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 1.22 by root, Fri Jul 11 10:54:50 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
177static mutex_t wrklock = X_MUTEX_INIT; 296static xmutex_t wrklock;
178static mutex_t reslock = X_MUTEX_INIT; 297static xmutex_t reslock;
179static mutex_t reqlock = X_MUTEX_INIT; 298static xmutex_t reqlock;
180static cond_t reqwait = X_COND_INIT; 299static xcond_t reqwait;
181 300
182#if !HAVE_PREADWRITE 301#if !HAVE_PREADWRITE
183/* 302/*
184 * make our pread/pwrite emulation safe against themselves, but not against 303 * make our pread/pwrite emulation safe against themselves, but not against
185 * normal read/write by using a mutex. slows down execution a lot, 304 * normal read/write by using a mutex. slows down execution a lot,
186 * but that's your problem, not mine. 305 * but that's your problem, not mine.
187 */ 306 */
188static mutex_t preadwritelock = X_MUTEX_INIT; 307static xmutex_t preadwritelock = X_MUTEX_INIT;
189#endif 308#endif
190 309
191typedef struct etp_worker 310typedef struct etp_worker
192{ 311{
193 /* locked by wrklock */ 312 /* locked by wrklock */
194 struct etp_worker *prev, *next; 313 struct etp_worker *prev, *next;
195 314
196 thread_t tid; 315 xthread_t tid;
197 316
198 /* locked by reslock, reqlock or wrklock */ 317 /* locked by reslock, reqlock or wrklock */
199 ETP_REQ *req; /* currently processed request */ 318 ETP_REQ *req; /* currently processed request */
200 319
201 ETP_WORKER_COMMON 320 ETP_WORKER_COMMON
206#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock) 325#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
207#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock) 326#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
208 327
209/* worker threads management */ 328/* worker threads management */
210 329
330static void ecb_cold
211static void etp_worker_clear (etp_worker *wrk) 331etp_worker_clear (etp_worker *wrk)
212{ 332{
213 ETP_WORKER_CLEAR (wrk); 333 ETP_WORKER_CLEAR (wrk);
214} 334}
215 335
336static void ecb_cold
216static void etp_worker_free (etp_worker *wrk) 337etp_worker_free (etp_worker *wrk)
217{ 338{
218 wrk->next->prev = wrk->prev; 339 wrk->next->prev = wrk->prev;
219 wrk->prev->next = wrk->next; 340 wrk->prev->next = wrk->next;
220 341
221 free (wrk); 342 free (wrk);
222} 343}
223 344
224static unsigned int etp_nreqs (void) 345static unsigned int
346etp_nreqs (void)
225{ 347{
226 int retval; 348 int retval;
227 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 349 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
228 retval = nreqs; 350 retval = nreqs;
229 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 351 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
230 return retval; 352 return retval;
231} 353}
232 354
233static unsigned int etp_nready (void) 355static unsigned int
356etp_nready (void)
234{ 357{
235 unsigned int retval; 358 unsigned int retval;
236 359
237 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 360 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
238 retval = nready; 361 retval = nready;
239 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 362 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
240 363
241 return retval; 364 return retval;
242} 365}
243 366
244static unsigned int etp_npending (void) 367static unsigned int
368etp_npending (void)
245{ 369{
246 unsigned int retval; 370 unsigned int retval;
247 371
248 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 372 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
249 retval = npending; 373 retval = npending;
250 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 374 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
251 375
252 return retval; 376 return retval;
253} 377}
254 378
255static unsigned int etp_nthreads (void) 379static unsigned int
380etp_nthreads (void)
256{ 381{
257 unsigned int retval; 382 unsigned int retval;
258 383
259 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 384 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
260 retval = started; 385 retval = started;
274} etp_reqq; 399} etp_reqq;
275 400
276static etp_reqq req_queue; 401static etp_reqq req_queue;
277static etp_reqq res_queue; 402static etp_reqq res_queue;
278 403
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
279static int reqq_push (etp_reqq *q, ETP_REQ *req) 416reqq_push (etp_reqq *q, ETP_REQ *req)
280{ 417{
281 int pri = req->pri; 418 int pri = req->pri;
282 req->next = 0; 419 req->next = 0;
283 420
284 if (q->qe[pri]) 421 if (q->qe[pri])
290 q->qe[pri] = q->qs[pri] = req; 427 q->qe[pri] = q->qs[pri] = req;
291 428
292 return q->size++; 429 return q->size++;
293} 430}
294 431
432static ETP_REQ * ecb_noinline
295static ETP_REQ *reqq_shift (etp_reqq *q) 433reqq_shift (etp_reqq *q)
296{ 434{
297 int pri; 435 int pri;
298 436
299 if (!q->size) 437 if (!q->size)
300 return 0; 438 return 0;
315 } 453 }
316 454
317 abort (); 455 abort ();
318} 456}
319 457
320static void etp_atfork_prepare (void) 458static void ecb_cold
459etp_thread_init (void)
321{ 460{
322 X_LOCK (wrklock);
323 X_LOCK (reqlock);
324 X_LOCK (reslock);
325#if !HAVE_PREADWRITE 461#if !HAVE_PREADWRITE
326 X_LOCK (preadwritelock);
327#endif
328}
329
330static void etp_atfork_parent (void)
331{
332#if !HAVE_PREADWRITE
333 X_UNLOCK (preadwritelock); 462 X_MUTEX_CREATE (preadwritelock);
334#endif 463#endif
335 X_UNLOCK (reslock); 464 X_MUTEX_CREATE (wrklock);
336 X_UNLOCK (reqlock); 465 X_MUTEX_CREATE (reslock);
337 X_UNLOCK (wrklock); 466 X_MUTEX_CREATE (reqlock);
467 X_COND_CREATE (reqwait);
338} 468}
339 469
470static void ecb_cold
340static void etp_atfork_child (void) 471etp_atfork_child (void)
341{ 472{
342 ETP_REQ *prv; 473 reqq_init (&req_queue);
474 reqq_init (&res_queue);
343 475
344 while ((prv = reqq_shift (&req_queue))) 476 wrk_first.next =
345 ETP_DESTROY (prv); 477 wrk_first.prev = &wrk_first;
346
347 while ((prv = reqq_shift (&res_queue)))
348 ETP_DESTROY (prv);
349
350 while (wrk_first.next != &wrk_first)
351 {
352 etp_worker *wrk = wrk_first.next;
353
354 if (wrk->req)
355 ETP_DESTROY (wrk->req);
356
357 etp_worker_clear (wrk);
358 etp_worker_free (wrk);
359 }
360 478
361 started = 0; 479 started = 0;
362 idle = 0; 480 idle = 0;
363 nreqs = 0; 481 nreqs = 0;
364 nready = 0; 482 nready = 0;
365 npending = 0; 483 npending = 0;
366 484
367 etp_atfork_parent (); 485 etp_thread_init ();
368} 486}
369 487
370static void 488static void ecb_cold
371etp_once_init (void) 489etp_once_init (void)
372{ 490{
373 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);
374} 493}
375 494
376static int 495static int ecb_cold
377etp_init (void (*want_poll)(void), void (*done_poll)(void)) 496etp_init (void (*want_poll)(void), void (*done_poll)(void))
378{ 497{
379 static pthread_once_t doinit = PTHREAD_ONCE_INIT; 498 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
380 499
381 pthread_once (&doinit, etp_once_init); 500 pthread_once (&doinit, etp_once_init);
386 return 0; 505 return 0;
387} 506}
388 507
389X_THREAD_PROC (etp_proc); 508X_THREAD_PROC (etp_proc);
390 509
510static void ecb_cold
391static void etp_start_thread (void) 511etp_start_thread (void)
392{ 512{
393 etp_worker *wrk = calloc (1, sizeof (etp_worker)); 513 etp_worker *wrk = calloc (1, sizeof (etp_worker));
394 514
395 /*TODO*/ 515 /*TODO*/
396 assert (("unable to allocate worker thread data", wrk)); 516 assert (("unable to allocate worker thread data", wrk));
409 free (wrk); 529 free (wrk);
410 530
411 X_UNLOCK (wrklock); 531 X_UNLOCK (wrklock);
412} 532}
413 533
534static void
414static void etp_maybe_start_thread (void) 535etp_maybe_start_thread (void)
415{ 536{
416 if (etp_nthreads () >= wanted) 537 if (ecb_expect_true (etp_nthreads () >= wanted))
417 return; 538 return;
418 539
419 /* todo: maybe use idle here, but might be less exact */ 540 /* todo: maybe use idle here, but might be less exact */
420 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 ()))
421 return; 542 return;
422 543
423 etp_start_thread (); 544 etp_start_thread ();
424} 545}
425 546
547static void ecb_cold
426static void etp_end_thread (void) 548etp_end_thread (void)
427{ 549{
428 eio_req *req = calloc (1, sizeof (eio_req)); 550 eio_req *req = calloc (1, sizeof (eio_req));
429 551
430 req->type = -1; 552 req->type = -1;
431 req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 553 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
438 X_LOCK (wrklock); 560 X_LOCK (wrklock);
439 --started; 561 --started;
440 X_UNLOCK (wrklock); 562 X_UNLOCK (wrklock);
441} 563}
442 564
443static int etp_poll (void) 565static int
566etp_poll (void)
444{ 567{
445 unsigned int maxreqs; 568 unsigned int maxreqs;
446 unsigned int maxtime; 569 unsigned int maxtime;
447 struct timeval tv_start, tv_now; 570 struct timeval tv_start, tv_now;
448 571
478 601
479 X_LOCK (reqlock); 602 X_LOCK (reqlock);
480 --nreqs; 603 --nreqs;
481 X_UNLOCK (reqlock); 604 X_UNLOCK (reqlock);
482 605
483 if (req->type == EIO_GROUP && req->size) 606 if (ecb_expect_false (req->type == EIO_GROUP && req->size))
484 { 607 {
485 req->int1 = 1; /* mark request as delayed */ 608 req->int1 = 1; /* mark request as delayed */
486 continue; 609 continue;
487 } 610 }
488 else 611 else
489 { 612 {
490 int res = ETP_FINISH (req); 613 int res = ETP_FINISH (req);
491 if (res) 614 if (ecb_expect_false (res))
492 return res; 615 return res;
493 } 616 }
494 617
495 if (maxreqs && !--maxreqs) 618 if (ecb_expect_false (maxreqs && !--maxreqs))
496 break; 619 break;
497 620
498 if (maxtime) 621 if (maxtime)
499 { 622 {
500 gettimeofday (&tv_now, 0); 623 gettimeofday (&tv_now, 0);
506 629
507 errno = EAGAIN; 630 errno = EAGAIN;
508 return -1; 631 return -1;
509} 632}
510 633
634static void
511static void etp_cancel (ETP_REQ *req) 635etp_cancel (ETP_REQ *req)
512{ 636{
513 X_LOCK (wrklock); 637 req->cancelled = 1;
514 req->flags |= EIO_FLAG_CANCELLED;
515 X_UNLOCK (wrklock);
516 638
517 eio_grp_cancel (req); 639 eio_grp_cancel (req);
518} 640}
519 641
642static void
520static void etp_submit (ETP_REQ *req) 643etp_submit (ETP_REQ *req)
521{ 644{
522 req->pri -= ETP_PRI_MIN; 645 req->pri -= ETP_PRI_MIN;
523 646
524 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;
525 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;
526 649
650 if (ecb_expect_false (req->type == EIO_GROUP))
651 {
652 /* I hope this is worth it :/ */
527 X_LOCK (reqlock); 653 X_LOCK (reqlock);
528 ++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;
529 ++nready; 670 ++nready;
530 reqq_push (&req_queue, req); 671 reqq_push (&req_queue, req);
531 X_COND_SIGNAL (reqwait); 672 X_COND_SIGNAL (reqwait);
532 X_UNLOCK (reqlock); 673 X_UNLOCK (reqlock);
533 674
534 etp_maybe_start_thread (); 675 etp_maybe_start_thread ();
676 }
535} 677}
536 678
679static void ecb_cold
537static void etp_set_max_poll_time (double nseconds) 680etp_set_max_poll_time (double nseconds)
538{ 681{
539 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 682 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
540 max_poll_time = nseconds; 683 max_poll_time = nseconds * EIO_TICKS;
541 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 684 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
542} 685}
543 686
687static void ecb_cold
544static void etp_set_max_poll_reqs (unsigned int maxreqs) 688etp_set_max_poll_reqs (unsigned int maxreqs)
545{ 689{
546 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 690 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
547 max_poll_reqs = maxreqs; 691 max_poll_reqs = maxreqs;
548 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 692 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
549} 693}
550 694
695static void ecb_cold
551static void etp_set_max_idle (unsigned int nthreads) 696etp_set_max_idle (unsigned int nthreads)
552{ 697{
553 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 698 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
554 max_idle = nthreads <= 0 ? 1 : nthreads; 699 max_idle = nthreads;
555 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 700 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
556} 701}
557 702
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
558static void etp_set_min_parallel (unsigned int nthreads) 712etp_set_min_parallel (unsigned int nthreads)
559{ 713{
560 if (wanted < nthreads) 714 if (wanted < nthreads)
561 wanted = nthreads; 715 wanted = nthreads;
562} 716}
563 717
718static void ecb_cold
564static void etp_set_max_parallel (unsigned int nthreads) 719etp_set_max_parallel (unsigned int nthreads)
565{ 720{
566 if (wanted > nthreads) 721 if (wanted > nthreads)
567 wanted = nthreads; 722 wanted = nthreads;
568 723
569 while (started > wanted) 724 while (started > wanted)
570 etp_end_thread (); 725 etp_end_thread ();
571} 726}
572 727
573/*****************************************************************************/ 728/*****************************************************************************/
574 729
730static void
575static void grp_try_feed (eio_req *grp) 731grp_try_feed (eio_req *grp)
576{ 732{
577 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 733 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
578 { 734 {
579 int old_len = grp->size; 735 grp->flags &= ~EIO_FLAG_GROUPADD;
580 736
581 EIO_FEED (grp); 737 EIO_FEED (grp);
582 738
583 /* stop if no progress has been made */ 739 /* stop if no progress has been made */
584 if (old_len == grp->size) 740 if (!(grp->flags & EIO_FLAG_GROUPADD))
585 { 741 {
586 grp->feed = 0; 742 grp->feed = 0;
587 break; 743 break;
588 } 744 }
589 } 745 }
590} 746}
591 747
748static int
592static int grp_dec (eio_req *grp) 749grp_dec (eio_req *grp)
593{ 750{
594 --grp->size; 751 --grp->size;
595 752
596 /* call feeder, if applicable */ 753 /* call feeder, if applicable */
597 grp_try_feed (grp); 754 grp_try_feed (grp);
601 return eio_finish (grp); 758 return eio_finish (grp);
602 else 759 else
603 return 0; 760 return 0;
604} 761}
605 762
763static void
606void eio_destroy (eio_req *req) 764eio_destroy (eio_req *req)
607{ 765{
608 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); 766 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
609 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); 767 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
610 768
611 EIO_DESTROY (req); 769 EIO_DESTROY (req);
612} 770}
613 771
772static int
614static int eio_finish (eio_req *req) 773eio_finish (eio_req *req)
615{ 774{
616 int res = EIO_FINISH (req); 775 int res = EIO_FINISH (req);
617 776
618 if (req->grp) 777 if (req->grp)
619 { 778 {
627 if (grp->grp_first == req) 786 if (grp->grp_first == req)
628 grp->grp_first = req->grp_next; 787 grp->grp_first = req->grp_next;
629 788
630 res2 = grp_dec (grp); 789 res2 = grp_dec (grp);
631 790
632 if (!res && res2) 791 if (!res)
633 res = res2; 792 res = res2;
634 } 793 }
635 794
636 eio_destroy (req); 795 eio_destroy (req);
637 796
638 return res; 797 return res;
639} 798}
640 799
800void
641void eio_grp_cancel (eio_req *grp) 801eio_grp_cancel (eio_req *grp)
642{ 802{
643 for (grp = grp->grp_first; grp; grp = grp->grp_next) 803 for (grp = grp->grp_first; grp; grp = grp->grp_next)
644 eio_cancel (grp); 804 eio_cancel (grp);
645} 805}
646 806
807void
647void eio_cancel (eio_req *req) 808eio_cancel (eio_req *req)
648{ 809{
649 etp_cancel (req); 810 etp_cancel (req);
650} 811}
651 812
813void
652void eio_submit (eio_req *req) 814eio_submit (eio_req *req)
653{ 815{
654 etp_submit (req); 816 etp_submit (req);
655} 817}
656 818
657unsigned int eio_nreqs (void) 819unsigned int
820eio_nreqs (void)
658{ 821{
659 return etp_nreqs (); 822 return etp_nreqs ();
660} 823}
661 824
662unsigned int eio_nready (void) 825unsigned int
826eio_nready (void)
663{ 827{
664 return etp_nready (); 828 return etp_nready ();
665} 829}
666 830
667unsigned int eio_npending (void) 831unsigned int
832eio_npending (void)
668{ 833{
669 return etp_npending (); 834 return etp_npending ();
670} 835}
671 836
672unsigned int eio_nthreads (void) 837unsigned int ecb_cold
838eio_nthreads (void)
673{ 839{
674 return etp_nthreads (); 840 return etp_nthreads ();
675} 841}
676 842
843void ecb_cold
677void eio_set_max_poll_time (double nseconds) 844eio_set_max_poll_time (double nseconds)
678{ 845{
679 etp_set_max_poll_time (nseconds); 846 etp_set_max_poll_time (nseconds);
680} 847}
681 848
849void ecb_cold
682void eio_set_max_poll_reqs (unsigned int maxreqs) 850eio_set_max_poll_reqs (unsigned int maxreqs)
683{ 851{
684 etp_set_max_poll_reqs (maxreqs); 852 etp_set_max_poll_reqs (maxreqs);
685} 853}
686 854
855void ecb_cold
687void eio_set_max_idle (unsigned int nthreads) 856eio_set_max_idle (unsigned int nthreads)
688{ 857{
689 etp_set_max_idle (nthreads); 858 etp_set_max_idle (nthreads);
690} 859}
691 860
861void ecb_cold
862eio_set_idle_timeout (unsigned int seconds)
863{
864 etp_set_idle_timeout (seconds);
865}
866
867void ecb_cold
692void eio_set_min_parallel (unsigned int nthreads) 868eio_set_min_parallel (unsigned int nthreads)
693{ 869{
694 etp_set_min_parallel (nthreads); 870 etp_set_min_parallel (nthreads);
695} 871}
696 872
873void ecb_cold
697void eio_set_max_parallel (unsigned int nthreads) 874eio_set_max_parallel (unsigned int nthreads)
698{ 875{
699 etp_set_max_parallel (nthreads); 876 etp_set_max_parallel (nthreads);
700} 877}
701 878
702int eio_poll (void) 879int eio_poll (void)
706 883
707/*****************************************************************************/ 884/*****************************************************************************/
708/* work around various missing functions */ 885/* work around various missing functions */
709 886
710#if !HAVE_PREADWRITE 887#if !HAVE_PREADWRITE
888# undef pread
889# undef pwrite
711# define pread eio__pread 890# define pread eio__pread
712# define pwrite eio__pwrite 891# define pwrite eio__pwrite
713 892
714static ssize_t 893static eio_ssize_t
715eio__pread (int fd, void *buf, size_t count, off_t offset) 894eio__pread (int fd, void *buf, size_t count, off_t offset)
716{ 895{
717 ssize_t res; 896 eio_ssize_t res;
718 off_t ooffset; 897 off_t ooffset;
719 898
720 X_LOCK (preadwritelock); 899 X_LOCK (preadwritelock);
721 ooffset = lseek (fd, 0, SEEK_CUR); 900 ooffset = lseek (fd, 0, SEEK_CUR);
722 lseek (fd, offset, SEEK_SET); 901 lseek (fd, offset, SEEK_SET);
725 X_UNLOCK (preadwritelock); 904 X_UNLOCK (preadwritelock);
726 905
727 return res; 906 return res;
728} 907}
729 908
730static ssize_t 909static eio_ssize_t
731eio__pwrite (int fd, void *buf, size_t count, off_t offset) 910eio__pwrite (int fd, void *buf, size_t count, off_t offset)
732{ 911{
733 ssize_t res; 912 eio_ssize_t res;
734 off_t ooffset; 913 off_t ooffset;
735 914
736 X_LOCK (preadwritelock); 915 X_LOCK (preadwritelock);
737 ooffset = lseek (fd, 0, SEEK_CUR); 916 ooffset = lseek (fd, 0, SEEK_CUR);
738 lseek (fd, offset, SEEK_SET); 917 lseek (fd, offset, SEEK_SET);
739 res = write (fd, buf, count); 918 res = write (fd, buf, count);
740 lseek (fd, offset, SEEK_SET); 919 lseek (fd, ooffset, SEEK_SET);
741 X_UNLOCK (preadwritelock); 920 X_UNLOCK (preadwritelock);
742 921
743 return res; 922 return res;
744} 923}
745#endif 924#endif
746 925
747#ifndef HAVE_FUTIMES 926#ifndef HAVE_UTIMES
748 927
928# undef utimes
749# define utimes(path,times) eio__utimes (path, times) 929# define utimes(path,times) eio__utimes (path, times)
750# define futimes(fd,times) eio__futimes (fd, times)
751 930
752static int 931static int
753eio__utimes (const char *filename, const struct timeval times[2]) 932eio__utimes (const char *filename, const struct timeval times[2])
754{ 933{
755 if (times) 934 if (times)
763 } 942 }
764 else 943 else
765 return utime (filename, 0); 944 return utime (filename, 0);
766} 945}
767 946
947#endif
948
949#ifndef HAVE_FUTIMES
950
951# undef futimes
952# define futimes(fd,times) eio__futimes (fd, times)
953
954static int
768static int eio__futimes (int fd, const struct timeval tv[2]) 955eio__futimes (int fd, const struct timeval tv[2])
769{ 956{
770 errno = ENOSYS; 957 errno = ENOSYS;
771 return -1; 958 return -1;
772} 959}
773 960
774#endif 961#endif
775 962
776#if !HAVE_FDATASYNC 963#if !HAVE_FDATASYNC
964# undef fdatasync
777# define fdatasync fsync 965# define fdatasync(fd) fsync (fd)
778#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}
779 1006
780#if !HAVE_READAHEAD 1007#if !HAVE_READAHEAD
1008# undef readahead
781# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 1009# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
782 1010
783static ssize_t 1011static eio_ssize_t
784eio__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)
785{ 1013{
786 size_t todo = count; 1014 size_t todo = count;
787 dBUF; 1015 dBUF;
788 1016
800} 1028}
801 1029
802#endif 1030#endif
803 1031
804/* sendfile always needs emulation */ 1032/* sendfile always needs emulation */
805static ssize_t 1033static eio_ssize_t
806eio__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)
807{ 1035{
1036 eio_ssize_t written = 0;
808 ssize_t res; 1037 eio_ssize_t res;
809 1038
810 if (!count) 1039 if (!count)
811 return 0; 1040 return 0;
812 1041
1042 for (;;)
1043 {
1044#ifdef __APPLE__
1045# undef HAVE_SENDFILE /* broken, as everything on os x */
1046#endif
813#if HAVE_SENDFILE 1047#if HAVE_SENDFILE
814# if __linux 1048# if __linux
1049 off_t soffset = offset;
815 res = sendfile (ofd, ifd, &offset, count); 1050 res = sendfile (ofd, ifd, &soffset, count);
816 1051
817# elif __freebsd 1052# elif __FreeBSD__
818 /* 1053 /*
819 * 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
820 * wasted on making it similar to other I/O functions. 1055 * wasted on making it similar to other I/O functions.
821 */ 1056 */
822 {
823 off_t sbytes; 1057 off_t sbytes;
824 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 1058 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
825 1059
826 if (res < 0 && sbytes) 1060 #if 0 /* according to the manpage, this is correct, but broken behaviour */
827 /* 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)
828 res = sbytes; 1065 res = sbytes;
829 } 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;
830 1079
831# elif __hpux 1080# elif __hpux
832 res = sendfile (ofd, ifd, offset, count, 0, 0); 1081 res = sendfile (ofd, ifd, offset, count, 0, 0);
833 1082
834# elif __solaris 1083# elif __solaris
835 {
836 struct sendfilevec vec; 1084 struct sendfilevec vec;
837 size_t sbytes; 1085 size_t sbytes;
838 1086
839 vec.sfv_fd = ifd; 1087 vec.sfv_fd = ifd;
840 vec.sfv_flag = 0; 1088 vec.sfv_flag = 0;
841 vec.sfv_off = offset; 1089 vec.sfv_off = offset;
842 vec.sfv_len = count; 1090 vec.sfv_len = count;
843 1091
844 res = sendfilev (ofd, &vec, 1, &sbytes); 1092 res = sendfilev (ofd, &vec, 1, &sbytes);
845 1093
846 if (res < 0 && sbytes) 1094 if (res < 0 && sbytes)
847 res = sbytes; 1095 res = sbytes;
848 }
849 1096
850# 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
851#else 1108#else
852 res = -1; 1109 res = -1;
853 errno = ENOSYS; 1110 errno = ENOSYS;
854#endif 1111#endif
855 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
856 if (res < 0 1136 if (res < 0
857 && (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
858#if __solaris 1145#if __solaris
859 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1146 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
860#endif 1147#endif
861 ) 1148 )
862 ) 1149 )
866 1153
867 res = 0; 1154 res = 0;
868 1155
869 while (count) 1156 while (count)
870 { 1157 {
871 ssize_t cnt; 1158 eio_ssize_t cnt;
872 1159
873 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);
874 1161
875 if (cnt <= 0) 1162 if (cnt <= 0)
876 { 1163 {
893 } 1180 }
894 1181
895 return res; 1182 return res;
896} 1183}
897 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
898/* read a full directory */ 1629/* read a full directory */
899static void 1630static void
900eio__scandir (eio_req *req, etp_worker *self) 1631eio__scandir (eio_req *req, etp_worker *self)
901{ 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
902 DIR *dirp; 1645 DIR *dirp;
903 EIO_STRUCT_DIRENT *entp; 1646 EIO_STRUCT_DIRENT *entp;
904 char *name, *names; 1647#endif
905 int memlen = 4096;
906 int memofs = 0;
907 int res = 0;
908 1648
909 X_LOCK (wrklock); 1649 req->result = -1;
910 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 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
911 self->dirp = dirp = opendir (req->ptr1); 1686 dirp = opendir (req->ptr1);
1687#endif
1688
912 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;
913 req->ptr2 = names = malloc (memlen); 1694 req->ptr2 = names = malloc (namesalloc);
914 X_UNLOCK (wrklock);
915 1695
916 if (dirp && names) 1696 if (dirp && names && (!flags || dents))
917 for (;;) 1697 for (;;)
918 { 1698 {
1699 int more;
1700
1701#ifdef _WIN32
1702 more = dirp;
1703#else
919 errno = 0; 1704 errno = 0;
920 entp = readdir (dirp); 1705 entp = readdir (dirp);
1706 more = entp;
1707#endif
921 1708
922 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
923 break; 1753 break;
1754 }
924 1755
925 name = entp->d_name; 1756 /* now add the entry to our list(s) */
1757 name = D_NAME (entp);
926 1758
1759 /* skip . and .. entries */
927 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1760 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
928 { 1761 {
929 int len = strlen (name) + 1; 1762 int len = D_NAMLEN (entp) + 1;
930 1763
931 res++; 1764 while (ecb_expect_false (namesoffs + len > namesalloc))
932
933 while (memofs + len > memlen)
934 { 1765 {
935 memlen *= 2; 1766 namesalloc *= 2;
936 X_LOCK (wrklock);
937 req->ptr2 = names = realloc (names, memlen); 1767 req->ptr2 = names = realloc (names, namesalloc);
938 X_UNLOCK (wrklock);
939 1768
940 if (!names) 1769 if (!names)
941 break; 1770 break;
942 } 1771 }
943 1772
944 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
945 memofs += len; 1863 namesoffs += len;
1864 ++dentoffs;
946 } 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
947 } 1880 }
948
949 if (errno)
950 res = -1;
951
952 req->result = res;
953} 1881}
954 1882
955/*****************************************************************************/ 1883/*****************************************************************************/
956 1884
957#define ALLOC(len) \ 1885#define ALLOC(len) \
973{ 1901{
974 ETP_REQ *req; 1902 ETP_REQ *req;
975 struct timespec ts; 1903 struct timespec ts;
976 etp_worker *self = (etp_worker *)thr_arg; 1904 etp_worker *self = (etp_worker *)thr_arg;
977 1905
978 /* try to distribute timeouts somewhat randomly */ 1906 /* try to distribute timeouts somewhat evenly */
979 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 1907 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
980 1908
981 for (;;) 1909 for (;;)
982 { 1910 {
1911 ts.tv_sec = 0;
1912
983 X_LOCK (reqlock); 1913 X_LOCK (reqlock);
984 1914
985 for (;;) 1915 for (;;)
986 { 1916 {
987 self->req = req = reqq_shift (&req_queue); 1917 self->req = req = reqq_shift (&req_queue);
988 1918
989 if (req) 1919 if (req)
990 break; 1920 break;
991 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
992 ++idle; 1931 ++idle;
993 1932
994 ts.tv_sec = time (0) + IDLE_TIMEOUT; 1933 if (idle <= max_idle)
995 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
996 { 1937 {
997 if (idle > max_idle) 1938 /* initialise timeout once */
998 { 1939 if (!ts.tv_sec)
999 --idle; 1940 ts.tv_sec = time (0) + idle_timeout;
1000 X_UNLOCK (reqlock);
1001 X_LOCK (wrklock);
1002 --started;
1003 X_UNLOCK (wrklock);
1004 goto quit;
1005 }
1006 1941
1007 /* we are allowed to idle, so do so without any timeout */
1008 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.,.. */
1009 } 1944 }
1010 1945
1011 --idle; 1946 --idle;
1012 } 1947 }
1013 1948
1016 X_UNLOCK (reqlock); 1951 X_UNLOCK (reqlock);
1017 1952
1018 if (req->type < 0) 1953 if (req->type < 0)
1019 goto quit; 1954 goto quit;
1020 1955
1021 if (!EIO_CANCELLED (req))
1022 ETP_EXECUTE (self, req); 1956 ETP_EXECUTE (self, req);
1023 1957
1024 X_LOCK (reslock); 1958 X_LOCK (reslock);
1025 1959
1026 ++npending; 1960 ++npending;
1027 1961
1042 return 0; 1976 return 0;
1043} 1977}
1044 1978
1045/*****************************************************************************/ 1979/*****************************************************************************/
1046 1980
1981int ecb_cold
1047int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1982eio_init (void (*want_poll)(void), void (*done_poll)(void))
1048{ 1983{
1049 return etp_init (want_poll, done_poll); 1984 return etp_init (want_poll, done_poll);
1050} 1985}
1051 1986
1987ecb_inline void
1052static void eio_api_destroy (eio_req *req) 1988eio_api_destroy (eio_req *req)
1053{ 1989{
1054 free (req); 1990 free (req);
1055} 1991}
1056 1992
1057#define REQ(rtype) \ 1993#define REQ(rtype) \
1076 { \ 2012 { \
1077 eio_api_destroy (req); \ 2013 eio_api_destroy (req); \
1078 return 0; \ 2014 return 0; \
1079 } 2015 }
1080 2016
2017static void
1081static void eio_execute (etp_worker *self, eio_req *req) 2018eio_execute (etp_worker *self, eio_req *req)
1082{ 2019{
1083 errno = 0; 2020 if (ecb_expect_false (EIO_CANCELLED (req)))
2021 {
2022 req->result = -1;
2023 req->errorno = ECANCELED;
2024 return;
2025 }
1084 2026
1085 switch (req->type) 2027 switch (req->type)
1086 { 2028 {
1087 case EIO_READ: ALLOC (req->size); 2029 case EIO_READ: ALLOC (req->size);
1088 req->result = req->offs >= 0 2030 req->result = req->offs >= 0
1100 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2042 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1101 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2043 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1102 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2044 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1103 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; 2045 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1104 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
1105 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;
1106 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;
1107 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;
1108 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;
1109 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break; 2056 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1116 case EIO_RMDIR: req->result = rmdir (req->ptr1); break; 2063 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1117 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;
1118 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; 2065 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1119 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; 2066 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1120 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; 2067 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1121 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;
1122 2069
2070 case EIO_REALPATH: eio__realpath (req, self); break;
2071
1123 case EIO_READLINK: ALLOC (NAME_MAX); 2072 case EIO_READLINK: ALLOC (PATH_MAX);
1124 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; 2073 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1125 2074
1126 case EIO_SYNC: req->result = 0; sync (); break; 2075 case EIO_SYNC: req->result = 0; sync (); break;
1127 case EIO_FSYNC: req->result = fsync (req->int1); break; 2076 case EIO_FSYNC: req->result = fsync (req->int1); break;
1128 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;
1129 2084
1130 case EIO_READDIR: eio__scandir (req, self); break; 2085 case EIO_READDIR: eio__scandir (req, self); break;
1131 2086
1132 case EIO_BUSY: 2087 case EIO_BUSY:
1133#ifdef _WIN32 2088#ifdef _WIN32
1134 Sleep (req->nv1 * 1000.); 2089 Sleep (req->nv1 * 1e3);
1135#else 2090#else
1136 { 2091 {
1137 struct timeval tv; 2092 struct timeval tv;
1138 2093
1139 tv.tv_sec = req->nv1; 2094 tv.tv_sec = req->nv1;
1140 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; 2095 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1141 2096
1142 req->result = select (0, 0, 0, 0, &tv); 2097 req->result = select (0, 0, 0, 0, &tv);
1143 } 2098 }
1144#endif 2099#endif
1145 break; 2100 break;
1160 times = tv; 2115 times = tv;
1161 } 2116 }
1162 else 2117 else
1163 times = 0; 2118 times = 0;
1164 2119
1165
1166 req->result = req->type == EIO_FUTIME 2120 req->result = req->type == EIO_FUTIME
1167 ? futimes (req->int1, times) 2121 ? futimes (req->int1, times)
1168 : utimes (req->ptr1, times); 2122 : utimes (req->ptr1, times);
1169 } 2123 }
2124 break;
1170 2125
1171 case EIO_GROUP: 2126 case EIO_GROUP:
2127 abort (); /* handled in eio_request */
2128
1172 case EIO_NOP: 2129 case EIO_NOP:
1173 req->result = 0; 2130 req->result = 0;
1174 break; 2131 break;
1175 2132
1176 case EIO_CUSTOM: 2133 case EIO_CUSTOM:
1177 req->feed (req); 2134 req->feed (req);
1178 break; 2135 break;
1179 2136
1180 default: 2137 default:
2138 errno = ENOSYS;
1181 req->result = -1; 2139 req->result = -1;
1182 break; 2140 break;
1183 } 2141 }
1184 2142
1185 req->errorno = errno; 2143 req->errorno = errno;
1205eio_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)
1206{ 2164{
1207 REQ (EIO_FSYNC); req->int1 = fd; SEND; 2165 REQ (EIO_FSYNC); req->int1 = fd; SEND;
1208} 2166}
1209 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
1210eio_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)
1211{ 2199{
1212 REQ (EIO_FDATASYNC); req->int1 = fd; SEND; 2200 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1213} 2201}
1214 2202
1235eio_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)
1236{ 2224{
1237 REQ (EIO_FSTAT); req->int1 = fd; SEND; 2225 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1238} 2226}
1239 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
1240eio_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)
1241{ 2234{
1242 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;
1243} 2236}
1244 2237
1250eio_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)
1251{ 2244{
1252 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2245 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1253} 2246}
1254 2247
1255eio_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)
1256{ 2249{
1257 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;
1258} 2251}
1259 2252
1260eio_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)
1280eio_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)
1281{ 2274{
1282 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2275 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1283} 2276}
1284 2277
1285eio_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)
1286{ 2279{
1287 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;
1288} 2281}
1289 2282
1290eio_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)
1306eio_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)
1307{ 2300{
1308 return eio__1path (EIO_READLINK, path, pri, cb, data); 2301 return eio__1path (EIO_READLINK, path, pri, cb, data);
1309} 2302}
1310 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
1311eio_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)
1312{ 2310{
1313 return eio__1path (EIO_STAT, path, pri, cb, data); 2311 return eio__1path (EIO_STAT, path, pri, cb, data);
1314} 2312}
1315 2313
1316eio_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)
1317{ 2315{
1318 return eio__1path (EIO_LSTAT, path, pri, cb, data); 2316 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1319} 2317}
1320 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
1321eio_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)
1322{ 2325{
1323 return eio__1path (EIO_UNLINK, path, pri, cb, data); 2326 return eio__1path (EIO_UNLINK, path, pri, cb, data);
1324} 2327}
1325 2328
1326eio_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)
1327{ 2330{
1328 return eio__1path (EIO_RMDIR, path, pri, cb, data); 2331 return eio__1path (EIO_RMDIR, path, pri, cb, data);
1329} 2332}
1330 2333
1331eio_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)
1332{ 2335{
1333 return eio__1path (EIO_READDIR, path, pri, cb, data); 2336 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1334} 2337}
1335 2338
1336eio_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)
1337{ 2340{
1338 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;
1339} 2342}
1340 2343
1341static eio_req * 2344static eio_req *
1342eio__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)
1343{ 2346{
1367eio_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)
1368{ 2371{
1369 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2372 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1370} 2373}
1371 2374
1372eio_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)
1373{ 2376{
1374 REQ (EIO_CUSTOM); req->feed = execute; SEND; 2377 REQ (EIO_CUSTOM); req->feed = execute; SEND;
1375} 2378}
1376 2379
1377#endif 2380#endif
1388#undef SEND 2391#undef SEND
1389 2392
1390/*****************************************************************************/ 2393/*****************************************************************************/
1391/* grp functions */ 2394/* grp functions */
1392 2395
2396void
1393void 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)
1394{ 2398{
1395 grp->int2 = limit; 2399 grp->int2 = limit;
1396 grp->feed = feed; 2400 grp->feed = feed;
1397 2401
1398 grp_try_feed (grp); 2402 grp_try_feed (grp);
1399} 2403}
1400 2404
2405void
1401void eio_grp_limit (eio_req *grp, int limit) 2406eio_grp_limit (eio_req *grp, int limit)
1402{ 2407{
1403 grp->int2 = limit; 2408 grp->int2 = limit;
1404 2409
1405 grp_try_feed (grp); 2410 grp_try_feed (grp);
1406} 2411}
1407 2412
2413void
1408void eio_grp_add (eio_req *grp, eio_req *req) 2414eio_grp_add (eio_req *grp, eio_req *req)
1409{ 2415{
1410 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;
1411 2419
1412 ++grp->size; 2420 ++grp->size;
1413 req->grp = grp; 2421 req->grp = grp;
1414 2422
1415 req->grp_prev = 0; 2423 req->grp_prev = 0;
1422} 2430}
1423 2431
1424/*****************************************************************************/ 2432/*****************************************************************************/
1425/* misc garbage */ 2433/* misc garbage */
1426 2434
2435eio_ssize_t
1427ssize_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)
1428{ 2437{
1429 etp_worker wrk; 2438 etp_worker wrk;
2439 eio_ssize_t ret;
1430 2440
1431 wrk.dbuf = 0; 2441 wrk.dbuf = 0;
1432 2442
1433 eio__sendfile (ofd, ifd, offset, count, &wrk); 2443 ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
1434 2444
1435 if (wrk.dbuf) 2445 if (wrk.dbuf)
1436 free (wrk.dbuf); 2446 free (wrk.dbuf);
1437}
1438 2447
2448 return ret;
2449}
2450

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines