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

Comparing libeio/eio.c (file contents):
Revision 1.27 by root, Wed Oct 22 18:15:36 2008 UTC vs.
Revision 1.95 by root, Tue Jul 19 04:56:43 2011 UTC

1/* 1/*
2 * libeio implementation 2 * libeio implementation
3 * 3 *
4 * Copyright (c) 2007,2008 Marc Alexander Lehmann <libeio@schmorp.de> 4 * Copyright (c) 2007,2008,2009,2010,2011 Marc Alexander Lehmann <libeio@schmorp.de>
5 * All rights reserved. 5 * All rights reserved.
6 * 6 *
7 * Redistribution and use in source and binary forms, with or without modifica- 7 * Redistribution and use in source and binary forms, with or without modifica-
8 * tion, are permitted provided that the following conditions are met: 8 * tion, are permitted provided that the following conditions are met:
9 * 9 *
35 * and other provisions required by the GPL. If you do not delete the 35 * and other provisions required by the GPL. If you do not delete the
36 * provisions above, a recipient may use your version of this file under 36 * provisions above, a recipient may use your version of this file under
37 * either the BSD or the GPL. 37 * either the BSD or the GPL.
38 */ 38 */
39 39
40#ifndef _WIN32
41# include "config.h"
42#endif
43
40#include "eio.h" 44#include "eio.h"
45#include "ecb.h"
46
47#ifdef EIO_STACKSIZE
48# define XTHREAD_STACKSIZE EIO_STACKSIZE
49#endif
41#include "xthread.h" 50#include "xthread.h"
42 51
43#include <errno.h> 52#include <errno.h>
44#include <stddef.h> 53#include <stddef.h>
45#include <stdlib.h> 54#include <stdlib.h>
49#include <sys/stat.h> 58#include <sys/stat.h>
50#include <limits.h> 59#include <limits.h>
51#include <fcntl.h> 60#include <fcntl.h>
52#include <assert.h> 61#include <assert.h>
53 62
63/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
64/* intptr_t only comes from stdint.h, says idiot openbsd coder */
65#if HAVE_STDINT_H
66# include <stdint.h>
67#endif
68
69#ifndef ECANCELED
70# define ECANCELED EDOM
71#endif
72#ifndef ELOOP
73# define ELOOP EDOM
74#endif
75
76#if !defined(ENOTSOCK) && defined(WSAENOTSOCK)
77# define ENOTSOCK WSAENOTSOCK
78#endif
79
80static void eio_destroy (eio_req *req);
81
54#ifndef EIO_FINISH 82#ifndef EIO_FINISH
55# 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
56#endif 84#endif
57 85
58#ifndef EIO_DESTROY 86#ifndef EIO_DESTROY
61 89
62#ifndef EIO_FEED 90#ifndef EIO_FEED
63# 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)
64#endif 92#endif
65 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
66#ifdef _WIN32 105#ifdef _WIN32
67 106
68 /*doh*/ 107 #undef PAGESIZE
108 #define PAGESIZE 4096 /* GetSystemInfo? */
109
110 #ifdef EIO_STRUCT_STATI64
111 #define stat(path,buf) _stati64 (path,buf)
112 #define fstat(fd,buf) _fstati64 (fd,buf)
113 #endif
114 #define lstat(path,buf) stat (path,buf)
115 #define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
116 #define mkdir(path,mode) _mkdir (path)
117 #define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
118
119 #define chmod(path,mode) _chmod (path, mode)
120 #define dup(fd) _dup (fd)
121 #define dup2(fd1,fd2) _dup2 (fd1, fd2)
122
123 #define fchmod(fd,mode) EIO_ENOSYS ()
124 #define chown(path,uid,gid) EIO_ENOSYS ()
125 #define fchown(fd,uid,gid) EIO_ENOSYS ()
126 #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
127 #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
128 #define mknod(path,mode,dev) EIO_ENOSYS ()
129 #define sync() EIO_ENOSYS ()
130 #define readlink(path,buf,s) EIO_ENOSYS ()
131 #define statvfs(path,buf) EIO_ENOSYS ()
132 #define fstatvfs(fd,buf) EIO_ENOSYS ()
133
134 /* we could even stat and see if it exists */
135 static int
136 symlink (const char *old, const char *neu)
137 {
138 #if WINVER >= 0x0600
139 if (CreateSymbolicLink (neu, old, 1))
140 return 0;
141
142 if (CreateSymbolicLink (neu, old, 0))
143 return 0;
144 #endif
145
146 return EIO_ERRNO (ENOENT, -1);
147 }
148
149 /* POSIX API only */
150 #define CreateHardLink(neu,old,flags) 0
151 #define CreateSymbolicLink(neu,old,flags) 0
152
153 struct statvfs
154 {
155 int dummy;
156 };
157
158 #define DT_DIR EIO_DT_DIR
159 #define DT_REG EIO_DT_REG
160 #define D_NAME(entp) entp.cFileName
161 #define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
69 162
70#else 163#else
71 164
72# include "config.h"
73# include <sys/time.h> 165 #include <sys/time.h>
74# include <sys/select.h> 166 #include <sys/select.h>
167 #include <sys/statvfs.h>
75# include <unistd.h> 168 #include <unistd.h>
76# include <utime.h> 169 #include <utime.h>
77# include <signal.h> 170 #include <signal.h>
78# include <dirent.h> 171 #include <dirent.h>
79 172
173 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
174 #include <sys/mman.h>
175 #endif
176
177 #define D_NAME(entp) entp->d_name
178
179 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
180 #if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
181 #define _DIRENT_HAVE_D_TYPE /* sigh */
182 #define D_INO(de) (de)->d_fileno
183 #define D_NAMLEN(de) (de)->d_namlen
184 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
185 #define D_INO(de) (de)->d_ino
186 #endif
187
188 #ifdef _D_EXACT_NAMLEN
189 #undef D_NAMLEN
190 #define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
191 #endif
192
193 #ifdef _DIRENT_HAVE_D_TYPE
194 #define D_TYPE(de) (de)->d_type
195 #endif
196
80# ifndef EIO_STRUCT_DIRENT 197 #ifndef EIO_STRUCT_DIRENT
81# define EIO_STRUCT_DIRENT struct dirent 198 #define EIO_STRUCT_DIRENT struct dirent
82# endif 199 #endif
83 200
84#endif 201#endif
85 202
86#if HAVE_SENDFILE 203#if HAVE_SENDFILE
87# if __linux 204# if __linux
88# include <sys/sendfile.h> 205# include <sys/sendfile.h>
89# elif __freebsd 206# elif __FreeBSD__ || defined __APPLE__
90# include <sys/socket.h> 207# include <sys/socket.h>
91# include <sys/uio.h> 208# include <sys/uio.h>
92# elif __hpux 209# elif __hpux
93# include <sys/socket.h> 210# include <sys/socket.h>
94# elif __solaris /* not yet */ 211# elif __solaris
95# include <sys/sendfile.h> 212# include <sys/sendfile.h>
96# else 213# else
97# error sendfile support requested but not available 214# error sendfile support requested but not available
98# endif 215# endif
99#endif 216#endif
100 217
101/* number of seconds after which an idle threads exit */ 218#ifndef D_TYPE
102#define IDLE_TIMEOUT 10 219# define D_TYPE(de) 0
220#endif
221#ifndef D_INO
222# define D_INO(de) 0
223#endif
224#ifndef D_NAMLEN
225# define D_NAMLEN(entp) strlen (D_NAME (entp))
226#endif
103 227
104/* used for struct dirent, AIX doesn't provide it */ 228/* used for struct dirent, AIX doesn't provide it */
105#ifndef NAME_MAX 229#ifndef NAME_MAX
106# define NAME_MAX 4096 230# define NAME_MAX 4096
107#endif 231#endif
108 232
233/* used for readlink etc. */
234#ifndef PATH_MAX
235# define PATH_MAX 4096
236#endif
237
109/* buffer size for various temporary buffers */ 238/* buffer size for various temporary buffers */
110#define EIO_BUFSIZE 65536 239#define EIO_BUFSIZE 65536
111 240
112#define dBUF \ 241#define dBUF \
113 char *eio_buf; \
114 ETP_WORKER_LOCK (self); \
115 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 242 char *eio_buf = malloc (EIO_BUFSIZE); \
116 ETP_WORKER_UNLOCK (self); \
117 errno = ENOMEM; \ 243 errno = ENOMEM; \
118 if (!eio_buf) \ 244 if (!eio_buf) \
119 return -1; 245 return -1
246
247#define FUBd \
248 free (eio_buf)
120 249
121#define EIO_TICKS ((1000000 + 1023) >> 10) 250#define EIO_TICKS ((1000000 + 1023) >> 10)
122
123/*****************************************************************************/
124
125#if __GNUC__ >= 3
126# define expect(expr,value) __builtin_expect ((expr),(value))
127#else
128# define expect(expr,value) (expr)
129#endif
130
131#define expect_false(expr) expect ((expr) != 0, 0)
132#define expect_true(expr) expect ((expr) != 0, 1)
133
134/*****************************************************************************/
135 251
136#define ETP_PRI_MIN EIO_PRI_MIN 252#define ETP_PRI_MIN EIO_PRI_MIN
137#define ETP_PRI_MAX EIO_PRI_MAX 253#define ETP_PRI_MAX EIO_PRI_MAX
138 254
139struct etp_worker; 255struct etp_worker;
143static int eio_finish (eio_req *req); 259static int eio_finish (eio_req *req);
144#define ETP_FINISH(req) eio_finish (req) 260#define ETP_FINISH(req) eio_finish (req)
145static void eio_execute (struct etp_worker *self, eio_req *req); 261static void eio_execute (struct etp_worker *self, eio_req *req);
146#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) 262#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
147 263
148#define ETP_WORKER_CLEAR(req) \
149 if (wrk->dbuf) \
150 { \
151 free (wrk->dbuf); \
152 wrk->dbuf = 0; \
153 } \
154 \
155 if (wrk->dirp) \
156 { \
157 closedir (wrk->dirp); \
158 wrk->dirp = 0; \
159 }
160#define ETP_WORKER_COMMON \
161 void *dbuf; \
162 DIR *dirp;
163
164/*****************************************************************************/ 264/*****************************************************************************/
165 265
166#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) 266#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
167 267
168/* calculcate time difference in ~1/EIO_TICKS of a second */ 268/* calculate time difference in ~1/EIO_TICKS of a second */
269ecb_inline int
169static int tvdiff (struct timeval *tv1, struct timeval *tv2) 270tvdiff (struct timeval *tv1, struct timeval *tv2)
170{ 271{
171 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 272 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
172 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 273 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
173} 274}
174 275
181static unsigned int max_poll_reqs; /* reslock */ 282static unsigned int max_poll_reqs; /* reslock */
182 283
183static volatile unsigned int nreqs; /* reqlock */ 284static volatile unsigned int nreqs; /* reqlock */
184static volatile unsigned int nready; /* reqlock */ 285static volatile unsigned int nready; /* reqlock */
185static volatile unsigned int npending; /* reqlock */ 286static volatile unsigned int npending; /* reqlock */
186static volatile unsigned int max_idle = 4; 287static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
288static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
187 289
188static mutex_t wrklock = X_MUTEX_INIT; 290static xmutex_t wrklock;
189static mutex_t reslock = X_MUTEX_INIT; 291static xmutex_t reslock;
190static mutex_t reqlock = X_MUTEX_INIT; 292static xmutex_t reqlock;
191static cond_t reqwait = X_COND_INIT; 293static xcond_t reqwait;
192 294
193#if !HAVE_PREADWRITE 295#if !HAVE_PREADWRITE
194/* 296/*
195 * make our pread/pwrite emulation safe against themselves, but not against 297 * make our pread/pwrite emulation safe against themselves, but not against
196 * normal read/write by using a mutex. slows down execution a lot, 298 * normal read/write by using a mutex. slows down execution a lot,
197 * but that's your problem, not mine. 299 * but that's your problem, not mine.
198 */ 300 */
199static mutex_t preadwritelock = X_MUTEX_INIT; 301static xmutex_t preadwritelock;
200#endif 302#endif
201 303
202typedef struct etp_worker 304typedef struct etp_worker
203{ 305{
204 /* locked by wrklock */ 306 /* locked by wrklock */
205 struct etp_worker *prev, *next; 307 struct etp_worker *prev, *next;
206 308
207 thread_t tid; 309 xthread_t tid;
208 310
209 /* locked by reslock, reqlock or wrklock */ 311 /* locked by reslock, reqlock or wrklock */
210 ETP_REQ *req; /* currently processed request */ 312 ETP_REQ *req; /* currently processed request */
211 313
314#ifdef ETP_WORKER_COMMON
212 ETP_WORKER_COMMON 315 ETP_WORKER_COMMON
316#endif
213} etp_worker; 317} etp_worker;
214 318
215static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */ 319static etp_worker wrk_first; /* NOT etp */
216 320
217#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock) 321#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
218#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock) 322#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
219 323
220/* worker threads management */ 324/* worker threads management */
221 325
326static void ecb_cold
222static void etp_worker_clear (etp_worker *wrk) 327etp_worker_clear (etp_worker *wrk)
223{ 328{
224 ETP_WORKER_CLEAR (wrk);
225} 329}
226 330
331static void ecb_cold
227static void etp_worker_free (etp_worker *wrk) 332etp_worker_free (etp_worker *wrk)
228{ 333{
229 wrk->next->prev = wrk->prev; 334 wrk->next->prev = wrk->prev;
230 wrk->prev->next = wrk->next; 335 wrk->prev->next = wrk->next;
231 336
232 free (wrk); 337 free (wrk);
233} 338}
234 339
235static unsigned int etp_nreqs (void) 340static unsigned int
341etp_nreqs (void)
236{ 342{
237 int retval; 343 int retval;
238 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 344 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
239 retval = nreqs; 345 retval = nreqs;
240 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 346 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
241 return retval; 347 return retval;
242} 348}
243 349
244static unsigned int etp_nready (void) 350static unsigned int
351etp_nready (void)
245{ 352{
246 unsigned int retval; 353 unsigned int retval;
247 354
248 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 355 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
249 retval = nready; 356 retval = nready;
250 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 357 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
251 358
252 return retval; 359 return retval;
253} 360}
254 361
255static unsigned int etp_npending (void) 362static unsigned int
363etp_npending (void)
256{ 364{
257 unsigned int retval; 365 unsigned int retval;
258 366
259 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 367 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
260 retval = npending; 368 retval = npending;
261 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 369 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
262 370
263 return retval; 371 return retval;
264} 372}
265 373
266static unsigned int etp_nthreads (void) 374static unsigned int
375etp_nthreads (void)
267{ 376{
268 unsigned int retval; 377 unsigned int retval;
269 378
270 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 379 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
271 retval = started; 380 retval = started;
285} etp_reqq; 394} etp_reqq;
286 395
287static etp_reqq req_queue; 396static etp_reqq req_queue;
288static etp_reqq res_queue; 397static etp_reqq res_queue;
289 398
399static void ecb_noinline ecb_cold
400reqq_init (etp_reqq *q)
401{
402 int pri;
403
404 for (pri = 0; pri < ETP_NUM_PRI; ++pri)
405 q->qs[pri] = q->qe[pri] = 0;
406
407 q->size = 0;
408}
409
410static int ecb_noinline
290static int reqq_push (etp_reqq *q, ETP_REQ *req) 411reqq_push (etp_reqq *q, ETP_REQ *req)
291{ 412{
292 int pri = req->pri; 413 int pri = req->pri;
293 req->next = 0; 414 req->next = 0;
294 415
295 if (q->qe[pri]) 416 if (q->qe[pri])
301 q->qe[pri] = q->qs[pri] = req; 422 q->qe[pri] = q->qs[pri] = req;
302 423
303 return q->size++; 424 return q->size++;
304} 425}
305 426
427static ETP_REQ * ecb_noinline
306static ETP_REQ *reqq_shift (etp_reqq *q) 428reqq_shift (etp_reqq *q)
307{ 429{
308 int pri; 430 int pri;
309 431
310 if (!q->size) 432 if (!q->size)
311 return 0; 433 return 0;
326 } 448 }
327 449
328 abort (); 450 abort ();
329} 451}
330 452
331static void etp_atfork_prepare (void) 453static int ecb_cold
454etp_init (void (*want_poll)(void), void (*done_poll)(void))
332{ 455{
333 X_LOCK (wrklock); 456 X_MUTEX_CREATE (wrklock);
334 X_LOCK (reqlock); 457 X_MUTEX_CREATE (reslock);
335 X_LOCK (reslock); 458 X_MUTEX_CREATE (reqlock);
336#if !HAVE_PREADWRITE 459 X_COND_CREATE (reqwait);
337 X_LOCK (preadwritelock);
338#endif
339}
340 460
341static void etp_atfork_parent (void) 461 reqq_init (&req_queue);
342{ 462 reqq_init (&res_queue);
343#if !HAVE_PREADWRITE
344 X_UNLOCK (preadwritelock);
345#endif
346 X_UNLOCK (reslock);
347 X_UNLOCK (reqlock);
348 X_UNLOCK (wrklock);
349}
350 463
351static void etp_atfork_child (void) 464 wrk_first.next =
352{ 465 wrk_first.prev = &wrk_first;
353 ETP_REQ *prv;
354
355 while ((prv = reqq_shift (&req_queue)))
356 ETP_DESTROY (prv);
357
358 while ((prv = reqq_shift (&res_queue)))
359 ETP_DESTROY (prv);
360
361 while (wrk_first.next != &wrk_first)
362 {
363 etp_worker *wrk = wrk_first.next;
364
365 if (wrk->req)
366 ETP_DESTROY (wrk->req);
367
368 etp_worker_clear (wrk);
369 etp_worker_free (wrk);
370 }
371 466
372 started = 0; 467 started = 0;
373 idle = 0; 468 idle = 0;
374 nreqs = 0; 469 nreqs = 0;
375 nready = 0; 470 nready = 0;
376 npending = 0; 471 npending = 0;
377 472
378 etp_atfork_parent ();
379}
380
381static void
382etp_once_init (void)
383{
384 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
385}
386
387static int
388etp_init (void (*want_poll)(void), void (*done_poll)(void))
389{
390 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
391
392 pthread_once (&doinit, etp_once_init);
393
394 want_poll_cb = want_poll; 473 want_poll_cb = want_poll;
395 done_poll_cb = done_poll; 474 done_poll_cb = done_poll;
396 475
397 return 0; 476 return 0;
398} 477}
399 478
400X_THREAD_PROC (etp_proc); 479X_THREAD_PROC (etp_proc);
401 480
481static void ecb_cold
402static void etp_start_thread (void) 482etp_start_thread (void)
403{ 483{
404 etp_worker *wrk = calloc (1, sizeof (etp_worker)); 484 etp_worker *wrk = calloc (1, sizeof (etp_worker));
405 485
406 /*TODO*/ 486 /*TODO*/
407 assert (("unable to allocate worker thread data", wrk)); 487 assert (("unable to allocate worker thread data", wrk));
420 free (wrk); 500 free (wrk);
421 501
422 X_UNLOCK (wrklock); 502 X_UNLOCK (wrklock);
423} 503}
424 504
505static void
425static void etp_maybe_start_thread (void) 506etp_maybe_start_thread (void)
426{ 507{
427 if (expect_true (etp_nthreads () >= wanted)) 508 if (ecb_expect_true (etp_nthreads () >= wanted))
428 return; 509 return;
429 510
430 /* todo: maybe use idle here, but might be less exact */ 511 /* todo: maybe use idle here, but might be less exact */
431 if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ())) 512 if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
432 return; 513 return;
433 514
434 etp_start_thread (); 515 etp_start_thread ();
435} 516}
436 517
518static void ecb_cold
437static void etp_end_thread (void) 519etp_end_thread (void)
438{ 520{
439 eio_req *req = calloc (1, sizeof (eio_req)); 521 eio_req *req = calloc (1, sizeof (eio_req));
440 522
441 req->type = -1; 523 req->type = -1;
442 req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 524 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
449 X_LOCK (wrklock); 531 X_LOCK (wrklock);
450 --started; 532 --started;
451 X_UNLOCK (wrklock); 533 X_UNLOCK (wrklock);
452} 534}
453 535
454static int etp_poll (void) 536static int
537etp_poll (void)
455{ 538{
456 unsigned int maxreqs; 539 unsigned int maxreqs;
457 unsigned int maxtime; 540 unsigned int maxtime;
458 struct timeval tv_start, tv_now; 541 struct timeval tv_start, tv_now;
459 542
489 572
490 X_LOCK (reqlock); 573 X_LOCK (reqlock);
491 --nreqs; 574 --nreqs;
492 X_UNLOCK (reqlock); 575 X_UNLOCK (reqlock);
493 576
494 if (expect_false (req->type == EIO_GROUP && req->size)) 577 if (ecb_expect_false (req->type == EIO_GROUP && req->size))
495 { 578 {
496 req->int1 = 1; /* mark request as delayed */ 579 req->int1 = 1; /* mark request as delayed */
497 continue; 580 continue;
498 } 581 }
499 else 582 else
500 { 583 {
501 int res = ETP_FINISH (req); 584 int res = ETP_FINISH (req);
502 if (expect_false (res)) 585 if (ecb_expect_false (res))
503 return res; 586 return res;
504 } 587 }
505 588
506 if (expect_false (maxreqs && !--maxreqs)) 589 if (ecb_expect_false (maxreqs && !--maxreqs))
507 break; 590 break;
508 591
509 if (maxtime) 592 if (maxtime)
510 { 593 {
511 gettimeofday (&tv_now, 0); 594 gettimeofday (&tv_now, 0);
517 600
518 errno = EAGAIN; 601 errno = EAGAIN;
519 return -1; 602 return -1;
520} 603}
521 604
605static void
522static void etp_cancel (ETP_REQ *req) 606etp_cancel (ETP_REQ *req)
523{ 607{
524 X_LOCK (wrklock); 608 req->cancelled = 1;
525 req->flags |= EIO_FLAG_CANCELLED;
526 X_UNLOCK (wrklock);
527 609
528 eio_grp_cancel (req); 610 eio_grp_cancel (req);
529} 611}
530 612
613static void
531static void etp_submit (ETP_REQ *req) 614etp_submit (ETP_REQ *req)
532{ 615{
533 req->pri -= ETP_PRI_MIN; 616 req->pri -= ETP_PRI_MIN;
534 617
535 if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN; 618 if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
536 if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 619 if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
537 620
538 if (expect_false (req->type == EIO_GROUP)) 621 if (ecb_expect_false (req->type == EIO_GROUP))
539 { 622 {
540 /* I hope this is worth it :/ */ 623 /* I hope this is worth it :/ */
541 X_LOCK (reqlock); 624 X_LOCK (reqlock);
542 ++nreqs; 625 ++nreqs;
543 X_UNLOCK (reqlock); 626 X_UNLOCK (reqlock);
562 645
563 etp_maybe_start_thread (); 646 etp_maybe_start_thread ();
564 } 647 }
565} 648}
566 649
650static void ecb_cold
567static void etp_set_max_poll_time (double nseconds) 651etp_set_max_poll_time (double nseconds)
568{ 652{
569 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 653 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
570 max_poll_time = nseconds; 654 max_poll_time = nseconds * EIO_TICKS;
571 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 655 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
572} 656}
573 657
658static void ecb_cold
574static void etp_set_max_poll_reqs (unsigned int maxreqs) 659etp_set_max_poll_reqs (unsigned int maxreqs)
575{ 660{
576 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 661 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
577 max_poll_reqs = maxreqs; 662 max_poll_reqs = maxreqs;
578 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 663 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
579} 664}
580 665
666static void ecb_cold
581static void etp_set_max_idle (unsigned int nthreads) 667etp_set_max_idle (unsigned int nthreads)
582{ 668{
583 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 669 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
584 max_idle = nthreads <= 0 ? 1 : nthreads; 670 max_idle = nthreads;
585 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 671 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
586} 672}
587 673
674static void ecb_cold
675etp_set_idle_timeout (unsigned int seconds)
676{
677 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
678 idle_timeout = seconds;
679 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
680}
681
682static void ecb_cold
588static void etp_set_min_parallel (unsigned int nthreads) 683etp_set_min_parallel (unsigned int nthreads)
589{ 684{
590 if (wanted < nthreads) 685 if (wanted < nthreads)
591 wanted = nthreads; 686 wanted = nthreads;
592} 687}
593 688
689static void ecb_cold
594static void etp_set_max_parallel (unsigned int nthreads) 690etp_set_max_parallel (unsigned int nthreads)
595{ 691{
596 if (wanted > nthreads) 692 if (wanted > nthreads)
597 wanted = nthreads; 693 wanted = nthreads;
598 694
599 while (started > wanted) 695 while (started > wanted)
600 etp_end_thread (); 696 etp_end_thread ();
601} 697}
602 698
603/*****************************************************************************/ 699/*****************************************************************************/
604 700
701static void
605static void grp_try_feed (eio_req *grp) 702grp_try_feed (eio_req *grp)
606{ 703{
607 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 704 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
608 { 705 {
609 grp->flags &= ~EIO_FLAG_GROUPADD; 706 grp->flags &= ~EIO_FLAG_GROUPADD;
610 707
617 break; 714 break;
618 } 715 }
619 } 716 }
620} 717}
621 718
719static int
622static int grp_dec (eio_req *grp) 720grp_dec (eio_req *grp)
623{ 721{
624 --grp->size; 722 --grp->size;
625 723
626 /* call feeder, if applicable */ 724 /* call feeder, if applicable */
627 grp_try_feed (grp); 725 grp_try_feed (grp);
631 return eio_finish (grp); 729 return eio_finish (grp);
632 else 730 else
633 return 0; 731 return 0;
634} 732}
635 733
734static void
636void eio_destroy (eio_req *req) 735eio_destroy (eio_req *req)
637{ 736{
638 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); 737 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
639 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); 738 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
640 739
641 EIO_DESTROY (req); 740 EIO_DESTROY (req);
642} 741}
643 742
743static int
644static int eio_finish (eio_req *req) 744eio_finish (eio_req *req)
645{ 745{
646 int res = EIO_FINISH (req); 746 int res = EIO_FINISH (req);
647 747
648 if (req->grp) 748 if (req->grp)
649 { 749 {
657 if (grp->grp_first == req) 757 if (grp->grp_first == req)
658 grp->grp_first = req->grp_next; 758 grp->grp_first = req->grp_next;
659 759
660 res2 = grp_dec (grp); 760 res2 = grp_dec (grp);
661 761
662 if (!res && res2) 762 if (!res)
663 res = res2; 763 res = res2;
664 } 764 }
665 765
666 eio_destroy (req); 766 eio_destroy (req);
667 767
668 return res; 768 return res;
669} 769}
670 770
771void
671void eio_grp_cancel (eio_req *grp) 772eio_grp_cancel (eio_req *grp)
672{ 773{
673 for (grp = grp->grp_first; grp; grp = grp->grp_next) 774 for (grp = grp->grp_first; grp; grp = grp->grp_next)
674 eio_cancel (grp); 775 eio_cancel (grp);
675} 776}
676 777
778void
677void eio_cancel (eio_req *req) 779eio_cancel (eio_req *req)
678{ 780{
679 etp_cancel (req); 781 etp_cancel (req);
680} 782}
681 783
784void
682void eio_submit (eio_req *req) 785eio_submit (eio_req *req)
683{ 786{
684 etp_submit (req); 787 etp_submit (req);
685} 788}
686 789
687unsigned int eio_nreqs (void) 790unsigned int
791eio_nreqs (void)
688{ 792{
689 return etp_nreqs (); 793 return etp_nreqs ();
690} 794}
691 795
692unsigned int eio_nready (void) 796unsigned int
797eio_nready (void)
693{ 798{
694 return etp_nready (); 799 return etp_nready ();
695} 800}
696 801
697unsigned int eio_npending (void) 802unsigned int
803eio_npending (void)
698{ 804{
699 return etp_npending (); 805 return etp_npending ();
700} 806}
701 807
702unsigned int eio_nthreads (void) 808unsigned int ecb_cold
809eio_nthreads (void)
703{ 810{
704 return etp_nthreads (); 811 return etp_nthreads ();
705} 812}
706 813
814void ecb_cold
707void eio_set_max_poll_time (double nseconds) 815eio_set_max_poll_time (double nseconds)
708{ 816{
709 etp_set_max_poll_time (nseconds); 817 etp_set_max_poll_time (nseconds);
710} 818}
711 819
820void ecb_cold
712void eio_set_max_poll_reqs (unsigned int maxreqs) 821eio_set_max_poll_reqs (unsigned int maxreqs)
713{ 822{
714 etp_set_max_poll_reqs (maxreqs); 823 etp_set_max_poll_reqs (maxreqs);
715} 824}
716 825
826void ecb_cold
717void eio_set_max_idle (unsigned int nthreads) 827eio_set_max_idle (unsigned int nthreads)
718{ 828{
719 etp_set_max_idle (nthreads); 829 etp_set_max_idle (nthreads);
720} 830}
721 831
832void ecb_cold
833eio_set_idle_timeout (unsigned int seconds)
834{
835 etp_set_idle_timeout (seconds);
836}
837
838void ecb_cold
722void eio_set_min_parallel (unsigned int nthreads) 839eio_set_min_parallel (unsigned int nthreads)
723{ 840{
724 etp_set_min_parallel (nthreads); 841 etp_set_min_parallel (nthreads);
725} 842}
726 843
844void ecb_cold
727void eio_set_max_parallel (unsigned int nthreads) 845eio_set_max_parallel (unsigned int nthreads)
728{ 846{
729 etp_set_max_parallel (nthreads); 847 etp_set_max_parallel (nthreads);
730} 848}
731 849
732int eio_poll (void) 850int eio_poll (void)
741# undef pread 859# undef pread
742# undef pwrite 860# undef pwrite
743# define pread eio__pread 861# define pread eio__pread
744# define pwrite eio__pwrite 862# define pwrite eio__pwrite
745 863
746static ssize_t 864static eio_ssize_t
747eio__pread (int fd, void *buf, size_t count, off_t offset) 865eio__pread (int fd, void *buf, size_t count, off_t offset)
748{ 866{
749 ssize_t res; 867 eio_ssize_t res;
750 off_t ooffset; 868 off_t ooffset;
751 869
752 X_LOCK (preadwritelock); 870 X_LOCK (preadwritelock);
753 ooffset = lseek (fd, 0, SEEK_CUR); 871 ooffset = lseek (fd, 0, SEEK_CUR);
754 lseek (fd, offset, SEEK_SET); 872 lseek (fd, offset, SEEK_SET);
757 X_UNLOCK (preadwritelock); 875 X_UNLOCK (preadwritelock);
758 876
759 return res; 877 return res;
760} 878}
761 879
762static ssize_t 880static eio_ssize_t
763eio__pwrite (int fd, void *buf, size_t count, off_t offset) 881eio__pwrite (int fd, void *buf, size_t count, off_t offset)
764{ 882{
765 ssize_t res; 883 eio_ssize_t res;
766 off_t ooffset; 884 off_t ooffset;
767 885
768 X_LOCK (preadwritelock); 886 X_LOCK (preadwritelock);
769 ooffset = lseek (fd, 0, SEEK_CUR); 887 ooffset = lseek (fd, 0, SEEK_CUR);
770 lseek (fd, offset, SEEK_SET); 888 lseek (fd, offset, SEEK_SET);
771 res = write (fd, buf, count); 889 res = write (fd, buf, count);
772 lseek (fd, offset, SEEK_SET); 890 lseek (fd, ooffset, SEEK_SET);
773 X_UNLOCK (preadwritelock); 891 X_UNLOCK (preadwritelock);
774 892
775 return res; 893 return res;
776} 894}
777#endif 895#endif
778 896
779#ifndef HAVE_FUTIMES 897#ifndef HAVE_UTIMES
780 898
781# undef utimes 899# undef utimes
782# undef futimes
783# define utimes(path,times) eio__utimes (path, times) 900# define utimes(path,times) eio__utimes (path, times)
784# define futimes(fd,times) eio__futimes (fd, times)
785 901
786static int 902static int
787eio__utimes (const char *filename, const struct timeval times[2]) 903eio__utimes (const char *filename, const struct timeval times[2])
788{ 904{
789 if (times) 905 if (times)
797 } 913 }
798 else 914 else
799 return utime (filename, 0); 915 return utime (filename, 0);
800} 916}
801 917
918#endif
919
920#ifndef HAVE_FUTIMES
921
922# undef futimes
923# define futimes(fd,times) eio__futimes (fd, times)
924
925static int
802static int eio__futimes (int fd, const struct timeval tv[2]) 926eio__futimes (int fd, const struct timeval tv[2])
803{ 927{
804 errno = ENOSYS; 928 errno = ENOSYS;
805 return -1; 929 return -1;
806} 930}
807 931
811# undef fdatasync 935# undef fdatasync
812# define fdatasync(fd) fsync (fd) 936# define fdatasync(fd) fsync (fd)
813#endif 937#endif
814 938
815/* sync_file_range always needs emulation */ 939/* sync_file_range always needs emulation */
816int 940static int
817eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags) 941eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
818{ 942{
819#if HAVE_SYNC_FILE_RANGE 943#if HAVE_SYNC_FILE_RANGE
820 int res; 944 int res;
821 945
829 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0); 953 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
830 } 954 }
831 955
832 res = sync_file_range (fd, offset, nbytes, flags); 956 res = sync_file_range (fd, offset, nbytes, flags);
833 957
834 if (res != ENOSYS) 958 if (!res || errno != ENOSYS)
835 return res; 959 return res;
836#endif 960#endif
837 961
838 /* even though we could play tricks with the flags, it's better to always 962 /* even though we could play tricks with the flags, it's better to always
839 * call fdatasync, as thta matches the expectation of it's users best */ 963 * call fdatasync, as that matches the expectation of its users best */
840 return fdatasync (fd); 964 return fdatasync (fd);
965}
966
967static int
968eio__fallocate (int fd, int mode, off_t offset, size_t len)
969{
970#if HAVE_FALLOCATE
971 return fallocate (fd, mode, offset, len);
972#else
973 errno = ENOSYS;
974 return -1;
975#endif
841} 976}
842 977
843#if !HAVE_READAHEAD 978#if !HAVE_READAHEAD
844# undef readahead 979# undef readahead
845# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 980# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
846 981
847static ssize_t 982static eio_ssize_t
848eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) 983eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
849{ 984{
850 size_t todo = count; 985 size_t todo = count;
851 dBUF; 986 dBUF;
852 987
857 pread (fd, eio_buf, len, offset); 992 pread (fd, eio_buf, len, offset);
858 offset += len; 993 offset += len;
859 todo -= len; 994 todo -= len;
860 } 995 }
861 996
997 FUBd;
998
862 errno = 0; 999 errno = 0;
863 return count; 1000 return count;
864} 1001}
865 1002
866#endif 1003#endif
867 1004
868/* sendfile always needs emulation */ 1005/* sendfile always needs emulation */
869static ssize_t 1006static eio_ssize_t
870eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) 1007eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
871{ 1008{
1009 eio_ssize_t written = 0;
872 ssize_t res; 1010 eio_ssize_t res;
873 1011
874 if (!count) 1012 if (!count)
875 return 0; 1013 return 0;
876 1014
1015 for (;;)
1016 {
1017#ifdef __APPLE__
1018# undef HAVE_SENDFILE /* broken, as everything on os x */
1019#endif
877#if HAVE_SENDFILE 1020#if HAVE_SENDFILE
878# if __linux 1021# if __linux
1022 off_t soffset = offset;
879 res = sendfile (ofd, ifd, &offset, count); 1023 res = sendfile (ofd, ifd, &soffset, count);
880 1024
881# elif __freebsd 1025# elif __FreeBSD__
882 /* 1026 /*
883 * Of course, the freebsd sendfile is a dire hack with no thoughts 1027 * Of course, the freebsd sendfile is a dire hack with no thoughts
884 * wasted on making it similar to other I/O functions. 1028 * wasted on making it similar to other I/O functions.
885 */ 1029 */
886 {
887 off_t sbytes; 1030 off_t sbytes;
888 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 1031 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
889 1032
890 if (res < 0 && sbytes) 1033 #if 0 /* according to the manpage, this is correct, but broken behaviour */
891 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 1034 /* freebsd' sendfile will return 0 on success */
1035 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
1036 /* not on e.g. EIO or EPIPE - sounds broken */
1037 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
892 res = sbytes; 1038 res = sbytes;
893 } 1039 #endif
1040
1041 /* according to source inspection, this is correct, and useful behaviour */
1042 if (sbytes)
1043 res = sbytes;
1044
1045# elif defined (__APPLE__)
1046 off_t sbytes = count;
1047 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
1048
1049 /* according to the manpage, sbytes is always valid */
1050 if (sbytes)
1051 res = sbytes;
894 1052
895# elif __hpux 1053# elif __hpux
896 res = sendfile (ofd, ifd, offset, count, 0, 0); 1054 res = sendfile (ofd, ifd, offset, count, 0, 0);
897 1055
898# elif __solaris 1056# elif __solaris
899 {
900 struct sendfilevec vec; 1057 struct sendfilevec vec;
901 size_t sbytes; 1058 size_t sbytes;
902 1059
903 vec.sfv_fd = ifd; 1060 vec.sfv_fd = ifd;
904 vec.sfv_flag = 0; 1061 vec.sfv_flag = 0;
905 vec.sfv_off = offset; 1062 vec.sfv_off = offset;
906 vec.sfv_len = count; 1063 vec.sfv_len = count;
907 1064
908 res = sendfilev (ofd, &vec, 1, &sbytes); 1065 res = sendfilev (ofd, &vec, 1, &sbytes);
909 1066
910 if (res < 0 && sbytes) 1067 if (res < 0 && sbytes)
911 res = sbytes; 1068 res = sbytes;
912 }
913 1069
914# endif 1070# endif
1071
1072#elif defined (_WIN32) && 0
1073 /* does not work, just for documentation of what would need to be done */
1074 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
1075 /* libeio guarantees that the file offset does not change, and windows */
1076 /* has no way to get an independent handle to the same file description */
1077 HANDLE h = TO_SOCKET (ifd);
1078 SetFilePointer (h, offset, 0, FILE_BEGIN);
1079 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1080
915#else 1081#else
916 res = -1; 1082 res = -1;
917 errno = ENOSYS; 1083 errno = ENOSYS;
918#endif 1084#endif
919 1085
1086 /* we assume sendfile can copy at least 128mb in one go */
1087 if (res <= 128 * 1024 * 1024)
1088 {
1089 if (res > 0)
1090 written += res;
1091
1092 if (written)
1093 return written;
1094
1095 break;
1096 }
1097 else
1098 {
1099 /* if we requested more, then probably the kernel was lazy */
1100 written += res;
1101 offset += res;
1102 count -= res;
1103
1104 if (!count)
1105 return written;
1106 }
1107 }
1108
920 if (res < 0 1109 if (res < 0
921 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1110 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1111 /* BSDs */
1112#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1113 || errno == ENOTSUP
1114#endif
1115#ifdef EOPNOTSUPP /* windows */
1116 || errno == EOPNOTSUPP /* BSDs */
1117#endif
922#if __solaris 1118#if __solaris
923 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1119 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
924#endif 1120#endif
925 ) 1121 )
926 ) 1122 )
930 1126
931 res = 0; 1127 res = 0;
932 1128
933 while (count) 1129 while (count)
934 { 1130 {
935 ssize_t cnt; 1131 eio_ssize_t cnt;
936 1132
937 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 1133 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
938 1134
939 if (cnt <= 0) 1135 if (cnt <= 0)
940 { 1136 {
952 1148
953 offset += cnt; 1149 offset += cnt;
954 res += cnt; 1150 res += cnt;
955 count -= cnt; 1151 count -= cnt;
956 } 1152 }
1153
1154 FUBd;
957 } 1155 }
958 1156
959 return res; 1157 return res;
1158}
1159
1160#ifdef PAGESIZE
1161# define eio_pagesize() PAGESIZE
1162#else
1163static intptr_t
1164eio_pagesize (void)
1165{
1166 static intptr_t page;
1167
1168 if (!page)
1169 page = sysconf (_SC_PAGESIZE);
1170
1171 return page;
1172}
1173#endif
1174
1175static void
1176eio_page_align (void **addr, size_t *length)
1177{
1178 intptr_t mask = eio_pagesize () - 1;
1179
1180 /* round down addr */
1181 intptr_t adj = mask & (intptr_t)*addr;
1182
1183 *addr = (void *)((intptr_t)*addr - adj);
1184 *length += adj;
1185
1186 /* round up length */
1187 *length = (*length + mask) & ~mask;
1188}
1189
1190#if !_POSIX_MEMLOCK
1191# define eio__mlockall(a) EIO_ENOSYS ()
1192#else
1193
1194static int
1195eio__mlockall (int flags)
1196{
1197 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1198 extern int mallopt (int, int);
1199 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1200 #endif
1201
1202 if (EIO_MCL_CURRENT != MCL_CURRENT
1203 || EIO_MCL_FUTURE != MCL_FUTURE)
1204 {
1205 flags = 0
1206 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1207 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1208 }
1209
1210 return mlockall (flags);
1211}
1212#endif
1213
1214#if !_POSIX_MEMLOCK_RANGE
1215# define eio__mlock(a,b) EIO_ENOSYS ()
1216#else
1217
1218static int
1219eio__mlock (void *addr, size_t length)
1220{
1221 eio_page_align (&addr, &length);
1222
1223 return mlock (addr, length);
1224}
1225
1226#endif
1227
1228#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1229# define eio__msync(a,b,c) EIO_ENOSYS ()
1230#else
1231
1232static int
1233eio__msync (void *mem, size_t len, int flags)
1234{
1235 eio_page_align (&mem, &len);
1236
1237 if (EIO_MS_ASYNC != MS_SYNC
1238 || EIO_MS_INVALIDATE != MS_INVALIDATE
1239 || EIO_MS_SYNC != MS_SYNC)
1240 {
1241 flags = 0
1242 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1243 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1244 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1245 }
1246
1247 return msync (mem, len, flags);
1248}
1249
1250#endif
1251
1252static int
1253eio__mtouch (eio_req *req)
1254{
1255 void *mem = req->ptr2;
1256 size_t len = req->size;
1257 int flags = req->int1;
1258
1259 eio_page_align (&mem, &len);
1260
1261 {
1262 intptr_t addr = (intptr_t)mem;
1263 intptr_t end = addr + len;
1264 intptr_t page = eio_pagesize ();
1265
1266 if (addr < end)
1267 if (flags & EIO_MT_MODIFY) /* modify */
1268 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1269 else
1270 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1271 }
1272
1273 return 0;
1274}
1275
1276/*****************************************************************************/
1277/* requests implemented outside eio_execute, because they are so large */
1278
1279static void
1280eio__realpath (eio_req *req, etp_worker *self)
1281{
1282 char *rel = req->ptr1;
1283 char *res;
1284 char *tmp1, *tmp2;
1285#if SYMLOOP_MAX > 32
1286 int symlinks = SYMLOOP_MAX;
1287#else
1288 int symlinks = 32;
1289#endif
1290
1291 req->result = -1;
1292
1293 errno = EINVAL;
1294 if (!rel)
1295 return;
1296
1297 errno = ENOENT;
1298 if (!*rel)
1299 return;
1300
1301 if (!req->ptr2)
1302 {
1303 X_LOCK (wrklock);
1304 req->flags |= EIO_FLAG_PTR2_FREE;
1305 X_UNLOCK (wrklock);
1306 req->ptr2 = malloc (PATH_MAX * 3);
1307
1308 errno = ENOMEM;
1309 if (!req->ptr2)
1310 return;
1311 }
1312
1313 res = req->ptr2;
1314 tmp1 = res + PATH_MAX;
1315 tmp2 = tmp1 + PATH_MAX;
1316
1317#if 0 /* disabled, the musl way to do things is just too racy */
1318#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1319 /* on linux we may be able to ask the kernel */
1320 {
1321 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1322
1323 if (fd >= 0)
1324 {
1325 sprintf (tmp1, "/proc/self/fd/%d", fd);
1326 req->result = readlink (tmp1, res, PATH_MAX);
1327 close (fd);
1328
1329 /* here we should probably stat the open file and the disk file, to make sure they still match */
1330
1331 if (req->result > 0)
1332 goto done;
1333 }
1334 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1335 return;
1336 }
1337#endif
1338#endif
1339
1340 if (*rel != '/')
1341 {
1342 if (!getcwd (res, PATH_MAX))
1343 return;
1344
1345 if (res [1]) /* only use if not / */
1346 res += strlen (res);
1347 }
1348
1349 while (*rel)
1350 {
1351 eio_ssize_t len, linklen;
1352 char *beg = rel;
1353
1354 while (*rel && *rel != '/')
1355 ++rel;
1356
1357 len = rel - beg;
1358
1359 if (!len) /* skip slashes */
1360 {
1361 ++rel;
1362 continue;
1363 }
1364
1365 if (beg [0] == '.')
1366 {
1367 if (len == 1)
1368 continue; /* . - nop */
1369
1370 if (beg [1] == '.' && len == 2)
1371 {
1372 /* .. - back up one component, if possible */
1373
1374 while (res != req->ptr2)
1375 if (*--res == '/')
1376 break;
1377
1378 continue;
1379 }
1380 }
1381
1382 errno = ENAMETOOLONG;
1383 if (res + 1 + len + 1 >= tmp1)
1384 return;
1385
1386 /* copy one component */
1387 *res = '/';
1388 memcpy (res + 1, beg, len);
1389
1390 /* zero-terminate, for readlink */
1391 res [len + 1] = 0;
1392
1393 /* now check if it's a symlink */
1394 linklen = readlink (req->ptr2, tmp1, PATH_MAX);
1395
1396 if (linklen < 0)
1397 {
1398 if (errno != EINVAL)
1399 return;
1400
1401 /* it's a normal directory. hopefully */
1402 res += len + 1;
1403 }
1404 else
1405 {
1406 /* yay, it was a symlink - build new path in tmp2 */
1407 int rellen = strlen (rel);
1408
1409 errno = ENAMETOOLONG;
1410 if (linklen + 1 + rellen >= PATH_MAX)
1411 return;
1412
1413 errno = ELOOP;
1414 if (!--symlinks)
1415 return;
1416
1417 if (*tmp1 == '/')
1418 res = req->ptr2; /* symlink resolves to an absolute path */
1419
1420 /* we need to be careful, as rel might point into tmp2 already */
1421 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1422 tmp2 [linklen] = '/';
1423 memcpy (tmp2, tmp1, linklen);
1424
1425 rel = tmp2;
1426 }
1427 }
1428
1429 /* special case for the lone root path */
1430 if (res == req->ptr2)
1431 *res++ = '/';
1432
1433 req->result = res - (char *)req->ptr2;
1434
1435done:
1436 req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
1437}
1438
1439static signed char
1440eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1441{
1442 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1443 : a->inode < b->inode ? -1
1444 : a->inode > b->inode ? 1
1445 : 0;
1446}
1447
1448#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1449
1450#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1451#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1452
1453static void
1454eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1455{
1456 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1457 unsigned char *bit = bits;
1458
1459 assert (CHAR_BIT == 8);
1460 assert (sizeof (eio_dirent) * 8 < 256);
1461 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1462 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1463
1464 if (size <= EIO_SORT_FAST)
1465 return;
1466
1467 /* first prepare an array of bits to test in our radix sort */
1468 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1469 /* inode_bits must contain all inodes ORed together */
1470 /* which is used to skip bits that are 0 everywhere, which is very common */
1471 {
1472 eio_ino_t endianness;
1473 int i, j;
1474
1475 /* we store the byte offset of byte n into byte n of "endianness" */
1476 for (i = 0; i < sizeof (eio_ino_t); ++i)
1477 ((unsigned char *)&endianness)[i] = i;
1478
1479 *bit++ = 0;
1480
1481 for (i = 0; i < sizeof (eio_ino_t); ++i)
1482 {
1483 /* shifting off the byte offsets out of "endianness" */
1484 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1485 endianness >>= 8;
1486
1487 for (j = 0; j < 8; ++j)
1488 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1489 *bit++ = offs + j;
1490 }
1491
1492 for (j = 0; j < 8; ++j)
1493 if (score_bits & (1 << j))
1494 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1495 }
1496
1497 /* now actually do the sorting (a variant of MSD radix sort) */
1498 {
1499 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1500 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1501 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1502 int stk_idx = 0;
1503
1504 base_stk [stk_idx] = dents;
1505 end_stk [stk_idx] = dents + size;
1506 bit_stk [stk_idx] = bit - 1;
1507
1508 do
1509 {
1510 base = base_stk [stk_idx];
1511 end = end_stk [stk_idx];
1512 bit = bit_stk [stk_idx];
1513
1514 for (;;)
1515 {
1516 unsigned char O = *bit >> 3;
1517 unsigned char M = 1 << (*bit & 7);
1518
1519 eio_dirent *a = base;
1520 eio_dirent *b = end;
1521
1522 if (b - a < EIO_SORT_CUTOFF)
1523 break;
1524
1525 /* now bit-partition the array on the bit */
1526 /* this ugly asymmetric loop seems to perform much better than typical */
1527 /* partition algos found in the literature */
1528 do
1529 if (!(((unsigned char *)a)[O] & M))
1530 ++a;
1531 else if (!(((unsigned char *)--b)[O] & M))
1532 {
1533 eio_dirent tmp = *a; *a = *b; *b = tmp;
1534 ++a;
1535 }
1536 while (b > a);
1537
1538 /* next bit, or stop, if no bits left in this path */
1539 if (!*--bit)
1540 break;
1541
1542 base_stk [stk_idx] = a;
1543 end_stk [stk_idx] = end;
1544 bit_stk [stk_idx] = bit;
1545 ++stk_idx;
1546
1547 end = a;
1548 }
1549 }
1550 while (stk_idx--);
1551 }
1552}
1553
1554static void
1555eio_dent_insertion_sort (eio_dirent *dents, int size)
1556{
1557 /* first move the smallest element to the front, to act as a sentinel */
1558 {
1559 int i;
1560 eio_dirent *min = dents;
1561
1562 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1563 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1564 if (EIO_DENT_CMP (dents [i], <, *min))
1565 min = &dents [i];
1566
1567 /* swap elements 0 and j (minimum) */
1568 {
1569 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1570 }
1571 }
1572
1573 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1574 {
1575 eio_dirent *i, *j;
1576
1577 for (i = dents + 1; i < dents + size; ++i)
1578 {
1579 eio_dirent value = *i;
1580
1581 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1582 j [1] = j [0];
1583
1584 j [1] = value;
1585 }
1586 }
1587}
1588
1589static void
1590eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1591{
1592 if (size <= 1)
1593 return; /* our insertion sort relies on size > 0 */
1594
1595 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1596 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1597 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1598
1599 /* use an insertion sort at the end, or for small arrays, */
1600 /* as insertion sort is more efficient for small partitions */
1601 eio_dent_insertion_sort (dents, size);
960} 1602}
961 1603
962/* read a full directory */ 1604/* read a full directory */
963static void 1605static void
964eio__scandir (eio_req *req, etp_worker *self) 1606eio__scandir (eio_req *req, etp_worker *self)
965{ 1607{
1608 char *name, *names;
1609 int namesalloc = 4096 - sizeof (void *) * 4;
1610 int namesoffs = 0;
1611 int flags = req->int1;
1612 eio_dirent *dents = 0;
1613 int dentalloc = 128;
1614 int dentoffs = 0;
1615 eio_ino_t inode_bits = 0;
1616#ifdef _WIN32
1617 HANDLE dirp;
1618 WIN32_FIND_DATA entp;
1619#else
966 DIR *dirp; 1620 DIR *dirp;
967 EIO_STRUCT_DIRENT *entp; 1621 EIO_STRUCT_DIRENT *entp;
968 char *name, *names; 1622#endif
969 int memlen = 4096;
970 int memofs = 0;
971 int res = 0;
972 1623
973 X_LOCK (wrklock); 1624 req->result = -1;
974 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 1625
1626 if (!(flags & EIO_READDIR_DENTS))
1627 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1628
1629#ifdef _WIN32
1630 {
1631 int len = strlen ((const char *)req->ptr1);
1632 char *path = malloc (MAX_PATH);
1633 const char *fmt;
1634
1635 if (!len)
1636 fmt = "./*";
1637 else if (((const char *)req->ptr1)[len - 1] == '/' || ((const char *)req->ptr1)[len - 1] == '\\')
1638 fmt = "%s*";
1639 else
1640 fmt = "%s/*";
1641
1642 _snprintf (path, MAX_PATH, fmt, (const char *)req->ptr1);
1643 dirp = FindFirstFile (path, &entp);
1644 free (path);
1645
1646 if (dirp == INVALID_HANDLE_VALUE)
1647 {
1648 dirp = 0;
1649
1650 switch (GetLastError ())
1651 {
1652 case ERROR_FILE_NOT_FOUND:
1653 req->result = 0;
1654 break;
1655
1656 case ERROR_INVALID_NAME:
1657 case ERROR_PATH_NOT_FOUND:
1658 case ERROR_NO_MORE_FILES:
1659 errno = ENOENT;
1660 break;
1661
1662 case ERROR_NOT_ENOUGH_MEMORY:
1663 errno = ENOMEM;
1664 break;
1665
1666 default:
1667 errno = EINVAL;
1668 break;
1669 }
1670 }
1671 }
1672#else
975 self->dirp = dirp = opendir (req->ptr1); 1673 dirp = opendir (req->ptr1);
1674#endif
1675
976 req->flags |= EIO_FLAG_PTR2_FREE; 1676 if (req->flags & EIO_FLAG_PTR1_FREE)
1677 free (req->ptr1);
1678
1679 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1680 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
977 req->ptr2 = names = malloc (memlen); 1681 req->ptr2 = names = malloc (namesalloc);
978 X_UNLOCK (wrklock);
979 1682
980 if (dirp && names) 1683 if (dirp && names && (!flags || dents))
981 for (;;) 1684 for (;;)
982 { 1685 {
1686 int done;
1687
1688#ifdef _WIN32
1689 done = !dirp;
1690#else
983 errno = 0; 1691 errno = 0;
984 entp = readdir (dirp); 1692 entp = readdir (dirp);
1693 done = !entp;
1694#endif
985 1695
986 if (!entp) 1696 if (done)
1697 {
1698#ifndef _WIN32
1699 int old_errno = errno;
1700 closedir (dirp);
1701 errno = old_errno;
1702
1703 if (errno)
1704 break;
1705#endif
1706
1707 /* sort etc. */
1708 req->int1 = flags;
1709 req->result = dentoffs;
1710
1711 if (flags & EIO_READDIR_STAT_ORDER)
1712 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1713 else if (flags & EIO_READDIR_DIRS_FIRST)
1714 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1715 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1716 else
1717 {
1718 /* in this case, all is known, and we just put dirs first and sort them */
1719 eio_dirent *oth = dents + dentoffs;
1720 eio_dirent *dir = dents;
1721
1722 /* now partition dirs to the front, and non-dirs to the back */
1723 /* by walking from both sides and swapping if necessary */
1724 while (oth > dir)
1725 {
1726 if (dir->type == EIO_DT_DIR)
1727 ++dir;
1728 else if ((--oth)->type == EIO_DT_DIR)
1729 {
1730 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1731
1732 ++dir;
1733 }
1734 }
1735
1736 /* now sort the dirs only (dirs all have the same score) */
1737 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1738 }
1739
987 break; 1740 break;
1741 }
988 1742
989 name = entp->d_name; 1743 /* now add the entry to our list(s) */
1744 name = D_NAME (entp);
990 1745
1746 /* skip . and .. entries */
991 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1747 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
992 { 1748 {
993 int len = strlen (name) + 1; 1749 int len = D_NAMLEN (entp) + 1;
994 1750
995 res++; 1751 while (ecb_expect_false (namesoffs + len > namesalloc))
996
997 while (memofs + len > memlen)
998 { 1752 {
999 memlen *= 2; 1753 namesalloc *= 2;
1000 X_LOCK (wrklock);
1001 req->ptr2 = names = realloc (names, memlen); 1754 req->ptr2 = names = realloc (names, namesalloc);
1002 X_UNLOCK (wrklock);
1003 1755
1004 if (!names) 1756 if (!names)
1005 break; 1757 break;
1006 } 1758 }
1007 1759
1008 memcpy (names + memofs, name, len); 1760 memcpy (names + namesoffs, name, len);
1761
1762 if (dents)
1763 {
1764 struct eio_dirent *ent;
1765
1766 if (ecb_expect_false (dentoffs == dentalloc))
1767 {
1768 dentalloc *= 2;
1769 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1770
1771 if (!dents)
1772 break;
1773 }
1774
1775 ent = dents + dentoffs;
1776
1777 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1778 ent->namelen = len - 1;
1779 ent->inode = D_INO (entp);
1780
1781 inode_bits |= ent->inode;
1782
1783 switch (D_TYPE (entp))
1784 {
1785 default:
1786 ent->type = EIO_DT_UNKNOWN;
1787 flags |= EIO_READDIR_FOUND_UNKNOWN;
1788 break;
1789
1790 #ifdef DT_FIFO
1791 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1792 #endif
1793 #ifdef DT_CHR
1794 case DT_CHR: ent->type = EIO_DT_CHR; break;
1795 #endif
1796 #ifdef DT_MPC
1797 case DT_MPC: ent->type = EIO_DT_MPC; break;
1798 #endif
1799 #ifdef DT_DIR
1800 case DT_DIR: ent->type = EIO_DT_DIR; break;
1801 #endif
1802 #ifdef DT_NAM
1803 case DT_NAM: ent->type = EIO_DT_NAM; break;
1804 #endif
1805 #ifdef DT_BLK
1806 case DT_BLK: ent->type = EIO_DT_BLK; break;
1807 #endif
1808 #ifdef DT_MPB
1809 case DT_MPB: ent->type = EIO_DT_MPB; break;
1810 #endif
1811 #ifdef DT_REG
1812 case DT_REG: ent->type = EIO_DT_REG; break;
1813 #endif
1814 #ifdef DT_NWK
1815 case DT_NWK: ent->type = EIO_DT_NWK; break;
1816 #endif
1817 #ifdef DT_CMP
1818 case DT_CMP: ent->type = EIO_DT_CMP; break;
1819 #endif
1820 #ifdef DT_LNK
1821 case DT_LNK: ent->type = EIO_DT_LNK; break;
1822 #endif
1823 #ifdef DT_SOCK
1824 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1825 #endif
1826 #ifdef DT_DOOR
1827 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1828 #endif
1829 #ifdef DT_WHT
1830 case DT_WHT: ent->type = EIO_DT_WHT; break;
1831 #endif
1832 }
1833
1834 ent->score = 7;
1835
1836 if (flags & EIO_READDIR_DIRS_FIRST)
1837 {
1838 if (ent->type == EIO_DT_UNKNOWN)
1839 {
1840 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1841 ent->score = 1;
1842 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1843 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1844 }
1845 else if (ent->type == EIO_DT_DIR)
1846 ent->score = 0;
1847 }
1848 }
1849
1009 memofs += len; 1850 namesoffs += len;
1851 ++dentoffs;
1010 } 1852 }
1853
1854 if (EIO_CANCELLED (req))
1855 {
1856 errno = ECANCELED;
1857 break;
1858 }
1859
1860#ifdef _WIN32
1861 if (!FindNextFile (dirp, &entp))
1862 {
1863 FindClose (dirp);
1864 dirp = 0;
1865 }
1866#endif
1011 } 1867 }
1012
1013 if (errno)
1014 res = -1;
1015
1016 req->result = res;
1017}
1018
1019#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1020# undef msync
1021# define msync(a,b,c) ENOSYS
1022#endif
1023
1024int
1025eio__mtouch (void *mem, size_t len, int flags)
1026{
1027 intptr_t addr = (intptr_t)mem;
1028 intptr_t end = addr + len;
1029#ifdef PAGESIZE
1030 const intptr_t page = PAGESIZE;
1031#else
1032 static intptr_t page;
1033
1034 if (!page)
1035 page = sysconf (_SC_PAGESIZE);
1036#endif
1037
1038 addr &= ~(page - 1); /* assume page size is always a power of two */
1039
1040 if (addr < end)
1041 if (flags) /* modify */
1042 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
1043 else
1044 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
1045
1046 return 0;
1047} 1868}
1048 1869
1049/*****************************************************************************/ 1870/*****************************************************************************/
1050 1871
1051#define ALLOC(len) \ 1872#define ALLOC(len) \
1067{ 1888{
1068 ETP_REQ *req; 1889 ETP_REQ *req;
1069 struct timespec ts; 1890 struct timespec ts;
1070 etp_worker *self = (etp_worker *)thr_arg; 1891 etp_worker *self = (etp_worker *)thr_arg;
1071 1892
1072 /* try to distribute timeouts somewhat randomly */ 1893 /* try to distribute timeouts somewhat evenly */
1073 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 1894 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1074 1895
1075 for (;;) 1896 for (;;)
1076 { 1897 {
1898 ts.tv_sec = 0;
1899
1077 X_LOCK (reqlock); 1900 X_LOCK (reqlock);
1078 1901
1079 for (;;) 1902 for (;;)
1080 { 1903 {
1081 self->req = req = reqq_shift (&req_queue); 1904 self->req = req = reqq_shift (&req_queue);
1082 1905
1083 if (req) 1906 if (req)
1084 break; 1907 break;
1085 1908
1909 if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
1910 {
1911 X_UNLOCK (reqlock);
1912 X_LOCK (wrklock);
1913 --started;
1914 X_UNLOCK (wrklock);
1915 goto quit;
1916 }
1917
1086 ++idle; 1918 ++idle;
1087 1919
1088 ts.tv_sec = time (0) + IDLE_TIMEOUT; 1920 if (idle <= max_idle)
1089 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 1921 /* we are allowed to idle, so do so without any timeout */
1922 X_COND_WAIT (reqwait, reqlock);
1923 else
1090 { 1924 {
1091 if (idle > max_idle) 1925 /* initialise timeout once */
1092 { 1926 if (!ts.tv_sec)
1093 --idle; 1927 ts.tv_sec = time (0) + idle_timeout;
1094 X_UNLOCK (reqlock);
1095 X_LOCK (wrklock);
1096 --started;
1097 X_UNLOCK (wrklock);
1098 goto quit;
1099 }
1100 1928
1101 /* we are allowed to idle, so do so without any timeout */
1102 X_COND_WAIT (reqwait, reqlock); 1929 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
1930 ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1103 } 1931 }
1104 1932
1105 --idle; 1933 --idle;
1106 } 1934 }
1107 1935
1110 X_UNLOCK (reqlock); 1938 X_UNLOCK (reqlock);
1111 1939
1112 if (req->type < 0) 1940 if (req->type < 0)
1113 goto quit; 1941 goto quit;
1114 1942
1115 if (!EIO_CANCELLED (req))
1116 ETP_EXECUTE (self, req); 1943 ETP_EXECUTE (self, req);
1117 1944
1118 X_LOCK (reslock); 1945 X_LOCK (reslock);
1119 1946
1120 ++npending; 1947 ++npending;
1121 1948
1136 return 0; 1963 return 0;
1137} 1964}
1138 1965
1139/*****************************************************************************/ 1966/*****************************************************************************/
1140 1967
1968int ecb_cold
1141int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1969eio_init (void (*want_poll)(void), void (*done_poll)(void))
1142{ 1970{
1971#if !HAVE_PREADWRITE
1972 X_MUTEX_CREATE (preadwritelock);
1973#endif
1974
1143 return etp_init (want_poll, done_poll); 1975 return etp_init (want_poll, done_poll);
1144} 1976}
1145 1977
1978ecb_inline void
1146static void eio_api_destroy (eio_req *req) 1979eio_api_destroy (eio_req *req)
1147{ 1980{
1148 free (req); 1981 free (req);
1149} 1982}
1150 1983
1151#define REQ(rtype) \ 1984#define REQ(rtype) \
1170 { \ 2003 { \
1171 eio_api_destroy (req); \ 2004 eio_api_destroy (req); \
1172 return 0; \ 2005 return 0; \
1173 } 2006 }
1174 2007
2008static void
1175static void eio_execute (etp_worker *self, eio_req *req) 2009eio_execute (etp_worker *self, eio_req *req)
1176{ 2010{
1177 errno = 0; 2011 if (ecb_expect_false (EIO_CANCELLED (req)))
2012 {
2013 req->result = -1;
2014 req->errorno = ECANCELED;
2015 return;
2016 }
1178 2017
1179 switch (req->type) 2018 switch (req->type)
1180 { 2019 {
1181 case EIO_READ: ALLOC (req->size); 2020 case EIO_READ: ALLOC (req->size);
1182 req->result = req->offs >= 0 2021 req->result = req->offs >= 0
1185 case EIO_WRITE: req->result = req->offs >= 0 2024 case EIO_WRITE: req->result = req->offs >= 0
1186 ? pwrite (req->int1, req->ptr2, req->size, req->offs) 2025 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1187 : write (req->int1, req->ptr2, req->size); break; 2026 : write (req->int1, req->ptr2, req->size); break;
1188 2027
1189 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break; 2028 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1190 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break; 2029 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
1191 2030
1192 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2031 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1193 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2032 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1194 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2033 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1195 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2034 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1196 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2035 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1197 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; 2036 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
2037
2038 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2039 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2040 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2041 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1198 2042
1199 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; 2043 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1200 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; 2044 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1201 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break; 2045 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1202 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break; 2046 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1210 case EIO_RMDIR: req->result = rmdir (req->ptr1); break; 2054 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1211 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; 2055 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1212 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; 2056 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1213 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; 2057 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1214 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; 2058 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1215 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break; 2059 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1216 2060
2061 case EIO_REALPATH: eio__realpath (req, self); break;
2062
1217 case EIO_READLINK: ALLOC (NAME_MAX); 2063 case EIO_READLINK: ALLOC (PATH_MAX);
1218 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; 2064 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1219 2065
1220 case EIO_SYNC: req->result = 0; sync (); break; 2066 case EIO_SYNC: req->result = 0; sync (); break;
1221 case EIO_FSYNC: req->result = fsync (req->int1); break; 2067 case EIO_FSYNC: req->result = fsync (req->int1); break;
1222 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 2068 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1223 case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break; 2069 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
2070 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1224 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break; 2071 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
2072 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1225 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break; 2073 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
2074 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
1226 2075
1227 case EIO_READDIR: eio__scandir (req, self); break; 2076 case EIO_READDIR: eio__scandir (req, self); break;
1228 2077
1229 case EIO_BUSY: 2078 case EIO_BUSY:
1230#ifdef _WIN32 2079#ifdef _WIN32
1231 Sleep (req->nv1 * 1000.); 2080 Sleep (req->nv1 * 1e3);
1232#else 2081#else
1233 { 2082 {
1234 struct timeval tv; 2083 struct timeval tv;
1235 2084
1236 tv.tv_sec = req->nv1; 2085 tv.tv_sec = req->nv1;
1237 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; 2086 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1238 2087
1239 req->result = select (0, 0, 0, 0, &tv); 2088 req->result = select (0, 0, 0, 0, &tv);
1240 } 2089 }
1241#endif 2090#endif
1242 break; 2091 break;
1257 times = tv; 2106 times = tv;
1258 } 2107 }
1259 else 2108 else
1260 times = 0; 2109 times = 0;
1261 2110
1262
1263 req->result = req->type == EIO_FUTIME 2111 req->result = req->type == EIO_FUTIME
1264 ? futimes (req->int1, times) 2112 ? futimes (req->int1, times)
1265 : utimes (req->ptr1, times); 2113 : utimes (req->ptr1, times);
1266 } 2114 }
1267 break; 2115 break;
1272 case EIO_NOP: 2120 case EIO_NOP:
1273 req->result = 0; 2121 req->result = 0;
1274 break; 2122 break;
1275 2123
1276 case EIO_CUSTOM: 2124 case EIO_CUSTOM:
1277 ((void (*)(eio_req *))req->feed) (req); 2125 req->feed (req);
1278 break; 2126 break;
1279 2127
1280 default: 2128 default:
2129 errno = ENOSYS;
1281 req->result = -1; 2130 req->result = -1;
1282 break; 2131 break;
1283 } 2132 }
1284 2133
1285 req->errorno = errno; 2134 req->errorno = errno;
1315eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2164eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1316{ 2165{
1317 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2166 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1318} 2167}
1319 2168
2169eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
2170{
2171 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
2172}
2173
2174eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
2175{
2176 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2177}
2178
1320eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) 2179eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
1321{ 2180{
1322 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; 2181 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
1323} 2182}
1324 2183
2184eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2185{
2186 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
2187}
2188
1325eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) 2189eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1326{ 2190{
1327 REQ (EIO_FDATASYNC); req->int1 = fd; SEND; 2191 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1328} 2192}
1329 2193
1350eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) 2214eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1351{ 2215{
1352 REQ (EIO_FSTAT); req->int1 = fd; SEND; 2216 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1353} 2217}
1354 2218
2219eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
2220{
2221 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
2222}
2223
1355eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) 2224eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1356{ 2225{
1357 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; 2226 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1358} 2227}
1359 2228
1365eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) 2234eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1366{ 2235{
1367 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2236 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1368} 2237}
1369 2238
1370eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2239eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1371{ 2240{
1372 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2241 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1373} 2242}
1374 2243
1375eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) 2244eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
1395eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data) 2264eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1396{ 2265{
1397 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2266 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1398} 2267}
1399 2268
1400eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2269eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1401{ 2270{
1402 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2271 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1403} 2272}
1404 2273
1405eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) 2274eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
1421eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data) 2290eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1422{ 2291{
1423 return eio__1path (EIO_READLINK, path, pri, cb, data); 2292 return eio__1path (EIO_READLINK, path, pri, cb, data);
1424} 2293}
1425 2294
2295eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2296{
2297 return eio__1path (EIO_REALPATH, path, pri, cb, data);
2298}
2299
1426eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) 2300eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1427{ 2301{
1428 return eio__1path (EIO_STAT, path, pri, cb, data); 2302 return eio__1path (EIO_STAT, path, pri, cb, data);
1429} 2303}
1430 2304
1431eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) 2305eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1432{ 2306{
1433 return eio__1path (EIO_LSTAT, path, pri, cb, data); 2307 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1434} 2308}
1435 2309
2310eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
2311{
2312 return eio__1path (EIO_STATVFS, path, pri, cb, data);
2313}
2314
1436eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) 2315eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1437{ 2316{
1438 return eio__1path (EIO_UNLINK, path, pri, cb, data); 2317 return eio__1path (EIO_UNLINK, path, pri, cb, data);
1439} 2318}
1440 2319
1441eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data) 2320eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1442{ 2321{
1443 return eio__1path (EIO_RMDIR, path, pri, cb, data); 2322 return eio__1path (EIO_RMDIR, path, pri, cb, data);
1444} 2323}
1445 2324
1446eio_req *eio_readdir (const char *path, int pri, eio_cb cb, void *data) 2325eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1447{ 2326{
1448 return eio__1path (EIO_READDIR, path, pri, cb, data); 2327 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1449} 2328}
1450 2329
1451eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) 2330eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1452{ 2331{
1453 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; 2332 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1454} 2333}
1455 2334
1456static eio_req * 2335static eio_req *
1457eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2336eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1458{ 2337{
1482eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2361eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1483{ 2362{
1484 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2363 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1485} 2364}
1486 2365
1487eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) 2366eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
1488{ 2367{
1489 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND; 2368 REQ (EIO_CUSTOM); req->feed = execute; SEND;
1490} 2369}
1491 2370
1492#endif 2371#endif
1493 2372
1494eio_req *eio_grp (eio_cb cb, void *data) 2373eio_req *eio_grp (eio_cb cb, void *data)
1503#undef SEND 2382#undef SEND
1504 2383
1505/*****************************************************************************/ 2384/*****************************************************************************/
1506/* grp functions */ 2385/* grp functions */
1507 2386
2387void
1508void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2388eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1509{ 2389{
1510 grp->int2 = limit; 2390 grp->int2 = limit;
1511 grp->feed = feed; 2391 grp->feed = feed;
1512 2392
1513 grp_try_feed (grp); 2393 grp_try_feed (grp);
1514} 2394}
1515 2395
2396void
1516void eio_grp_limit (eio_req *grp, int limit) 2397eio_grp_limit (eio_req *grp, int limit)
1517{ 2398{
1518 grp->int2 = limit; 2399 grp->int2 = limit;
1519 2400
1520 grp_try_feed (grp); 2401 grp_try_feed (grp);
1521} 2402}
1522 2403
2404void
1523void eio_grp_add (eio_req *grp, eio_req *req) 2405eio_grp_add (eio_req *grp, eio_req *req)
1524{ 2406{
1525 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2407 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1526 2408
1527 grp->flags |= EIO_FLAG_GROUPADD; 2409 grp->flags |= EIO_FLAG_GROUPADD;
1528 2410
1539} 2421}
1540 2422
1541/*****************************************************************************/ 2423/*****************************************************************************/
1542/* misc garbage */ 2424/* misc garbage */
1543 2425
2426eio_ssize_t
1544ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2427eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1545{ 2428{
1546 etp_worker wrk;
1547
1548 wrk.dbuf = 0;
1549
1550 eio__sendfile (ofd, ifd, offset, count, &wrk); 2429 return eio__sendfile (ofd, ifd, offset, count);
1551
1552 if (wrk.dbuf)
1553 free (wrk.dbuf);
1554} 2430}
1555 2431

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines