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 1.19 by root, Tue Jun 17 23:34:20 2008 UTC vs.
Revision 1.91 by root, Sun Jul 17 04:20:04 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
76static void eio_destroy (eio_req *req);
77
54#ifndef EIO_FINISH 78#ifndef EIO_FINISH
55# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0 79# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
56#endif 80#endif
57 81
58#ifndef EIO_DESTROY 82#ifndef EIO_DESTROY
61 85
62#ifndef EIO_FEED 86#ifndef EIO_FEED
63# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0) 87# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)
64#endif 88#endif
65 89
90#ifndef EIO_FD_TO_WIN32_HANDLE
91# define EIO_FD_TO_WIN32_HANDLE(fd) _get_osfhandle (fd)
92#endif
93#ifndef EIO_WIN32_HANDLE_TO_FD
94# define EIO_WIN32_HANDLE_TO_FD(handle) _open_osfhandle (handle, 0)
95#endif
96
97#define EIO_ERRNO(errval,retval) ((errno = errval), retval)
98
99#define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1)
100
66#ifdef _WIN32 101#ifdef _WIN32
67 102
68 /*doh*/ 103 #undef PAGESIZE
104 #define PAGESIZE 4096 /* GetSystemInfo? */
105
106 #ifdef EIO_STRUCT_STATI64
107 #define stat(path,buf) _stati64 (path,buf)
108 #define fstat(fd,buf) _fstati64 (path,buf)
109 #endif
110 #define lstat(path,buf) stat (path,buf)
111 #define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
112 #define mkdir(path,mode) _mkdir (path)
113 #define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
114
115 #define chmod(path,mode) _chmod (path, mode)
116 #define dup(fd) _dup (fd)
117 #define dup2(fd1,fd2) _dup2 (fd1, fd2)
118
119 #define fchmod(fd,mode) EIO_ENOSYS ()
120 #define chown(path,uid,gid) EIO_ENOSYS ()
121 #define fchown(fd,uid,gid) EIO_ENOSYS ()
122 #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
123 #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
124 #define mknod(path,mode,dev) EIO_ENOSYS ()
125 #define sync() EIO_ENOSYS ()
126 #define readlink(path,buf,s) EIO_ENOSYS ()
127 #define statvfs(path,buf) EIO_ENOSYS ()
128 #define fstatvfs(fd,buf) EIO_ENOSYS ()
129
130 /* we could even stat and see if it exists */
131 static int
132 symlink (const char *old, const char *neu)
133 {
134 #if WINVER >= 0x0600
135 if (CreateSymbolicLink (neu, old, 1))
136 return 0;
137
138 if (CreateSymbolicLink (neu, old, 0))
139 return 0;
140 #endif
141
142 return EIO_ERRNO (ENOENT, -1);
143 }
144
145 /* POSIX API only */
146 #define CreateHardLink(neu,old,flags) 0
147 #define CreateSymbolicLink(neu,old,flags) 0
148
149 struct statvfs
150 {
151 int dummy;
152 };
153
154 #define DT_DIR EIO_DT_DIR
155 #define DT_REG EIO_DT_REG
156 #define D_NAME(entp) entp.cFileName
157 #define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
69 158
70#else 159#else
71 160
72# include "config.h"
73# include <sys/time.h> 161 #include <sys/time.h>
74# include <sys/select.h> 162 #include <sys/select.h>
163 #include <sys/statvfs.h>
75# include <unistd.h> 164 #include <unistd.h>
76# include <utime.h> 165 #include <utime.h>
77# include <signal.h> 166 #include <signal.h>
78# include <dirent.h> 167 #include <dirent.h>
79 168
169 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
170 #include <sys/mman.h>
171 #endif
172
173 #define D_NAME(entp) entp->d_name
174
175 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
176 #if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
177 #define _DIRENT_HAVE_D_TYPE /* sigh */
178 #define D_INO(de) (de)->d_fileno
179 #define D_NAMLEN(de) (de)->d_namlen
180 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
181 #define D_INO(de) (de)->d_ino
182 #endif
183
184 #ifdef _D_EXACT_NAMLEN
185 #undef D_NAMLEN
186 #define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
187 #endif
188
189 #ifdef _DIRENT_HAVE_D_TYPE
190 #define D_TYPE(de) (de)->d_type
191 #endif
192
80# ifndef EIO_STRUCT_DIRENT 193 #ifndef EIO_STRUCT_DIRENT
81# define EIO_STRUCT_DIRENT struct dirent 194 #define EIO_STRUCT_DIRENT struct dirent
82# endif 195 #endif
83 196
84#endif 197#endif
85 198
86#if HAVE_SENDFILE 199#if HAVE_SENDFILE
87# if __linux 200# if __linux
88# include <sys/sendfile.h> 201# include <sys/sendfile.h>
89# elif __freebsd 202# elif __FreeBSD__ || defined __APPLE__
90# include <sys/socket.h> 203# include <sys/socket.h>
91# include <sys/uio.h> 204# include <sys/uio.h>
92# elif __hpux 205# elif __hpux
93# include <sys/socket.h> 206# include <sys/socket.h>
94# elif __solaris /* not yet */ 207# elif __solaris
95# include <sys/sendfile.h> 208# include <sys/sendfile.h>
96# else 209# else
97# error sendfile support requested but not available 210# error sendfile support requested but not available
98# endif 211# endif
99#endif 212#endif
100 213
101/* number of seconds after which an idle threads exit */ 214#ifndef D_TYPE
102#define IDLE_TIMEOUT 10 215# define D_TYPE(de) 0
216#endif
217#ifndef D_INO
218# define D_INO(de) 0
219#endif
220#ifndef D_NAMLEN
221# define D_NAMLEN(entp) strlen (D_NAME (entp))
222#endif
103 223
104/* used for struct dirent, AIX doesn't provide it */ 224/* used for struct dirent, AIX doesn't provide it */
105#ifndef NAME_MAX 225#ifndef NAME_MAX
106# define NAME_MAX 4096 226# define NAME_MAX 4096
227#endif
228
229/* used for readlink etc. */
230#ifndef PATH_MAX
231# define PATH_MAX 4096
107#endif 232#endif
108 233
109/* buffer size for various temporary buffers */ 234/* buffer size for various temporary buffers */
110#define EIO_BUFSIZE 65536 235#define EIO_BUFSIZE 65536
111 236
118 if (!eio_buf) \ 243 if (!eio_buf) \
119 return -1; 244 return -1;
120 245
121#define EIO_TICKS ((1000000 + 1023) >> 10) 246#define EIO_TICKS ((1000000 + 1023) >> 10)
122 247
123/*****************************************************************************/
124
125#define ETP_PRI_MIN EIO_PRI_MIN 248#define ETP_PRI_MIN EIO_PRI_MIN
126#define ETP_PRI_MAX EIO_PRI_MAX 249#define ETP_PRI_MAX EIO_PRI_MAX
127 250
128struct etp_worker; 251struct etp_worker;
129 252
137#define ETP_WORKER_CLEAR(req) \ 260#define ETP_WORKER_CLEAR(req) \
138 if (wrk->dbuf) \ 261 if (wrk->dbuf) \
139 { \ 262 { \
140 free (wrk->dbuf); \ 263 free (wrk->dbuf); \
141 wrk->dbuf = 0; \ 264 wrk->dbuf = 0; \
142 } \
143 \
144 if (wrk->dirp) \
145 { \
146 closedir (wrk->dirp); \
147 wrk->dirp = 0; \
148 } 265 }
266
149#define ETP_WORKER_COMMON \ 267#define ETP_WORKER_COMMON \
150 void *dbuf; \ 268 void *dbuf;
151 DIR *dirp;
152 269
153/*****************************************************************************/ 270/*****************************************************************************/
154 271
155#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) 272#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
156 273
157/* calculcate time difference in ~1/EIO_TICKS of a second */ 274/* calculate time difference in ~1/EIO_TICKS of a second */
275ecb_inline int
158static int tvdiff (struct timeval *tv1, struct timeval *tv2) 276tvdiff (struct timeval *tv1, struct timeval *tv2)
159{ 277{
160 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 278 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
161 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 279 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
162} 280}
163 281
170static unsigned int max_poll_reqs; /* reslock */ 288static unsigned int max_poll_reqs; /* reslock */
171 289
172static volatile unsigned int nreqs; /* reqlock */ 290static volatile unsigned int nreqs; /* reqlock */
173static volatile unsigned int nready; /* reqlock */ 291static volatile unsigned int nready; /* reqlock */
174static volatile unsigned int npending; /* reqlock */ 292static volatile unsigned int npending; /* reqlock */
175static volatile unsigned int max_idle = 4; 293static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
294static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
176 295
296static xmutex_t wrklock;
297static xmutex_t reslock;
298static xmutex_t reqlock;
299static xcond_t reqwait;
300
301#if !HAVE_PREADWRITE
302/*
303 * make our pread/pwrite emulation safe against themselves, but not against
304 * normal read/write by using a mutex. slows down execution a lot,
305 * but that's your problem, not mine.
306 */
177static mutex_t wrklock = X_MUTEX_INIT; 307static xmutex_t preadwritelock = X_MUTEX_INIT;
178static mutex_t reslock = X_MUTEX_INIT; 308#endif
179static mutex_t reqlock = X_MUTEX_INIT;
180static cond_t reqwait = X_COND_INIT;
181 309
182typedef struct etp_worker 310typedef struct etp_worker
183{ 311{
184 /* locked by wrklock */ 312 /* locked by wrklock */
185 struct etp_worker *prev, *next; 313 struct etp_worker *prev, *next;
186 314
187 thread_t tid; 315 xthread_t tid;
188 316
189 /* locked by reslock, reqlock or wrklock */ 317 /* locked by reslock, reqlock or wrklock */
190 ETP_REQ *req; /* currently processed request */ 318 ETP_REQ *req; /* currently processed request */
191 319
192 ETP_WORKER_COMMON 320 ETP_WORKER_COMMON
197#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock) 325#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
198#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock) 326#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
199 327
200/* worker threads management */ 328/* worker threads management */
201 329
330static void ecb_cold
202static void etp_worker_clear (etp_worker *wrk) 331etp_worker_clear (etp_worker *wrk)
203{ 332{
204 ETP_WORKER_CLEAR (wrk); 333 ETP_WORKER_CLEAR (wrk);
205} 334}
206 335
336static void ecb_cold
207static void etp_worker_free (etp_worker *wrk) 337etp_worker_free (etp_worker *wrk)
208{ 338{
209 wrk->next->prev = wrk->prev; 339 wrk->next->prev = wrk->prev;
210 wrk->prev->next = wrk->next; 340 wrk->prev->next = wrk->next;
211 341
212 free (wrk); 342 free (wrk);
213} 343}
214 344
215static unsigned int etp_nreqs (void) 345static unsigned int
346etp_nreqs (void)
216{ 347{
217 int retval; 348 int retval;
218 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 349 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
219 retval = nreqs; 350 retval = nreqs;
220 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 351 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
221 return retval; 352 return retval;
222} 353}
223 354
224static unsigned int etp_nready (void) 355static unsigned int
356etp_nready (void)
225{ 357{
226 unsigned int retval; 358 unsigned int retval;
227 359
228 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 360 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
229 retval = nready; 361 retval = nready;
230 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 362 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
231 363
232 return retval; 364 return retval;
233} 365}
234 366
235static unsigned int etp_npending (void) 367static unsigned int
368etp_npending (void)
236{ 369{
237 unsigned int retval; 370 unsigned int retval;
238 371
239 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 372 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
240 retval = npending; 373 retval = npending;
241 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 374 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
242 375
243 return retval; 376 return retval;
244} 377}
245 378
246static unsigned int etp_nthreads (void) 379static unsigned int
380etp_nthreads (void)
247{ 381{
248 unsigned int retval; 382 unsigned int retval;
249 383
250 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 384 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
251 retval = started; 385 retval = started;
265} etp_reqq; 399} etp_reqq;
266 400
267static etp_reqq req_queue; 401static etp_reqq req_queue;
268static etp_reqq res_queue; 402static etp_reqq res_queue;
269 403
404static void ecb_noinline ecb_cold
405reqq_init (etp_reqq *q)
406{
407 int pri;
408
409 for (pri = 0; pri < ETP_NUM_PRI; ++pri)
410 q->qs[pri] = q->qe[pri] = 0;
411
412 q->size = 0;
413}
414
415static int ecb_noinline
270static int reqq_push (etp_reqq *q, ETP_REQ *req) 416reqq_push (etp_reqq *q, ETP_REQ *req)
271{ 417{
272 int pri = req->pri; 418 int pri = req->pri;
273 req->next = 0; 419 req->next = 0;
274 420
275 if (q->qe[pri]) 421 if (q->qe[pri])
281 q->qe[pri] = q->qs[pri] = req; 427 q->qe[pri] = q->qs[pri] = req;
282 428
283 return q->size++; 429 return q->size++;
284} 430}
285 431
432static ETP_REQ * ecb_noinline
286static ETP_REQ *reqq_shift (etp_reqq *q) 433reqq_shift (etp_reqq *q)
287{ 434{
288 int pri; 435 int pri;
289 436
290 if (!q->size) 437 if (!q->size)
291 return 0; 438 return 0;
306 } 453 }
307 454
308 abort (); 455 abort ();
309} 456}
310 457
311static void etp_atfork_prepare (void) 458static void ecb_cold
459etp_thread_init (void)
312{ 460{
313 X_LOCK (wrklock);
314 X_LOCK (reqlock);
315 X_LOCK (reslock);
316#if !HAVE_PREADWRITE 461#if !HAVE_PREADWRITE
317 X_LOCK (preadwritelock);
318#endif
319}
320
321static void etp_atfork_parent (void)
322{
323#if !HAVE_PREADWRITE
324 X_UNLOCK (preadwritelock); 462 X_MUTEX_CREATE (preadwritelock);
325#endif 463#endif
326 X_UNLOCK (reslock); 464 X_MUTEX_CREATE (wrklock);
327 X_UNLOCK (reqlock); 465 X_MUTEX_CREATE (reslock);
328 X_UNLOCK (wrklock); 466 X_MUTEX_CREATE (reqlock);
467 X_COND_CREATE (reqwait);
329} 468}
330 469
470static void ecb_cold
331static void etp_atfork_child (void) 471etp_atfork_child (void)
332{ 472{
333 ETP_REQ *prv; 473 reqq_init (&req_queue);
474 reqq_init (&res_queue);
334 475
335 while (prv = reqq_shift (&req_queue)) 476 wrk_first.next =
336 ETP_DESTROY (prv); 477 wrk_first.prev = &wrk_first;
337
338 while (prv = reqq_shift (&res_queue))
339 ETP_DESTROY (prv);
340
341 while (wrk_first.next != &wrk_first)
342 {
343 etp_worker *wrk = wrk_first.next;
344
345 if (wrk->req)
346 ETP_DESTROY (wrk->req);
347
348 etp_worker_clear (wrk);
349 etp_worker_free (wrk);
350 }
351 478
352 started = 0; 479 started = 0;
353 idle = 0; 480 idle = 0;
354 nreqs = 0; 481 nreqs = 0;
355 nready = 0; 482 nready = 0;
356 npending = 0; 483 npending = 0;
357 484
358 etp_atfork_parent (); 485 etp_thread_init ();
359} 486}
360 487
361static void 488static void ecb_cold
362etp_once_init (void) 489etp_once_init (void)
363{ 490{
364 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child); 491 etp_thread_init ();
492 X_THREAD_ATFORK (0, 0, etp_atfork_child);
365} 493}
366 494
367static int 495static int ecb_cold
368etp_init (void (*want_poll)(void), void (*done_poll)(void)) 496etp_init (void (*want_poll)(void), void (*done_poll)(void))
369{ 497{
370 static pthread_once_t doinit = PTHREAD_ONCE_INIT; 498 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
371 499
372 pthread_once (&doinit, etp_once_init); 500 pthread_once (&doinit, etp_once_init);
377 return 0; 505 return 0;
378} 506}
379 507
380X_THREAD_PROC (etp_proc); 508X_THREAD_PROC (etp_proc);
381 509
510static void ecb_cold
382static void etp_start_thread (void) 511etp_start_thread (void)
383{ 512{
384 etp_worker *wrk = calloc (1, sizeof (etp_worker)); 513 etp_worker *wrk = calloc (1, sizeof (etp_worker));
385 514
386 /*TODO*/ 515 /*TODO*/
387 assert (("unable to allocate worker thread data", wrk)); 516 assert (("unable to allocate worker thread data", wrk));
400 free (wrk); 529 free (wrk);
401 530
402 X_UNLOCK (wrklock); 531 X_UNLOCK (wrklock);
403} 532}
404 533
534static void
405static void etp_maybe_start_thread (void) 535etp_maybe_start_thread (void)
406{ 536{
407 if (etp_nthreads () >= wanted) 537 if (ecb_expect_true (etp_nthreads () >= wanted))
408 return; 538 return;
409 539
410 /* todo: maybe use idle here, but might be less exact */ 540 /* todo: maybe use idle here, but might be less exact */
411 if (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()) 541 if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
412 return; 542 return;
413 543
414 etp_start_thread (); 544 etp_start_thread ();
415} 545}
416 546
547static void ecb_cold
417static void etp_end_thread (void) 548etp_end_thread (void)
418{ 549{
419 eio_req *req = calloc (1, sizeof (eio_req)); 550 eio_req *req = calloc (1, sizeof (eio_req));
420 551
421 req->type = -1; 552 req->type = -1;
422 req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 553 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
429 X_LOCK (wrklock); 560 X_LOCK (wrklock);
430 --started; 561 --started;
431 X_UNLOCK (wrklock); 562 X_UNLOCK (wrklock);
432} 563}
433 564
434static int etp_poll (void) 565static int
566etp_poll (void)
435{ 567{
436 unsigned int maxreqs; 568 unsigned int maxreqs;
437 unsigned int maxtime; 569 unsigned int maxtime;
438 struct timeval tv_start, tv_now; 570 struct timeval tv_start, tv_now;
439 571
469 601
470 X_LOCK (reqlock); 602 X_LOCK (reqlock);
471 --nreqs; 603 --nreqs;
472 X_UNLOCK (reqlock); 604 X_UNLOCK (reqlock);
473 605
474 if (req->type == EIO_GROUP && req->size) 606 if (ecb_expect_false (req->type == EIO_GROUP && req->size))
475 { 607 {
476 req->int1 = 1; /* mark request as delayed */ 608 req->int1 = 1; /* mark request as delayed */
477 continue; 609 continue;
478 } 610 }
479 else 611 else
480 { 612 {
481 int res = ETP_FINISH (req); 613 int res = ETP_FINISH (req);
482 if (res) 614 if (ecb_expect_false (res))
483 return res; 615 return res;
484 } 616 }
485 617
486 if (maxreqs && !--maxreqs) 618 if (ecb_expect_false (maxreqs && !--maxreqs))
487 break; 619 break;
488 620
489 if (maxtime) 621 if (maxtime)
490 { 622 {
491 gettimeofday (&tv_now, 0); 623 gettimeofday (&tv_now, 0);
497 629
498 errno = EAGAIN; 630 errno = EAGAIN;
499 return -1; 631 return -1;
500} 632}
501 633
634static void
502static void etp_cancel (ETP_REQ *req) 635etp_cancel (ETP_REQ *req)
503{ 636{
504 X_LOCK (wrklock); 637 req->cancelled = 1;
505 req->flags |= EIO_FLAG_CANCELLED;
506 X_UNLOCK (wrklock);
507 638
508 eio_grp_cancel (req); 639 eio_grp_cancel (req);
509} 640}
510 641
642static void
511static void etp_submit (ETP_REQ *req) 643etp_submit (ETP_REQ *req)
512{ 644{
513 req->pri -= ETP_PRI_MIN; 645 req->pri -= ETP_PRI_MIN;
514 646
515 if (req->pri < ETP_PRI_MIN - ETP_PRI_MIN) req->pri = ETP_PRI_MIN - ETP_PRI_MIN; 647 if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
516 if (req->pri > ETP_PRI_MAX - ETP_PRI_MIN) req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 648 if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
517 649
650 if (ecb_expect_false (req->type == EIO_GROUP))
651 {
652 /* I hope this is worth it :/ */
518 X_LOCK (reqlock); 653 X_LOCK (reqlock);
519 ++nreqs; 654 ++nreqs;
655 X_UNLOCK (reqlock);
656
657 X_LOCK (reslock);
658
659 ++npending;
660
661 if (!reqq_push (&res_queue, req) && want_poll_cb)
662 want_poll_cb ();
663
664 X_UNLOCK (reslock);
665 }
666 else
667 {
668 X_LOCK (reqlock);
669 ++nreqs;
520 ++nready; 670 ++nready;
521 reqq_push (&req_queue, req); 671 reqq_push (&req_queue, req);
522 X_COND_SIGNAL (reqwait); 672 X_COND_SIGNAL (reqwait);
523 X_UNLOCK (reqlock); 673 X_UNLOCK (reqlock);
524 674
525 etp_maybe_start_thread (); 675 etp_maybe_start_thread ();
676 }
526} 677}
527 678
679static void ecb_cold
528static void etp_set_max_poll_time (double nseconds) 680etp_set_max_poll_time (double nseconds)
529{ 681{
530 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 682 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
531 max_poll_time = nseconds; 683 max_poll_time = nseconds * EIO_TICKS;
532 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 684 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
533} 685}
534 686
687static void ecb_cold
535static void etp_set_max_poll_reqs (unsigned int maxreqs) 688etp_set_max_poll_reqs (unsigned int maxreqs)
536{ 689{
537 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 690 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
538 max_poll_reqs = maxreqs; 691 max_poll_reqs = maxreqs;
539 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 692 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
540} 693}
541 694
695static void ecb_cold
542static void etp_set_max_idle (unsigned int nthreads) 696etp_set_max_idle (unsigned int nthreads)
543{ 697{
544 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 698 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
545 max_idle = nthreads <= 0 ? 1 : nthreads; 699 max_idle = nthreads;
546 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 700 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
547} 701}
548 702
703static void ecb_cold
704etp_set_idle_timeout (unsigned int seconds)
705{
706 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
707 idle_timeout = seconds;
708 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
709}
710
711static void ecb_cold
549static void etp_set_min_parallel (unsigned int nthreads) 712etp_set_min_parallel (unsigned int nthreads)
550{ 713{
551 if (wanted < nthreads) 714 if (wanted < nthreads)
552 wanted = nthreads; 715 wanted = nthreads;
553} 716}
554 717
718static void ecb_cold
555static void etp_set_max_parallel (unsigned int nthreads) 719etp_set_max_parallel (unsigned int nthreads)
556{ 720{
557 if (wanted > nthreads) 721 if (wanted > nthreads)
558 wanted = nthreads; 722 wanted = nthreads;
559 723
560 while (started > wanted) 724 while (started > wanted)
561 etp_end_thread (); 725 etp_end_thread ();
562} 726}
563 727
564/*****************************************************************************/ 728/*****************************************************************************/
565 729
730static void
566static void grp_try_feed (eio_req *grp) 731grp_try_feed (eio_req *grp)
567{ 732{
568 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 733 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
569 { 734 {
570 int old_len = grp->size; 735 grp->flags &= ~EIO_FLAG_GROUPADD;
571 736
572 EIO_FEED (grp); 737 EIO_FEED (grp);
573 738
574 /* stop if no progress has been made */ 739 /* stop if no progress has been made */
575 if (old_len == grp->size) 740 if (!(grp->flags & EIO_FLAG_GROUPADD))
576 { 741 {
577 grp->feed = 0; 742 grp->feed = 0;
578 break; 743 break;
579 } 744 }
580 } 745 }
581} 746}
582 747
748static int
583static int grp_dec (eio_req *grp) 749grp_dec (eio_req *grp)
584{ 750{
585 --grp->size; 751 --grp->size;
586 752
587 /* call feeder, if applicable */ 753 /* call feeder, if applicable */
588 grp_try_feed (grp); 754 grp_try_feed (grp);
592 return eio_finish (grp); 758 return eio_finish (grp);
593 else 759 else
594 return 0; 760 return 0;
595} 761}
596 762
763static void
597void eio_destroy (eio_req *req) 764eio_destroy (eio_req *req)
598{ 765{
599 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); 766 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
600 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); 767 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
601 768
602 EIO_DESTROY (req); 769 EIO_DESTROY (req);
603} 770}
604 771
772static int
605static int eio_finish (eio_req *req) 773eio_finish (eio_req *req)
606{ 774{
607 int res = EIO_FINISH (req); 775 int res = EIO_FINISH (req);
608 776
609 if (req->grp) 777 if (req->grp)
610 { 778 {
618 if (grp->grp_first == req) 786 if (grp->grp_first == req)
619 grp->grp_first = req->grp_next; 787 grp->grp_first = req->grp_next;
620 788
621 res2 = grp_dec (grp); 789 res2 = grp_dec (grp);
622 790
623 if (!res && res2) 791 if (!res)
624 res = res2; 792 res = res2;
625 } 793 }
626 794
627 eio_destroy (req); 795 eio_destroy (req);
628 796
629 return res; 797 return res;
630} 798}
631 799
800void
632void eio_grp_cancel (eio_req *grp) 801eio_grp_cancel (eio_req *grp)
633{ 802{
634 for (grp = grp->grp_first; grp; grp = grp->grp_next) 803 for (grp = grp->grp_first; grp; grp = grp->grp_next)
635 eio_cancel (grp); 804 eio_cancel (grp);
636} 805}
637 806
807void
638void eio_cancel (eio_req *req) 808eio_cancel (eio_req *req)
639{ 809{
640 etp_cancel (req); 810 etp_cancel (req);
641} 811}
642 812
813void
643void eio_submit (eio_req *req) 814eio_submit (eio_req *req)
644{ 815{
645 etp_submit (req); 816 etp_submit (req);
646} 817}
647 818
648unsigned int eio_nreqs (void) 819unsigned int
820eio_nreqs (void)
649{ 821{
650 return etp_nreqs (); 822 return etp_nreqs ();
651} 823}
652 824
653unsigned int eio_nready (void) 825unsigned int
826eio_nready (void)
654{ 827{
655 return etp_nready (); 828 return etp_nready ();
656} 829}
657 830
658unsigned int eio_npending (void) 831unsigned int
832eio_npending (void)
659{ 833{
660 return etp_npending (); 834 return etp_npending ();
661} 835}
662 836
663unsigned int eio_nthreads (void) 837unsigned int ecb_cold
838eio_nthreads (void)
664{ 839{
665 return etp_nthreads (); 840 return etp_nthreads ();
666} 841}
667 842
843void ecb_cold
668void eio_set_max_poll_time (double nseconds) 844eio_set_max_poll_time (double nseconds)
669{ 845{
670 etp_set_max_poll_time (nseconds); 846 etp_set_max_poll_time (nseconds);
671} 847}
672 848
849void ecb_cold
673void eio_set_max_poll_reqs (unsigned int maxreqs) 850eio_set_max_poll_reqs (unsigned int maxreqs)
674{ 851{
675 etp_set_max_poll_reqs (maxreqs); 852 etp_set_max_poll_reqs (maxreqs);
676} 853}
677 854
855void ecb_cold
678void eio_set_max_idle (unsigned int nthreads) 856eio_set_max_idle (unsigned int nthreads)
679{ 857{
680 etp_set_max_idle (nthreads); 858 etp_set_max_idle (nthreads);
681} 859}
682 860
861void ecb_cold
862eio_set_idle_timeout (unsigned int seconds)
863{
864 etp_set_idle_timeout (seconds);
865}
866
867void ecb_cold
683void eio_set_min_parallel (unsigned int nthreads) 868eio_set_min_parallel (unsigned int nthreads)
684{ 869{
685 etp_set_min_parallel (nthreads); 870 etp_set_min_parallel (nthreads);
686} 871}
687 872
873void ecb_cold
688void eio_set_max_parallel (unsigned int nthreads) 874eio_set_max_parallel (unsigned int nthreads)
689{ 875{
690 etp_set_max_parallel (nthreads); 876 etp_set_max_parallel (nthreads);
691} 877}
692 878
693int eio_poll (void) 879int eio_poll (void)
697 883
698/*****************************************************************************/ 884/*****************************************************************************/
699/* work around various missing functions */ 885/* work around various missing functions */
700 886
701#if !HAVE_PREADWRITE 887#if !HAVE_PREADWRITE
888# undef pread
889# undef pwrite
702# define pread eio__pread 890# define pread eio__pread
703# define pwrite eio__pwrite 891# define pwrite eio__pwrite
704 892
705/*
706 * make our pread/pwrite safe against themselves, but not against
707 * normal read/write by using a mutex. slows down execution a lot,
708 * but that's your problem, not mine.
709 */
710static mutex_t preadwritelock = X_MUTEX_INIT;
711
712static ssize_t 893static eio_ssize_t
713eio__pread (int fd, void *buf, size_t count, off_t offset) 894eio__pread (int fd, void *buf, size_t count, off_t offset)
714{ 895{
715 ssize_t res; 896 eio_ssize_t res;
716 off_t ooffset; 897 off_t ooffset;
717 898
718 X_LOCK (preadwritelock); 899 X_LOCK (preadwritelock);
719 ooffset = lseek (fd, 0, SEEK_CUR); 900 ooffset = lseek (fd, 0, SEEK_CUR);
720 lseek (fd, offset, SEEK_SET); 901 lseek (fd, offset, SEEK_SET);
723 X_UNLOCK (preadwritelock); 904 X_UNLOCK (preadwritelock);
724 905
725 return res; 906 return res;
726} 907}
727 908
728static ssize_t 909static eio_ssize_t
729eio__pwrite (int fd, void *buf, size_t count, off_t offset) 910eio__pwrite (int fd, void *buf, size_t count, off_t offset)
730{ 911{
731 ssize_t res; 912 eio_ssize_t res;
732 off_t ooffset; 913 off_t ooffset;
733 914
734 X_LOCK (preadwritelock); 915 X_LOCK (preadwritelock);
735 ooffset = lseek (fd, 0, SEEK_CUR); 916 ooffset = lseek (fd, 0, SEEK_CUR);
736 lseek (fd, offset, SEEK_SET); 917 lseek (fd, offset, SEEK_SET);
737 res = write (fd, buf, count); 918 res = write (fd, buf, count);
738 lseek (fd, offset, SEEK_SET); 919 lseek (fd, ooffset, SEEK_SET);
739 X_UNLOCK (preadwritelock); 920 X_UNLOCK (preadwritelock);
740 921
741 return res; 922 return res;
742} 923}
743#endif 924#endif
744 925
745#ifndef HAVE_FUTIMES 926#ifndef HAVE_UTIMES
746 927
928# undef utimes
747# define utimes(path,times) eio__utimes (path, times) 929# define utimes(path,times) eio__utimes (path, times)
748# define futimes(fd,times) eio__futimes (fd, times)
749 930
750static int 931static int
751eio__utimes (const char *filename, const struct timeval times[2]) 932eio__utimes (const char *filename, const struct timeval times[2])
752{ 933{
753 if (times) 934 if (times)
761 } 942 }
762 else 943 else
763 return utime (filename, 0); 944 return utime (filename, 0);
764} 945}
765 946
947#endif
948
949#ifndef HAVE_FUTIMES
950
951# undef futimes
952# define futimes(fd,times) eio__futimes (fd, times)
953
954static int
766static int eio__futimes (int fd, const struct timeval tv[2]) 955eio__futimes (int fd, const struct timeval tv[2])
767{ 956{
768 errno = ENOSYS; 957 errno = ENOSYS;
769 return -1; 958 return -1;
770} 959}
771 960
772#endif 961#endif
773 962
774#if !HAVE_FDATASYNC 963#if !HAVE_FDATASYNC
964# undef fdatasync
775# define fdatasync fsync 965# define fdatasync(fd) fsync (fd)
776#endif 966#endif
967
968/* sync_file_range always needs emulation */
969static int
970eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
971{
972#if HAVE_SYNC_FILE_RANGE
973 int res;
974
975 if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
976 || EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
977 || EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
978 {
979 flags = 0
980 | (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
981 | (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
982 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
983 }
984
985 res = sync_file_range (fd, offset, nbytes, flags);
986
987 if (!res || errno != ENOSYS)
988 return res;
989#endif
990
991 /* even though we could play tricks with the flags, it's better to always
992 * call fdatasync, as that matches the expectation of its users best */
993 return fdatasync (fd);
994}
995
996static int
997eio__fallocate (int fd, int mode, off_t offset, size_t len)
998{
999#if HAVE_FALLOCATE
1000 return fallocate (fd, mode, offset, len);
1001#else
1002 errno = ENOSYS;
1003 return -1;
1004#endif
1005}
777 1006
778#if !HAVE_READAHEAD 1007#if !HAVE_READAHEAD
1008# undef readahead
779# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 1009# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
780 1010
781static ssize_t 1011static eio_ssize_t
782eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) 1012eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
783{ 1013{
784 size_t todo = count; 1014 size_t todo = count;
785 dBUF; 1015 dBUF;
786 1016
798} 1028}
799 1029
800#endif 1030#endif
801 1031
802/* sendfile always needs emulation */ 1032/* sendfile always needs emulation */
803static ssize_t 1033static eio_ssize_t
804eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) 1034eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)
805{ 1035{
1036 eio_ssize_t written = 0;
806 ssize_t res; 1037 eio_ssize_t res;
807 1038
808 if (!count) 1039 if (!count)
809 return 0; 1040 return 0;
810 1041
1042 for (;;)
1043 {
1044#ifdef __APPLE__
1045# undef HAVE_SENDFILE /* broken, as everything on os x */
1046#endif
811#if HAVE_SENDFILE 1047#if HAVE_SENDFILE
812# if __linux 1048# if __linux
1049 off_t soffset = offset;
813 res = sendfile (ofd, ifd, &offset, count); 1050 res = sendfile (ofd, ifd, &soffset, count);
814 1051
815# elif __freebsd 1052# elif __FreeBSD__
816 /* 1053 /*
817 * Of course, the freebsd sendfile is a dire hack with no thoughts 1054 * Of course, the freebsd sendfile is a dire hack with no thoughts
818 * wasted on making it similar to other I/O functions. 1055 * wasted on making it similar to other I/O functions.
819 */ 1056 */
820 {
821 off_t sbytes; 1057 off_t sbytes;
822 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 1058 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
823 1059
824 if (res < 0 && sbytes) 1060 #if 0 /* according to the manpage, this is correct, but broken behaviour */
825 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 1061 /* freebsd' sendfile will return 0 on success */
1062 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
1063 /* not on e.g. EIO or EPIPE - sounds broken */
1064 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
826 res = sbytes; 1065 res = sbytes;
827 } 1066 #endif
1067
1068 /* according to source inspection, this is correct, and useful behaviour */
1069 if (sbytes)
1070 res = sbytes;
1071
1072# elif defined (__APPLE__)
1073 off_t sbytes = count;
1074 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
1075
1076 /* according to the manpage, sbytes is always valid */
1077 if (sbytes)
1078 res = sbytes;
828 1079
829# elif __hpux 1080# elif __hpux
830 res = sendfile (ofd, ifd, offset, count, 0, 0); 1081 res = sendfile (ofd, ifd, offset, count, 0, 0);
831 1082
832# elif __solaris 1083# elif __solaris
833 {
834 struct sendfilevec vec; 1084 struct sendfilevec vec;
835 size_t sbytes; 1085 size_t sbytes;
836 1086
837 vec.sfv_fd = ifd; 1087 vec.sfv_fd = ifd;
838 vec.sfv_flag = 0; 1088 vec.sfv_flag = 0;
839 vec.sfv_off = offset; 1089 vec.sfv_off = offset;
840 vec.sfv_len = count; 1090 vec.sfv_len = count;
841 1091
842 res = sendfilev (ofd, &vec, 1, &sbytes); 1092 res = sendfilev (ofd, &vec, 1, &sbytes);
843 1093
844 if (res < 0 && sbytes) 1094 if (res < 0 && sbytes)
845 res = sbytes; 1095 res = sbytes;
846 }
847 1096
848# endif 1097# endif
1098
1099#elif defined (_WIN32) && 0
1100 /* does not work, just for documentation of what would need to be done */
1101 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
1102 /* libeio guarantees that the file offset does not change, and windows */
1103 /* has no way to get an independent handle to the same file description */
1104 HANDLE h = TO_SOCKET (ifd);
1105 SetFilePointer (h, offset, 0, FILE_BEGIN);
1106 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1107
849#else 1108#else
850 res = -1; 1109 res = -1;
851 errno = ENOSYS; 1110 errno = ENOSYS;
852#endif 1111#endif
853 1112
1113 /* we assume sendfile can copy at least 128mb in one go */
1114 if (res <= 128 * 1024 * 1024)
1115 {
1116 if (res > 0)
1117 written += res;
1118
1119 if (written)
1120 return written;
1121
1122 break;
1123 }
1124 else
1125 {
1126 /* if we requested more, then probably the kernel was lazy */
1127 written += res;
1128 offset += res;
1129 count -= res;
1130
1131 if (!count)
1132 return written;
1133 }
1134 }
1135
854 if (res < 0 1136 if (res < 0
855 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1137 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1138 /* BSDs */
1139#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1140 || errno == ENOTSUP
1141#endif
1142#ifdef EOPNOTSUPP /* windows */
1143 || errno == EOPNOTSUPP /* BSDs */
1144#endif
856#if __solaris 1145#if __solaris
857 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1146 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
858#endif 1147#endif
859 ) 1148 )
860 ) 1149 )
864 1153
865 res = 0; 1154 res = 0;
866 1155
867 while (count) 1156 while (count)
868 { 1157 {
869 ssize_t cnt; 1158 eio_ssize_t cnt;
870 1159
871 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 1160 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
872 1161
873 if (cnt <= 0) 1162 if (cnt <= 0)
874 { 1163 {
891 } 1180 }
892 1181
893 return res; 1182 return res;
894} 1183}
895 1184
1185#ifdef PAGESIZE
1186# define eio_pagesize() PAGESIZE
1187#else
1188static intptr_t
1189eio_pagesize (void)
1190{
1191 static intptr_t page;
1192
1193 if (!page)
1194 page = sysconf (_SC_PAGESIZE);
1195
1196 return page;
1197}
1198#endif
1199
1200static void
1201eio_page_align (void **addr, size_t *length)
1202{
1203 intptr_t mask = eio_pagesize () - 1;
1204
1205 /* round down addr */
1206 intptr_t adj = mask & (intptr_t)*addr;
1207
1208 *addr = (void *)((intptr_t)*addr - adj);
1209 *length += adj;
1210
1211 /* round up length */
1212 *length = (*length + mask) & ~mask;
1213}
1214
1215#if !_POSIX_MEMLOCK
1216# define eio__mlockall(a) EIO_ENOSYS ()
1217#else
1218
1219static int
1220eio__mlockall (int flags)
1221{
1222 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1223 extern int mallopt (int, int);
1224 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1225 #endif
1226
1227 if (EIO_MCL_CURRENT != MCL_CURRENT
1228 || EIO_MCL_FUTURE != MCL_FUTURE)
1229 {
1230 flags = 0
1231 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1232 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1233 }
1234
1235 return mlockall (flags);
1236}
1237#endif
1238
1239#if !_POSIX_MEMLOCK_RANGE
1240# define eio__mlock(a,b) EIO_ENOSYS ()
1241#else
1242
1243static int
1244eio__mlock (void *addr, size_t length)
1245{
1246 eio_page_align (&addr, &length);
1247
1248 return mlock (addr, length);
1249}
1250
1251#endif
1252
1253#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1254# define eio__msync(a,b,c) EIO_ENOSYS ()
1255#else
1256
1257static int
1258eio__msync (void *mem, size_t len, int flags)
1259{
1260 eio_page_align (&mem, &len);
1261
1262 if (EIO_MS_ASYNC != MS_SYNC
1263 || EIO_MS_INVALIDATE != MS_INVALIDATE
1264 || EIO_MS_SYNC != MS_SYNC)
1265 {
1266 flags = 0
1267 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1268 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1269 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1270 }
1271
1272 return msync (mem, len, flags);
1273}
1274
1275#endif
1276
1277static int
1278eio__mtouch (eio_req *req)
1279{
1280 void *mem = req->ptr2;
1281 size_t len = req->size;
1282 int flags = req->int1;
1283
1284 eio_page_align (&mem, &len);
1285
1286 {
1287 intptr_t addr = (intptr_t)mem;
1288 intptr_t end = addr + len;
1289 intptr_t page = eio_pagesize ();
1290
1291 if (addr < end)
1292 if (flags & EIO_MT_MODIFY) /* modify */
1293 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1294 else
1295 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1296 }
1297
1298 return 0;
1299}
1300
1301/*****************************************************************************/
1302/* requests implemented outside eio_execute, because they are so large */
1303
1304static void
1305eio__realpath (eio_req *req, etp_worker *self)
1306{
1307 char *rel = req->ptr1;
1308 char *res;
1309 char *tmp1, *tmp2;
1310#if SYMLOOP_MAX > 32
1311 int symlinks = SYMLOOP_MAX;
1312#else
1313 int symlinks = 32;
1314#endif
1315
1316 req->result = -1;
1317
1318 errno = EINVAL;
1319 if (!rel)
1320 return;
1321
1322 errno = ENOENT;
1323 if (!*rel)
1324 return;
1325
1326 if (!req->ptr2)
1327 {
1328 X_LOCK (wrklock);
1329 req->flags |= EIO_FLAG_PTR2_FREE;
1330 X_UNLOCK (wrklock);
1331 req->ptr2 = malloc (PATH_MAX * 3);
1332
1333 errno = ENOMEM;
1334 if (!req->ptr2)
1335 return;
1336 }
1337
1338 res = req->ptr2;
1339 tmp1 = res + PATH_MAX;
1340 tmp2 = tmp1 + PATH_MAX;
1341
1342#if 0 /* disabled, the musl way to do things is just too racy */
1343#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1344 /* on linux we may be able to ask the kernel */
1345 {
1346 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1347
1348 if (fd >= 0)
1349 {
1350 sprintf (tmp1, "/proc/self/fd/%d", fd);
1351 req->result = readlink (tmp1, res, PATH_MAX);
1352 close (fd);
1353
1354 /* here we should probably stat the open file and the disk file, to make sure they still match */
1355
1356 if (req->result > 0)
1357 goto done;
1358 }
1359 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1360 return;
1361 }
1362#endif
1363#endif
1364
1365 if (*rel != '/')
1366 {
1367 if (!getcwd (res, PATH_MAX))
1368 return;
1369
1370 if (res [1]) /* only use if not / */
1371 res += strlen (res);
1372 }
1373
1374 while (*rel)
1375 {
1376 eio_ssize_t len, linklen;
1377 char *beg = rel;
1378
1379 while (*rel && *rel != '/')
1380 ++rel;
1381
1382 len = rel - beg;
1383
1384 if (!len) /* skip slashes */
1385 {
1386 ++rel;
1387 continue;
1388 }
1389
1390 if (beg [0] == '.')
1391 {
1392 if (len == 1)
1393 continue; /* . - nop */
1394
1395 if (beg [1] == '.' && len == 2)
1396 {
1397 /* .. - back up one component, if possible */
1398
1399 while (res != req->ptr2)
1400 if (*--res == '/')
1401 break;
1402
1403 continue;
1404 }
1405 }
1406
1407 errno = ENAMETOOLONG;
1408 if (res + 1 + len + 1 >= tmp1)
1409 return;
1410
1411 /* copy one component */
1412 *res = '/';
1413 memcpy (res + 1, beg, len);
1414
1415 /* zero-terminate, for readlink */
1416 res [len + 1] = 0;
1417
1418 /* now check if it's a symlink */
1419 linklen = readlink (req->ptr2, tmp1, PATH_MAX);
1420
1421 if (linklen < 0)
1422 {
1423 if (errno != EINVAL)
1424 return;
1425
1426 /* it's a normal directory. hopefully */
1427 res += len + 1;
1428 }
1429 else
1430 {
1431 /* yay, it was a symlink - build new path in tmp2 */
1432 int rellen = strlen (rel);
1433
1434 errno = ENAMETOOLONG;
1435 if (linklen + 1 + rellen >= PATH_MAX)
1436 return;
1437
1438 errno = ELOOP;
1439 if (!--symlinks)
1440 return;
1441
1442 if (*tmp1 == '/')
1443 res = req->ptr2; /* symlink resolves to an absolute path */
1444
1445 /* we need to be careful, as rel might point into tmp2 already */
1446 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1447 tmp2 [linklen] = '/';
1448 memcpy (tmp2, tmp1, linklen);
1449
1450 rel = tmp2;
1451 }
1452 }
1453
1454 /* special case for the lone root path */
1455 if (res == req->ptr2)
1456 *res++ = '/';
1457
1458 req->result = res - (char *)req->ptr2;
1459
1460done:
1461 req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
1462}
1463
1464static signed char
1465eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1466{
1467 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1468 : a->inode < b->inode ? -1
1469 : a->inode > b->inode ? 1
1470 : 0;
1471}
1472
1473#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1474
1475#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1476#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1477
1478static void
1479eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1480{
1481 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1482 unsigned char *bit = bits;
1483
1484 assert (CHAR_BIT == 8);
1485 assert (sizeof (eio_dirent) * 8 < 256);
1486 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1487 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1488
1489 if (size <= EIO_SORT_FAST)
1490 return;
1491
1492 /* first prepare an array of bits to test in our radix sort */
1493 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1494 /* inode_bits must contain all inodes ORed together */
1495 /* which is used to skip bits that are 0 everywhere, which is very common */
1496 {
1497 eio_ino_t endianness;
1498 int i, j;
1499
1500 /* we store the byte offset of byte n into byte n of "endianness" */
1501 for (i = 0; i < sizeof (eio_ino_t); ++i)
1502 ((unsigned char *)&endianness)[i] = i;
1503
1504 *bit++ = 0;
1505
1506 for (i = 0; i < sizeof (eio_ino_t); ++i)
1507 {
1508 /* shifting off the byte offsets out of "endianness" */
1509 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1510 endianness >>= 8;
1511
1512 for (j = 0; j < 8; ++j)
1513 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1514 *bit++ = offs + j;
1515 }
1516
1517 for (j = 0; j < 8; ++j)
1518 if (score_bits & (1 << j))
1519 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1520 }
1521
1522 /* now actually do the sorting (a variant of MSD radix sort) */
1523 {
1524 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1525 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1526 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1527 int stk_idx = 0;
1528
1529 base_stk [stk_idx] = dents;
1530 end_stk [stk_idx] = dents + size;
1531 bit_stk [stk_idx] = bit - 1;
1532
1533 do
1534 {
1535 base = base_stk [stk_idx];
1536 end = end_stk [stk_idx];
1537 bit = bit_stk [stk_idx];
1538
1539 for (;;)
1540 {
1541 unsigned char O = *bit >> 3;
1542 unsigned char M = 1 << (*bit & 7);
1543
1544 eio_dirent *a = base;
1545 eio_dirent *b = end;
1546
1547 if (b - a < EIO_SORT_CUTOFF)
1548 break;
1549
1550 /* now bit-partition the array on the bit */
1551 /* this ugly asymmetric loop seems to perform much better than typical */
1552 /* partition algos found in the literature */
1553 do
1554 if (!(((unsigned char *)a)[O] & M))
1555 ++a;
1556 else if (!(((unsigned char *)--b)[O] & M))
1557 {
1558 eio_dirent tmp = *a; *a = *b; *b = tmp;
1559 ++a;
1560 }
1561 while (b > a);
1562
1563 /* next bit, or stop, if no bits left in this path */
1564 if (!*--bit)
1565 break;
1566
1567 base_stk [stk_idx] = a;
1568 end_stk [stk_idx] = end;
1569 bit_stk [stk_idx] = bit;
1570 ++stk_idx;
1571
1572 end = a;
1573 }
1574 }
1575 while (stk_idx--);
1576 }
1577}
1578
1579static void
1580eio_dent_insertion_sort (eio_dirent *dents, int size)
1581{
1582 /* first move the smallest element to the front, to act as a sentinel */
1583 {
1584 int i;
1585 eio_dirent *min = dents;
1586
1587 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1588 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1589 if (EIO_DENT_CMP (dents [i], <, *min))
1590 min = &dents [i];
1591
1592 /* swap elements 0 and j (minimum) */
1593 {
1594 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1595 }
1596 }
1597
1598 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1599 {
1600 eio_dirent *i, *j;
1601
1602 for (i = dents + 1; i < dents + size; ++i)
1603 {
1604 eio_dirent value = *i;
1605
1606 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1607 j [1] = j [0];
1608
1609 j [1] = value;
1610 }
1611 }
1612}
1613
1614static void
1615eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1616{
1617 if (size <= 1)
1618 return; /* our insertion sort relies on size > 0 */
1619
1620 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1621 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1622 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1623
1624 /* use an insertion sort at the end, or for small arrays, */
1625 /* as insertion sort is more efficient for small partitions */
1626 eio_dent_insertion_sort (dents, size);
1627}
1628
896/* read a full directory */ 1629/* read a full directory */
897static void 1630static void
898eio__scandir (eio_req *req, etp_worker *self) 1631eio__scandir (eio_req *req, etp_worker *self)
899{ 1632{
1633 char *name, *names;
1634 int namesalloc = 4096 - sizeof (void *) * 4;
1635 int namesoffs = 0;
1636 int flags = req->int1;
1637 eio_dirent *dents = 0;
1638 int dentalloc = 128;
1639 int dentoffs = 0;
1640 eio_ino_t inode_bits = 0;
1641#ifdef _WIN32
1642 HANDLE dirp;
1643 WIN32_FIND_DATA entp;
1644#else
900 DIR *dirp; 1645 DIR *dirp;
901 EIO_STRUCT_DIRENT *entp; 1646 EIO_STRUCT_DIRENT *entp;
902 char *name, *names; 1647#endif
903 int memlen = 4096;
904 int memofs = 0;
905 int res = 0;
906 1648
907 X_LOCK (wrklock); 1649 req->result = -1;
1650
1651 if (!(flags & EIO_READDIR_DENTS))
1652 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1653
1654#ifdef _WIN32
1655 {
1656 char *path = malloc (MAX_PATH);
1657 _snprintf (path, MAX_PATH, "%s/*", (const char *)req->ptr1);
1658 dirp = FindFirstFile (path, &entp);
1659 free (path);
1660
1661 if (!dirp)
1662 {
1663 switch (GetLastError ())
1664 {
1665 case ERROR_FILE_NOT_FOUND:
1666 req->result = 0;
1667 break;
1668
1669 case ERROR_PATH_NOT_FOUND:
1670 case ERROR_NO_MORE_FILES:
1671 errno = ENOENT;
1672 break;
1673
1674 case ERROR_NOT_ENOUGH_MEMORY:
1675 errno = ENOMEM;
1676 break;
1677
1678 default:
1679 errno = EINVAL;
1680 break;
1681 }
1682
1683 }
1684 }
1685#else
908 self->dirp = dirp = opendir (req->ptr1); 1686 dirp = opendir (req->ptr1);
1687#endif
1688
909 req->flags |= EIO_FLAG_PTR2_FREE; 1689 if (req->flags & EIO_FLAG_PTR1_FREE)
1690 free (req->ptr1);
1691
1692 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1693 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
910 req->ptr2 = names = malloc (memlen); 1694 req->ptr2 = names = malloc (namesalloc);
911 X_UNLOCK (wrklock);
912 1695
913 if (dirp && names) 1696 if (dirp && names && (!flags || dents))
914 for (;;) 1697 for (;;)
915 { 1698 {
1699 int more;
1700
1701#ifdef _WIN32
1702 more = dirp;
1703#else
916 errno = 0; 1704 errno = 0;
917 entp = readdir (dirp); 1705 entp = readdir (dirp);
1706 more = entp;
1707#endif
918 1708
919 if (!entp) 1709 if (!more)
1710 {
1711#ifndef _WIN32
1712 int old_errno = errno;
1713 closedir (dirp);
1714 errno = old_errno;
1715
1716 if (errno)
1717 break;
1718#endif
1719
1720 /* sort etc. */
1721 req->int1 = flags;
1722 req->result = dentoffs;
1723
1724 if (flags & EIO_READDIR_STAT_ORDER)
1725 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1726 else if (flags & EIO_READDIR_DIRS_FIRST)
1727 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1728 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1729 else
1730 {
1731 /* in this case, all is known, and we just put dirs first and sort them */
1732 eio_dirent *oth = dents + dentoffs;
1733 eio_dirent *dir = dents;
1734
1735 /* now partition dirs to the front, and non-dirs to the back */
1736 /* by walking from both sides and swapping if necessary */
1737 while (oth > dir)
1738 {
1739 if (dir->type == EIO_DT_DIR)
1740 ++dir;
1741 else if ((--oth)->type == EIO_DT_DIR)
1742 {
1743 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1744
1745 ++dir;
1746 }
1747 }
1748
1749 /* now sort the dirs only (dirs all have the same score) */
1750 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1751 }
1752
920 break; 1753 break;
1754 }
921 1755
922 name = entp->d_name; 1756 /* now add the entry to our list(s) */
1757 name = D_NAME (entp);
923 1758
1759 /* skip . and .. entries */
924 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1760 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
925 { 1761 {
926 int len = strlen (name) + 1; 1762 int len = D_NAMLEN (entp) + 1;
927 1763
928 res++; 1764 while (ecb_expect_false (namesoffs + len > namesalloc))
929
930 while (memofs + len > memlen)
931 { 1765 {
932 memlen *= 2; 1766 namesalloc *= 2;
933 X_LOCK (wrklock);
934 req->ptr2 = names = realloc (names, memlen); 1767 req->ptr2 = names = realloc (names, namesalloc);
935 X_UNLOCK (wrklock);
936 1768
937 if (!names) 1769 if (!names)
938 break; 1770 break;
939 } 1771 }
940 1772
941 memcpy (names + memofs, name, len); 1773 memcpy (names + namesoffs, name, len);
1774
1775 if (dents)
1776 {
1777 struct eio_dirent *ent;
1778
1779 if (ecb_expect_false (dentoffs == dentalloc))
1780 {
1781 dentalloc *= 2;
1782 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1783
1784 if (!dents)
1785 break;
1786 }
1787
1788 ent = dents + dentoffs;
1789
1790 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1791 ent->namelen = len - 1;
1792 ent->inode = D_INO (entp);
1793
1794 inode_bits |= ent->inode;
1795
1796 switch (D_TYPE (entp))
1797 {
1798 default:
1799 ent->type = EIO_DT_UNKNOWN;
1800 flags |= EIO_READDIR_FOUND_UNKNOWN;
1801 break;
1802
1803 #ifdef DT_FIFO
1804 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1805 #endif
1806 #ifdef DT_CHR
1807 case DT_CHR: ent->type = EIO_DT_CHR; break;
1808 #endif
1809 #ifdef DT_MPC
1810 case DT_MPC: ent->type = EIO_DT_MPC; break;
1811 #endif
1812 #ifdef DT_DIR
1813 case DT_DIR: ent->type = EIO_DT_DIR; break;
1814 #endif
1815 #ifdef DT_NAM
1816 case DT_NAM: ent->type = EIO_DT_NAM; break;
1817 #endif
1818 #ifdef DT_BLK
1819 case DT_BLK: ent->type = EIO_DT_BLK; break;
1820 #endif
1821 #ifdef DT_MPB
1822 case DT_MPB: ent->type = EIO_DT_MPB; break;
1823 #endif
1824 #ifdef DT_REG
1825 case DT_REG: ent->type = EIO_DT_REG; break;
1826 #endif
1827 #ifdef DT_NWK
1828 case DT_NWK: ent->type = EIO_DT_NWK; break;
1829 #endif
1830 #ifdef DT_CMP
1831 case DT_CMP: ent->type = EIO_DT_CMP; break;
1832 #endif
1833 #ifdef DT_LNK
1834 case DT_LNK: ent->type = EIO_DT_LNK; break;
1835 #endif
1836 #ifdef DT_SOCK
1837 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1838 #endif
1839 #ifdef DT_DOOR
1840 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1841 #endif
1842 #ifdef DT_WHT
1843 case DT_WHT: ent->type = EIO_DT_WHT; break;
1844 #endif
1845 }
1846
1847 ent->score = 7;
1848
1849 if (flags & EIO_READDIR_DIRS_FIRST)
1850 {
1851 if (ent->type == EIO_DT_UNKNOWN)
1852 {
1853 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1854 ent->score = 1;
1855 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1856 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1857 }
1858 else if (ent->type == EIO_DT_DIR)
1859 ent->score = 0;
1860 }
1861 }
1862
942 memofs += len; 1863 namesoffs += len;
1864 ++dentoffs;
943 } 1865 }
1866
1867 if (EIO_CANCELLED (req))
1868 {
1869 errno = ECANCELED;
1870 break;
1871 }
1872
1873#ifdef _WIN32
1874 if (!FindNextFile (dirp, &entp))
1875 {
1876 FindClose (dirp);
1877 dirp = 0;
1878 }
1879#endif
944 } 1880 }
945
946 if (errno)
947 res = -1;
948
949 req->result = res;
950} 1881}
951 1882
952/*****************************************************************************/ 1883/*****************************************************************************/
953 1884
954#define ALLOC(len) \ 1885#define ALLOC(len) \
970{ 1901{
971 ETP_REQ *req; 1902 ETP_REQ *req;
972 struct timespec ts; 1903 struct timespec ts;
973 etp_worker *self = (etp_worker *)thr_arg; 1904 etp_worker *self = (etp_worker *)thr_arg;
974 1905
975 /* try to distribute timeouts somewhat randomly */ 1906 /* try to distribute timeouts somewhat evenly */
976 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 1907 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
977 1908
978 for (;;) 1909 for (;;)
979 { 1910 {
1911 ts.tv_sec = 0;
1912
980 X_LOCK (reqlock); 1913 X_LOCK (reqlock);
981 1914
982 for (;;) 1915 for (;;)
983 { 1916 {
984 self->req = req = reqq_shift (&req_queue); 1917 self->req = req = reqq_shift (&req_queue);
985 1918
986 if (req) 1919 if (req)
987 break; 1920 break;
988 1921
1922 if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
1923 {
1924 X_UNLOCK (reqlock);
1925 X_LOCK (wrklock);
1926 --started;
1927 X_UNLOCK (wrklock);
1928 goto quit;
1929 }
1930
989 ++idle; 1931 ++idle;
990 1932
991 ts.tv_sec = time (0) + IDLE_TIMEOUT; 1933 if (idle <= max_idle)
992 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 1934 /* we are allowed to idle, so do so without any timeout */
1935 X_COND_WAIT (reqwait, reqlock);
1936 else
993 { 1937 {
994 if (idle > max_idle) 1938 /* initialise timeout once */
995 { 1939 if (!ts.tv_sec)
996 --idle; 1940 ts.tv_sec = time (0) + idle_timeout;
997 X_UNLOCK (reqlock);
998 X_LOCK (wrklock);
999 --started;
1000 X_UNLOCK (wrklock);
1001 goto quit;
1002 }
1003 1941
1004 /* we are allowed to idle, so do so without any timeout */
1005 X_COND_WAIT (reqwait, reqlock); 1942 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
1943 ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1006 } 1944 }
1007 1945
1008 --idle; 1946 --idle;
1009 } 1947 }
1010 1948
1013 X_UNLOCK (reqlock); 1951 X_UNLOCK (reqlock);
1014 1952
1015 if (req->type < 0) 1953 if (req->type < 0)
1016 goto quit; 1954 goto quit;
1017 1955
1018 if (!EIO_CANCELLED (req))
1019 ETP_EXECUTE (self, req); 1956 ETP_EXECUTE (self, req);
1020 1957
1021 X_LOCK (reslock); 1958 X_LOCK (reslock);
1022 1959
1023 ++npending; 1960 ++npending;
1024 1961
1039 return 0; 1976 return 0;
1040} 1977}
1041 1978
1042/*****************************************************************************/ 1979/*****************************************************************************/
1043 1980
1981int ecb_cold
1044int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1982eio_init (void (*want_poll)(void), void (*done_poll)(void))
1045{ 1983{
1046 return etp_init (want_poll, done_poll); 1984 return etp_init (want_poll, done_poll);
1047} 1985}
1048 1986
1987ecb_inline void
1049static void eio_api_destroy (eio_req *req) 1988eio_api_destroy (eio_req *req)
1050{ 1989{
1051 free (req); 1990 free (req);
1052} 1991}
1053 1992
1054#define REQ(rtype) \ 1993#define REQ(rtype) \
1073 { \ 2012 { \
1074 eio_api_destroy (req); \ 2013 eio_api_destroy (req); \
1075 return 0; \ 2014 return 0; \
1076 } 2015 }
1077 2016
2017static void
1078static void eio_execute (etp_worker *self, eio_req *req) 2018eio_execute (etp_worker *self, eio_req *req)
1079{ 2019{
1080 errno = 0; 2020 if (ecb_expect_false (EIO_CANCELLED (req)))
2021 {
2022 req->result = -1;
2023 req->errorno = ECANCELED;
2024 return;
2025 }
1081 2026
1082 switch (req->type) 2027 switch (req->type)
1083 { 2028 {
1084 case EIO_READ: ALLOC (req->size); 2029 case EIO_READ: ALLOC (req->size);
1085 req->result = req->offs >= 0 2030 req->result = req->offs >= 0
1097 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2042 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1098 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2043 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1099 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2044 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1100 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; 2045 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1101 2046
2047 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2048 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2049 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2050 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2051
1102 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; 2052 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1103 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; 2053 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1104 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break; 2054 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1105 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break; 2055 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1106 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break; 2056 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1113 case EIO_RMDIR: req->result = rmdir (req->ptr1); break; 2063 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1114 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; 2064 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1115 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; 2065 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1116 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; 2066 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1117 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; 2067 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1118 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break; 2068 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1119 2069
2070 case EIO_REALPATH: eio__realpath (req, self); break;
2071
1120 case EIO_READLINK: ALLOC (NAME_MAX); 2072 case EIO_READLINK: ALLOC (PATH_MAX);
1121 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; 2073 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1122 2074
1123 case EIO_SYNC: req->result = 0; sync (); break; 2075 case EIO_SYNC: req->result = 0; sync (); break;
1124 case EIO_FSYNC: req->result = fsync (req->int1); break; 2076 case EIO_FSYNC: req->result = fsync (req->int1); break;
1125 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 2077 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
2078 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
2079 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
2080 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
2081 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
2082 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
2083 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
1126 2084
1127 case EIO_READDIR: eio__scandir (req, self); break; 2085 case EIO_READDIR: eio__scandir (req, self); break;
1128 2086
1129 case EIO_BUSY: 2087 case EIO_BUSY:
1130#ifdef _WIN32 2088#ifdef _WIN32
1131 Sleep (req->nv1 * 1000.); 2089 Sleep (req->nv1 * 1e3);
1132#else 2090#else
1133 { 2091 {
1134 struct timeval tv; 2092 struct timeval tv;
1135 2093
1136 tv.tv_sec = req->nv1; 2094 tv.tv_sec = req->nv1;
1137 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; 2095 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1138 2096
1139 req->result = select (0, 0, 0, 0, &tv); 2097 req->result = select (0, 0, 0, 0, &tv);
1140 } 2098 }
1141#endif 2099#endif
1142 break; 2100 break;
1157 times = tv; 2115 times = tv;
1158 } 2116 }
1159 else 2117 else
1160 times = 0; 2118 times = 0;
1161 2119
1162
1163 req->result = req->type == EIO_FUTIME 2120 req->result = req->type == EIO_FUTIME
1164 ? futimes (req->int1, times) 2121 ? futimes (req->int1, times)
1165 : utimes (req->ptr1, times); 2122 : utimes (req->ptr1, times);
1166 } 2123 }
2124 break;
1167 2125
1168 case EIO_GROUP: 2126 case EIO_GROUP:
2127 abort (); /* handled in eio_request */
2128
1169 case EIO_NOP: 2129 case EIO_NOP:
1170 req->result = 0; 2130 req->result = 0;
1171 break; 2131 break;
1172 2132
1173 case EIO_CUSTOM: 2133 case EIO_CUSTOM:
1174 req->feed (req); 2134 req->feed (req);
1175 break; 2135 break;
1176 2136
1177 default: 2137 default:
2138 errno = ENOSYS;
1178 req->result = -1; 2139 req->result = -1;
1179 break; 2140 break;
1180 } 2141 }
1181 2142
1182 req->errorno = errno; 2143 req->errorno = errno;
1202eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data) 2163eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
1203{ 2164{
1204 REQ (EIO_FSYNC); req->int1 = fd; SEND; 2165 REQ (EIO_FSYNC); req->int1 = fd; SEND;
1205} 2166}
1206 2167
2168eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2169{
2170 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2171}
2172
2173eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2174{
2175 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2176}
2177
2178eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
2179{
2180 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
2181}
2182
2183eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
2184{
2185 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2186}
2187
2188eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2189{
2190 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2191}
2192
2193eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2194{
2195 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
2196}
2197
1207eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) 2198eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1208{ 2199{
1209 REQ (EIO_FDATASYNC); req->int1 = fd; SEND; 2200 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1210} 2201}
1211 2202
1232eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) 2223eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1233{ 2224{
1234 REQ (EIO_FSTAT); req->int1 = fd; SEND; 2225 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1235} 2226}
1236 2227
2228eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
2229{
2230 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
2231}
2232
1237eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) 2233eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1238{ 2234{
1239 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; 2235 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1240} 2236}
1241 2237
1247eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) 2243eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1248{ 2244{
1249 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2245 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1250} 2246}
1251 2247
1252eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2248eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1253{ 2249{
1254 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2250 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1255} 2251}
1256 2252
1257eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) 2253eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
1277eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data) 2273eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1278{ 2274{
1279 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2275 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1280} 2276}
1281 2277
1282eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2278eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1283{ 2279{
1284 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2280 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1285} 2281}
1286 2282
1287eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) 2283eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
1303eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data) 2299eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1304{ 2300{
1305 return eio__1path (EIO_READLINK, path, pri, cb, data); 2301 return eio__1path (EIO_READLINK, path, pri, cb, data);
1306} 2302}
1307 2303
2304eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2305{
2306 return eio__1path (EIO_REALPATH, path, pri, cb, data);
2307}
2308
1308eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) 2309eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1309{ 2310{
1310 return eio__1path (EIO_STAT, path, pri, cb, data); 2311 return eio__1path (EIO_STAT, path, pri, cb, data);
1311} 2312}
1312 2313
1313eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) 2314eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1314{ 2315{
1315 return eio__1path (EIO_LSTAT, path, pri, cb, data); 2316 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1316} 2317}
1317 2318
2319eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
2320{
2321 return eio__1path (EIO_STATVFS, path, pri, cb, data);
2322}
2323
1318eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) 2324eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1319{ 2325{
1320 return eio__1path (EIO_UNLINK, path, pri, cb, data); 2326 return eio__1path (EIO_UNLINK, path, pri, cb, data);
1321} 2327}
1322 2328
1323eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data) 2329eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1324{ 2330{
1325 return eio__1path (EIO_RMDIR, path, pri, cb, data); 2331 return eio__1path (EIO_RMDIR, path, pri, cb, data);
1326} 2332}
1327 2333
1328eio_req *eio_readdir (const char *path, int pri, eio_cb cb, void *data) 2334eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1329{ 2335{
1330 return eio__1path (EIO_READDIR, path, pri, cb, data); 2336 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1331} 2337}
1332 2338
1333eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) 2339eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1334{ 2340{
1335 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; 2341 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1336} 2342}
1337 2343
1338static eio_req * 2344static eio_req *
1339eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2345eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1340{ 2346{
1364eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2370eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1365{ 2371{
1366 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2372 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1367} 2373}
1368 2374
1369eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) 2375eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
1370{ 2376{
1371 REQ (EIO_CUSTOM); req->feed = execute; SEND; 2377 REQ (EIO_CUSTOM); req->feed = execute; SEND;
1372} 2378}
1373 2379
1374#endif 2380#endif
1385#undef SEND 2391#undef SEND
1386 2392
1387/*****************************************************************************/ 2393/*****************************************************************************/
1388/* grp functions */ 2394/* grp functions */
1389 2395
2396void
1390void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2397eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1391{ 2398{
1392 grp->int2 = limit; 2399 grp->int2 = limit;
1393 grp->feed = feed; 2400 grp->feed = feed;
1394 2401
1395 grp_try_feed (grp); 2402 grp_try_feed (grp);
1396} 2403}
1397 2404
2405void
1398void eio_grp_limit (eio_req *grp, int limit) 2406eio_grp_limit (eio_req *grp, int limit)
1399{ 2407{
1400 grp->int2 = limit; 2408 grp->int2 = limit;
1401 2409
1402 grp_try_feed (grp); 2410 grp_try_feed (grp);
1403} 2411}
1404 2412
2413void
1405void eio_grp_add (eio_req *grp, eio_req *req) 2414eio_grp_add (eio_req *grp, eio_req *req)
1406{ 2415{
1407 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2416 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
2417
2418 grp->flags |= EIO_FLAG_GROUPADD;
1408 2419
1409 ++grp->size; 2420 ++grp->size;
1410 req->grp = grp; 2421 req->grp = grp;
1411 2422
1412 req->grp_prev = 0; 2423 req->grp_prev = 0;
1419} 2430}
1420 2431
1421/*****************************************************************************/ 2432/*****************************************************************************/
1422/* misc garbage */ 2433/* misc garbage */
1423 2434
2435eio_ssize_t
1424ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2436eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1425{ 2437{
1426 etp_worker wrk; 2438 etp_worker wrk;
2439 eio_ssize_t ret;
1427 2440
1428 wrk.dbuf = 0; 2441 wrk.dbuf = 0;
1429 2442
1430 eio__sendfile (ofd, ifd, offset, count, &wrk); 2443 ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
1431 2444
1432 if (wrk.dbuf) 2445 if (wrk.dbuf)
1433 free (wrk.dbuf); 2446 free (wrk.dbuf);
1434}
1435 2447
2448 return ret;
2449}
2450

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines