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

Comparing cvsroot/libeio/eio.c (file contents):
Revision 1.70 by root, Fri Jun 10 06:55:10 2011 UTC vs.
Revision 1.104 by root, Mon Sep 26 16:54:25 2011 UTC

40#ifndef _WIN32 40#ifndef _WIN32
41# include "config.h" 41# include "config.h"
42#endif 42#endif
43 43
44#include "eio.h" 44#include "eio.h"
45#include "ecb.h"
45 46
46#ifdef EIO_STACKSIZE 47#ifdef EIO_STACKSIZE
47# define XTHREAD_STACKSIZE EIO_STACKSIZE 48# define X_STACKSIZE EIO_STACKSIZE
48#endif 49#endif
49#include "xthread.h" 50#include "xthread.h"
50 51
51#include <errno.h> 52#include <errno.h>
52#include <stddef.h> 53#include <stddef.h>
53#include <stdlib.h> 54#include <stdlib.h>
54#include <string.h> 55#include <string.h>
55#include <errno.h> 56#include <errno.h>
56#include <sys/types.h> 57#include <sys/types.h>
57#include <sys/stat.h> 58#include <sys/stat.h>
58#include <sys/statvfs.h>
59#include <limits.h> 59#include <limits.h>
60#include <fcntl.h> 60#include <fcntl.h>
61#include <assert.h> 61#include <assert.h>
62 62
63/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */ 63/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
64/* intptr_t only comes form stdint.h, says idiot openbsd coder */ 64/* intptr_t only comes from stdint.h, says idiot openbsd coder */
65#if HAVE_STDINT_H 65#if HAVE_STDINT_H
66# include <stdint.h> 66# include <stdint.h>
67#endif 67#endif
68 68
69#ifndef ECANCELED
70# define ECANCELED EDOM
71#endif
72#ifndef ELOOP
73# define ELOOP EDOM
74#endif
75
76#if !defined(ENOTSOCK) && defined(WSAENOTSOCK)
77# define ENOTSOCK WSAENOTSOCK
78#endif
79
80static void eio_destroy (eio_req *req);
81
69#ifndef EIO_FINISH 82#ifndef EIO_FINISH
70# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0 83# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
71#endif 84#endif
72 85
73#ifndef EIO_DESTROY 86#ifndef EIO_DESTROY
76 89
77#ifndef EIO_FEED 90#ifndef EIO_FEED
78# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0) 91# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)
79#endif 92#endif
80 93
94#ifndef EIO_FD_TO_WIN32_HANDLE
95# define EIO_FD_TO_WIN32_HANDLE(fd) _get_osfhandle (fd)
96#endif
97#ifndef EIO_WIN32_HANDLE_TO_FD
98# define EIO_WIN32_HANDLE_TO_FD(handle) _open_osfhandle (handle, 0)
99#endif
100
101#define EIO_ERRNO(errval,retval) ((errno = errval), retval)
102
103#define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1)
104
81#ifdef _WIN32 105#ifdef _WIN32
82 106
83 /*doh*/ 107 #undef PAGESIZE
108 #define PAGESIZE 4096 /* GetSystemInfo? */
109
110 /* TODO: look at how perl does stat (non-sloppy), unlink (ro-files), utime, link */
111
112 #ifdef EIO_STRUCT_STATI64
113 /* look at perl's non-sloppy stat */
114 #define stat(path,buf) _stati64 (path,buf)
115 #define fstat(fd,buf) _fstati64 (fd,buf)
116 #endif
117 #define lstat(path,buf) stat (path,buf)
118 #define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
119 #define mkdir(path,mode) _mkdir (path)
120 #define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
121
122 #define chmod(path,mode) _chmod (path, mode)
123 #define dup(fd) _dup (fd)
124 #define dup2(fd1,fd2) _dup2 (fd1, fd2)
125
126 #define fchmod(fd,mode) EIO_ENOSYS ()
127 #define chown(path,uid,gid) EIO_ENOSYS ()
128 #define fchown(fd,uid,gid) EIO_ENOSYS ()
129 #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
130 #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
131 #define mknod(path,mode,dev) EIO_ENOSYS ()
132 #define sync() EIO_ENOSYS ()
133 #define readlink(path,buf,s) EIO_ENOSYS ()
134 #define statvfs(path,buf) EIO_ENOSYS ()
135 #define fstatvfs(fd,buf) EIO_ENOSYS ()
136
137 /* rename() uses MoveFile, which fails to overwrite */
138 #define rename(old,neu) eio__rename (old, neu)
139
140 static int
141 eio__rename (const char *old, const char *neu)
142 {
143 if (MoveFileEx (old, neu, MOVEFILE_REPLACE_EXISTING))
144 return 0;
145
146 /* should steal _dosmaperr */
147 switch (GetLastError ())
148 {
149 case ERROR_FILE_NOT_FOUND:
150 case ERROR_PATH_NOT_FOUND:
151 case ERROR_INVALID_DRIVE:
152 case ERROR_NO_MORE_FILES:
153 case ERROR_BAD_NETPATH:
154 case ERROR_BAD_NET_NAME:
155 case ERROR_BAD_PATHNAME:
156 case ERROR_FILENAME_EXCED_RANGE:
157 errno = ENOENT;
158 break;
159
160 default:
161 errno = EACCES;
162 break;
163 }
164
165 return -1;
166 }
167
168 /* we could even stat and see if it exists */
169 static int
170 symlink (const char *old, const char *neu)
171 {
172 #if WINVER >= 0x0600
173 if (CreateSymbolicLink (neu, old, 1))
174 return 0;
175
176 if (CreateSymbolicLink (neu, old, 0))
177 return 0;
178 #endif
179
180 return EIO_ERRNO (ENOENT, -1);
181 }
182
183 /* POSIX API only */
184 #define CreateHardLink(neu,old,flags) 0
185 #define CreateSymbolicLink(neu,old,flags) 0
186
187 struct statvfs
188 {
189 int dummy;
190 };
191
192 #define DT_DIR EIO_DT_DIR
193 #define DT_REG EIO_DT_REG
194 #define D_NAME(entp) entp.cFileName
195 #define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
196
84#else 197#else
85 198
86# include <sys/time.h> 199 #include <sys/time.h>
87# include <sys/select.h> 200 #include <sys/select.h>
201 #include <sys/statvfs.h>
88# include <unistd.h> 202 #include <unistd.h>
203 #include <signal.h>
204 #include <dirent.h>
205
206 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
207 #include <sys/mman.h>
208 #endif
209
210 #define D_NAME(entp) entp->d_name
211
212 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
213 #if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
214 #define _DIRENT_HAVE_D_TYPE /* sigh */
215 #define D_INO(de) (de)->d_fileno
216 #define D_NAMLEN(de) (de)->d_namlen
217 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
218 #define D_INO(de) (de)->d_ino
219 #endif
220
221 #ifdef _D_EXACT_NAMLEN
222 #undef D_NAMLEN
223 #define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
224 #endif
225
226 #ifdef _DIRENT_HAVE_D_TYPE
227 #define D_TYPE(de) (de)->d_type
228 #endif
229
230 #ifndef EIO_STRUCT_DIRENT
231 #define EIO_STRUCT_DIRENT struct dirent
232 #endif
233
234#endif
235
236#if HAVE_UTIMES
89# include <utime.h> 237# include <utime.h>
90# include <signal.h>
91# include <dirent.h>
92
93#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
94# include <sys/mman.h>
95#endif
96
97/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
98# if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
99# define _DIRENT_HAVE_D_TYPE /* sigh */
100# define D_INO(de) (de)->d_fileno
101# define D_NAMLEN(de) (de)->d_namlen
102# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
103# define D_INO(de) (de)->d_ino
104# endif 238#endif
105 239
106#ifdef _D_EXACT_NAMLEN 240#if HAVE_SYS_SYSCALL_H
107# undef D_NAMLEN 241# include <sys/syscall.h>
108# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
109#endif
110
111# ifdef _DIRENT_HAVE_D_TYPE
112# define D_TYPE(de) (de)->d_type
113# endif 242#endif
114 243
115# ifndef EIO_STRUCT_DIRENT 244#if HAVE_SYS_PRCTL_H
116# define EIO_STRUCT_DIRENT struct dirent 245# include <sys/prctl.h>
117# endif
118
119#endif 246#endif
120 247
121#if HAVE_SENDFILE 248#if HAVE_SENDFILE
122# if __linux 249# if __linux
123# include <sys/sendfile.h> 250# include <sys/sendfile.h>
138#endif 265#endif
139#ifndef D_INO 266#ifndef D_INO
140# define D_INO(de) 0 267# define D_INO(de) 0
141#endif 268#endif
142#ifndef D_NAMLEN 269#ifndef D_NAMLEN
143# define D_NAMLEN(de) strlen ((de)->d_name) 270# define D_NAMLEN(entp) strlen (D_NAME (entp))
144#endif 271#endif
145 272
146/* used for struct dirent, AIX doesn't provide it */ 273/* used for struct dirent, AIX doesn't provide it */
147#ifndef NAME_MAX 274#ifndef NAME_MAX
148# define NAME_MAX 4096 275# define NAME_MAX 4096
155 282
156/* buffer size for various temporary buffers */ 283/* buffer size for various temporary buffers */
157#define EIO_BUFSIZE 65536 284#define EIO_BUFSIZE 65536
158 285
159#define dBUF \ 286#define dBUF \
160 char *eio_buf; \
161 ETP_WORKER_LOCK (self); \
162 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 287 char *eio_buf = malloc (EIO_BUFSIZE); \
163 ETP_WORKER_UNLOCK (self); \
164 errno = ENOMEM; \ 288 errno = ENOMEM; \
165 if (!eio_buf) \ 289 if (!eio_buf) \
166 return -1; 290 return -1
291
292#define FUBd \
293 free (eio_buf)
167 294
168#define EIO_TICKS ((1000000 + 1023) >> 10) 295#define EIO_TICKS ((1000000 + 1023) >> 10)
169 296
170/*****************************************************************************/ 297/*****************************************************************************/
171 298
172#if __GNUC__ >= 3 299struct tmpbuf
173# define expect(expr,value) __builtin_expect ((expr),(value)) 300{
174#else 301 void *ptr;
175# define expect(expr,value) (expr) 302 int len;
176#endif 303};
177 304
178#define expect_false(expr) expect ((expr) != 0, 0) 305static void *
179#define expect_true(expr) expect ((expr) != 0, 1) 306tmpbuf_get (struct tmpbuf *buf, int len)
307{
308 if (buf->len < len)
309 {
310 free (buf->ptr);
311 buf->ptr = malloc (buf->len = len);
312 }
313
314 return buf->ptr;
315}
180 316
181/*****************************************************************************/ 317/*****************************************************************************/
182 318
183#define ETP_PRI_MIN EIO_PRI_MIN 319#define ETP_PRI_MIN EIO_PRI_MIN
184#define ETP_PRI_MAX EIO_PRI_MAX 320#define ETP_PRI_MAX EIO_PRI_MAX
190static int eio_finish (eio_req *req); 326static int eio_finish (eio_req *req);
191#define ETP_FINISH(req) eio_finish (req) 327#define ETP_FINISH(req) eio_finish (req)
192static void eio_execute (struct etp_worker *self, eio_req *req); 328static void eio_execute (struct etp_worker *self, eio_req *req);
193#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) 329#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
194 330
195#define ETP_WORKER_CLEAR(req) \
196 if (wrk->dbuf) \
197 { \
198 free (wrk->dbuf); \
199 wrk->dbuf = 0; \
200 } \
201 \
202 if (wrk->dirp) \
203 { \
204 closedir (wrk->dirp); \
205 wrk->dirp = 0; \
206 }
207
208#define ETP_WORKER_COMMON \
209 void *dbuf; \
210 DIR *dirp;
211
212/*****************************************************************************/ 331/*****************************************************************************/
213 332
214#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) 333#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
215 334
216/* calculate time difference in ~1/EIO_TICKS of a second */ 335/* calculate time difference in ~1/EIO_TICKS of a second */
336ecb_inline int
217static int tvdiff (struct timeval *tv1, struct timeval *tv2) 337tvdiff (struct timeval *tv1, struct timeval *tv2)
218{ 338{
219 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 339 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
220 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 340 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
221} 341}
222 342
226static void (*done_poll_cb) (void); 346static void (*done_poll_cb) (void);
227 347
228static unsigned int max_poll_time; /* reslock */ 348static unsigned int max_poll_time; /* reslock */
229static unsigned int max_poll_reqs; /* reslock */ 349static unsigned int max_poll_reqs; /* reslock */
230 350
231static volatile unsigned int nreqs; /* reqlock */ 351static unsigned int nreqs; /* reqlock */
232static volatile unsigned int nready; /* reqlock */ 352static unsigned int nready; /* reqlock */
233static volatile unsigned int npending; /* reqlock */ 353static unsigned int npending; /* reqlock */
234static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */ 354static unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
235static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */ 355static unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
236 356
237static xmutex_t wrklock; 357static xmutex_t wrklock;
238static xmutex_t reslock; 358static xmutex_t reslock;
239static xmutex_t reqlock; 359static xmutex_t reqlock;
240static xcond_t reqwait; 360static xcond_t reqwait;
243/* 363/*
244 * make our pread/pwrite emulation safe against themselves, but not against 364 * make our pread/pwrite emulation safe against themselves, but not against
245 * normal read/write by using a mutex. slows down execution a lot, 365 * normal read/write by using a mutex. slows down execution a lot,
246 * but that's your problem, not mine. 366 * but that's your problem, not mine.
247 */ 367 */
248static xmutex_t preadwritelock = X_MUTEX_INIT; 368static xmutex_t preadwritelock;
249#endif 369#endif
250 370
251typedef struct etp_worker 371typedef struct etp_worker
252{ 372{
373 struct tmpbuf tmpbuf;
374
253 /* locked by wrklock */ 375 /* locked by wrklock */
254 struct etp_worker *prev, *next; 376 struct etp_worker *prev, *next;
255 377
256 xthread_t tid; 378 xthread_t tid;
257 379
258 /* locked by reslock, reqlock or wrklock */ 380#ifdef ETP_WORKER_COMMON
259 ETP_REQ *req; /* currently processed request */
260
261 ETP_WORKER_COMMON 381 ETP_WORKER_COMMON
382#endif
262} etp_worker; 383} etp_worker;
263 384
264static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */ 385static etp_worker wrk_first; /* NOT etp */
265 386
266#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock) 387#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
267#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock) 388#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
268 389
269/* worker threads management */ 390/* worker threads management */
270 391
392static void ecb_cold
271static void etp_worker_clear (etp_worker *wrk) 393etp_worker_clear (etp_worker *wrk)
272{ 394{
273 ETP_WORKER_CLEAR (wrk);
274} 395}
275 396
397static void ecb_cold
276static void etp_worker_free (etp_worker *wrk) 398etp_worker_free (etp_worker *wrk)
277{ 399{
400 free (wrk->tmpbuf.ptr);
401
278 wrk->next->prev = wrk->prev; 402 wrk->next->prev = wrk->prev;
279 wrk->prev->next = wrk->next; 403 wrk->prev->next = wrk->next;
280 404
281 free (wrk); 405 free (wrk);
282} 406}
283 407
284static unsigned int etp_nreqs (void) 408static unsigned int
409etp_nreqs (void)
285{ 410{
286 int retval; 411 int retval;
287 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 412 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
288 retval = nreqs; 413 retval = nreqs;
289 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 414 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
290 return retval; 415 return retval;
291} 416}
292 417
293static unsigned int etp_nready (void) 418static unsigned int
419etp_nready (void)
294{ 420{
295 unsigned int retval; 421 unsigned int retval;
296 422
297 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 423 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
298 retval = nready; 424 retval = nready;
299 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 425 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
300 426
301 return retval; 427 return retval;
302} 428}
303 429
304static unsigned int etp_npending (void) 430static unsigned int
431etp_npending (void)
305{ 432{
306 unsigned int retval; 433 unsigned int retval;
307 434
308 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 435 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
309 retval = npending; 436 retval = npending;
310 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 437 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
311 438
312 return retval; 439 return retval;
313} 440}
314 441
315static unsigned int etp_nthreads (void) 442static unsigned int
443etp_nthreads (void)
316{ 444{
317 unsigned int retval; 445 unsigned int retval;
318 446
319 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 447 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
320 retval = started; 448 retval = started;
334} etp_reqq; 462} etp_reqq;
335 463
336static etp_reqq req_queue; 464static etp_reqq req_queue;
337static etp_reqq res_queue; 465static etp_reqq res_queue;
338 466
467static void ecb_noinline ecb_cold
468reqq_init (etp_reqq *q)
469{
470 int pri;
471
472 for (pri = 0; pri < ETP_NUM_PRI; ++pri)
473 q->qs[pri] = q->qe[pri] = 0;
474
475 q->size = 0;
476}
477
478static int ecb_noinline
339static int reqq_push (etp_reqq *q, ETP_REQ *req) 479reqq_push (etp_reqq *q, ETP_REQ *req)
340{ 480{
341 int pri = req->pri; 481 int pri = req->pri;
342 req->next = 0; 482 req->next = 0;
343 483
344 if (q->qe[pri]) 484 if (q->qe[pri])
350 q->qe[pri] = q->qs[pri] = req; 490 q->qe[pri] = q->qs[pri] = req;
351 491
352 return q->size++; 492 return q->size++;
353} 493}
354 494
495static ETP_REQ * ecb_noinline
355static ETP_REQ *reqq_shift (etp_reqq *q) 496reqq_shift (etp_reqq *q)
356{ 497{
357 int pri; 498 int pri;
358 499
359 if (!q->size) 500 if (!q->size)
360 return 0; 501 return 0;
375 } 516 }
376 517
377 abort (); 518 abort ();
378} 519}
379 520
380static void etp_thread_init (void) 521static int ecb_cold
522etp_init (void (*want_poll)(void), void (*done_poll)(void))
381{ 523{
382 X_MUTEX_CREATE (wrklock); 524 X_MUTEX_CREATE (wrklock);
383 X_MUTEX_CREATE (reslock); 525 X_MUTEX_CREATE (reslock);
384 X_MUTEX_CREATE (reqlock); 526 X_MUTEX_CREATE (reqlock);
385 X_COND_CREATE (reqwait); 527 X_COND_CREATE (reqwait);
386}
387 528
388static void etp_atfork_prepare (void) 529 reqq_init (&req_queue);
389{ 530 reqq_init (&res_queue);
390 X_LOCK (wrklock);
391 X_LOCK (reqlock);
392 X_LOCK (reslock);
393#if !HAVE_PREADWRITE
394 X_LOCK (preadwritelock);
395#endif
396}
397 531
398static void etp_atfork_parent (void) 532 wrk_first.next =
399{ 533 wrk_first.prev = &wrk_first;
400#if !HAVE_PREADWRITE
401 X_UNLOCK (preadwritelock);
402#endif
403 X_UNLOCK (reslock);
404 X_UNLOCK (reqlock);
405 X_UNLOCK (wrklock);
406}
407
408static void etp_atfork_child (void)
409{
410 ETP_REQ *prv;
411
412 while ((prv = reqq_shift (&req_queue)))
413 ETP_DESTROY (prv);
414
415 while ((prv = reqq_shift (&res_queue)))
416 ETP_DESTROY (prv);
417
418 while (wrk_first.next != &wrk_first)
419 {
420 etp_worker *wrk = wrk_first.next;
421
422 if (wrk->req)
423 ETP_DESTROY (wrk->req);
424
425 etp_worker_clear (wrk);
426 etp_worker_free (wrk);
427 }
428 534
429 started = 0; 535 started = 0;
430 idle = 0; 536 idle = 0;
431 nreqs = 0; 537 nreqs = 0;
432 nready = 0; 538 nready = 0;
433 npending = 0; 539 npending = 0;
434 540
435 etp_thread_init ();
436}
437
438static void
439etp_once_init (void)
440{
441 etp_thread_init ();
442 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
443}
444
445static int
446etp_init (void (*want_poll)(void), void (*done_poll)(void))
447{
448 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
449
450 pthread_once (&doinit, etp_once_init);
451
452 want_poll_cb = want_poll; 541 want_poll_cb = want_poll;
453 done_poll_cb = done_poll; 542 done_poll_cb = done_poll;
454 543
455 return 0; 544 return 0;
456} 545}
457 546
458X_THREAD_PROC (etp_proc); 547X_THREAD_PROC (etp_proc);
459 548
549static void ecb_cold
460static void etp_start_thread (void) 550etp_start_thread (void)
461{ 551{
462 etp_worker *wrk = calloc (1, sizeof (etp_worker)); 552 etp_worker *wrk = calloc (1, sizeof (etp_worker));
463 553
464 /*TODO*/ 554 /*TODO*/
465 assert (("unable to allocate worker thread data", wrk)); 555 assert (("unable to allocate worker thread data", wrk));
478 free (wrk); 568 free (wrk);
479 569
480 X_UNLOCK (wrklock); 570 X_UNLOCK (wrklock);
481} 571}
482 572
573static void
483static void etp_maybe_start_thread (void) 574etp_maybe_start_thread (void)
484{ 575{
485 if (expect_true (etp_nthreads () >= wanted)) 576 if (ecb_expect_true (etp_nthreads () >= wanted))
486 return; 577 return;
487 578
488 /* todo: maybe use idle here, but might be less exact */ 579 /* todo: maybe use idle here, but might be less exact */
489 if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ())) 580 if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
490 return; 581 return;
491 582
492 etp_start_thread (); 583 etp_start_thread ();
493} 584}
494 585
586static void ecb_cold
495static void etp_end_thread (void) 587etp_end_thread (void)
496{ 588{
497 eio_req *req = calloc (1, sizeof (eio_req)); 589 eio_req *req = calloc (1, sizeof (eio_req));
498 590
499 req->type = -1; 591 req->type = -1;
500 req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 592 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
507 X_LOCK (wrklock); 599 X_LOCK (wrklock);
508 --started; 600 --started;
509 X_UNLOCK (wrklock); 601 X_UNLOCK (wrklock);
510} 602}
511 603
512static int etp_poll (void) 604static int
605etp_poll (void)
513{ 606{
514 unsigned int maxreqs; 607 unsigned int maxreqs;
515 unsigned int maxtime; 608 unsigned int maxtime;
516 struct timeval tv_start, tv_now; 609 struct timeval tv_start, tv_now;
517 610
547 640
548 X_LOCK (reqlock); 641 X_LOCK (reqlock);
549 --nreqs; 642 --nreqs;
550 X_UNLOCK (reqlock); 643 X_UNLOCK (reqlock);
551 644
552 if (expect_false (req->type == EIO_GROUP && req->size)) 645 if (ecb_expect_false (req->type == EIO_GROUP && req->size))
553 { 646 {
554 req->int1 = 1; /* mark request as delayed */ 647 req->int1 = 1; /* mark request as delayed */
555 continue; 648 continue;
556 } 649 }
557 else 650 else
558 { 651 {
559 int res = ETP_FINISH (req); 652 int res = ETP_FINISH (req);
560 if (expect_false (res)) 653 if (ecb_expect_false (res))
561 return res; 654 return res;
562 } 655 }
563 656
564 if (expect_false (maxreqs && !--maxreqs)) 657 if (ecb_expect_false (maxreqs && !--maxreqs))
565 break; 658 break;
566 659
567 if (maxtime) 660 if (maxtime)
568 { 661 {
569 gettimeofday (&tv_now, 0); 662 gettimeofday (&tv_now, 0);
575 668
576 errno = EAGAIN; 669 errno = EAGAIN;
577 return -1; 670 return -1;
578} 671}
579 672
673static void
580static void etp_cancel (ETP_REQ *req) 674etp_cancel (ETP_REQ *req)
581{ 675{
582 X_LOCK (wrklock); 676 req->cancelled = 1;
583 req->flags |= EIO_FLAG_CANCELLED;
584 X_UNLOCK (wrklock);
585 677
586 eio_grp_cancel (req); 678 eio_grp_cancel (req);
587} 679}
588 680
681static void
589static void etp_submit (ETP_REQ *req) 682etp_submit (ETP_REQ *req)
590{ 683{
591 req->pri -= ETP_PRI_MIN; 684 req->pri -= ETP_PRI_MIN;
592 685
593 if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN; 686 if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
594 if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 687 if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
595 688
596 if (expect_false (req->type == EIO_GROUP)) 689 if (ecb_expect_false (req->type == EIO_GROUP))
597 { 690 {
598 /* I hope this is worth it :/ */ 691 /* I hope this is worth it :/ */
599 X_LOCK (reqlock); 692 X_LOCK (reqlock);
600 ++nreqs; 693 ++nreqs;
601 X_UNLOCK (reqlock); 694 X_UNLOCK (reqlock);
620 713
621 etp_maybe_start_thread (); 714 etp_maybe_start_thread ();
622 } 715 }
623} 716}
624 717
718static void ecb_cold
625static void etp_set_max_poll_time (double nseconds) 719etp_set_max_poll_time (double nseconds)
626{ 720{
627 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 721 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
628 max_poll_time = nseconds * EIO_TICKS; 722 max_poll_time = nseconds * EIO_TICKS;
629 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 723 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
630} 724}
631 725
726static void ecb_cold
632static void etp_set_max_poll_reqs (unsigned int maxreqs) 727etp_set_max_poll_reqs (unsigned int maxreqs)
633{ 728{
634 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 729 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
635 max_poll_reqs = maxreqs; 730 max_poll_reqs = maxreqs;
636 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 731 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
637} 732}
638 733
734static void ecb_cold
639static void etp_set_max_idle (unsigned int nthreads) 735etp_set_max_idle (unsigned int nthreads)
640{ 736{
641 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 737 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
642 max_idle = nthreads; 738 max_idle = nthreads;
643 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 739 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
644} 740}
645 741
742static void ecb_cold
646static void etp_set_idle_timeout (unsigned int seconds) 743etp_set_idle_timeout (unsigned int seconds)
647{ 744{
648 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 745 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
649 idle_timeout = seconds; 746 idle_timeout = seconds;
650 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 747 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
651} 748}
652 749
750static void ecb_cold
653static void etp_set_min_parallel (unsigned int nthreads) 751etp_set_min_parallel (unsigned int nthreads)
654{ 752{
655 if (wanted < nthreads) 753 if (wanted < nthreads)
656 wanted = nthreads; 754 wanted = nthreads;
657} 755}
658 756
757static void ecb_cold
659static void etp_set_max_parallel (unsigned int nthreads) 758etp_set_max_parallel (unsigned int nthreads)
660{ 759{
661 if (wanted > nthreads) 760 if (wanted > nthreads)
662 wanted = nthreads; 761 wanted = nthreads;
663 762
664 while (started > wanted) 763 while (started > wanted)
665 etp_end_thread (); 764 etp_end_thread ();
666} 765}
667 766
668/*****************************************************************************/ 767/*****************************************************************************/
669 768
769static void
670static void grp_try_feed (eio_req *grp) 770grp_try_feed (eio_req *grp)
671{ 771{
672 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 772 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
673 { 773 {
674 grp->flags &= ~EIO_FLAG_GROUPADD; 774 grp->flags &= ~EIO_FLAG_GROUPADD;
675 775
682 break; 782 break;
683 } 783 }
684 } 784 }
685} 785}
686 786
787static int
687static int grp_dec (eio_req *grp) 788grp_dec (eio_req *grp)
688{ 789{
689 --grp->size; 790 --grp->size;
690 791
691 /* call feeder, if applicable */ 792 /* call feeder, if applicable */
692 grp_try_feed (grp); 793 grp_try_feed (grp);
696 return eio_finish (grp); 797 return eio_finish (grp);
697 else 798 else
698 return 0; 799 return 0;
699} 800}
700 801
802static void
701void eio_destroy (eio_req *req) 803eio_destroy (eio_req *req)
702{ 804{
703 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); 805 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
704 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); 806 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
705 807
706 EIO_DESTROY (req); 808 EIO_DESTROY (req);
707} 809}
708 810
811static int
709static int eio_finish (eio_req *req) 812eio_finish (eio_req *req)
710{ 813{
711 int res = EIO_FINISH (req); 814 int res = EIO_FINISH (req);
712 815
713 if (req->grp) 816 if (req->grp)
714 { 817 {
722 if (grp->grp_first == req) 825 if (grp->grp_first == req)
723 grp->grp_first = req->grp_next; 826 grp->grp_first = req->grp_next;
724 827
725 res2 = grp_dec (grp); 828 res2 = grp_dec (grp);
726 829
727 if (!res && res2) 830 if (!res)
728 res = res2; 831 res = res2;
729 } 832 }
730 833
731 eio_destroy (req); 834 eio_destroy (req);
732 835
733 return res; 836 return res;
734} 837}
735 838
839void
736void eio_grp_cancel (eio_req *grp) 840eio_grp_cancel (eio_req *grp)
737{ 841{
738 for (grp = grp->grp_first; grp; grp = grp->grp_next) 842 for (grp = grp->grp_first; grp; grp = grp->grp_next)
739 eio_cancel (grp); 843 eio_cancel (grp);
740} 844}
741 845
846void
742void eio_cancel (eio_req *req) 847eio_cancel (eio_req *req)
743{ 848{
744 etp_cancel (req); 849 etp_cancel (req);
745} 850}
746 851
852void
747void eio_submit (eio_req *req) 853eio_submit (eio_req *req)
748{ 854{
749 etp_submit (req); 855 etp_submit (req);
750} 856}
751 857
752unsigned int eio_nreqs (void) 858unsigned int
859eio_nreqs (void)
753{ 860{
754 return etp_nreqs (); 861 return etp_nreqs ();
755} 862}
756 863
757unsigned int eio_nready (void) 864unsigned int
865eio_nready (void)
758{ 866{
759 return etp_nready (); 867 return etp_nready ();
760} 868}
761 869
762unsigned int eio_npending (void) 870unsigned int
871eio_npending (void)
763{ 872{
764 return etp_npending (); 873 return etp_npending ();
765} 874}
766 875
767unsigned int eio_nthreads (void) 876unsigned int ecb_cold
877eio_nthreads (void)
768{ 878{
769 return etp_nthreads (); 879 return etp_nthreads ();
770} 880}
771 881
882void ecb_cold
772void eio_set_max_poll_time (double nseconds) 883eio_set_max_poll_time (double nseconds)
773{ 884{
774 etp_set_max_poll_time (nseconds); 885 etp_set_max_poll_time (nseconds);
775} 886}
776 887
888void ecb_cold
777void eio_set_max_poll_reqs (unsigned int maxreqs) 889eio_set_max_poll_reqs (unsigned int maxreqs)
778{ 890{
779 etp_set_max_poll_reqs (maxreqs); 891 etp_set_max_poll_reqs (maxreqs);
780} 892}
781 893
894void ecb_cold
782void eio_set_max_idle (unsigned int nthreads) 895eio_set_max_idle (unsigned int nthreads)
783{ 896{
784 etp_set_max_idle (nthreads); 897 etp_set_max_idle (nthreads);
785} 898}
786 899
900void ecb_cold
787void eio_set_idle_timeout (unsigned int seconds) 901eio_set_idle_timeout (unsigned int seconds)
788{ 902{
789 etp_set_idle_timeout (seconds); 903 etp_set_idle_timeout (seconds);
790} 904}
791 905
906void ecb_cold
792void eio_set_min_parallel (unsigned int nthreads) 907eio_set_min_parallel (unsigned int nthreads)
793{ 908{
794 etp_set_min_parallel (nthreads); 909 etp_set_min_parallel (nthreads);
795} 910}
796 911
912void ecb_cold
797void eio_set_max_parallel (unsigned int nthreads) 913eio_set_max_parallel (unsigned int nthreads)
798{ 914{
799 etp_set_max_parallel (nthreads); 915 etp_set_max_parallel (nthreads);
800} 916}
801 917
802int eio_poll (void) 918int eio_poll (void)
811# undef pread 927# undef pread
812# undef pwrite 928# undef pwrite
813# define pread eio__pread 929# define pread eio__pread
814# define pwrite eio__pwrite 930# define pwrite eio__pwrite
815 931
816static ssize_t 932static eio_ssize_t
817eio__pread (int fd, void *buf, size_t count, off_t offset) 933eio__pread (int fd, void *buf, size_t count, off_t offset)
818{ 934{
819 ssize_t res; 935 eio_ssize_t res;
820 off_t ooffset; 936 off_t ooffset;
821 937
822 X_LOCK (preadwritelock); 938 X_LOCK (preadwritelock);
823 ooffset = lseek (fd, 0, SEEK_CUR); 939 ooffset = lseek (fd, 0, SEEK_CUR);
824 lseek (fd, offset, SEEK_SET); 940 lseek (fd, offset, SEEK_SET);
827 X_UNLOCK (preadwritelock); 943 X_UNLOCK (preadwritelock);
828 944
829 return res; 945 return res;
830} 946}
831 947
832static ssize_t 948static eio_ssize_t
833eio__pwrite (int fd, void *buf, size_t count, off_t offset) 949eio__pwrite (int fd, void *buf, size_t count, off_t offset)
834{ 950{
835 ssize_t res; 951 eio_ssize_t res;
836 off_t ooffset; 952 off_t ooffset;
837 953
838 X_LOCK (preadwritelock); 954 X_LOCK (preadwritelock);
839 ooffset = lseek (fd, 0, SEEK_CUR); 955 ooffset = lseek (fd, 0, SEEK_CUR);
840 lseek (fd, offset, SEEK_SET); 956 lseek (fd, offset, SEEK_SET);
872#ifndef HAVE_FUTIMES 988#ifndef HAVE_FUTIMES
873 989
874# undef futimes 990# undef futimes
875# define futimes(fd,times) eio__futimes (fd, times) 991# define futimes(fd,times) eio__futimes (fd, times)
876 992
993static int
877static int eio__futimes (int fd, const struct timeval tv[2]) 994eio__futimes (int fd, const struct timeval tv[2])
878{ 995{
879 errno = ENOSYS; 996 errno = ENOSYS;
880 return -1; 997 return -1;
881} 998}
882 999
885#if !HAVE_FDATASYNC 1002#if !HAVE_FDATASYNC
886# undef fdatasync 1003# undef fdatasync
887# define fdatasync(fd) fsync (fd) 1004# define fdatasync(fd) fsync (fd)
888#endif 1005#endif
889 1006
1007static int
1008eio__syncfs (int fd)
1009{
1010 int res;
1011
1012#if HAVE_SYS_SYNCFS
1013 res = (int)syscall (__NR_syncfs, (int)(fd));
1014#else
1015 res = -1;
1016 errno = ENOSYS;
1017#endif
1018
1019 if (res < 0 && errno == ENOSYS && fd >= 0)
1020 sync ();
1021
1022 return res;
1023}
1024
890/* sync_file_range always needs emulation */ 1025/* sync_file_range always needs emulation */
891int 1026static int
892eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags) 1027eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
893{ 1028{
894#if HAVE_SYNC_FILE_RANGE 1029#if HAVE_SYNC_FILE_RANGE
895 int res; 1030 int res;
896 1031
913 /* even though we could play tricks with the flags, it's better to always 1048 /* even though we could play tricks with the flags, it's better to always
914 * call fdatasync, as that matches the expectation of its users best */ 1049 * call fdatasync, as that matches the expectation of its users best */
915 return fdatasync (fd); 1050 return fdatasync (fd);
916} 1051}
917 1052
1053static int
1054eio__fallocate (int fd, int mode, off_t offset, size_t len)
1055{
1056#if HAVE_FALLOCATE
1057 return fallocate (fd, mode, offset, len);
1058#else
1059 errno = ENOSYS;
1060 return -1;
1061#endif
1062}
1063
918#if !HAVE_READAHEAD 1064#if !HAVE_READAHEAD
919# undef readahead 1065# undef readahead
920# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 1066# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
921 1067
922static ssize_t 1068static eio_ssize_t
923eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) 1069eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
924{ 1070{
925 size_t todo = count; 1071 size_t todo = count;
926 dBUF; 1072 dBUF;
927 1073
932 pread (fd, eio_buf, len, offset); 1078 pread (fd, eio_buf, len, offset);
933 offset += len; 1079 offset += len;
934 todo -= len; 1080 todo -= len;
935 } 1081 }
936 1082
1083 FUBd;
1084
937 errno = 0; 1085 errno = 0;
938 return count; 1086 return count;
939} 1087}
940 1088
941#endif 1089#endif
942 1090
943/* sendfile always needs emulation */ 1091/* sendfile always needs emulation */
944static ssize_t 1092static eio_ssize_t
945eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) 1093eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
946{ 1094{
947 ssize_t written = 0; 1095 eio_ssize_t written = 0;
948 ssize_t res; 1096 eio_ssize_t res;
949 1097
950 if (!count) 1098 if (!count)
951 return 0; 1099 return 0;
952 1100
953 for (;;) 1101 for (;;)
954 { 1102 {
1103#ifdef __APPLE__
1104# undef HAVE_SENDFILE /* broken, as everything on os x */
1105#endif
955#if HAVE_SENDFILE 1106#if HAVE_SENDFILE
956# if __linux 1107# if __linux
957 off_t soffset = offset; 1108 off_t soffset = offset;
958 res = sendfile (ofd, ifd, &soffset, count); 1109 res = sendfile (ofd, ifd, &soffset, count);
959 1110
1002 if (res < 0 && sbytes) 1153 if (res < 0 && sbytes)
1003 res = sbytes; 1154 res = sbytes;
1004 1155
1005# endif 1156# endif
1006 1157
1007#elif defined (_WIN32) 1158#elif defined (_WIN32) && 0
1008 /* does not work, just for documentation of what would need to be done */ 1159 /* does not work, just for documentation of what would need to be done */
1009 /* actually, cannot be done like this, as TransmitFile changes the file offset, */ 1160 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
1010 /* libeio guarantees that the file offset does not change, and windows */ 1161 /* libeio guarantees that the file offset does not change, and windows */
1011 /* has no way to get an independent handle to the same file description */ 1162 /* has no way to get an independent handle to the same file description */
1012 HANDLE h = TO_SOCKET (ifd); 1163 HANDLE h = TO_SOCKET (ifd);
1045 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1196 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1046 /* BSDs */ 1197 /* BSDs */
1047#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */ 1198#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1048 || errno == ENOTSUP 1199 || errno == ENOTSUP
1049#endif 1200#endif
1201#ifdef EOPNOTSUPP /* windows */
1050 || errno == EOPNOTSUPP /* BSDs */ 1202 || errno == EOPNOTSUPP /* BSDs */
1203#endif
1051#if __solaris 1204#if __solaris
1052 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1205 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
1053#endif 1206#endif
1054 ) 1207 )
1055 ) 1208 )
1059 1212
1060 res = 0; 1213 res = 0;
1061 1214
1062 while (count) 1215 while (count)
1063 { 1216 {
1064 ssize_t cnt; 1217 eio_ssize_t cnt;
1065 1218
1066 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 1219 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
1067 1220
1068 if (cnt <= 0) 1221 if (cnt <= 0)
1069 { 1222 {
1081 1234
1082 offset += cnt; 1235 offset += cnt;
1083 res += cnt; 1236 res += cnt;
1084 count -= cnt; 1237 count -= cnt;
1085 } 1238 }
1239
1240 FUBd;
1086 } 1241 }
1087 1242
1088 return res; 1243 return res;
1244}
1245
1246#ifdef PAGESIZE
1247# define eio_pagesize() PAGESIZE
1248#else
1249static intptr_t
1250eio_pagesize (void)
1251{
1252 static intptr_t page;
1253
1254 if (!page)
1255 page = sysconf (_SC_PAGESIZE);
1256
1257 return page;
1258}
1259#endif
1260
1261static void
1262eio_page_align (void **addr, size_t *length)
1263{
1264 intptr_t mask = eio_pagesize () - 1;
1265
1266 /* round down addr */
1267 intptr_t adj = mask & (intptr_t)*addr;
1268
1269 *addr = (void *)((intptr_t)*addr - adj);
1270 *length += adj;
1271
1272 /* round up length */
1273 *length = (*length + mask) & ~mask;
1274}
1275
1276#if !_POSIX_MEMLOCK
1277# define eio__mlockall(a) EIO_ENOSYS ()
1278#else
1279
1280static int
1281eio__mlockall (int flags)
1282{
1283 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1284 extern int mallopt (int, int);
1285 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1286 #endif
1287
1288 if (EIO_MCL_CURRENT != MCL_CURRENT
1289 || EIO_MCL_FUTURE != MCL_FUTURE)
1290 {
1291 flags = 0
1292 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1293 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1294 }
1295
1296 return mlockall (flags);
1297}
1298#endif
1299
1300#if !_POSIX_MEMLOCK_RANGE
1301# define eio__mlock(a,b) EIO_ENOSYS ()
1302#else
1303
1304static int
1305eio__mlock (void *addr, size_t length)
1306{
1307 eio_page_align (&addr, &length);
1308
1309 return mlock (addr, length);
1310}
1311
1312#endif
1313
1314#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1315# define eio__msync(a,b,c) EIO_ENOSYS ()
1316#else
1317
1318static int
1319eio__msync (void *mem, size_t len, int flags)
1320{
1321 eio_page_align (&mem, &len);
1322
1323 if (EIO_MS_ASYNC != MS_SYNC
1324 || EIO_MS_INVALIDATE != MS_INVALIDATE
1325 || EIO_MS_SYNC != MS_SYNC)
1326 {
1327 flags = 0
1328 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1329 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1330 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1331 }
1332
1333 return msync (mem, len, flags);
1334}
1335
1336#endif
1337
1338static int
1339eio__mtouch (eio_req *req)
1340{
1341 void *mem = req->ptr2;
1342 size_t len = req->size;
1343 int flags = req->int1;
1344
1345 eio_page_align (&mem, &len);
1346
1347 {
1348 intptr_t addr = (intptr_t)mem;
1349 intptr_t end = addr + len;
1350 intptr_t page = eio_pagesize ();
1351
1352 if (addr < end)
1353 if (flags & EIO_MT_MODIFY) /* modify */
1354 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1355 else
1356 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1357 }
1358
1359 return 0;
1360}
1361
1362/*****************************************************************************/
1363/* requests implemented outside eio_execute, because they are so large */
1364
1365/* copies some absolute path to tmpbuf */
1366static char *
1367eio__getwd (struct tmpbuf *tmpbuf, eio_wd wd)
1368{
1369 if (wd == EIO_CWD)
1370 return getcwd (tmpbuf->ptr, PATH_MAX);
1371
1372#if HAVE_AT
1373 abort (); /*TODO*/
1374#else
1375 strcpy (tmpbuf->ptr, wd);
1376#endif
1377 return tmpbuf->ptr;
1378}
1379
1380/* result will always end up in tmpbuf, there is always space for adding a 0-byte */
1381static int
1382eio__realpath (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
1383{
1384 const char *rel = path;
1385 char *res;
1386 char *tmp1, *tmp2;
1387#if SYMLOOP_MAX > 32
1388 int symlinks = SYMLOOP_MAX;
1389#else
1390 int symlinks = 32;
1391#endif
1392
1393 /*D*/ /*TODO: wd ignored */
1394
1395 errno = EINVAL;
1396 if (!rel)
1397 return -1;
1398
1399 errno = ENOENT;
1400 if (!*rel)
1401 return -1;
1402
1403 res = tmpbuf_get (tmpbuf, PATH_MAX * 3);
1404 tmp1 = res + PATH_MAX;
1405 tmp2 = tmp1 + PATH_MAX;
1406
1407#if 0 /* disabled, the musl way to do things is just too racy */
1408#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1409 /* on linux we may be able to ask the kernel */
1410 {
1411 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1412
1413 if (fd >= 0)
1414 {
1415 sprintf (tmp1, "/proc/self/fd/%d", fd);
1416 req->result = readlink (tmp1, res, PATH_MAX);
1417 close (fd);
1418
1419 /* here we should probably stat the open file and the disk file, to make sure they still match */
1420
1421 if (req->result > 0)
1422 goto done;
1423 }
1424 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1425 return;
1426 }
1427#endif
1428#endif
1429
1430 if (*rel != '/')
1431 {
1432 if (!eio__getwd (tmpbuf, wd))
1433 return -1;
1434
1435 if (res [1]) /* only use if not / */
1436 res += strlen (res);
1437 }
1438
1439 while (*rel)
1440 {
1441 eio_ssize_t len, linklen;
1442 const char *beg = rel;
1443
1444 while (*rel && *rel != '/')
1445 ++rel;
1446
1447 len = rel - beg;
1448
1449 if (!len) /* skip slashes */
1450 {
1451 ++rel;
1452 continue;
1453 }
1454
1455 if (beg [0] == '.')
1456 {
1457 if (len == 1)
1458 continue; /* . - nop */
1459
1460 if (beg [1] == '.' && len == 2)
1461 {
1462 /* .. - back up one component, if possible */
1463
1464 while (res != tmpbuf->ptr)
1465 if (*--res == '/')
1466 break;
1467
1468 continue;
1469 }
1470 }
1471
1472 errno = ENAMETOOLONG;
1473 if (res + 1 + len + 1 >= tmp1)
1474 return;
1475
1476 /* copy one component */
1477 *res = '/';
1478 memcpy (res + 1, beg, len);
1479
1480 /* zero-terminate, for readlink */
1481 res [len + 1] = 0;
1482
1483 /* now check if it's a symlink */
1484 linklen = readlink (tmpbuf->ptr, tmp1, PATH_MAX);
1485
1486 if (linklen < 0)
1487 {
1488 if (errno != EINVAL)
1489 return -1;
1490
1491 /* it's a normal directory. hopefully */
1492 res += len + 1;
1493 }
1494 else
1495 {
1496 /* yay, it was a symlink - build new path in tmp2 */
1497 int rellen = strlen (rel);
1498
1499 errno = ENAMETOOLONG;
1500 if (linklen + 1 + rellen >= PATH_MAX)
1501 return -1;
1502
1503 errno = ELOOP;
1504 if (!--symlinks)
1505 return -1;
1506
1507 if (*tmp1 == '/')
1508 res = tmpbuf->ptr; /* symlink resolves to an absolute path */
1509
1510 /* we need to be careful, as rel might point into tmp2 already */
1511 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1512 tmp2 [linklen] = '/';
1513 memcpy (tmp2, tmp1, linklen);
1514
1515 rel = tmp2;
1516 }
1517 }
1518
1519 /* special case for the lone root path */
1520 if (res == tmpbuf->ptr)
1521 *res++ = '/';
1522
1523 return res - (char *)tmpbuf->ptr;
1089} 1524}
1090 1525
1091static signed char 1526static signed char
1092eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) 1527eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1093{ 1528{
1101 1536
1102#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */ 1537#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1103#define EIO_SORT_FAST 60 /* when to only use insertion sort */ 1538#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1104 1539
1105static void 1540static void
1106eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1541eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1107{ 1542{
1108 unsigned char bits [9 + sizeof (ino_t) * 8]; 1543 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1109 unsigned char *bit = bits; 1544 unsigned char *bit = bits;
1110 1545
1111 assert (CHAR_BIT == 8); 1546 assert (CHAR_BIT == 8);
1112 assert (sizeof (eio_dirent) * 8 < 256); 1547 assert (sizeof (eio_dirent) * 8 < 256);
1113 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */ 1548 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1115 1550
1116 if (size <= EIO_SORT_FAST) 1551 if (size <= EIO_SORT_FAST)
1117 return; 1552 return;
1118 1553
1119 /* first prepare an array of bits to test in our radix sort */ 1554 /* first prepare an array of bits to test in our radix sort */
1120 /* try to take endianness into account, as well as differences in ino_t sizes */ 1555 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1121 /* inode_bits must contain all inodes ORed together */ 1556 /* inode_bits must contain all inodes ORed together */
1122 /* which is used to skip bits that are 0 everywhere, which is very common */ 1557 /* which is used to skip bits that are 0 everywhere, which is very common */
1123 { 1558 {
1124 ino_t endianness; 1559 eio_ino_t endianness;
1125 int i, j; 1560 int i, j;
1126 1561
1127 /* we store the byte offset of byte n into byte n of "endianness" */ 1562 /* we store the byte offset of byte n into byte n of "endianness" */
1128 for (i = 0; i < sizeof (ino_t); ++i) 1563 for (i = 0; i < sizeof (eio_ino_t); ++i)
1129 ((unsigned char *)&endianness)[i] = i; 1564 ((unsigned char *)&endianness)[i] = i;
1130 1565
1131 *bit++ = 0; 1566 *bit++ = 0;
1132 1567
1133 for (i = 0; i < sizeof (ino_t); ++i) 1568 for (i = 0; i < sizeof (eio_ino_t); ++i)
1134 { 1569 {
1135 /* shifting off the byte offsets out of "endianness" */ 1570 /* shifting off the byte offsets out of "endianness" */
1136 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8; 1571 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1137 endianness >>= 8; 1572 endianness >>= 8;
1138 1573
1139 for (j = 0; j < 8; ++j) 1574 for (j = 0; j < 8; ++j)
1140 if (inode_bits & (((ino_t)1) << (i * 8 + j))) 1575 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1141 *bit++ = offs + j; 1576 *bit++ = offs + j;
1142 } 1577 }
1143 1578
1144 for (j = 0; j < 8; ++j) 1579 for (j = 0; j < 8; ++j)
1145 if (score_bits & (1 << j)) 1580 if (score_bits & (1 << j))
1146 *bit++ = offsetof (eio_dirent, score) * 8 + j; 1581 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1147 } 1582 }
1148 1583
1149 /* now actually do the sorting (a variant of MSD radix sort) */ 1584 /* now actually do the sorting (a variant of MSD radix sort) */
1150 { 1585 {
1151 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base; 1586 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1152 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end; 1587 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1153 unsigned char *bit_stk [9 + sizeof (ino_t) * 8]; 1588 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1154 int stk_idx = 0; 1589 int stk_idx = 0;
1155 1590
1156 base_stk [stk_idx] = dents; 1591 base_stk [stk_idx] = dents;
1157 end_stk [stk_idx] = dents + size; 1592 end_stk [stk_idx] = dents + size;
1158 bit_stk [stk_idx] = bit - 1; 1593 bit_stk [stk_idx] = bit - 1;
1237 } 1672 }
1238 } 1673 }
1239} 1674}
1240 1675
1241static void 1676static void
1242eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1677eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1243{ 1678{
1244 if (size <= 1) 1679 if (size <= 1)
1245 return; /* our insertion sort relies on size > 0 */ 1680 return; /* our insertion sort relies on size > 0 */
1246 1681
1247 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */ 1682 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1253 eio_dent_insertion_sort (dents, size); 1688 eio_dent_insertion_sort (dents, size);
1254} 1689}
1255 1690
1256/* read a full directory */ 1691/* read a full directory */
1257static void 1692static void
1258eio__scandir (eio_req *req, etp_worker *self) 1693eio__scandir (eio_req *req)
1259{ 1694{
1260 DIR *dirp;
1261 EIO_STRUCT_DIRENT *entp;
1262 char *name, *names; 1695 char *name, *names;
1263 int namesalloc = 4096; 1696 int namesalloc = 4096 - sizeof (void *) * 4;
1264 int namesoffs = 0; 1697 int namesoffs = 0;
1265 int flags = req->int1; 1698 int flags = req->int1;
1266 eio_dirent *dents = 0; 1699 eio_dirent *dents = 0;
1267 int dentalloc = 128; 1700 int dentalloc = 128;
1268 int dentoffs = 0; 1701 int dentoffs = 0;
1269 ino_t inode_bits = 0; 1702 eio_ino_t inode_bits = 0;
1703#ifdef _WIN32
1704 HANDLE dirp;
1705 WIN32_FIND_DATA entp;
1706#else
1707 DIR *dirp;
1708 EIO_STRUCT_DIRENT *entp;
1709#endif
1270 1710
1271 req->result = -1; 1711 req->result = -1;
1272 1712
1273 if (!(flags & EIO_READDIR_DENTS)) 1713 if (!(flags & EIO_READDIR_DENTS))
1274 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); 1714 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1275 1715
1276 X_LOCK (wrklock); 1716#ifdef _WIN32
1277 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 1717 {
1718 int len = strlen ((const char *)req->ptr1);
1719 char *path = malloc (MAX_PATH);
1720 const char *fmt;
1721
1722 if (!len)
1723 fmt = "./*";
1724 else if (((const char *)req->ptr1)[len - 1] == '/' || ((const char *)req->ptr1)[len - 1] == '\\')
1725 fmt = "%s*";
1726 else
1727 fmt = "%s/*";
1728
1729 _snprintf (path, MAX_PATH, fmt, (const char *)req->ptr1);
1730 dirp = FindFirstFile (path, &entp);
1731 free (path);
1732
1733 if (dirp == INVALID_HANDLE_VALUE)
1734 {
1735 /* should steal _dosmaperr */
1736 switch (GetLastError ())
1737 {
1738 case ERROR_FILE_NOT_FOUND:
1739 req->result = 0;
1740 break;
1741
1742 case ERROR_INVALID_NAME:
1743 case ERROR_PATH_NOT_FOUND:
1744 case ERROR_NO_MORE_FILES:
1745 errno = ENOENT;
1746 break;
1747
1748 case ERROR_NOT_ENOUGH_MEMORY:
1749 errno = ENOMEM;
1750 break;
1751
1752 default:
1753 errno = EINVAL;
1754 break;
1755 }
1756
1757 return;
1758 }
1759 }
1760#else
1278 self->dirp = dirp = opendir (req->ptr1); 1761 dirp = opendir (req->ptr1);
1762
1763 if (!dirp)
1764 return;
1765#endif
1766
1767 if (req->flags & EIO_FLAG_PTR1_FREE)
1768 free (req->ptr1);
1279 1769
1280 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1770 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1281 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; 1771 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1282 req->ptr2 = names = malloc (namesalloc); 1772 req->ptr2 = names = malloc (namesalloc);
1283 X_UNLOCK (wrklock);
1284 1773
1285 if (dirp && names && (!flags || dents)) 1774 if (!names || (flags && !dents))
1775 return;
1776
1286 for (;;) 1777 for (;;)
1287 { 1778 {
1779 int done;
1780
1781#ifdef _WIN32
1782 done = !dirp;
1783#else
1288 errno = 0; 1784 errno = 0;
1289 entp = readdir (dirp); 1785 entp = readdir (dirp);
1786 done = !entp;
1787#endif
1290 1788
1291 if (!entp) 1789 if (done)
1292 { 1790 {
1791#ifndef _WIN32
1792 int old_errno = errno;
1793 closedir (dirp);
1794 errno = old_errno;
1795
1293 if (errno) 1796 if (errno)
1294 break; 1797 break;
1798#endif
1295 1799
1296 /* sort etc. */ 1800 /* sort etc. */
1297 req->int1 = flags; 1801 req->int1 = flags;
1298 req->result = dentoffs; 1802 req->result = dentoffs;
1299 1803
1300 if (flags & EIO_READDIR_STAT_ORDER) 1804 if (flags & EIO_READDIR_STAT_ORDER)
1301 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits); 1805 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1302 else if (flags & EIO_READDIR_DIRS_FIRST) 1806 else if (flags & EIO_READDIR_DIRS_FIRST)
1303 if (flags & EIO_READDIR_FOUND_UNKNOWN) 1807 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1304 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */ 1808 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1305 else 1809 else
1810 {
1811 /* in this case, all is known, and we just put dirs first and sort them */
1812 eio_dirent *oth = dents + dentoffs;
1813 eio_dirent *dir = dents;
1814
1815 /* now partition dirs to the front, and non-dirs to the back */
1816 /* by walking from both sides and swapping if necessary */
1817 while (oth > dir)
1818 {
1819 if (dir->type == EIO_DT_DIR)
1820 ++dir;
1821 else if ((--oth)->type == EIO_DT_DIR)
1822 {
1823 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1824
1825 ++dir;
1826 }
1827 }
1828
1829 /* now sort the dirs only (dirs all have the same score) */
1830 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1831 }
1832
1833 break;
1834 }
1835
1836 /* now add the entry to our list(s) */
1837 name = D_NAME (entp);
1838
1839 /* skip . and .. entries */
1840 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1841 {
1842 int len = D_NAMLEN (entp) + 1;
1843
1844 while (ecb_expect_false (namesoffs + len > namesalloc))
1845 {
1846 namesalloc *= 2;
1847 req->ptr2 = names = realloc (names, namesalloc);
1848
1849 if (!names)
1850 break;
1851 }
1852
1853 memcpy (names + namesoffs, name, len);
1854
1855 if (dents)
1856 {
1857 struct eio_dirent *ent;
1858
1859 if (ecb_expect_false (dentoffs == dentalloc))
1306 { 1860 {
1307 /* in this case, all is known, and we just put dirs first and sort them */ 1861 dentalloc *= 2;
1862 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1863
1864 if (!dents)
1865 break;
1866 }
1867
1308 eio_dirent *oth = dents + dentoffs; 1868 ent = dents + dentoffs;
1309 eio_dirent *dir = dents;
1310 1869
1311 /* now partition dirs to the front, and non-dirs to the back */ 1870 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1312 /* by walking from both sides and swapping if necessary */ 1871 ent->namelen = len - 1;
1313 while (oth > dir) 1872 ent->inode = D_INO (entp);
1873
1874 inode_bits |= ent->inode;
1875
1876 switch (D_TYPE (entp))
1877 {
1878 default:
1879 ent->type = EIO_DT_UNKNOWN;
1880 flags |= EIO_READDIR_FOUND_UNKNOWN;
1881 break;
1882
1883 #ifdef DT_FIFO
1884 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1885 #endif
1886 #ifdef DT_CHR
1887 case DT_CHR: ent->type = EIO_DT_CHR; break;
1888 #endif
1889 #ifdef DT_MPC
1890 case DT_MPC: ent->type = EIO_DT_MPC; break;
1891 #endif
1892 #ifdef DT_DIR
1893 case DT_DIR: ent->type = EIO_DT_DIR; break;
1894 #endif
1895 #ifdef DT_NAM
1896 case DT_NAM: ent->type = EIO_DT_NAM; break;
1897 #endif
1898 #ifdef DT_BLK
1899 case DT_BLK: ent->type = EIO_DT_BLK; break;
1900 #endif
1901 #ifdef DT_MPB
1902 case DT_MPB: ent->type = EIO_DT_MPB; break;
1903 #endif
1904 #ifdef DT_REG
1905 case DT_REG: ent->type = EIO_DT_REG; break;
1906 #endif
1907 #ifdef DT_NWK
1908 case DT_NWK: ent->type = EIO_DT_NWK; break;
1909 #endif
1910 #ifdef DT_CMP
1911 case DT_CMP: ent->type = EIO_DT_CMP; break;
1912 #endif
1913 #ifdef DT_LNK
1914 case DT_LNK: ent->type = EIO_DT_LNK; break;
1915 #endif
1916 #ifdef DT_SOCK
1917 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1918 #endif
1919 #ifdef DT_DOOR
1920 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1921 #endif
1922 #ifdef DT_WHT
1923 case DT_WHT: ent->type = EIO_DT_WHT; break;
1924 #endif
1925 }
1926
1927 ent->score = 7;
1928
1929 if (flags & EIO_READDIR_DIRS_FIRST)
1930 {
1931 if (ent->type == EIO_DT_UNKNOWN)
1314 { 1932 {
1315 if (dir->type == EIO_DT_DIR) 1933 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1316 ++dir; 1934 ent->score = 1;
1317 else if ((--oth)->type == EIO_DT_DIR) 1935 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1318 { 1936 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1319 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1320
1321 ++dir;
1322 }
1323 } 1937 }
1324 1938 else if (ent->type == EIO_DT_DIR)
1325 /* now sort the dirs only (dirs all have the same score) */ 1939 ent->score = 0;
1326 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1327 } 1940 }
1328
1329 break;
1330 }
1331
1332 /* now add the entry to our list(s) */
1333 name = entp->d_name;
1334
1335 /* skip . and .. entries */
1336 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1337 {
1338 int len = D_NAMLEN (entp) + 1;
1339
1340 while (expect_false (namesoffs + len > namesalloc))
1341 {
1342 namesalloc *= 2;
1343 X_LOCK (wrklock);
1344 req->ptr2 = names = realloc (names, namesalloc);
1345 X_UNLOCK (wrklock);
1346
1347 if (!names)
1348 break;
1349 } 1941 }
1350 1942
1351 memcpy (names + namesoffs, name, len);
1352
1353 if (dents)
1354 {
1355 struct eio_dirent *ent;
1356
1357 if (expect_false (dentoffs == dentalloc))
1358 {
1359 dentalloc *= 2;
1360 X_LOCK (wrklock);
1361 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1362 X_UNLOCK (wrklock);
1363
1364 if (!dents)
1365 break;
1366 }
1367
1368 ent = dents + dentoffs;
1369
1370 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1371 ent->namelen = len - 1;
1372 ent->inode = D_INO (entp);
1373
1374 inode_bits |= ent->inode;
1375
1376 switch (D_TYPE (entp))
1377 {
1378 default:
1379 ent->type = EIO_DT_UNKNOWN;
1380 flags |= EIO_READDIR_FOUND_UNKNOWN;
1381 break;
1382
1383 #ifdef DT_FIFO
1384 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1385 #endif
1386 #ifdef DT_CHR
1387 case DT_CHR: ent->type = EIO_DT_CHR; break;
1388 #endif
1389 #ifdef DT_MPC
1390 case DT_MPC: ent->type = EIO_DT_MPC; break;
1391 #endif
1392 #ifdef DT_DIR
1393 case DT_DIR: ent->type = EIO_DT_DIR; break;
1394 #endif
1395 #ifdef DT_NAM
1396 case DT_NAM: ent->type = EIO_DT_NAM; break;
1397 #endif
1398 #ifdef DT_BLK
1399 case DT_BLK: ent->type = EIO_DT_BLK; break;
1400 #endif
1401 #ifdef DT_MPB
1402 case DT_MPB: ent->type = EIO_DT_MPB; break;
1403 #endif
1404 #ifdef DT_REG
1405 case DT_REG: ent->type = EIO_DT_REG; break;
1406 #endif
1407 #ifdef DT_NWK
1408 case DT_NWK: ent->type = EIO_DT_NWK; break;
1409 #endif
1410 #ifdef DT_CMP
1411 case DT_CMP: ent->type = EIO_DT_CMP; break;
1412 #endif
1413 #ifdef DT_LNK
1414 case DT_LNK: ent->type = EIO_DT_LNK; break;
1415 #endif
1416 #ifdef DT_SOCK
1417 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1418 #endif
1419 #ifdef DT_DOOR
1420 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1421 #endif
1422 #ifdef DT_WHT
1423 case DT_WHT: ent->type = EIO_DT_WHT; break;
1424 #endif
1425 }
1426
1427 ent->score = 7;
1428
1429 if (flags & EIO_READDIR_DIRS_FIRST)
1430 {
1431 if (ent->type == EIO_DT_UNKNOWN)
1432 {
1433 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1434 ent->score = 1;
1435 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1436 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1437 }
1438 else if (ent->type == EIO_DT_DIR)
1439 ent->score = 0;
1440 }
1441 }
1442
1443 namesoffs += len; 1943 namesoffs += len;
1444 ++dentoffs; 1944 ++dentoffs;
1445 } 1945 }
1446 1946
1447 if (EIO_CANCELLED (req)) 1947 if (EIO_CANCELLED (req))
1448 { 1948 {
1449 errno = ECANCELED; 1949 errno = ECANCELED;
1450 break; 1950 break;
1451 } 1951 }
1952
1953#ifdef _WIN32
1954 if (!FindNextFile (dirp, &entp))
1955 {
1956 FindClose (dirp);
1957 dirp = 0;
1452 } 1958 }
1959#endif
1960 }
1453} 1961}
1454 1962
1455#ifdef PAGESIZE 1963/*****************************************************************************/
1456# define eio_pagesize() PAGESIZE 1964/* working directory stuff */
1965
1966#if HAVE_AT
1967
1968#define WD2FD(wd) (wd ? ((int)wd) - 1 : AT_FDCWD)
1969
1970eio_wd
1971eio_wd_open_sync (eio_wd wd, const char *path)
1972{
1973 int fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1974
1975 return fd >= 0 ? (eio_wd)(fd + 1) : EIO_INVALID_WD;
1976}
1977
1978void
1979eio_wd_close_sync (eio_wd wd)
1980{
1981 int fd = WD2FD (wd);
1982
1983 if (fd >= 0)
1984 close (fd);
1985}
1986
1457#else 1987#else
1458static intptr_t
1459eio_pagesize (void)
1460{
1461 static intptr_t page;
1462 1988
1463 if (!page) 1989/* on legacy systems, we represent the working directories simply by their path strings */
1464 page = sysconf (_SC_PAGESIZE);
1465 1990
1991static const char *
1992wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
1993{
1994 if (!wd || *path == '/')
1466 return page; 1995 return path;
1467}
1468#endif
1469 1996
1470static void 1997 {
1471eio_page_align (void **addr, size_t *length) 1998 int l1 = strlen ((const char *)wd);
1472{ 1999 int l2 = strlen (path);
1473 intptr_t mask = eio_pagesize () - 1;
1474 2000
1475 /* round down addr */ 2001 char *res = tmpbuf_get (tmpbuf, l1 + l2 + 2);
1476 intptr_t adj = mask & (intptr_t)*addr;
1477 2002
1478 *addr = (void *)((intptr_t)*addr - adj); 2003 memcpy (res, wd, l1);
1479 *length += adj; 2004 res [l1] = '/';
2005 memcpy (res + l1 + 1, path, l2 + 1);
1480 2006
1481 /* round up length */ 2007 return res;
1482 *length = (*length + mask) & ~mask; 2008 }
1483} 2009}
1484 2010
1485#if !_POSIX_MEMLOCK 2011eio_wd
1486# define eio__mlockall(a) ((errno = ENOSYS), -1) 2012eio__wd_open_sync (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
1487#else
1488
1489static int
1490eio__mlockall (int flags)
1491{ 2013{
1492 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7 2014 if (*path == '/') /* absolute paths ignore wd */
1493 extern int mallopt (int, int); 2015 path = strdup (path);
1494 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */ 2016 else if (path [0] == '.' && !path [1]) /* special case '.', as it is common */
1495 #endif 2017 return wd;
1496 2018 else
1497 if (EIO_MCL_CURRENT != MCL_CURRENT
1498 || EIO_MCL_FUTURE != MCL_FUTURE)
1499 { 2019 {
1500 flags = 0 2020 int len = eio__realpath (tmpbuf, wd, path);
1501 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1502 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1503 }
1504 2021
1505 return mlockall (flags); 2022 path = EIO_INVALID_WD;
1506}
1507#endif
1508 2023
1509#if !_POSIX_MEMLOCK_RANGE 2024 if (len >= 0)
1510# define eio__mlock(a,b) ((errno = ENOSYS), -1) 2025 {
1511#else 2026 ((char *)tmpbuf->ptr)[len] = 0;
1512 2027 path = strdup (tmpbuf->ptr);
1513static int 2028 }
1514eio__mlock (void *addr, size_t length)
1515{
1516 eio_page_align (&addr, &length);
1517
1518 return mlock (addr, length);
1519}
1520
1521#endif
1522
1523#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1524# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
1525#else
1526
1527int
1528eio__msync (void *mem, size_t len, int flags)
1529{
1530 eio_page_align (&mem, &len);
1531
1532 if (EIO_MS_ASYNC != MS_SYNC
1533 || EIO_MS_INVALIDATE != MS_INVALIDATE
1534 || EIO_MS_SYNC != MS_SYNC)
1535 { 2029 }
1536 flags = 0
1537 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1538 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1539 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1540 }
1541 2030
1542 return msync (mem, len, flags); 2031 if (!path)
1543} 2032 path = EIO_INVALID_WD;
1544 2033
1545#endif 2034 return (eio_wd)path;
2035}
1546 2036
1547int 2037eio_wd
1548eio__mtouch (eio_req *req) 2038eio_wd_open_sync (eio_wd wd, const char *path)
1549{ 2039{
1550 void *mem = req->ptr2; 2040 struct tmpbuf tmpbuf = { 0 };
1551 size_t len = req->size; 2041 wd = eio__wd_open_sync (&tmpbuf, wd, path);
1552 int flags = req->int1; 2042 free (tmpbuf.ptr);
1553 2043
1554 eio_page_align (&mem, &len);
1555
1556 {
1557 intptr_t addr = (intptr_t)mem;
1558 intptr_t end = addr + len;
1559 intptr_t page = eio_pagesize ();
1560
1561 if (addr < end)
1562 if (flags & EIO_MT_MODIFY) /* modify */
1563 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1564 else
1565 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1566 }
1567
1568 return 0; 2044 return wd;
1569} 2045}
2046
2047void
2048eio_wd_close_sync (eio_wd wd)
2049{
2050 if (wd != EIO_INVALID_WD)
2051 free (wd);
2052}
2053
2054#endif
1570 2055
1571/*****************************************************************************/ 2056/*****************************************************************************/
1572 2057
1573#define ALLOC(len) \ 2058#define ALLOC(len) \
1574 if (!req->ptr2) \ 2059 if (!req->ptr2) \
1589{ 2074{
1590 ETP_REQ *req; 2075 ETP_REQ *req;
1591 struct timespec ts; 2076 struct timespec ts;
1592 etp_worker *self = (etp_worker *)thr_arg; 2077 etp_worker *self = (etp_worker *)thr_arg;
1593 2078
2079#if HAVE_PRCTL_SET_NAME
2080 prctl (PR_SET_NAME, (unsigned long)"eio_thread", 0, 0, 0);
2081#endif
2082
1594 /* try to distribute timeouts somewhat randomly */ 2083 /* try to distribute timeouts somewhat evenly */
1595 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 2084 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1596 2085
1597 for (;;) 2086 for (;;)
1598 { 2087 {
2088 ts.tv_sec = 0;
2089
1599 X_LOCK (reqlock); 2090 X_LOCK (reqlock);
1600 2091
1601 for (;;) 2092 for (;;)
1602 { 2093 {
1603 self->req = req = reqq_shift (&req_queue); 2094 req = reqq_shift (&req_queue);
1604 2095
1605 if (req) 2096 if (req)
1606 break; 2097 break;
1607 2098
2099 if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
2100 {
2101 X_UNLOCK (reqlock);
2102 X_LOCK (wrklock);
2103 --started;
2104 X_UNLOCK (wrklock);
2105 goto quit;
2106 }
2107
1608 ++idle; 2108 ++idle;
1609 2109
1610 ts.tv_sec = time (0) + idle_timeout; 2110 if (idle <= max_idle)
1611 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 2111 /* we are allowed to idle, so do so without any timeout */
2112 X_COND_WAIT (reqwait, reqlock);
2113 else
1612 { 2114 {
1613 if (idle > max_idle) 2115 /* initialise timeout once */
1614 { 2116 if (!ts.tv_sec)
1615 --idle; 2117 ts.tv_sec = time (0) + idle_timeout;
1616 X_UNLOCK (reqlock);
1617 X_LOCK (wrklock);
1618 --started;
1619 X_UNLOCK (wrklock);
1620 goto quit;
1621 }
1622 2118
1623 /* we are allowed to idle, so do so without any timeout */
1624 X_COND_WAIT (reqwait, reqlock); 2119 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
2120 ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1625 } 2121 }
1626 2122
1627 --idle; 2123 --idle;
1628 } 2124 }
1629 2125
1632 X_UNLOCK (reqlock); 2128 X_UNLOCK (reqlock);
1633 2129
1634 if (req->type < 0) 2130 if (req->type < 0)
1635 goto quit; 2131 goto quit;
1636 2132
1637 if (!EIO_CANCELLED (req))
1638 ETP_EXECUTE (self, req); 2133 ETP_EXECUTE (self, req);
1639 2134
1640 X_LOCK (reslock); 2135 X_LOCK (reslock);
1641 2136
1642 ++npending; 2137 ++npending;
1643 2138
1644 if (!reqq_push (&res_queue, req) && want_poll_cb) 2139 if (!reqq_push (&res_queue, req) && want_poll_cb)
1645 want_poll_cb (); 2140 want_poll_cb ();
1646 2141
1647 self->req = 0;
1648 etp_worker_clear (self); 2142 etp_worker_clear (self);
1649 2143
1650 X_UNLOCK (reslock); 2144 X_UNLOCK (reslock);
1651 } 2145 }
1652 2146
1653quit: 2147quit:
2148 free (req);
2149
1654 X_LOCK (wrklock); 2150 X_LOCK (wrklock);
1655 etp_worker_free (self); 2151 etp_worker_free (self);
1656 X_UNLOCK (wrklock); 2152 X_UNLOCK (wrklock);
1657 2153
1658 return 0; 2154 return 0;
1659} 2155}
1660 2156
1661/*****************************************************************************/ 2157/*****************************************************************************/
1662 2158
2159int ecb_cold
1663int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 2160eio_init (void (*want_poll)(void), void (*done_poll)(void))
1664{ 2161{
2162#if !HAVE_PREADWRITE
2163 X_MUTEX_CREATE (preadwritelock);
2164#endif
2165
1665 return etp_init (want_poll, done_poll); 2166 return etp_init (want_poll, done_poll);
1666} 2167}
1667 2168
2169ecb_inline void
1668static void eio_api_destroy (eio_req *req) 2170eio_api_destroy (eio_req *req)
1669{ 2171{
1670 free (req); 2172 free (req);
1671} 2173}
1672 2174
1673#define REQ(rtype) \ 2175#define REQ(rtype) \
1692 { \ 2194 { \
1693 eio_api_destroy (req); \ 2195 eio_api_destroy (req); \
1694 return 0; \ 2196 return 0; \
1695 } 2197 }
1696 2198
2199static void
1697static void eio_execute (etp_worker *self, eio_req *req) 2200eio_execute (etp_worker *self, eio_req *req)
1698{ 2201{
2202#if HAVE_AT
2203 int dirfd;
2204#else
2205 const char *path;
2206#endif
2207
2208 if (ecb_expect_false (EIO_CANCELLED (req)))
2209 {
2210 req->result = -1;
2211 req->errorno = ECANCELED;
2212 return;
2213 }
2214
2215 if (ecb_expect_false (req->wd == EIO_INVALID_WD))
2216 {
2217 req->result = -1;
2218 req->errorno = ENOENT;
2219 return;
2220 }
2221
2222 if (req->type >= EIO_OPEN)
2223 {
2224 #if HAVE_AT
2225 dirfd = WD2FD (req->wd);
2226 #else
2227 path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
2228 #endif
2229 }
2230
1699 switch (req->type) 2231 switch (req->type)
1700 { 2232 {
2233 case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
2234 req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
2235 break;
2236 case EIO_WD_CLOSE: eio_wd_close_sync (req->wd); break;
2237
1701 case EIO_READ: ALLOC (req->size); 2238 case EIO_READ: ALLOC (req->size);
1702 req->result = req->offs >= 0 2239 req->result = req->offs >= 0
1703 ? pread (req->int1, req->ptr2, req->size, req->offs) 2240 ? pread (req->int1, req->ptr2, req->size, req->offs)
1704 : read (req->int1, req->ptr2, req->size); break; 2241 : read (req->int1, req->ptr2, req->size); break;
1705 case EIO_WRITE: req->result = req->offs >= 0 2242 case EIO_WRITE: req->result = req->offs >= 0
1706 ? pwrite (req->int1, req->ptr2, req->size, req->offs) 2243 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1707 : write (req->int1, req->ptr2, req->size); break; 2244 : write (req->int1, req->ptr2, req->size); break;
1708 2245
1709 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break; 2246 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1710 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break; 2247 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
1711 2248
2249#if HAVE_AT
2250 case EIO_GETPATH: abort ();
1712 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2251 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1713 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2252 req->result = fstatat (dirfd, path, (EIO_STRUCT_STAT *)req->ptr2); break;
1714 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2253 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1715 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2254 req->result = lstat (dirfd, path, (EIO_STRUCT_STAT *)req->ptr2); break;
2255#if 0/*D*/
2256 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2257 req->result = statvfs (dirfd, path, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2258#endif
2259 case EIO_CHOWN: req->result = chown (dirfd, path, req->int2, req->int3); break;
2260 case EIO_CHMOD: req->result = chmod (dirfd, path, (mode_t)req->int2); break;
2261 case EIO_TRUNCATE: req->result = truncate (dirfd, path, req->offs); break;
2262 case EIO_OPEN: req->result = open (dirfd, path, req->int1, (mode_t)req->int2); break;
2263
2264 case EIO_UNLINK: req->result = unlink (dirfd, path); break;
2265 case EIO_RMDIR: req->result = rmdir (dirfd, path); break;
2266 case EIO_MKDIR: req->result = mkdir (dirfd, path, (mode_t)req->int2); break;
2267 case EIO_RENAME: req->result = rename (dirfd, path, req->ptr2); break;
2268 case EIO_LINK: req->result = link (dirfd, path, req->ptr2); break;
2269 case EIO_SYMLINK: req->result = symlink (dirfd, path, req->ptr2); break;
2270 case EIO_MKNOD: req->result = mknod (dirfd, path, (mode_t)req->int2, (dev_t)req->offs); break;
2271#else
2272 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2273 req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
2274 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2275 req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
2276 case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
2277 case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
2278 case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
2279 case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
2280
2281 case EIO_UNLINK: req->result = unlink (path ); break;
2282 case EIO_RMDIR: req->result = rmdir (path ); break;
2283 case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
2284 case EIO_RENAME: req->result = rename (path , req->ptr2); break;
2285 case EIO_LINK: req->result = link (path , req->ptr2); break;
2286 case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
2287 case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
2288
2289 case EIO_READLINK: ALLOC (PATH_MAX);
2290 req->result = readlink (path, req->ptr2, PATH_MAX); break;
2291#endif
2292
2293 case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
2294 {
2295 ALLOC (req->result);
2296 memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
2297 }
2298 break;
2299
2300 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS)); /*D*/
2301 req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
2302
1716 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2303 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1717 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; 2304 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1718 2305
1719 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1720 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1721 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS)); 2306 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1722 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break; 2307 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1723 2308
1724 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1725 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; 2309 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1726 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1727 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break; 2310 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1728 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1729 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break; 2311 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1730 2312
1731 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
1732 case EIO_CLOSE: req->result = close (req->int1); break; 2313 case EIO_CLOSE: req->result = close (req->int1); break;
1733 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break; 2314 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1734 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
1735 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1736 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1737 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1738 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1739 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1740 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1741
1742 case EIO_READLINK: ALLOC (PATH_MAX);
1743 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1744
1745 case EIO_SYNC: req->result = 0; sync (); break; 2315 case EIO_SYNC: req->result = 0; sync (); break;
1746 case EIO_FSYNC: req->result = fsync (req->int1); break; 2316 case EIO_FSYNC: req->result = fsync (req->int1); break;
1747 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 2317 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
2318 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
2319 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1748 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break; 2320 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1749 case EIO_MTOUCH: req->result = eio__mtouch (req); break; 2321 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1750 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break; 2322 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1751 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break; 2323 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1752 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break; 2324 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
1753 2325
1754 case EIO_READDIR: eio__scandir (req, self); break; 2326 case EIO_READDIR: eio__scandir (req); break;
1755 2327
1756 case EIO_BUSY: 2328 case EIO_BUSY:
1757#ifdef _WIN32 2329#ifdef _WIN32
1758 Sleep (req->nv1 * 1e3); 2330 Sleep (req->nv1 * 1e3);
1759#else 2331#else
1837eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2409eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1838{ 2410{
1839 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2411 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1840} 2412}
1841 2413
2414eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2415{
2416 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2417}
2418
2419eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
2420{
2421 REQ (EIO_SYNCFS); req->int1 = fd; SEND;
2422}
2423
2424eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2425{
2426 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2427}
2428
1842eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2429eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1843{ 2430{
1844 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2431 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1845} 2432}
1846 2433
1852eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data) 2439eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
1853{ 2440{
1854 REQ (EIO_MLOCKALL); req->int1 = flags; SEND; 2441 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
1855} 2442}
1856 2443
1857eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) 2444eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
1858{ 2445{
1859 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; 2446 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
1860}
1861
1862eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1863{
1864 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1865} 2447}
1866 2448
1867eio_req *eio_close (int fd, int pri, eio_cb cb, void *data) 2449eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
1868{ 2450{
1869 REQ (EIO_CLOSE); req->int1 = fd; SEND; 2451 REQ (EIO_CLOSE); req->int1 = fd; SEND;
1907eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) 2489eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1908{ 2490{
1909 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2491 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1910} 2492}
1911 2493
1912eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2494eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1913{ 2495{
1914 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2496 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1915} 2497}
1916 2498
1917eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) 2499eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
1937eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data) 2519eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1938{ 2520{
1939 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2521 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1940} 2522}
1941 2523
1942eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2524eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1943{ 2525{
1944 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2526 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1945} 2527}
1946 2528
1947eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) 2529eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
1961} 2543}
1962 2544
1963eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data) 2545eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1964{ 2546{
1965 return eio__1path (EIO_READLINK, path, pri, cb, data); 2547 return eio__1path (EIO_READLINK, path, pri, cb, data);
2548}
2549
2550eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2551{
2552 return eio__1path (EIO_REALPATH, path, pri, cb, data);
1966} 2553}
1967 2554
1968eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) 2555eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1969{ 2556{
1970 return eio__1path (EIO_STAT, path, pri, cb, data); 2557 return eio__1path (EIO_STAT, path, pri, cb, data);
2029eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2616eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2030{ 2617{
2031 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2618 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
2032} 2619}
2033 2620
2034eio_req *eio_custom (void (*)(eio_req *) execute, int pri, eio_cb cb, void *data); 2621eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
2035{ 2622{
2036 REQ (EIO_CUSTOM); req->feed = execute; SEND; 2623 REQ (EIO_CUSTOM); req->feed = execute; SEND;
2037} 2624}
2038 2625
2039#endif 2626#endif
2050#undef SEND 2637#undef SEND
2051 2638
2052/*****************************************************************************/ 2639/*****************************************************************************/
2053/* grp functions */ 2640/* grp functions */
2054 2641
2642void
2055void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2643eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
2056{ 2644{
2057 grp->int2 = limit; 2645 grp->int2 = limit;
2058 grp->feed = feed; 2646 grp->feed = feed;
2059 2647
2060 grp_try_feed (grp); 2648 grp_try_feed (grp);
2061} 2649}
2062 2650
2651void
2063void eio_grp_limit (eio_req *grp, int limit) 2652eio_grp_limit (eio_req *grp, int limit)
2064{ 2653{
2065 grp->int2 = limit; 2654 grp->int2 = limit;
2066 2655
2067 grp_try_feed (grp); 2656 grp_try_feed (grp);
2068} 2657}
2069 2658
2659void
2070void eio_grp_add (eio_req *grp, eio_req *req) 2660eio_grp_add (eio_req *grp, eio_req *req)
2071{ 2661{
2072 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2662 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
2073 2663
2074 grp->flags |= EIO_FLAG_GROUPADD; 2664 grp->flags |= EIO_FLAG_GROUPADD;
2075 2665
2086} 2676}
2087 2677
2088/*****************************************************************************/ 2678/*****************************************************************************/
2089/* misc garbage */ 2679/* misc garbage */
2090 2680
2681eio_ssize_t
2091ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2682eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2092{ 2683{
2093 etp_worker wrk;
2094 ssize_t ret;
2095
2096 wrk.dbuf = 0;
2097
2098 ret = eio__sendfile (ofd, ifd, offset, count, &wrk); 2684 return eio__sendfile (ofd, ifd, offset, count);
2099
2100 if (wrk.dbuf)
2101 free (wrk.dbuf);
2102
2103 return ret;
2104} 2685}
2105 2686

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines