed (eio_req *grp) 776grp_try_feed (eio_req *grp)
639{ 777{
640 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 778 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
641 { 779 {
642 grp->flags &= ~EIO_FLAG_GROUPADD; 780 grp->flags &= ~EIO_FLAG_GROUPADD;
643 781
650 break; 788 break;
651 } 789 }
652 } 790 }
653} 791}
654 792
793static int
655static int grp_dec (eio_req *grp) 794grp_dec (eio_req *grp)
656{ 795{
657 --grp->size; 796 --grp->size;
658 797
659 /* call feeder, if applicable */ 798 /* call feeder, if applicable */
660 grp_try_feed (grp); 799 grp_try_feed (grp);
664 return eio_finish (grp); 803 return eio_finish (grp);
665 else 804 else
666 return 0; 805 return 0;
667} 806}
668 807
808static void
669void eio_destroy (eio_req *req) 809eio_destroy (eio_req *req)
670{ 810{
671 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); 811 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
672 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); 812 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
673 813
674 EIO_DESTROY (req); 814 EIO_DESTROY (req);
675} 815}
676 816
817static int
677static int eio_finish (eio_req *req) 818eio_finish (eio_req *req)
678{ 819{
679 int res = EIO_FINISH (req); 820 int res = EIO_FINISH (req);
680 821
681 if (req->grp) 822 if (req->grp)
682 { 823 {
690 if (grp->grp_first == req) 831 if (grp->grp_first == req)
691 grp->grp_first = req->grp_next; 832 grp->grp_first = req->grp_next;
692 833
693 res2 = grp_dec (grp); 834 res2 = grp_dec (grp);
694 835
695 if (!res && res2) 836 if (!res)
696 res = res2; 837 res = res2;
697 } 838 }
698 839
699 eio_destroy (req); 840 eio_destroy (req);
700 841
701 return res; 842 return res;
702} 843}
703 844
845void
704void eio_grp_cancel (eio_req *grp) 846eio_grp_cancel (eio_req *grp)
705{ 847{
706 for (grp = grp->grp_first; grp; grp = grp->grp_next) 848 for (grp = grp->grp_first; grp; grp = grp->grp_next)
707 eio_cancel (grp); 849 eio_cancel (grp);
708} 850}
709 851
852void
710void eio_cancel (eio_req *req) 853eio_cancel (eio_req *req)
711{ 854{
712 etp_cancel (req); 855 etp_cancel (req);
713} 856}
714 857
858void
715void eio_submit (eio_req *req) 859eio_submit (eio_req *req)
716{ 860{
717 etp_submit (req); 861 etp_submit (req);
718} 862}
719 863
720unsigned int eio_nreqs (void) 864unsigned int
865eio_nreqs (void)
721{ 866{
722 return etp_nreqs (); 867 return etp_nreqs ();
723} 868}
724 869
725unsigned int eio_nready (void) 870unsigned int
871eio_nready (void)
726{ 872{
727 return etp_nready (); 873 return etp_nready ();
728} 874}
729 875
730unsigned int eio_npending (void) 876unsigned int
877eio_npending (void)
731{ 878{
732 return etp_npending (); 879 return etp_npending ();
733} 880}
734 881
735unsigned int eio_nthreads (void) 882unsigned int ecb_cold
883eio_nthreads (void)
736{ 884{
737 return etp_nthreads (); 885 return etp_nthreads ();
738} 886}
739 887
888void ecb_cold
740void eio_set_max_poll_time (double nseconds) 889eio_set_max_poll_time (double nseconds)
741{ 890{
742 etp_set_max_poll_time (nseconds); 891 etp_set_max_poll_time (nseconds);
743} 892}
744 893
894void ecb_cold
745void eio_set_max_poll_reqs (unsigned int maxreqs) 895eio_set_max_poll_reqs (unsigned int maxreqs)
746{ 896{
747 etp_set_max_poll_reqs (maxreqs); 897 etp_set_max_poll_reqs (maxreqs);
748} 898}
749 899
900void ecb_cold
750void eio_set_max_idle (unsigned int nthreads) 901eio_set_max_idle (unsigned int nthreads)
751{ 902{
752 etp_set_max_idle (nthreads); 903 etp_set_max_idle (nthreads);
753} 904}
754 905
906void ecb_cold
907eio_set_idle_timeout (unsigned int seconds)
908{
909 etp_set_idle_timeout (seconds);
910}
911
912void ecb_cold
755void eio_set_min_parallel (unsigned int nthreads) 913eio_set_min_parallel (unsigned int nthreads)
756{ 914{
757 etp_set_min_parallel (nthreads); 915 etp_set_min_parallel (nthreads);
758} 916}
759 917
918void ecb_cold
760void eio_set_max_parallel (unsigned int nthreads) 919eio_set_max_parallel (unsigned int nthreads)
761{ 920{
762 etp_set_max_parallel (nthreads); 921 etp_set_max_parallel (nthreads);
763} 922}
764 923
765int eio_poll (void) 924int eio_poll (void)
774# undef pread 933# undef pread
775# undef pwrite 934# undef pwrite
776# define pread eio__pread 935# define pread eio__pread
777# define pwrite eio__pwrite 936# define pwrite eio__pwrite
778 937
779static ssize_t 938static eio_ssize_t
780eio__pread (int fd, void *buf, size_t count, off_t offset) 939eio__pread (int fd, void *buf, size_t count, off_t offset)
781{ 940{
782 ssize_t res; 941 eio_ssize_t res;
783 off_t ooffset; 942 off_t ooffset;
784 943
785 X_LOCK (preadwritelock); 944 X_LOCK (preadwritelock);
786 ooffset = lseek (fd, 0, SEEK_CUR); 945 ooffset = lseek (fd, 0, SEEK_CUR);
787 lseek (fd, offset, SEEK_SET); 946 lseek (fd, offset, SEEK_SET);
790 X_UNLOCK (preadwritelock); 949 X_UNLOCK (preadwritelock);
791 950
792 return res; 951 return res;
793} 952}
794 953
795static ssize_t 954static eio_ssize_t
796eio__pwrite (int fd, void *buf, size_t count, off_t offset) 955eio__pwrite (int fd, void *buf, size_t count, off_t offset)
797{ 956{
798 ssize_t res; 957 eio_ssize_t res;
799 off_t ooffset; 958 off_t ooffset;
800 959
801 X_LOCK (preadwritelock); 960 X_LOCK (preadwritelock);
802 ooffset = lseek (fd, 0, SEEK_CUR); 961 ooffset = lseek (fd, 0, SEEK_CUR);
803 lseek (fd, offset, SEEK_SET); 962 lseek (fd, offset, SEEK_SET);
807 966
808 return res; 967 return res;
809} 968}
810#endif 969#endif
811 970
812#ifndef HAVE_FUTIMES 971#ifndef HAVE_UTIMES
813 972
814# undef utimes 973# undef utimes
815# undef futimes
816# define utimes(path,times) eio__utimes (path, times) 974# define utimes(path,times) eio__utimes (path, times)
817# define futimes(fd,times) eio__futimes (fd, times)
818 975
819static int 976static int
820eio__utimes (const char *filename, const struct timeval times[2]) 977eio__utimes (const char *filename, const struct timeval times[2])
821{ 978{
822 if (times) 979 if (times)
830 } 987 }
831 else 988 else
832 return utime (filename, 0); 989 return utime (filename, 0);
833} 990}
834 991
992#endif
993
994#ifndef HAVE_FUTIMES
995
996# undef futimes
997# define futimes(fd,times) eio__futimes (fd, times)
998
999static int
835static int eio__futimes (int fd, const struct timeval tv[2]) 1000eio__futimes (int fd, const struct timeval tv[2])
836{ 1001{
837 errno = ENOSYS; 1002 errno = ENOSYS;
838 return -1; 1003 return -1;
839} 1004}
840 1005
843#if !HAVE_FDATASYNC 1008#if !HAVE_FDATASYNC
844# undef fdatasync 1009# undef fdatasync
845# define fdatasync(fd) fsync (fd) 1010# define fdatasync(fd) fsync (fd)
846#endif 1011#endif
847 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
848/* sync_file_range always needs emulation */ 1031/* sync_file_range always needs emulation */
849int 1032static int
850eio__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)
851{ 1034{
852#if HAVE_SYNC_FILE_RANGE 1035#if HAVE_SYNC_FILE_RANGE
853 int res; 1036 int res;
854 1037
871 /* 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
872 * call fdatasync, as that matches the expectation of its users best */ 1055 * call fdatasync, as that matches the expectation of its users best */
873 return fdatasync (fd); 1056 return fdatasync (fd);
874} 1057}
875 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
876#if !HAVE_READAHEAD 1070#if !HAVE_READAHEAD
877# undef readahead 1071# undef readahead
878# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 1072# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
879 1073
880static ssize_t 1074static eio_ssize_t
881eio__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)
882{ 1076{
883 size_t todo = count; 1077 size_t todo = count;
884 dBUF; 1078 dBUF;
885 1079
890 pread (fd, eio_buf, len, offset); 1084 pread (fd, eio_buf, len, offset);
891 offset += len; 1085 offset += len;
892 todo -= len; 1086 todo -= len;
893 } 1087 }
894 1088
1089 FUBd;
1090
895 errno = 0; 1091 errno = 0;
896 return count; 1092 return count;
897} 1093}
898 1094
899#endif 1095#endif
900 1096
901/* sendfile always needs emulation */ 1097/* sendfile always needs emulation */
902static ssize_t 1098static eio_ssize_t
903eio__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)
904{ 1100{
1101 eio_ssize_t written = 0;
905 ssize_t res; 1102 eio_ssize_t res;
906 1103
907 if (!count) 1104 if (!count)
908 return 0; 1105 return 0;
909 1106
1107 for (;;)
1108 {
1109#ifdef __APPLE__
1110# undef HAVE_SENDFILE /* broken, as everything on os x */
1111#endif
910#if HAVE_SENDFILE 1112#if HAVE_SENDFILE
911# if __linux 1113# if __linux
1114 off_t soffset = offset;
912 res = sendfile (ofd, ifd, &offset, count); 1115 res = sendfile (ofd, ifd, &soffset, count);
913 1116
914# elif __freebsd 1117# elif __FreeBSD__
915 /* 1118 /*
916 * 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
917 * wasted on making it similar to other I/O functions. 1120 * wasted on making it similar to other I/O functions.
918 */ 1121 */
919 {
920 off_t sbytes; 1122 off_t sbytes;
921 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 1123 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
922 1124
923 if (res < 0 && sbytes) 1125 #if 0 /* according to the manpage, this is correct, but broken behaviour */
924 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 1126 /* freebsd' sendfile will return 0 on success */
1127 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
1128 /* not on e.g. EIO or EPIPE - sounds broken */
1129 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
925 res = sbytes; 1130 res = sbytes;
926 } 1131 #endif
1132
1133 /* according to source inspection, this is correct, and useful behaviour */
1134 if (sbytes)
1135 res = sbytes;
927 1136
928# elif defined (__APPLE__) 1137# elif defined (__APPLE__)
929
930 {
931 off_t sbytes = count; 1138 off_t sbytes = count;
932 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0); 1139 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
933 1140
934 if (res < 0 && errno == EAGAIN && sbytes) 1141 /* according to the manpage, sbytes is always valid */
1142 if (sbytes)
935 res = sbytes; 1143 res = sbytes;
936 }
937 1144
938# elif __hpux 1145# elif __hpux
939 res = sendfile (ofd, ifd, offset, count, 0, 0); 1146 res = sendfile (ofd, ifd, offset, count, 0, 0);
940 1147
941# elif __solaris 1148# elif __solaris
942 {
943 struct sendfilevec vec; 1149 struct sendfilevec vec;
944 size_t sbytes; 1150 size_t sbytes;
945 1151
946 vec.sfv_fd = ifd; 1152 vec.sfv_fd = ifd;
947 vec.sfv_flag = 0; 1153 vec.sfv_flag = 0;
948 vec.sfv_off = offset; 1154 vec.sfv_off = offset;
949 vec.sfv_len = count; 1155 vec.sfv_len = count;
950 1156
951 res = sendfilev (ofd, &vec, 1, &sbytes); 1157 res = sendfilev (ofd, &vec, 1, &sbytes);
952 1158
953 if (res < 0 && sbytes) 1159 if (res < 0 && sbytes)
954 res = sbytes; 1160 res = sbytes;
955 }
956 1161
957# endif 1162# endif
958 1163
959#elif defined (_WIN32) 1164#elif defined (_WIN32) && 0
960
961 /* 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 */
962 { 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 */
963 HANDLE h = TO_SOCKET (ifd); 1169 HANDLE h = TO_SOCKET (ifd);
964 SetFilePointer (h, offset, 0, FILE_BEGIN); 1170 SetFilePointer (h, offset, 0, FILE_BEGIN);
965 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0); 1171 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
966 }
967 1172
968#else 1173#else
969 res = -1; 1174 res = -1;
970 errno = ENOSYS; 1175 errno = ENOSYS;
971#endif 1176#endif
972 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
973 if (res < 0 1201 if (res < 0
974 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1202 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
975 /* BSDs */ 1203 /* BSDs */
976#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... */
977 || errno == ENOTSUP 1205 || errno == ENOTSUP
978#endif 1206#endif
1207#ifdef EOPNOTSUPP /* windows */
979 || errno == EOPNOTSUPP /* BSDs */ 1208 || errno == EOPNOTSUPP /* BSDs */
1209#endif
980#if __solaris 1210#if __solaris
981 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1211 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
982#endif 1212#endif
983 ) 1213 )
984 ) 1214 )
988 1218
989 res = 0; 1219 res = 0;
990 1220
991 while (count) 1221 while (count)
992 { 1222 {
993 ssize_t cnt; 1223 eio_ssize_t cnt;
994 1224
995 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);
996 1226
997 if (cnt <= 0) 1227 if (cnt <= 0)
998 { 1228 {
1010 1240
1011 offset += cnt; 1241 offset += cnt;
1012 res += cnt; 1242 res += cnt;
1013 count -= cnt; 1243 count -= cnt;
1014 } 1244 }
1245
1246 FUBd;
1015 } 1247 }
1016 1248
1017 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;
1018} 1530}
1019 1531
1020static signed char 1532static signed char
1021eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) 1533eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1022{ 1534{
1023 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 */
1024 : 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;
1025} 1539}
1026 1540
1027#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
1028 1542
1029#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 */
1030#define EIO_SORT_FAST 60 /* when to only use insertion sort */ 1544#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1031 1545
1032static void 1546static void
1033eio_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)
1034{ 1548{
1035 unsigned char bits [9 + sizeof (ino_t) * 8]; 1549 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1036 unsigned char *bit = bits; 1550 unsigned char *bit = bits;
1037 1551
1038 assert (CHAR_BIT == 8); 1552 assert (CHAR_BIT == 8);
1039 assert (sizeof (eio_dirent) * 8 < 256); 1553 assert (sizeof (eio_dirent) * 8 < 256);
1040 assert (offsetof (eio_dirent, inode)); /* we use 0 as sentinel */ 1554 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1041 assert (offsetof (eio_dirent, score)); /* we use 0 as sentinel */ 1555 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1042 1556
1043 if (size <= EIO_SORT_FAST) 1557 if (size <= EIO_SORT_FAST)
1044 return; 1558 return;
1045 1559
1046 /* 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 */
1047 /* 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 */
1048 /* inode_bits must contain all inodes ORed together */ 1562 /* inode_bits must contain all inodes ORed together */
1049 /* 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 */
1050 { 1564 {
1051 ino_t endianness; 1565 eio_ino_t endianness;
1052 int i, j; 1566 int i, j;
1053 1567
1054 /* 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" */
1055 for (i = 0; i < sizeof (ino_t); ++i) 1569 for (i = 0; i < sizeof (eio_ino_t); ++i)
1056 ((unsigned char *)&endianness)[i] = i; 1570 ((unsigned char *)&endianness)[i] = i;
1057 1571
1058 *bit++ = 0; 1572 *bit++ = 0;
1059 1573
1060 for (i = 0; i < sizeof (ino_t); ++i) 1574 for (i = 0; i < sizeof (eio_ino_t); ++i)
1061 { 1575 {
1062 /* shifting off the byte offsets out of "endianness" */ 1576 /* shifting off the byte offsets out of "endianness" */
1063 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8; 1577 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1064 endianness >>= 8; 1578 endianness >>= 8;
1065 1579
1066 for (j = 0; j < 8; ++j) 1580 for (j = 0; j < 8; ++j)
1067 if (inode_bits & (((ino_t)1) << (i * 8 + j))) 1581 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1068 *bit++ = offs + j; 1582 *bit++ = offs + j;
1069 } 1583 }
1070 1584
1071 for (j = 0; j < 8; ++j) 1585 for (j = 0; j < 8; ++j)
1072 if (score_bits & (1 << j)) 1586 if (score_bits & (1 << j))
1073 *bit++ = offsetof (eio_dirent, score) * 8 + j; 1587 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1074 } 1588 }
1075 1589
1076 /* now actually do the sorting (a variant of MSD radix sort) */ 1590 /* now actually do the sorting (a variant of MSD radix sort) */
1077 { 1591 {
1078 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base; 1592 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1079 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end; 1593 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1080 unsigned char *bit_stk [9 + sizeof (ino_t) * 8]; 1594 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1081 int stk_idx = 0; 1595 int stk_idx = 0;
1082 1596
1083 base_stk [stk_idx] = dents; 1597 base_stk [stk_idx] = dents;
1084 end_stk [stk_idx] = dents + size; 1598 end_stk [stk_idx] = dents + size;
1085 bit_stk [stk_idx] = bit - 1; 1599 bit_stk [stk_idx] = bit - 1;
1164 } 1678 }
1165 } 1679 }
1166} 1680}
1167 1681
1168static void 1682static void
1169eio_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)
1170{ 1684{
1171 if (size <= 1) 1685 if (size <= 1)
1172 return; /* our insertion sort relies on size > 0 */ 1686 return; /* our insertion sort relies on size > 0 */
1173 1687
1174 /* 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 */
1180 eio_dent_insertion_sort (dents, size); 1694 eio_dent_insertion_sort (dents, size);
1181} 1695}
1182 1696
1183/* read a full directory */ 1697/* read a full directory */
1184static void 1698static void
1185eio__scandir (eio_req *req, etp_worker *self) 1699eio__scandir (eio_req *req)
1186{ 1700{
1187 DIR *dirp;
1188 EIO_STRUCT_DIRENT *entp;
1189 char *name, *names; 1701 char *name, *names;
1190 int namesalloc = 4096; 1702 int namesalloc = 4096 - sizeof (void *) * 4;
1191 int namesoffs = 0; 1703 int namesoffs = 0;
1192 int flags = req->int1; 1704 int flags = req->int1;
1193 eio_dirent *dents = 0; 1705 eio_dirent *dents = 0;
1194 int dentalloc = 128; 1706 int dentalloc = 128;
1195 int dentoffs = 0; 1707 int dentoffs = 0;
1196 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
1197 1716
1198 req->result = -1; 1717 req->result = -1;
1199 1718
1200 if (!(flags & EIO_READDIR_DENTS)) 1719 if (!(flags & EIO_READDIR_DENTS))
1201 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); 1720 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1202 1721
1203 X_LOCK (wrklock); 1722#ifdef _WIN32
1204 /* 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
1205 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
1206 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1776 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1207 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; 1777 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1208 req->ptr2 = names = malloc (namesalloc); 1778 req->ptr2 = names = malloc (namesalloc);
1209 X_UNLOCK (wrklock);
1210 1779
1211 if (dirp && names && (!flags || dents)) 1780 if (!names || (flags && !dents))
1781 return;
1782
1212 for (;;) 1783 for (;;)
1213 { 1784 {
1785 int done;
1786
1787#ifdef _WIN32
1788 done = !dirp;
1789#else
1214 errno = 0; 1790 errno = 0;
1215 entp = readdir (dirp); 1791 entp = readdir (dirp);
1792 done = !entp;
1793#endif
1216 1794
1217 if (!entp) 1795 if (done)
1218 { 1796 {
1797#ifndef _WIN32
1798 int old_errno = errno;
1799 closedir (dirp);
1800 errno = old_errno;
1801
1219 if (errno) 1802 if (errno)
1220 break; 1803 break;
1804#endif
1221 1805
1222 /* sort etc. */ 1806 /* sort etc. */
1223 req->int1 = flags; 1807 req->int1 = flags;
1224 req->result = dentoffs; 1808 req->result = dentoffs;
1225 1809
1226 if (flags & EIO_READDIR_STAT_ORDER) 1810 if (flags & EIO_READDIR_STAT_ORDER)
1227 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);
1228 else if (flags & EIO_READDIR_DIRS_FIRST) 1812 else if (flags & EIO_READDIR_DIRS_FIRST)
1229 if (flags & EIO_READDIR_FOUND_UNKNOWN) 1813 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1230 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 */
1231 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))
1232 { 1866 {
1233 /* 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
1234 eio_dirent *oth = dents + dentoffs; 1874 ent = dents + dentoffs;
1235 eio_dirent *dir = dents;
1236 1875
1237 /* 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 */
1238 /* by walking from both sides and swapping if necessary */ 1877 ent->namelen = len - 1;
1239 /* also clear score, so it doesn't influence sorting */ 1878 ent->inode = D_INO (entp);
1240 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)
1241 { 1938 {
1242 if (dir->type == EIO_DT_DIR) 1939 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1243 ++dir; 1940 ent->score = 1;
1244 else if ((--oth)->type == EIO_DT_DIR) 1941 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1245 { 1942 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1246 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1247
1248 ++dir;
1249 }
1250 } 1943 }
1251 1944 else if (ent->type == EIO_DT_DIR)
1252 /* now sort the dirs only */ 1945 ent->score = 0;
1253 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1254 } 1946 }
1255
1256 break;
1257 }
1258
1259 /* now add the entry to our list(s) */
1260 name = entp->d_name;
1261
1262 /* skip . and .. entries */
1263 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1264 {
1265 int len = D_NAMLEN (entp) + 1;
1266
1267 while (expect_false (namesoffs + len > namesalloc))
1268 {
1269 namesalloc *= 2;
1270 X_LOCK (wrklock);
1271 req->ptr2 = names = realloc (names, namesalloc);
1272 X_UNLOCK (wrklock);
1273
1274 if (!names)
1275 break;
1276 } 1947 }
1277 1948
1278 memcpy (names + namesoffs, name, len);
1279
1280 if (dents)
1281 {
1282 struct eio_dirent *ent;
1283
1284 if (expect_false (dentoffs == dentalloc))
1285 {
1286 dentalloc *= 2;
1287 X_LOCK (wrklock);
1288 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1289 X_UNLOCK (wrklock);
1290
1291 if (!dents)
1292 break;
1293 }
1294
1295 ent = dents + dentoffs;
1296
1297 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1298 ent->namelen = len - 1;
1299 ent->inode = D_INO (entp);
1300
1301 inode_bits |= ent->inode;
1302
1303 switch (D_TYPE (entp))
1304 {
1305 default:
1306 ent->type = EIO_DT_UNKNOWN;
1307 flags |= EIO_READDIR_FOUND_UNKNOWN;
1308 break;
1309
1310 #ifdef DT_FIFO
1311 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1312 #endif
1313 #ifdef DT_CHR
1314 case DT_CHR: ent->type = EIO_DT_CHR; break;
1315 #endif
1316 #ifdef DT_MPC
1317 case DT_MPC: ent->type = EIO_DT_MPC; break;
1318 #endif
1319 #ifdef DT_DIR
1320 case DT_DIR: ent->type = EIO_DT_DIR; break;
1321 #endif
1322 #ifdef DT_NAM
1323 case DT_NAM: ent->type = EIO_DT_NAM; break;
1324 #endif
1325 #ifdef DT_BLK
1326 case DT_BLK: ent->type = EIO_DT_BLK; break;
1327 #endif
1328 #ifdef DT_MPB
1329 case DT_MPB: ent->type = EIO_DT_MPB; break;
1330 #endif
1331 #ifdef DT_REG
1332 case DT_REG: ent->type = EIO_DT_REG; break;
1333 #endif
1334 #ifdef DT_NWK
1335 case DT_NWK: ent->type = EIO_DT_NWK; break;
1336 #endif
1337 #ifdef DT_CMP
1338 case DT_CMP: ent->type = EIO_DT_CMP; break;
1339 #endif
1340 #ifdef DT_LNK
1341 case DT_LNK: ent->type = EIO_DT_LNK; break;
1342 #endif
1343 #ifdef DT_SOCK
1344 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1345 #endif
1346 #ifdef DT_DOOR
1347 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1348 #endif
1349 #ifdef DT_WHT
1350 case DT_WHT: ent->type = EIO_DT_WHT; break;
1351 #endif
1352 }
1353
1354 ent->score = 7;
1355
1356 if (flags & EIO_READDIR_DIRS_FIRST)
1357 {
1358 if (ent->type == EIO_DT_UNKNOWN)
1359 {
1360 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1361 ent->score = 1;
1362 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1363 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1364 }
1365 else if (ent->type == EIO_DT_DIR)
1366 ent->score = 0;
1367 }
1368 }
1369
1370 namesoffs += len; 1949 namesoffs += len;
1371 ++dentoffs; 1950 ++dentoffs;
1372 } 1951 }
1373 1952
1374 if (EIO_CANCELLED (req)) 1953 if (EIO_CANCELLED (req))
1375 { 1954 {
1376 errno = ECANCELED; 1955 errno = ECANCELED;
1377 break; 1956 break;
1378 } 1957 }
1958
1959#ifdef _WIN32
1960 if (!FindNextFile (dirp, &entp))
1961 {
1962 FindClose (dirp);
1963 dirp = 0;
1379 } 1964 }
1965#endif
1966 }
1380} 1967}
1381 1968
1382#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) 1969/*****************************************************************************/
1383# 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
1384#else 2032#else
1385 2033
1386int 2034/* on legacy systems, we represent the working directories simply by their path strings */
1387eio__msync (void *mem, size_t len, int flags) 2035
2036static const char *
2037wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
1388{ 2038{
1389 if (EIO_MS_ASYNC != MS_SYNC 2039 if (!wd || *path == '/')
1390 || EIO_MS_INVALIDATE != MS_INVALIDATE 2040 return path;
1391 || 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
1392 { 2064 {
1393 flags = 0 2065 int len = eio__realpath (tmpbuf, wd, path);
1394 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1395 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1396 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1397 }
1398 2066
1399 return msync (mem, len, flags); 2067 path = EIO_INVALID_WD;
1400}
1401 2068
1402#endif 2069 if (len >= 0)
2070 {
2071 ((char *)tmpbuf->ptr)[len] = 0;
2072 path = strdup (tmpbuf->ptr);
2073 }
2074 }
1403 2075
1404int
1405eio__mtouch (void *mem, size_t len, int flags)
1406{
1407 intptr_t addr = (intptr_t)mem;
1408 intptr_t end = addr + len;
1409#ifdef PAGESIZE
1410 const intptr_t page = PAGESIZE;
1411#else
1412 static intptr_t page;
1413
1414 if (!page) 2076 if (!path)
1415 page = sysconf (_SC_PAGESIZE); 2077 path = EIO_INVALID_WD;
1416#endif
1417 2078
1418 /* round down to start of page, although this is probably useless */ 2079 return (eio_wd)path;
1419 addr &= ~(page - 1); /* assume page size is always a power of two */ 2080}
1420 2081
1421 if (addr < end) 2082eio_wd
1422 if (flags & EIO_MT_MODIFY) /* modify */ 2083eio_wd_open_sync (eio_wd wd, const char *path)
1423 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len); 2084{
1424 else 2085 struct tmpbuf tmpbuf = { 0 };
1425 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len); 2086 wd = eio__wd_open_sync (&tmpbuf, wd, path);
2087 free (tmpbuf.ptr);
1426 2088
1427 return 0; 2089 return wd;
1428} 2090}
2091
2092void
2093eio_wd_close_sync (eio_wd wd)
2094{
2095 if (wd != EIO_INVALID_WD)
2096 free (wd);
2097}
2098
2099#endif
1429 2100
1430/*****************************************************************************/ 2101/*****************************************************************************/
1431 2102
1432#define ALLOC(len) \ 2103#define ALLOC(len) \
1433 if (!req->ptr2) \ 2104 if (!req->ptr2) \
1448{ 2119{
1449 ETP_REQ *req; 2120 ETP_REQ *req;
1450 struct timespec ts; 2121 struct timespec ts;
1451 etp_worker *self = (etp_worker *)thr_arg; 2122 etp_worker *self = (etp_worker *)thr_arg;
1452 2123
2124#if HAVE_PRCTL_SET_NAME
2125 prctl (PR_SET_NAME, (unsigned long)"eio_thread", 0, 0, 0);
2126#endif
2127
1453 /* try to distribute timeouts somewhat randomly */ 2128 /* try to distribute timeouts somewhat evenly */
1454 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 2129 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1455 2130
1456 for (;;) 2131 for (;;)
1457 { 2132 {
2133 ts.tv_sec = 0;
2134
1458 X_LOCK (reqlock); 2135 X_LOCK (reqlock);
1459 2136
1460 for (;;) 2137 for (;;)
1461 { 2138 {
1462 self->req = req = reqq_shift (&req_queue); 2139 req = reqq_shift (&req_queue);
1463 2140
1464 if (req) 2141 if (req)
1465 break; 2142 break;
1466 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
1467 ++idle; 2153 ++idle;
1468 2154
1469 ts.tv_sec = time (0) + IDLE_TIMEOUT; 2155 if (idle <= max_idle)
1470 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
1471 { 2159 {
1472 if (idle > max_idle) 2160 /* initialise timeout once */
1473 { 2161 if (!ts.tv_sec)
1474 --idle; 2162 ts.tv_sec = time (0) + idle_timeout;
1475 X_UNLOCK (reqlock);
1476 X_LOCK (wrklock);
1477 --started;
1478 X_UNLOCK (wrklock);
1479 goto quit;
1480 }
1481 2163
1482 /* we are allowed to idle, so do so without any timeout */
1483 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.,.. */
1484 } 2166 }
1485 2167
1486 --idle; 2168 --idle;
1487 } 2169 }
1488 2170
1491 X_UNLOCK (reqlock); 2173 X_UNLOCK (reqlock);
1492 2174
1493 if (req->type < 0) 2175 if (req->type < 0)
1494 goto quit; 2176 goto quit;
1495 2177
1496 if (!EIO_CANCELLED (req))
1497 ETP_EXECUTE (self, req); 2178 ETP_EXECUTE (self, req);
1498 2179
1499 X_LOCK (reslock); 2180 X_LOCK (reslock);
1500 2181
1501 ++npending; 2182 ++npending;
1502 2183
1503 if (!reqq_push (&res_queue, req) && want_poll_cb) 2184 if (!reqq_push (&res_queue, req) && want_poll_cb)
1504 want_poll_cb (); 2185 want_poll_cb ();
1505 2186
1506 self->req = 0;
1507 etp_worker_clear (self); 2187 etp_worker_clear (self);
1508 2188
1509 X_UNLOCK (reslock); 2189 X_UNLOCK (reslock);
1510 } 2190 }
1511 2191
1512quit: 2192quit:
2193 free (req);
2194
1513 X_LOCK (wrklock); 2195 X_LOCK (wrklock);
1514 etp_worker_free (self); 2196 etp_worker_free (self);
1515 X_UNLOCK (wrklock); 2197 X_UNLOCK (wrklock);
1516 2198
1517 return 0; 2199 return 0;
1518} 2200}
1519 2201
1520/*****************************************************************************/ 2202/*****************************************************************************/
1521 2203
2204int ecb_cold
1522int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 2205eio_init (void (*want_poll)(void), void (*done_poll)(void))
1523{ 2206{
2207#if !HAVE_PREADWRITE
2208 X_MUTEX_CREATE (preadwritelock);
2209#endif
2210
1524 return etp_init (want_poll, done_poll); 2211 return etp_init (want_poll, done_poll);
1525} 2212}
1526 2213
2214ecb_inline void
1527static void eio_api_destroy (eio_req *req) 2215eio_api_destroy (eio_req *req)
1528{ 2216{
1529 free (req); 2217 free (req);
1530} 2218}
1531 2219
1532#define REQ(rtype) \ 2220#define REQ(rtype) \
1551 { \ 2239 { \
1552 eio_api_destroy (req); \ 2240 eio_api_destroy (req); \
1553 return 0; \ 2241 return 0; \
1554 } 2242 }
1555 2243
2244static void
1556static void eio_execute (etp_worker *self, eio_req *req) 2245eio_execute (etp_worker *self, eio_req *req)
1557{ 2246{
1558 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 }
1559 2275
1560 switch (req->type) 2276 switch (req->type)
1561 { 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
1562 case EIO_READ: ALLOC (req->size); 2284 case EIO_READ: ALLOC (req->size);
1563 req->result = req->offs >= 0 2285 req->result = req->offs >= 0
1564 ? pread (req->int1, req->ptr2, req->size, req->offs) 2286 ? pread (req->int1, req->ptr2, req->size, req->offs)
1565 : read (req->int1, req->ptr2, req->size); break; 2287 : read (req->int1, req->ptr2, req->size); break;
1566 case EIO_WRITE: req->result = req->offs >= 0 2288 case EIO_WRITE: req->result = req->offs >= 0
1567 ? pwrite (req->int1, req->ptr2, req->size, req->offs) 2289 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1568 : write (req->int1, req->ptr2, req->size); break; 2290 : write (req->int1, req->ptr2, req->size); break;
1569 2291
1570 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;
1571 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
1572 2296
1573 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2297 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1574 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;
1575 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2299 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1576 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;
1577 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1578 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1579
1580 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; 2301 case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break;
1581 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1582 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break; 2302 case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break;
1583 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1584 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break; 2303 case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break;
1585 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1586
1587 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break; 2304 case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break;
1588 case EIO_CLOSE: req->result = close (req->int1); break; 2305
1589 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1590 case EIO_UNLINK: req->result = unlink (req->ptr1); break; 2306 case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
1591 case EIO_RMDIR: req->result = rmdir (req->ptr1); break; 2307 case EIO_RMDIR: req->result = unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break;
1592 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; 2308 case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); break;
1593 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; 2309 case EIO_RENAME: req->result = renameat (dirfd, req->ptr1, WD2FD (req->int3), req->ptr2); break;
1594 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; 2310 case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD (req->int3), req->ptr2, 0); break;
1595 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; 2311 case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
1596 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break; 2312 case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1597
1598 case EIO_READLINK: ALLOC (NAME_MAX); 2313 case EIO_READLINK: ALLOC (PATH_MAX);
1599 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; 2314 req->result = readlinkat (dirfd, req->ptr1, req->ptr2, PATH_MAX); break;
1600 2315 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1601 case EIO_SYNC: req->result = 0; sync (); break; 2316 req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1602 case EIO_FSYNC: req->result = fsync (req->int1); break;
1603 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1604 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1605 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;
1606 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1607
1608 case EIO_READDIR: eio__scandir (req, self); break;
1609
1610 case EIO_BUSY: 2317 case EIO_UTIME:
1611#ifdef _WIN32 2318 case EIO_FUTIME:
1612 Sleep (req->nv1 * 1000.); 2319 {
2320 struct timespec ts[2];
2321 struct timespec *times;
2322
2323 if (req->nv1 != -1. || req->nv2 != -1.)
2324 {
2325 ts[0].tv_sec = req->nv1;
2326 ts[0].tv_nsec = (req->nv1 - ts[0].tv_sec) * 1e9;
2327 ts[1].tv_sec = req->nv2;
2328 ts[1].tv_nsec = (req->nv2 - ts[1].tv_sec) * 1e9;
2329
2330 times = ts;
2331 }
2332 else
2333 times = 0;
2334
2335 req->result = req->type == EIO_FUTIME
2336 ? futimens (req->int1, times)
2337 : utimensat (dirfd, req->ptr1, times, 0);
2338 }
2339 break;
2340
1613#else 2341#else
1614 {
1615 struct timeval tv;
1616 2342
1617 tv.tv_sec = req->nv1; 2343 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1618 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; 2344 req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
2345 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2346 req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
2347 case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
2348 case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
2349 case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
2350 case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
1619 2351
1620 req->result = select (0, 0, 0, 0, &tv); 2352 case EIO_UNLINK: req->result = unlink (path ); break;
1621 } 2353 case EIO_RMDIR: req->result = rmdir (path ); break;
1622#endif 2354 case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
1623 break; 2355 case EIO_RENAME: req->result = rename (path , req->ptr2); break;
2356 case EIO_LINK: req->result = link (path , req->ptr2); break;
2357 case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
2358 case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
2359 case EIO_READLINK: ALLOC (PATH_MAX);
2360 req->result = readlink (path, req->ptr2, PATH_MAX); break;
2361 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2362 req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
1624 2363
1625 case EIO_UTIME: 2364 case EIO_UTIME:
1626 case EIO_FUTIME: 2365 case EIO_FUTIME:
1627 { 2366 {
1628 struct timeval tv[2]; 2367 struct timeval tv[2];
1629 struct timeval *times; 2368 struct timeval *times;
1630 2369
1631 if (req->nv1 != -1. || req->nv2 != -1.) 2370 if (req->nv1 != -1. || req->nv2 != -1.)
1632 { 2371 {
1633 tv[0].tv_sec = req->nv1; 2372 tv[0].tv_sec = req->nv1;
1634 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.; 2373 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
1635 tv[1].tv_sec = req->nv2; 2374 tv[1].tv_sec = req->nv2;
1636 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.; 2375 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
1637 2376
1638 times = tv; 2377 times = tv;
1639 } 2378 }
1640 else 2379 else
1641 times = 0; 2380 times = 0;
1642
1643 2381
1644 req->result = req->type == EIO_FUTIME 2382 req->result = req->type == EIO_FUTIME
1645 ? futimes (req->int1, times) 2383 ? futimes (req->int1, times)
1646 : utimes (req->ptr1, times); 2384 : utimes (req->ptr1, times);
1647 } 2385 }
1648 break; 2386 break;
1649 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
1650 case EIO_GROUP: 2437 case EIO_GROUP:
1651 abort (); /* handled in eio_request */ 2438 abort (); /* handled in eio_request */
1652 2439
1653 case EIO_NOP: 2440 case EIO_NOP:
1654 req->result = 0; 2441 req->result = 0;
1655 break; 2442 break;
1656 2443
1657 case EIO_CUSTOM: 2444 case EIO_CUSTOM:
1658 ((void (*)(eio_req *))req->feed) (req); 2445 req->feed (req);
1659 break; 2446 break;
1660 2447
1661 default: 2448 default:
2449 errno = ENOSYS;
1662 req->result = -1; 2450 req->result = -1;
1663 break; 2451 break;
1664 } 2452 }
1665 2453
1666 req->errorno = errno; 2454 req->errorno = errno;
1691eio_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)
1692{ 2480{
1693 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;
1694} 2482}
1695 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
1696eio_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)
1697{ 2500{
1698 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;
1699} 2502}
1700 2503
1701eio_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)
1702{ 2505{
1703 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;
1704} 2507}
1705 2508
1706eio_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)
1707{ 2510{
1708 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;
1709} 2517}
1710 2518
1711eio_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)
1712{ 2520{
1713 REQ (EIO_CLOSE); req->int1 = fd; SEND; 2521 REQ (EIO_CLOSE); req->int1 = fd; SEND;
1731eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) 2539eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1732{ 2540{
1733 REQ (EIO_FSTAT); req->int1 = fd; SEND; 2541 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1734} 2542}
1735 2543
2544eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
2545{
2546 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
2547}
2548
1736eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) 2549eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1737{ 2550{
1738 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; 2551 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1739} 2552}
1740 2553
1746eio_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)
1747{ 2560{
1748 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2561 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1749} 2562}
1750 2563
1751eio_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)
1752{ 2565{
1753 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;
1754} 2567}
1755 2568
1756eio_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)
1776eio_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)
1777{ 2590{
1778 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2591 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1779} 2592}
1780 2593
1781eio_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)
1782{ 2595{
1783 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;
1784} 2597}
1785 2598
1786eio_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)
1802eio_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)
1803{ 2616{
1804 return eio__1path (EIO_READLINK, path, pri, cb, data); 2617 return eio__1path (EIO_READLINK, path, pri, cb, data);
1805} 2618}
1806 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
1807eio_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)
1808{ 2626{
1809 return eio__1path (EIO_STAT, path, pri, cb, data); 2627 return eio__1path (EIO_STAT, path, pri, cb, data);
1810} 2628}
1811 2629
1812eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) 2630eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1813{ 2631{
1814 return eio__1path (EIO_LSTAT, path, pri, cb, data); 2632 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1815} 2633}
1816 2634
2635eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
2636{
2637 return eio__1path (EIO_STATVFS, path, pri, cb, data);
2638}
2639
1817eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) 2640eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1818{ 2641{
1819 return eio__1path (EIO_UNLINK, path, pri, cb, data); 2642 return eio__1path (EIO_UNLINK, path, pri, cb, data);
1820} 2643}
1821 2644
1829 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND; 2652 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1830} 2653}
1831 2654
1832eio_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)
1833{ 2656{
1834 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;
1835} 2658}
1836 2659
1837static eio_req * 2660static eio_req *
1838eio__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)
1839{ 2662{
1863eio_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)
1864{ 2687{
1865 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2688 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1866} 2689}
1867 2690
1868eio_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)
1869{ 2692{
1870 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND; 2693 REQ (EIO_CUSTOM); req->feed = execute; SEND;
1871} 2694}
1872 2695
1873#endif 2696#endif
1874 2697
1875eio_req *eio_grp (eio_cb cb, void *data) 2698eio_req *eio_grp (eio_cb cb, void *data)
1884#undef SEND 2707#undef SEND
1885 2708
1886/*****************************************************************************/ 2709/*****************************************************************************/
1887/* grp functions */ 2710/* grp functions */
1888 2711
2712void
1889void 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)
1890{ 2714{
1891 grp->int2 = limit; 2715 grp->int2 = limit;
1892 grp->feed = feed; 2716 grp->feed = feed;
1893 2717
1894 grp_try_feed (grp); 2718 grp_try_feed (grp);
1895} 2719}
1896 2720
2721void
1897void eio_grp_limit (eio_req *grp, int limit) 2722eio_grp_limit (eio_req *grp, int limit)
1898{ 2723{
1899 grp->int2 = limit; 2724 grp->int2 = limit;
1900 2725
1901 grp_try_feed (grp); 2726 grp_try_feed (grp);
1902} 2727}
1903 2728
2729void
1904void eio_grp_add (eio_req *grp, eio_req *req) 2730eio_grp_add (eio_req *grp, eio_req *req)
1905{ 2731{
1906 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));
1907 2733
1908 grp->flags |= EIO_FLAG_GROUPADD; 2734 grp->flags |= EIO_FLAG_GROUPADD;
1909 2735
1920} 2746}
1921 2747
1922/*****************************************************************************/ 2748/*****************************************************************************/
1923/* misc garbage */ 2749/* misc garbage */
1924 2750
2751eio_ssize_t
1925ssize_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)
1926{ 2753{
1927 etp_worker wrk;
1928 ssize_t ret;
1929
1930 wrk.dbuf = 0;
1931
1932 ret = eio__sendfile (ofd, ifd, offset, count, &wrk); 2754 return eio__sendfile (ofd, ifd, offset, count);
1933
1934 if (wrk.dbuf)
1935 free (wrk.dbuf);
1936
1937 return ret;
1938} 2755}
1939 2756

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines