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

Comparing libeio/eio.c (file contents):
Revision 1.15 by root, Sat May 17 12:26:57 2008 UTC vs.
Revision 1.89 by root, Thu Jul 14 22:36:17 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
73static void eio_destroy (eio_req *req);
74
54#ifndef EIO_FINISH 75#ifndef EIO_FINISH
55# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0 76# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
56#endif 77#endif
57 78
58#ifndef EIO_DESTROY 79#ifndef EIO_DESTROY
61 82
62#ifndef EIO_FEED 83#ifndef EIO_FEED
63# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0) 84# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)
64#endif 85#endif
65 86
87#ifndef EIO_FD_TO_WIN32_HANDLE
88# define EIO_FD_TO_WIN32_HANDLE(fd) _get_osfhandle (fd)
89#endif
90#ifndef EIO_WIN32_HANDLE_TO_FD
91# define EIO_WIN32_HANDLE_TO_FD(handle) _open_osfhandle (handle, 0)
92#endif
93
94#define EIO_ERRNO(errval,retval) ((errno = errval), retval)
95
96#define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1)
97
66#ifdef _WIN32 98#ifdef _WIN32
67 99
68 /*doh*/ 100 #define PAGESIZE 4096 /* GetSystemInfo? */
101
102 #ifdef EIO_STRUCT_STATI64
103 #define stat(path,buf) _stati64 (path,buf)
104 #define fstat(fd,buf) _fstati64 (path,buf)
105 #endif
106 #define lstat(path,buf) stat (path,buf)
107 #define fsync(fd) (FlushFileBuffers (EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
108 #define mkdir(path,mode) _mkdir (path)
109 #define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
110
111 #define chown(path,uid,gid) EIO_ENOSYS ()
112 #define fchown(fd,uid,gid) EIO_ENOSYS ()
113 #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
114 #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
115 #define mknod(path,mode,dev) EIO_ENOSYS ()
116 #define sync() EIO_ENOSYS ()
117
118 /* we could even stat and see if it exists */
119 static int
120 symlink (const char *old, const char *neu)
121 {
122 if (CreateSymbolicLink (neu, old, 1))
123 return 0;
124
125 if (CreateSymbolicLink (neu, old, 0))
126 return 0;
127
128 return EIO_ERRNO (ENOENT, -1);
129 }
69 130
70#else 131#else
71 132
72# include "config.h"
73# include <sys/time.h> 133 #include <sys/time.h>
74# include <sys/select.h> 134 #include <sys/select.h>
135 #include <sys/statvfs.h>
75# include <unistd.h> 136 #include <unistd.h>
76# include <utime.h> 137 #include <utime.h>
77# include <signal.h> 138 #include <signal.h>
78# include <dirent.h> 139 #include <dirent.h>
79 140
141 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
142 #include <sys/mman.h>
143 #endif
144
145 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
146 #if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
147 #define _DIRENT_HAVE_D_TYPE /* sigh */
148 #define D_INO(de) (de)->d_fileno
149 #define D_NAMLEN(de) (de)->d_namlen
150 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
151 #define D_INO(de) (de)->d_ino
152 #endif
153
154 #ifdef _D_EXACT_NAMLEN
155 #undef D_NAMLEN
156 #define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
157 #endif
158
159 #ifdef _DIRENT_HAVE_D_TYPE
160 #define D_TYPE(de) (de)->d_type
161 #endif
162
80# ifndef EIO_STRUCT_DIRENT 163 #ifndef EIO_STRUCT_DIRENT
81# define EIO_STRUCT_DIRENT struct dirent 164 #define EIO_STRUCT_DIRENT struct dirent
82# endif 165 #endif
83 166
84#endif 167#endif
85 168
86#if HAVE_SENDFILE 169#if HAVE_SENDFILE
87# if __linux 170# if __linux
88# include <sys/sendfile.h> 171# include <sys/sendfile.h>
89# elif __freebsd 172# elif __FreeBSD__ || defined __APPLE__
90# include <sys/socket.h> 173# include <sys/socket.h>
91# include <sys/uio.h> 174# include <sys/uio.h>
92# elif __hpux 175# elif __hpux
93# include <sys/socket.h> 176# include <sys/socket.h>
94# elif __solaris /* not yet */ 177# elif __solaris
95# include <sys/sendfile.h> 178# include <sys/sendfile.h>
96# else 179# else
97# error sendfile support requested but not available 180# error sendfile support requested but not available
98# endif 181# endif
99#endif 182#endif
100 183
101/* number of seconds after which an idle threads exit */ 184#ifndef D_TYPE
102#define IDLE_TIMEOUT 10 185# define D_TYPE(de) 0
186#endif
187#ifndef D_INO
188# define D_INO(de) 0
189#endif
190#ifndef D_NAMLEN
191# define D_NAMLEN(de) strlen ((de)->d_name)
192#endif
103 193
104/* used for struct dirent, AIX doesn't provide it */ 194/* used for struct dirent, AIX doesn't provide it */
105#ifndef NAME_MAX 195#ifndef NAME_MAX
106# define NAME_MAX 4096 196# define NAME_MAX 4096
197#endif
198
199/* used for readlink etc. */
200#ifndef PATH_MAX
201# define PATH_MAX 4096
107#endif 202#endif
108 203
109/* buffer size for various temporary buffers */ 204/* buffer size for various temporary buffers */
110#define EIO_BUFSIZE 65536 205#define EIO_BUFSIZE 65536
111 206
117 errno = ENOMEM; \ 212 errno = ENOMEM; \
118 if (!eio_buf) \ 213 if (!eio_buf) \
119 return -1; 214 return -1;
120 215
121#define EIO_TICKS ((1000000 + 1023) >> 10) 216#define EIO_TICKS ((1000000 + 1023) >> 10)
122
123/*****************************************************************************/
124 217
125#define ETP_PRI_MIN EIO_PRI_MIN 218#define ETP_PRI_MIN EIO_PRI_MIN
126#define ETP_PRI_MAX EIO_PRI_MAX 219#define ETP_PRI_MAX EIO_PRI_MAX
127 220
128struct etp_worker; 221struct etp_worker;
144 if (wrk->dirp) \ 237 if (wrk->dirp) \
145 { \ 238 { \
146 closedir (wrk->dirp); \ 239 closedir (wrk->dirp); \
147 wrk->dirp = 0; \ 240 wrk->dirp = 0; \
148 } 241 }
242
149#define ETP_WORKER_COMMON \ 243#define ETP_WORKER_COMMON \
150 void *dbuf; \ 244 void *dbuf; \
151 DIR *dirp; 245 DIR *dirp;
152 246
153/*****************************************************************************/ 247/*****************************************************************************/
154 248
155#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) 249#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
156 250
157/* calculcate time difference in ~1/EIO_TICKS of a second */ 251/* calculate time difference in ~1/EIO_TICKS of a second */
252ecb_inline int
158static int tvdiff (struct timeval *tv1, struct timeval *tv2) 253tvdiff (struct timeval *tv1, struct timeval *tv2)
159{ 254{
160 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 255 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
161 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 256 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
162} 257}
163 258
170static unsigned int max_poll_reqs; /* reslock */ 265static unsigned int max_poll_reqs; /* reslock */
171 266
172static volatile unsigned int nreqs; /* reqlock */ 267static volatile unsigned int nreqs; /* reqlock */
173static volatile unsigned int nready; /* reqlock */ 268static volatile unsigned int nready; /* reqlock */
174static volatile unsigned int npending; /* reqlock */ 269static volatile unsigned int npending; /* reqlock */
175static volatile unsigned int max_idle = 4; 270static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
271static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
176 272
273static xmutex_t wrklock;
274static xmutex_t reslock;
275static xmutex_t reqlock;
276static xcond_t reqwait;
277
278#if !HAVE_PREADWRITE
279/*
280 * make our pread/pwrite emulation safe against themselves, but not against
281 * normal read/write by using a mutex. slows down execution a lot,
282 * but that's your problem, not mine.
283 */
177static mutex_t wrklock = X_MUTEX_INIT; 284static xmutex_t preadwritelock = X_MUTEX_INIT;
178static mutex_t reslock = X_MUTEX_INIT; 285#endif
179static mutex_t reqlock = X_MUTEX_INIT;
180static cond_t reqwait = X_COND_INIT;
181 286
182typedef struct etp_worker 287typedef struct etp_worker
183{ 288{
184 /* locked by wrklock */ 289 /* locked by wrklock */
185 struct etp_worker *prev, *next; 290 struct etp_worker *prev, *next;
186 291
187 thread_t tid; 292 xthread_t tid;
188 293
189 /* locked by reslock, reqlock or wrklock */ 294 /* locked by reslock, reqlock or wrklock */
190 ETP_REQ *req; /* currently processed request */ 295 ETP_REQ *req; /* currently processed request */
191 296
192 ETP_WORKER_COMMON 297 ETP_WORKER_COMMON
197#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock) 302#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
198#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock) 303#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
199 304
200/* worker threads management */ 305/* worker threads management */
201 306
307static void ecb_cold
202static void etp_worker_clear (etp_worker *wrk) 308etp_worker_clear (etp_worker *wrk)
203{ 309{
204 ETP_WORKER_CLEAR (wrk); 310 ETP_WORKER_CLEAR (wrk);
205} 311}
206 312
313static void ecb_cold
207static void etp_worker_free (etp_worker *wrk) 314etp_worker_free (etp_worker *wrk)
208{ 315{
209 wrk->next->prev = wrk->prev; 316 wrk->next->prev = wrk->prev;
210 wrk->prev->next = wrk->next; 317 wrk->prev->next = wrk->next;
211 318
212 free (wrk); 319 free (wrk);
213} 320}
214 321
215static unsigned int etp_nreqs (void) 322static unsigned int
323etp_nreqs (void)
216{ 324{
217 int retval; 325 int retval;
218 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 326 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
219 retval = nreqs; 327 retval = nreqs;
220 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 328 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
221 return retval; 329 return retval;
222} 330}
223 331
224static unsigned int etp_nready (void) 332static unsigned int
333etp_nready (void)
225{ 334{
226 unsigned int retval; 335 unsigned int retval;
227 336
228 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 337 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
229 retval = nready; 338 retval = nready;
230 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 339 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
231 340
232 return retval; 341 return retval;
233} 342}
234 343
235static unsigned int etp_npending (void) 344static unsigned int
345etp_npending (void)
236{ 346{
237 unsigned int retval; 347 unsigned int retval;
238 348
239 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 349 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
240 retval = npending; 350 retval = npending;
241 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 351 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
242 352
243 return retval; 353 return retval;
244} 354}
245 355
246static unsigned int etp_nthreads (void) 356static unsigned int
357etp_nthreads (void)
247{ 358{
248 unsigned int retval; 359 unsigned int retval;
249 360
250 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 361 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
251 retval = started; 362 retval = started;
265} etp_reqq; 376} etp_reqq;
266 377
267static etp_reqq req_queue; 378static etp_reqq req_queue;
268static etp_reqq res_queue; 379static etp_reqq res_queue;
269 380
381static int ecb_noinline
270static int reqq_push (etp_reqq *q, ETP_REQ *req) 382reqq_push (etp_reqq *q, ETP_REQ *req)
271{ 383{
272 int pri = req->pri; 384 int pri = req->pri;
273 req->next = 0; 385 req->next = 0;
274 386
275 if (q->qe[pri]) 387 if (q->qe[pri])
281 q->qe[pri] = q->qs[pri] = req; 393 q->qe[pri] = q->qs[pri] = req;
282 394
283 return q->size++; 395 return q->size++;
284} 396}
285 397
398static ETP_REQ * ecb_noinline
286static ETP_REQ *reqq_shift (etp_reqq *q) 399reqq_shift (etp_reqq *q)
287{ 400{
288 int pri; 401 int pri;
289 402
290 if (!q->size) 403 if (!q->size)
291 return 0; 404 return 0;
306 } 419 }
307 420
308 abort (); 421 abort ();
309} 422}
310 423
311static void etp_atfork_prepare (void) 424static void ecb_cold
425etp_thread_init (void)
312{ 426{
313 X_LOCK (wrklock);
314 X_LOCK (reqlock);
315 X_LOCK (reslock);
316#if !HAVE_PREADWRITE 427#if !HAVE_PREADWRITE
317 X_LOCK (preadwritelock); 428 X_MUTEX_CREATE (preadwritelock);
318#endif 429#endif
319#if !HAVE_READDIR_R 430 X_MUTEX_CREATE (wrklock);
320 X_LOCK (readdirlock); 431 X_MUTEX_CREATE (reslock);
321#endif 432 X_MUTEX_CREATE (reqlock);
433 X_COND_CREATE (reqwait);
322} 434}
323 435
436static void ecb_cold
437etp_atfork_prepare (void)
438{
439}
440
441static void ecb_cold
324static void etp_atfork_parent (void) 442etp_atfork_parent (void)
325{ 443{
326#if !HAVE_READDIR_R
327 X_UNLOCK (readdirlock);
328#endif
329#if !HAVE_PREADWRITE
330 X_UNLOCK (preadwritelock);
331#endif
332 X_UNLOCK (reslock);
333 X_UNLOCK (reqlock);
334 X_UNLOCK (wrklock);
335} 444}
336 445
446static void ecb_cold
337static void etp_atfork_child (void) 447etp_atfork_child (void)
338{ 448{
339 ETP_REQ *prv; 449 ETP_REQ *prv;
340 450
341 while (prv = reqq_shift (&req_queue)) 451 while ((prv = reqq_shift (&req_queue)))
342 ETP_DESTROY (prv); 452 ETP_DESTROY (prv);
343 453
344 while (prv = reqq_shift (&res_queue)) 454 while ((prv = reqq_shift (&res_queue)))
345 ETP_DESTROY (prv); 455 ETP_DESTROY (prv);
346 456
347 while (wrk_first.next != &wrk_first) 457 while (wrk_first.next != &wrk_first)
348 { 458 {
349 etp_worker *wrk = wrk_first.next; 459 etp_worker *wrk = wrk_first.next;
359 idle = 0; 469 idle = 0;
360 nreqs = 0; 470 nreqs = 0;
361 nready = 0; 471 nready = 0;
362 npending = 0; 472 npending = 0;
363 473
364 etp_atfork_parent (); 474 etp_thread_init ();
365} 475}
366 476
367static void 477static void ecb_cold
368etp_once_init (void) 478etp_once_init (void)
369{ 479{
480 etp_thread_init ();
370 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child); 481 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
371} 482}
372 483
373static int 484static int ecb_cold
374etp_init (void (*want_poll)(void), void (*done_poll)(void)) 485etp_init (void (*want_poll)(void), void (*done_poll)(void))
375{ 486{
376 static pthread_once_t doinit = PTHREAD_ONCE_INIT; 487 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
377 488
378 pthread_once (&doinit, etp_once_init); 489 pthread_once (&doinit, etp_once_init);
379 490
380 want_poll_cb = want_poll; 491 want_poll_cb = want_poll;
381 done_poll_cb = done_poll; 492 done_poll_cb = done_poll;
493
494 return 0;
382} 495}
383 496
384X_THREAD_PROC (etp_proc); 497X_THREAD_PROC (etp_proc);
385 498
499static void ecb_cold
386static void etp_start_thread (void) 500etp_start_thread (void)
387{ 501{
388 etp_worker *wrk = calloc (1, sizeof (etp_worker)); 502 etp_worker *wrk = calloc (1, sizeof (etp_worker));
389 503
390 /*TODO*/ 504 /*TODO*/
391 assert (("unable to allocate worker thread data", wrk)); 505 assert (("unable to allocate worker thread data", wrk));
404 free (wrk); 518 free (wrk);
405 519
406 X_UNLOCK (wrklock); 520 X_UNLOCK (wrklock);
407} 521}
408 522
523static void
409static void etp_maybe_start_thread (void) 524etp_maybe_start_thread (void)
410{ 525{
411 if (etp_nthreads () >= wanted) 526 if (ecb_expect_true (etp_nthreads () >= wanted))
412 return; 527 return;
413 528
414 /* todo: maybe use idle here, but might be less exact */ 529 /* todo: maybe use idle here, but might be less exact */
415 if (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()) 530 if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
416 return; 531 return;
417 532
418 etp_start_thread (); 533 etp_start_thread ();
419} 534}
420 535
536static void ecb_cold
421static void etp_end_thread (void) 537etp_end_thread (void)
422{ 538{
423 eio_req *req = calloc (1, sizeof (eio_req)); 539 eio_req *req = calloc (1, sizeof (eio_req));
424 540
425 req->type = -1; 541 req->type = -1;
426 req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 542 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
433 X_LOCK (wrklock); 549 X_LOCK (wrklock);
434 --started; 550 --started;
435 X_UNLOCK (wrklock); 551 X_UNLOCK (wrklock);
436} 552}
437 553
438static int etp_poll (void) 554static int
555etp_poll (void)
439{ 556{
440 unsigned int maxreqs; 557 unsigned int maxreqs;
441 unsigned int maxtime; 558 unsigned int maxtime;
442 struct timeval tv_start, tv_now; 559 struct timeval tv_start, tv_now;
443 560
473 590
474 X_LOCK (reqlock); 591 X_LOCK (reqlock);
475 --nreqs; 592 --nreqs;
476 X_UNLOCK (reqlock); 593 X_UNLOCK (reqlock);
477 594
478 if (req->type == EIO_GROUP && req->size) 595 if (ecb_expect_false (req->type == EIO_GROUP && req->size))
479 { 596 {
480 req->int1 = 1; /* mark request as delayed */ 597 req->int1 = 1; /* mark request as delayed */
481 continue; 598 continue;
482 } 599 }
483 else 600 else
484 { 601 {
485 int res = ETP_FINISH (req); 602 int res = ETP_FINISH (req);
486 if (res) 603 if (ecb_expect_false (res))
487 return res; 604 return res;
488 } 605 }
489 606
490 if (maxreqs && !--maxreqs) 607 if (ecb_expect_false (maxreqs && !--maxreqs))
491 break; 608 break;
492 609
493 if (maxtime) 610 if (maxtime)
494 { 611 {
495 gettimeofday (&tv_now, 0); 612 gettimeofday (&tv_now, 0);
501 618
502 errno = EAGAIN; 619 errno = EAGAIN;
503 return -1; 620 return -1;
504} 621}
505 622
623static void
506static void etp_cancel (ETP_REQ *req) 624etp_cancel (ETP_REQ *req)
507{ 625{
508 X_LOCK (wrklock); 626 req->cancelled = 1;
509 req->flags |= EIO_FLAG_CANCELLED;
510 X_UNLOCK (wrklock);
511 627
512 eio_grp_cancel (req); 628 eio_grp_cancel (req);
513} 629}
514 630
631static void
515static void etp_submit (ETP_REQ *req) 632etp_submit (ETP_REQ *req)
516{ 633{
517 req->pri -= ETP_PRI_MIN; 634 req->pri -= ETP_PRI_MIN;
518 635
519 if (req->pri < ETP_PRI_MIN - ETP_PRI_MIN) req->pri = ETP_PRI_MIN - ETP_PRI_MIN; 636 if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
520 if (req->pri > ETP_PRI_MAX - ETP_PRI_MIN) req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 637 if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
521 638
639 if (ecb_expect_false (req->type == EIO_GROUP))
640 {
641 /* I hope this is worth it :/ */
522 X_LOCK (reqlock); 642 X_LOCK (reqlock);
523 ++nreqs; 643 ++nreqs;
644 X_UNLOCK (reqlock);
645
646 X_LOCK (reslock);
647
648 ++npending;
649
650 if (!reqq_push (&res_queue, req) && want_poll_cb)
651 want_poll_cb ();
652
653 X_UNLOCK (reslock);
654 }
655 else
656 {
657 X_LOCK (reqlock);
658 ++nreqs;
524 ++nready; 659 ++nready;
525 reqq_push (&req_queue, req); 660 reqq_push (&req_queue, req);
526 X_COND_SIGNAL (reqwait); 661 X_COND_SIGNAL (reqwait);
527 X_UNLOCK (reqlock); 662 X_UNLOCK (reqlock);
528 663
529 etp_maybe_start_thread (); 664 etp_maybe_start_thread ();
665 }
530} 666}
531 667
668static void ecb_cold
532static void etp_set_max_poll_time (double nseconds) 669etp_set_max_poll_time (double nseconds)
533{ 670{
534 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 671 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
535 max_poll_time = nseconds; 672 max_poll_time = nseconds * EIO_TICKS;
536 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 673 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
537} 674}
538 675
676static void ecb_cold
539static void etp_set_max_poll_reqs (unsigned int maxreqs) 677etp_set_max_poll_reqs (unsigned int maxreqs)
540{ 678{
541 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 679 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
542 max_poll_reqs = maxreqs; 680 max_poll_reqs = maxreqs;
543 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 681 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
544} 682}
545 683
684static void ecb_cold
546static void etp_set_max_idle (unsigned int nthreads) 685etp_set_max_idle (unsigned int nthreads)
547{ 686{
548 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 687 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
549 max_idle = nthreads <= 0 ? 1 : nthreads; 688 max_idle = nthreads;
550 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 689 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
551} 690}
552 691
692static void ecb_cold
693etp_set_idle_timeout (unsigned int seconds)
694{
695 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
696 idle_timeout = seconds;
697 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
698}
699
700static void ecb_cold
553static void etp_set_min_parallel (unsigned int nthreads) 701etp_set_min_parallel (unsigned int nthreads)
554{ 702{
555 if (wanted < nthreads) 703 if (wanted < nthreads)
556 wanted = nthreads; 704 wanted = nthreads;
557} 705}
558 706
707static void ecb_cold
559static void etp_set_max_parallel (unsigned int nthreads) 708etp_set_max_parallel (unsigned int nthreads)
560{ 709{
561 if (wanted > nthreads) 710 if (wanted > nthreads)
562 wanted = nthreads; 711 wanted = nthreads;
563 712
564 while (started > wanted) 713 while (started > wanted)
565 etp_end_thread (); 714 etp_end_thread ();
566} 715}
567 716
568/*****************************************************************************/ 717/*****************************************************************************/
569 718
719static void
570static void grp_try_feed (eio_req *grp) 720grp_try_feed (eio_req *grp)
571{ 721{
572 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 722 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
573 { 723 {
574 int old_len = grp->size; 724 grp->flags &= ~EIO_FLAG_GROUPADD;
575 725
576 EIO_FEED (grp); 726 EIO_FEED (grp);
577 727
578 /* stop if no progress has been made */ 728 /* stop if no progress has been made */
579 if (old_len == grp->size) 729 if (!(grp->flags & EIO_FLAG_GROUPADD))
580 { 730 {
581 grp->feed = 0; 731 grp->feed = 0;
582 break; 732 break;
583 } 733 }
584 } 734 }
585} 735}
586 736
737static int
587static int grp_dec (eio_req *grp) 738grp_dec (eio_req *grp)
588{ 739{
589 --grp->size; 740 --grp->size;
590 741
591 /* call feeder, if applicable */ 742 /* call feeder, if applicable */
592 grp_try_feed (grp); 743 grp_try_feed (grp);
596 return eio_finish (grp); 747 return eio_finish (grp);
597 else 748 else
598 return 0; 749 return 0;
599} 750}
600 751
752static void
601void eio_destroy (eio_req *req) 753eio_destroy (eio_req *req)
602{ 754{
603 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); 755 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
604 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); 756 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
605 757
606 EIO_DESTROY (req); 758 EIO_DESTROY (req);
607} 759}
608 760
761static int
609static int eio_finish (eio_req *req) 762eio_finish (eio_req *req)
610{ 763{
611 int res = EIO_FINISH (req); 764 int res = EIO_FINISH (req);
612 765
613 if (req->grp) 766 if (req->grp)
614 { 767 {
622 if (grp->grp_first == req) 775 if (grp->grp_first == req)
623 grp->grp_first = req->grp_next; 776 grp->grp_first = req->grp_next;
624 777
625 res2 = grp_dec (grp); 778 res2 = grp_dec (grp);
626 779
627 if (!res && res2) 780 if (!res)
628 res = res2; 781 res = res2;
629 } 782 }
630 783
631 eio_destroy (req); 784 eio_destroy (req);
632 785
633 return res; 786 return res;
634} 787}
635 788
789void
636void eio_grp_cancel (eio_req *grp) 790eio_grp_cancel (eio_req *grp)
637{ 791{
638 for (grp = grp->grp_first; grp; grp = grp->grp_next) 792 for (grp = grp->grp_first; grp; grp = grp->grp_next)
639 eio_cancel (grp); 793 eio_cancel (grp);
640} 794}
641 795
796void
642void eio_cancel (eio_req *req) 797eio_cancel (eio_req *req)
643{ 798{
644 etp_cancel (req); 799 etp_cancel (req);
645} 800}
646 801
802void
647void eio_submit (eio_req *req) 803eio_submit (eio_req *req)
648{ 804{
649 etp_submit (req); 805 etp_submit (req);
650} 806}
651 807
652unsigned int eio_nreqs (void) 808unsigned int
809eio_nreqs (void)
653{ 810{
654 return etp_nreqs (); 811 return etp_nreqs ();
655} 812}
656 813
657unsigned int eio_nready (void) 814unsigned int
815eio_nready (void)
658{ 816{
659 return etp_nready (); 817 return etp_nready ();
660} 818}
661 819
662unsigned int eio_npending (void) 820unsigned int
821eio_npending (void)
663{ 822{
664 return etp_npending (); 823 return etp_npending ();
665} 824}
666 825
667unsigned int eio_nthreads (void) 826unsigned int ecb_cold
827eio_nthreads (void)
668{ 828{
669 return etp_nthreads (); 829 return etp_nthreads ();
670} 830}
671 831
832void ecb_cold
672void eio_set_max_poll_time (double nseconds) 833eio_set_max_poll_time (double nseconds)
673{ 834{
674 etp_set_max_poll_time (nseconds); 835 etp_set_max_poll_time (nseconds);
675} 836}
676 837
838void ecb_cold
677void eio_set_max_poll_reqs (unsigned int maxreqs) 839eio_set_max_poll_reqs (unsigned int maxreqs)
678{ 840{
679 etp_set_max_poll_reqs (maxreqs); 841 etp_set_max_poll_reqs (maxreqs);
680} 842}
681 843
844void ecb_cold
682void eio_set_max_idle (unsigned int nthreads) 845eio_set_max_idle (unsigned int nthreads)
683{ 846{
684 etp_set_max_idle (nthreads); 847 etp_set_max_idle (nthreads);
685} 848}
686 849
850void ecb_cold
851eio_set_idle_timeout (unsigned int seconds)
852{
853 etp_set_idle_timeout (seconds);
854}
855
856void ecb_cold
687void eio_set_min_parallel (unsigned int nthreads) 857eio_set_min_parallel (unsigned int nthreads)
688{ 858{
689 etp_set_min_parallel (nthreads); 859 etp_set_min_parallel (nthreads);
690} 860}
691 861
862void ecb_cold
692void eio_set_max_parallel (unsigned int nthreads) 863eio_set_max_parallel (unsigned int nthreads)
693{ 864{
694 etp_set_max_parallel (nthreads); 865 etp_set_max_parallel (nthreads);
695} 866}
696 867
697int eio_poll (void) 868int eio_poll (void)
701 872
702/*****************************************************************************/ 873/*****************************************************************************/
703/* work around various missing functions */ 874/* work around various missing functions */
704 875
705#if !HAVE_PREADWRITE 876#if !HAVE_PREADWRITE
877# undef pread
878# undef pwrite
706# define pread eio__pread 879# define pread eio__pread
707# define pwrite eio__pwrite 880# define pwrite eio__pwrite
708 881
709/*
710 * make our pread/pwrite safe against themselves, but not against
711 * normal read/write by using a mutex. slows down execution a lot,
712 * but that's your problem, not mine.
713 */
714static mutex_t preadwritelock = X_MUTEX_INIT;
715
716static ssize_t 882static eio_ssize_t
717eio__pread (int fd, void *buf, size_t count, off_t offset) 883eio__pread (int fd, void *buf, size_t count, off_t offset)
718{ 884{
719 ssize_t res; 885 eio_ssize_t res;
720 off_t ooffset; 886 off_t ooffset;
721 887
722 X_LOCK (preadwritelock); 888 X_LOCK (preadwritelock);
723 ooffset = lseek (fd, 0, SEEK_CUR); 889 ooffset = lseek (fd, 0, SEEK_CUR);
724 lseek (fd, offset, SEEK_SET); 890 lseek (fd, offset, SEEK_SET);
727 X_UNLOCK (preadwritelock); 893 X_UNLOCK (preadwritelock);
728 894
729 return res; 895 return res;
730} 896}
731 897
732static ssize_t 898static eio_ssize_t
733eio__pwrite (int fd, void *buf, size_t count, off_t offset) 899eio__pwrite (int fd, void *buf, size_t count, off_t offset)
734{ 900{
735 ssize_t res; 901 eio_ssize_t res;
736 off_t ooffset; 902 off_t ooffset;
737 903
738 X_LOCK (preadwritelock); 904 X_LOCK (preadwritelock);
739 ooffset = lseek (fd, 0, SEEK_CUR); 905 ooffset = lseek (fd, 0, SEEK_CUR);
740 lseek (fd, offset, SEEK_SET); 906 lseek (fd, offset, SEEK_SET);
741 res = write (fd, buf, count); 907 res = write (fd, buf, count);
742 lseek (fd, offset, SEEK_SET); 908 lseek (fd, ooffset, SEEK_SET);
743 X_UNLOCK (preadwritelock); 909 X_UNLOCK (preadwritelock);
744 910
745 return res; 911 return res;
746} 912}
747#endif 913#endif
748 914
749#ifndef HAVE_FUTIMES 915#ifndef HAVE_UTIMES
750 916
917# undef utimes
751# define utimes(path,times) eio__utimes (path, times) 918# define utimes(path,times) eio__utimes (path, times)
752# define futimes(fd,times) eio__futimes (fd, times)
753 919
754static int 920static int
755eio__utimes (const char *filename, const struct timeval times[2]) 921eio__utimes (const char *filename, const struct timeval times[2])
756{ 922{
757 if (times) 923 if (times)
765 } 931 }
766 else 932 else
767 return utime (filename, 0); 933 return utime (filename, 0);
768} 934}
769 935
936#endif
937
938#ifndef HAVE_FUTIMES
939
940# undef futimes
941# define futimes(fd,times) eio__futimes (fd, times)
942
943static int
770static int eio__futimes (int fd, const struct timeval tv[2]) 944eio__futimes (int fd, const struct timeval tv[2])
771{ 945{
772 errno = ENOSYS; 946 errno = ENOSYS;
773 return -1; 947 return -1;
774} 948}
775 949
776#endif 950#endif
777 951
778#if !HAVE_FDATASYNC 952#if !HAVE_FDATASYNC
953# undef fdatasync
779# define fdatasync fsync 954# define fdatasync(fd) fsync (fd)
780#endif 955#endif
956
957/* sync_file_range always needs emulation */
958static int
959eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
960{
961#if HAVE_SYNC_FILE_RANGE
962 int res;
963
964 if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
965 || EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
966 || EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
967 {
968 flags = 0
969 | (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
970 | (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
971 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
972 }
973
974 res = sync_file_range (fd, offset, nbytes, flags);
975
976 if (!res || errno != ENOSYS)
977 return res;
978#endif
979
980 /* even though we could play tricks with the flags, it's better to always
981 * call fdatasync, as that matches the expectation of its users best */
982 return fdatasync (fd);
983}
984
985static int
986eio__fallocate (int fd, int mode, off_t offset, size_t len)
987{
988#if HAVE_FALLOCATE
989 return fallocate (fd, mode, offset, len);
990#else
991 errno = ENOSYS;
992 return -1;
993#endif
994}
781 995
782#if !HAVE_READAHEAD 996#if !HAVE_READAHEAD
997# undef readahead
783# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 998# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
784 999
785static ssize_t 1000static eio_ssize_t
786eio__readahead (int fd, off_t offset, size_t count, worker *self) 1001eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
787{ 1002{
788 size_t todo = count; 1003 size_t todo = count;
789 dBUF; 1004 dBUF;
790 1005
791 while (todo > 0) 1006 while (todo > 0)
801 return count; 1016 return count;
802} 1017}
803 1018
804#endif 1019#endif
805 1020
806#if !HAVE_READDIR_R
807# define readdir_r eio__readdir_r
808
809static mutex_t readdirlock = X_MUTEX_INIT;
810
811static int
812eio__readdir_r (DIR *dirp, EIO_STRUCT_DIRENT *ent, EIO_STRUCT_DIRENT **res)
813{
814 EIO_STRUCT_DIRENT *e;
815 int errorno;
816
817 X_LOCK (readdirlock);
818
819 e = readdir (dirp);
820 errorno = errno;
821
822 if (e)
823 {
824 *res = ent;
825 strcpy (ent->d_name, e->d_name);
826 }
827 else
828 *res = 0;
829
830 X_UNLOCK (readdirlock);
831
832 errno = errorno;
833 return e ? 0 : -1;
834}
835#endif
836
837/* sendfile always needs emulation */ 1021/* sendfile always needs emulation */
838static ssize_t 1022static eio_ssize_t
839eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) 1023eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)
840{ 1024{
1025 eio_ssize_t written = 0;
841 ssize_t res; 1026 eio_ssize_t res;
842 1027
843 if (!count) 1028 if (!count)
844 return 0; 1029 return 0;
845 1030
1031 for (;;)
1032 {
1033#ifdef __APPLE__
1034# undef HAVE_SENDFILE /* broken, as everything on os x */
1035#endif
846#if HAVE_SENDFILE 1036#if HAVE_SENDFILE
847# if __linux 1037# if __linux
1038 off_t soffset = offset;
848 res = sendfile (ofd, ifd, &offset, count); 1039 res = sendfile (ofd, ifd, &soffset, count);
849 1040
850# elif __freebsd 1041# elif __FreeBSD__
851 /* 1042 /*
852 * Of course, the freebsd sendfile is a dire hack with no thoughts 1043 * Of course, the freebsd sendfile is a dire hack with no thoughts
853 * wasted on making it similar to other I/O functions. 1044 * wasted on making it similar to other I/O functions.
854 */ 1045 */
855 {
856 off_t sbytes; 1046 off_t sbytes;
857 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 1047 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
858 1048
859 if (res < 0 && sbytes) 1049 #if 0 /* according to the manpage, this is correct, but broken behaviour */
860 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 1050 /* freebsd' sendfile will return 0 on success */
1051 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
1052 /* not on e.g. EIO or EPIPE - sounds broken */
1053 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
861 res = sbytes; 1054 res = sbytes;
862 } 1055 #endif
1056
1057 /* according to source inspection, this is correct, and useful behaviour */
1058 if (sbytes)
1059 res = sbytes;
1060
1061# elif defined (__APPLE__)
1062 off_t sbytes = count;
1063 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
1064
1065 /* according to the manpage, sbytes is always valid */
1066 if (sbytes)
1067 res = sbytes;
863 1068
864# elif __hpux 1069# elif __hpux
865 res = sendfile (ofd, ifd, offset, count, 0, 0); 1070 res = sendfile (ofd, ifd, offset, count, 0, 0);
866 1071
867# elif __solaris 1072# elif __solaris
868 {
869 struct sendfilevec vec; 1073 struct sendfilevec vec;
870 size_t sbytes; 1074 size_t sbytes;
871 1075
872 vec.sfv_fd = ifd; 1076 vec.sfv_fd = ifd;
873 vec.sfv_flag = 0; 1077 vec.sfv_flag = 0;
874 vec.sfv_off = offset; 1078 vec.sfv_off = offset;
875 vec.sfv_len = count; 1079 vec.sfv_len = count;
876 1080
877 res = sendfilev (ofd, &vec, 1, &sbytes); 1081 res = sendfilev (ofd, &vec, 1, &sbytes);
878 1082
879 if (res < 0 && sbytes) 1083 if (res < 0 && sbytes)
880 res = sbytes; 1084 res = sbytes;
881 }
882 1085
883# endif 1086# endif
1087
1088#elif defined (_WIN32) && 0
1089 /* does not work, just for documentation of what would need to be done */
1090 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
1091 /* libeio guarantees that the file offset does not change, and windows */
1092 /* has no way to get an independent handle to the same file description */
1093 HANDLE h = TO_SOCKET (ifd);
1094 SetFilePointer (h, offset, 0, FILE_BEGIN);
1095 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1096
884#else 1097#else
885 res = -1; 1098 res = -1;
886 errno = ENOSYS; 1099 errno = ENOSYS;
887#endif 1100#endif
888 1101
1102 /* we assume sendfile can copy at least 128mb in one go */
1103 if (res <= 128 * 1024 * 1024)
1104 {
1105 if (res > 0)
1106 written += res;
1107
1108 if (written)
1109 return written;
1110
1111 break;
1112 }
1113 else
1114 {
1115 /* if we requested more, then probably the kernel was lazy */
1116 written += res;
1117 offset += res;
1118 count -= res;
1119
1120 if (!count)
1121 return written;
1122 }
1123 }
1124
889 if (res < 0 1125 if (res < 0
890 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1126 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1127 /* BSDs */
1128#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1129 || errno == ENOTSUP
1130#endif
1131 || errno == EOPNOTSUPP /* BSDs */
891#if __solaris 1132#if __solaris
892 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1133 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
893#endif 1134#endif
894 ) 1135 )
895 ) 1136 )
899 1140
900 res = 0; 1141 res = 0;
901 1142
902 while (count) 1143 while (count)
903 { 1144 {
904 ssize_t cnt; 1145 eio_ssize_t cnt;
905 1146
906 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 1147 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
907 1148
908 if (cnt <= 0) 1149 if (cnt <= 0)
909 { 1150 {
926 } 1167 }
927 1168
928 return res; 1169 return res;
929} 1170}
930 1171
1172#ifdef PAGESIZE
1173# define eio_pagesize() PAGESIZE
1174#else
1175static intptr_t
1176eio_pagesize (void)
1177{
1178 static intptr_t page;
1179
1180 if (!page)
1181 page = sysconf (_SC_PAGESIZE);
1182
1183 return page;
1184}
1185#endif
1186
1187static void
1188eio_page_align (void **addr, size_t *length)
1189{
1190 intptr_t mask = eio_pagesize () - 1;
1191
1192 /* round down addr */
1193 intptr_t adj = mask & (intptr_t)*addr;
1194
1195 *addr = (void *)((intptr_t)*addr - adj);
1196 *length += adj;
1197
1198 /* round up length */
1199 *length = (*length + mask) & ~mask;
1200}
1201
1202#if !_POSIX_MEMLOCK
1203# define eio__mlockall(a) EIO_ENOSYS ()
1204#else
1205
1206static int
1207eio__mlockall (int flags)
1208{
1209 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1210 extern int mallopt (int, int);
1211 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1212 #endif
1213
1214 if (EIO_MCL_CURRENT != MCL_CURRENT
1215 || EIO_MCL_FUTURE != MCL_FUTURE)
1216 {
1217 flags = 0
1218 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1219 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1220 }
1221
1222 return mlockall (flags);
1223}
1224#endif
1225
1226#if !_POSIX_MEMLOCK_RANGE
1227# define eio__mlock(a,b) EIO_ENOSYS ()
1228#else
1229
1230static int
1231eio__mlock (void *addr, size_t length)
1232{
1233 eio_page_align (&addr, &length);
1234
1235 return mlock (addr, length);
1236}
1237
1238#endif
1239
1240#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1241# define eio__msync(a,b,c) EIO_ENOSYS ()
1242#else
1243
1244static int
1245eio__msync (void *mem, size_t len, int flags)
1246{
1247 eio_page_align (&mem, &len);
1248
1249 if (EIO_MS_ASYNC != MS_SYNC
1250 || EIO_MS_INVALIDATE != MS_INVALIDATE
1251 || EIO_MS_SYNC != MS_SYNC)
1252 {
1253 flags = 0
1254 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1255 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1256 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1257 }
1258
1259 return msync (mem, len, flags);
1260}
1261
1262#endif
1263
1264static int
1265eio__mtouch (eio_req *req)
1266{
1267 void *mem = req->ptr2;
1268 size_t len = req->size;
1269 int flags = req->int1;
1270
1271 eio_page_align (&mem, &len);
1272
1273 {
1274 intptr_t addr = (intptr_t)mem;
1275 intptr_t end = addr + len;
1276 intptr_t page = eio_pagesize ();
1277
1278 if (addr < end)
1279 if (flags & EIO_MT_MODIFY) /* modify */
1280 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1281 else
1282 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1283 }
1284
1285 return 0;
1286}
1287
1288/*****************************************************************************/
1289/* requests implemented outside eio_execute, because they are so large */
1290
1291static void
1292eio__realpath (eio_req *req, etp_worker *self)
1293{
1294 char *rel = req->ptr1;
1295 char *res;
1296 char *tmp1, *tmp2;
1297#if SYMLOOP_MAX > 32
1298 int symlinks = SYMLOOP_MAX;
1299#else
1300 int symlinks = 32;
1301#endif
1302
1303 req->result = -1;
1304
1305 errno = EINVAL;
1306 if (!rel)
1307 return;
1308
1309 errno = ENOENT;
1310 if (!*rel)
1311 return;
1312
1313 if (!req->ptr2)
1314 {
1315 X_LOCK (wrklock);
1316 req->flags |= EIO_FLAG_PTR2_FREE;
1317 X_UNLOCK (wrklock);
1318 req->ptr2 = malloc (PATH_MAX * 3);
1319
1320 errno = ENOMEM;
1321 if (!req->ptr2)
1322 return;
1323 }
1324
1325 res = req->ptr2;
1326 tmp1 = res + PATH_MAX;
1327 tmp2 = tmp1 + PATH_MAX;
1328
1329#if 0 /* disabled, the musl way to do things is just too racy */
1330#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1331 /* on linux we may be able to ask the kernel */
1332 {
1333 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1334
1335 if (fd >= 0)
1336 {
1337 sprintf (tmp1, "/proc/self/fd/%d", fd);
1338 req->result = readlink (tmp1, res, PATH_MAX);
1339 close (fd);
1340
1341 /* here we should probably stat the open file and the disk file, to make sure they still match */
1342
1343 if (req->result > 0)
1344 goto done;
1345 }
1346 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1347 return;
1348 }
1349#endif
1350#endif
1351
1352 if (*rel != '/')
1353 {
1354 if (!getcwd (res, PATH_MAX))
1355 return;
1356
1357 if (res [1]) /* only use if not / */
1358 res += strlen (res);
1359 }
1360
1361 while (*rel)
1362 {
1363 eio_ssize_t len, linklen;
1364 char *beg = rel;
1365
1366 while (*rel && *rel != '/')
1367 ++rel;
1368
1369 len = rel - beg;
1370
1371 if (!len) /* skip slashes */
1372 {
1373 ++rel;
1374 continue;
1375 }
1376
1377 if (beg [0] == '.')
1378 {
1379 if (len == 1)
1380 continue; /* . - nop */
1381
1382 if (beg [1] == '.' && len == 2)
1383 {
1384 /* .. - back up one component, if possible */
1385
1386 while (res != req->ptr2)
1387 if (*--res == '/')
1388 break;
1389
1390 continue;
1391 }
1392 }
1393
1394 errno = ENAMETOOLONG;
1395 if (res + 1 + len + 1 >= tmp1)
1396 return;
1397
1398 /* copy one component */
1399 *res = '/';
1400 memcpy (res + 1, beg, len);
1401
1402 /* zero-terminate, for readlink */
1403 res [len + 1] = 0;
1404
1405 /* now check if it's a symlink */
1406 linklen = readlink (req->ptr2, tmp1, PATH_MAX);
1407
1408 if (linklen < 0)
1409 {
1410 if (errno != EINVAL)
1411 return;
1412
1413 /* it's a normal directory. hopefully */
1414 res += len + 1;
1415 }
1416 else
1417 {
1418 /* yay, it was a symlink - build new path in tmp2 */
1419 int rellen = strlen (rel);
1420
1421 errno = ENAMETOOLONG;
1422 if (linklen + 1 + rellen >= PATH_MAX)
1423 return;
1424
1425 errno = ELOOP;
1426 if (!--symlinks)
1427 return;
1428
1429 if (*tmp1 == '/')
1430 res = req->ptr2; /* symlink resolves to an absolute path */
1431
1432 /* we need to be careful, as rel might point into tmp2 already */
1433 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1434 tmp2 [linklen] = '/';
1435 memcpy (tmp2, tmp1, linklen);
1436
1437 rel = tmp2;
1438 }
1439 }
1440
1441 /* special case for the lone root path */
1442 if (res == req->ptr2)
1443 *res++ = '/';
1444
1445 req->result = res - (char *)req->ptr2;
1446
1447done:
1448 req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
1449}
1450
1451static signed char
1452eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1453{
1454 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1455 : a->inode < b->inode ? -1
1456 : a->inode > b->inode ? 1
1457 : 0;
1458}
1459
1460#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1461
1462#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1463#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1464
1465static void
1466eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1467{
1468 unsigned char bits [9 + sizeof (ino_t) * 8];
1469 unsigned char *bit = bits;
1470
1471 assert (CHAR_BIT == 8);
1472 assert (sizeof (eio_dirent) * 8 < 256);
1473 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1474 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1475
1476 if (size <= EIO_SORT_FAST)
1477 return;
1478
1479 /* first prepare an array of bits to test in our radix sort */
1480 /* try to take endianness into account, as well as differences in ino_t sizes */
1481 /* inode_bits must contain all inodes ORed together */
1482 /* which is used to skip bits that are 0 everywhere, which is very common */
1483 {
1484 ino_t endianness;
1485 int i, j;
1486
1487 /* we store the byte offset of byte n into byte n of "endianness" */
1488 for (i = 0; i < sizeof (ino_t); ++i)
1489 ((unsigned char *)&endianness)[i] = i;
1490
1491 *bit++ = 0;
1492
1493 for (i = 0; i < sizeof (ino_t); ++i)
1494 {
1495 /* shifting off the byte offsets out of "endianness" */
1496 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1497 endianness >>= 8;
1498
1499 for (j = 0; j < 8; ++j)
1500 if (inode_bits & (((ino_t)1) << (i * 8 + j)))
1501 *bit++ = offs + j;
1502 }
1503
1504 for (j = 0; j < 8; ++j)
1505 if (score_bits & (1 << j))
1506 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1507 }
1508
1509 /* now actually do the sorting (a variant of MSD radix sort) */
1510 {
1511 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base;
1512 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end;
1513 unsigned char *bit_stk [9 + sizeof (ino_t) * 8];
1514 int stk_idx = 0;
1515
1516 base_stk [stk_idx] = dents;
1517 end_stk [stk_idx] = dents + size;
1518 bit_stk [stk_idx] = bit - 1;
1519
1520 do
1521 {
1522 base = base_stk [stk_idx];
1523 end = end_stk [stk_idx];
1524 bit = bit_stk [stk_idx];
1525
1526 for (;;)
1527 {
1528 unsigned char O = *bit >> 3;
1529 unsigned char M = 1 << (*bit & 7);
1530
1531 eio_dirent *a = base;
1532 eio_dirent *b = end;
1533
1534 if (b - a < EIO_SORT_CUTOFF)
1535 break;
1536
1537 /* now bit-partition the array on the bit */
1538 /* this ugly asymmetric loop seems to perform much better than typical */
1539 /* partition algos found in the literature */
1540 do
1541 if (!(((unsigned char *)a)[O] & M))
1542 ++a;
1543 else if (!(((unsigned char *)--b)[O] & M))
1544 {
1545 eio_dirent tmp = *a; *a = *b; *b = tmp;
1546 ++a;
1547 }
1548 while (b > a);
1549
1550 /* next bit, or stop, if no bits left in this path */
1551 if (!*--bit)
1552 break;
1553
1554 base_stk [stk_idx] = a;
1555 end_stk [stk_idx] = end;
1556 bit_stk [stk_idx] = bit;
1557 ++stk_idx;
1558
1559 end = a;
1560 }
1561 }
1562 while (stk_idx--);
1563 }
1564}
1565
1566static void
1567eio_dent_insertion_sort (eio_dirent *dents, int size)
1568{
1569 /* first move the smallest element to the front, to act as a sentinel */
1570 {
1571 int i;
1572 eio_dirent *min = dents;
1573
1574 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1575 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1576 if (EIO_DENT_CMP (dents [i], <, *min))
1577 min = &dents [i];
1578
1579 /* swap elements 0 and j (minimum) */
1580 {
1581 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1582 }
1583 }
1584
1585 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1586 {
1587 eio_dirent *i, *j;
1588
1589 for (i = dents + 1; i < dents + size; ++i)
1590 {
1591 eio_dirent value = *i;
1592
1593 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1594 j [1] = j [0];
1595
1596 j [1] = value;
1597 }
1598 }
1599}
1600
1601static void
1602eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1603{
1604 if (size <= 1)
1605 return; /* our insertion sort relies on size > 0 */
1606
1607 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1608 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1609 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1610
1611 /* use an insertion sort at the end, or for small arrays, */
1612 /* as insertion sort is more efficient for small partitions */
1613 eio_dent_insertion_sort (dents, size);
1614}
1615
931/* read a full directory */ 1616/* read a full directory */
932static void 1617static void
933eio__scandir (eio_req *req, etp_worker *self) 1618eio__scandir (eio_req *req, etp_worker *self)
934{ 1619{
935 DIR *dirp; 1620 DIR *dirp;
936 union
937 {
938 EIO_STRUCT_DIRENT d;
939 char b [offsetof (EIO_STRUCT_DIRENT, d_name) + NAME_MAX + 1];
940 } *u;
941 EIO_STRUCT_DIRENT *entp; 1621 EIO_STRUCT_DIRENT *entp;
942 char *name, *names; 1622 char *name, *names;
943 int memlen = 4096; 1623 int namesalloc = 4096;
944 int memofs = 0; 1624 int namesoffs = 0;
1625 int flags = req->int1;
1626 eio_dirent *dents = 0;
1627 int dentalloc = 128;
945 int res = 0; 1628 int dentoffs = 0;
1629 ino_t inode_bits = 0;
1630
1631 req->result = -1;
1632
1633 if (!(flags & EIO_READDIR_DENTS))
1634 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
946 1635
947 X_LOCK (wrklock); 1636 X_LOCK (wrklock);
1637 /* the corresponding closedir is in ETP_WORKER_CLEAR */
948 self->dirp = dirp = opendir (req->ptr1); 1638 self->dirp = dirp = opendir (req->ptr1);
949 self->dbuf = u = malloc (sizeof (*u)); 1639
950 req->flags |= EIO_FLAG_PTR2_FREE; 1640 if (req->flags & EIO_FLAG_PTR1_FREE)
1641 free (req->ptr1);
1642
1643 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1644 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
951 req->ptr2 = names = malloc (memlen); 1645 req->ptr2 = names = malloc (namesalloc);
952 X_UNLOCK (wrklock); 1646 X_UNLOCK (wrklock);
953 1647
954 if (dirp && u && names) 1648 if (dirp && names && (!flags || dents))
955 for (;;) 1649 for (;;)
956 { 1650 {
957 errno = 0; 1651 errno = 0;
958 readdir_r (dirp, &u->d, &entp); 1652 entp = readdir (dirp);
959 1653
960 if (!entp) 1654 if (!entp)
1655 {
1656 if (errno)
1657 break;
1658
1659 /* sort etc. */
1660 req->int1 = flags;
1661 req->result = dentoffs;
1662
1663 if (flags & EIO_READDIR_STAT_ORDER)
1664 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1665 else if (flags & EIO_READDIR_DIRS_FIRST)
1666 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1667 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1668 else
1669 {
1670 /* in this case, all is known, and we just put dirs first and sort them */
1671 eio_dirent *oth = dents + dentoffs;
1672 eio_dirent *dir = dents;
1673
1674 /* now partition dirs to the front, and non-dirs to the back */
1675 /* by walking from both sides and swapping if necessary */
1676 while (oth > dir)
1677 {
1678 if (dir->type == EIO_DT_DIR)
1679 ++dir;
1680 else if ((--oth)->type == EIO_DT_DIR)
1681 {
1682 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1683
1684 ++dir;
1685 }
1686 }
1687
1688 /* now sort the dirs only (dirs all have the same score) */
1689 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1690 }
1691
961 break; 1692 break;
1693 }
962 1694
1695 /* now add the entry to our list(s) */
963 name = entp->d_name; 1696 name = entp->d_name;
964 1697
1698 /* skip . and .. entries */
965 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1699 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
966 { 1700 {
967 int len = strlen (name) + 1; 1701 int len = D_NAMLEN (entp) + 1;
968 1702
969 res++; 1703 while (ecb_expect_false (namesoffs + len > namesalloc))
970
971 while (memofs + len > memlen)
972 { 1704 {
973 memlen *= 2; 1705 namesalloc *= 2;
974 X_LOCK (wrklock); 1706 X_LOCK (wrklock);
975 req->ptr2 = names = realloc (names, memlen); 1707 req->ptr2 = names = realloc (names, namesalloc);
976 X_UNLOCK (wrklock); 1708 X_UNLOCK (wrklock);
977 1709
978 if (!names) 1710 if (!names)
979 break; 1711 break;
980 } 1712 }
981 1713
982 memcpy (names + memofs, name, len); 1714 memcpy (names + namesoffs, name, len);
1715
1716 if (dents)
1717 {
1718 struct eio_dirent *ent;
1719
1720 if (ecb_expect_false (dentoffs == dentalloc))
1721 {
1722 dentalloc *= 2;
1723 X_LOCK (wrklock);
1724 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1725 X_UNLOCK (wrklock);
1726
1727 if (!dents)
1728 break;
1729 }
1730
1731 ent = dents + dentoffs;
1732
1733 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1734 ent->namelen = len - 1;
1735 ent->inode = D_INO (entp);
1736
1737 inode_bits |= ent->inode;
1738
1739 switch (D_TYPE (entp))
1740 {
1741 default:
1742 ent->type = EIO_DT_UNKNOWN;
1743 flags |= EIO_READDIR_FOUND_UNKNOWN;
1744 break;
1745
1746 #ifdef DT_FIFO
1747 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1748 #endif
1749 #ifdef DT_CHR
1750 case DT_CHR: ent->type = EIO_DT_CHR; break;
1751 #endif
1752 #ifdef DT_MPC
1753 case DT_MPC: ent->type = EIO_DT_MPC; break;
1754 #endif
1755 #ifdef DT_DIR
1756 case DT_DIR: ent->type = EIO_DT_DIR; break;
1757 #endif
1758 #ifdef DT_NAM
1759 case DT_NAM: ent->type = EIO_DT_NAM; break;
1760 #endif
1761 #ifdef DT_BLK
1762 case DT_BLK: ent->type = EIO_DT_BLK; break;
1763 #endif
1764 #ifdef DT_MPB
1765 case DT_MPB: ent->type = EIO_DT_MPB; break;
1766 #endif
1767 #ifdef DT_REG
1768 case DT_REG: ent->type = EIO_DT_REG; break;
1769 #endif
1770 #ifdef DT_NWK
1771 case DT_NWK: ent->type = EIO_DT_NWK; break;
1772 #endif
1773 #ifdef DT_CMP
1774 case DT_CMP: ent->type = EIO_DT_CMP; break;
1775 #endif
1776 #ifdef DT_LNK
1777 case DT_LNK: ent->type = EIO_DT_LNK; break;
1778 #endif
1779 #ifdef DT_SOCK
1780 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1781 #endif
1782 #ifdef DT_DOOR
1783 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1784 #endif
1785 #ifdef DT_WHT
1786 case DT_WHT: ent->type = EIO_DT_WHT; break;
1787 #endif
1788 }
1789
1790 ent->score = 7;
1791
1792 if (flags & EIO_READDIR_DIRS_FIRST)
1793 {
1794 if (ent->type == EIO_DT_UNKNOWN)
1795 {
1796 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1797 ent->score = 1;
1798 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1799 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1800 }
1801 else if (ent->type == EIO_DT_DIR)
1802 ent->score = 0;
1803 }
1804 }
1805
983 memofs += len; 1806 namesoffs += len;
1807 ++dentoffs;
1808 }
1809
1810 if (EIO_CANCELLED (req))
1811 {
1812 errno = ECANCELED;
1813 break;
984 } 1814 }
985 } 1815 }
986
987 if (errno)
988 res = -1;
989
990 req->result = res;
991} 1816}
992 1817
993/*****************************************************************************/ 1818/*****************************************************************************/
994 1819
995#define ALLOC(len) \ 1820#define ALLOC(len) \
1010X_THREAD_PROC (etp_proc) 1835X_THREAD_PROC (etp_proc)
1011{ 1836{
1012 ETP_REQ *req; 1837 ETP_REQ *req;
1013 struct timespec ts; 1838 struct timespec ts;
1014 etp_worker *self = (etp_worker *)thr_arg; 1839 etp_worker *self = (etp_worker *)thr_arg;
1840 int timeout;
1015 1841
1016 /* try to distribute timeouts somewhat randomly */ 1842 /* try to distribute timeouts somewhat evenly */
1017 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 1843 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1018 1844
1019 for (;;) 1845 for (;;)
1020 { 1846 {
1847 ts.tv_sec = 0;
1848
1021 X_LOCK (reqlock); 1849 X_LOCK (reqlock);
1022 1850
1023 for (;;) 1851 for (;;)
1024 { 1852 {
1025 self->req = req = reqq_shift (&req_queue); 1853 self->req = req = reqq_shift (&req_queue);
1026 1854
1027 if (req) 1855 if (req)
1028 break; 1856 break;
1029 1857
1858 if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
1859 {
1860 X_UNLOCK (reqlock);
1861 X_LOCK (wrklock);
1862 --started;
1863 X_UNLOCK (wrklock);
1864 goto quit;
1865 }
1866
1030 ++idle; 1867 ++idle;
1031 1868
1032 ts.tv_sec = time (0) + IDLE_TIMEOUT; 1869 if (idle <= max_idle)
1033 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 1870 /* we are allowed to idle, so do so without any timeout */
1871 X_COND_WAIT (reqwait, reqlock);
1872 else
1034 { 1873 {
1035 if (idle > max_idle) 1874 /* initialise timeout once */
1036 { 1875 if (!ts.tv_sec)
1037 --idle; 1876 ts.tv_sec = time (0) + idle_timeout;
1038 X_UNLOCK (reqlock);
1039 X_LOCK (wrklock);
1040 --started;
1041 X_UNLOCK (wrklock);
1042 goto quit;
1043 }
1044 1877
1045 /* we are allowed to idle, so do so without any timeout */
1046 X_COND_WAIT (reqwait, reqlock); 1878 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
1879 ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1047 } 1880 }
1048 1881
1049 --idle; 1882 --idle;
1050 } 1883 }
1051 1884
1054 X_UNLOCK (reqlock); 1887 X_UNLOCK (reqlock);
1055 1888
1056 if (req->type < 0) 1889 if (req->type < 0)
1057 goto quit; 1890 goto quit;
1058 1891
1059 if (!EIO_CANCELLED (req))
1060 ETP_EXECUTE (self, req); 1892 ETP_EXECUTE (self, req);
1061 1893
1062 X_LOCK (reslock); 1894 X_LOCK (reslock);
1063 1895
1064 ++npending; 1896 ++npending;
1065 1897
1080 return 0; 1912 return 0;
1081} 1913}
1082 1914
1083/*****************************************************************************/ 1915/*****************************************************************************/
1084 1916
1917int ecb_cold
1085int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1918eio_init (void (*want_poll)(void), void (*done_poll)(void))
1086{ 1919{
1087 etp_init (want_poll, done_poll); 1920 return etp_init (want_poll, done_poll);
1088} 1921}
1089 1922
1923ecb_inline void
1090static void eio_api_destroy (eio_req *req) 1924eio_api_destroy (eio_req *req)
1091{ 1925{
1092 free (req); 1926 free (req);
1093} 1927}
1094 1928
1095#define REQ(rtype) \ 1929#define REQ(rtype) \
1114 { \ 1948 { \
1115 eio_api_destroy (req); \ 1949 eio_api_destroy (req); \
1116 return 0; \ 1950 return 0; \
1117 } 1951 }
1118 1952
1953static void
1119static void eio_execute (etp_worker *self, eio_req *req) 1954eio_execute (etp_worker *self, eio_req *req)
1120{ 1955{
1121 errno = 0; 1956 if (ecb_expect_false (EIO_CANCELLED (req)))
1957 {
1958 req->result = -1;
1959 req->errorno = ECANCELED;
1960 return;
1961 }
1122 1962
1123 switch (req->type) 1963 switch (req->type)
1124 { 1964 {
1125 case EIO_READ: ALLOC (req->size); 1965 case EIO_READ: ALLOC (req->size);
1126 req->result = req->offs >= 0 1966 req->result = req->offs >= 0
1137 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 1977 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1138 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1978 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1139 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 1979 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1140 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1980 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1141 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; 1981 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1982
1983 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1984 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1985 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1986 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1142 1987
1143 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; 1988 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1144 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; 1989 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1145 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break; 1990 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1146 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break; 1991 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1156 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; 2001 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1157 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; 2002 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1158 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; 2003 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1159 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break; 2004 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1160 2005
2006 case EIO_REALPATH: eio__realpath (req, self); break;
2007
1161 case EIO_READLINK: ALLOC (NAME_MAX); 2008 case EIO_READLINK: ALLOC (PATH_MAX);
1162 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; 2009 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1163 2010
1164 case EIO_SYNC: req->result = 0; sync (); break; 2011 case EIO_SYNC: req->result = 0; sync (); break;
1165 case EIO_FSYNC: req->result = fsync (req->int1); break; 2012 case EIO_FSYNC: req->result = fsync (req->int1); break;
1166 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 2013 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
2014 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
2015 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
2016 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
2017 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
2018 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
2019 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
1167 2020
1168 case EIO_READDIR: eio__scandir (req, self); break; 2021 case EIO_READDIR: eio__scandir (req, self); break;
1169 2022
1170 case EIO_BUSY: 2023 case EIO_BUSY:
1171#ifdef _WIN32 2024#ifdef _WIN32
1172 Sleep (req->nv1 * 1000.); 2025 Sleep (req->nv1 * 1e3);
1173#else 2026#else
1174 { 2027 {
1175 struct timeval tv; 2028 struct timeval tv;
1176 2029
1177 tv.tv_sec = req->nv1; 2030 tv.tv_sec = req->nv1;
1178 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; 2031 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1179 2032
1180 req->result = select (0, 0, 0, 0, &tv); 2033 req->result = select (0, 0, 0, 0, &tv);
1181 } 2034 }
1182#endif 2035#endif
1183 break; 2036 break;
1198 times = tv; 2051 times = tv;
1199 } 2052 }
1200 else 2053 else
1201 times = 0; 2054 times = 0;
1202 2055
1203
1204 req->result = req->type == EIO_FUTIME 2056 req->result = req->type == EIO_FUTIME
1205 ? futimes (req->int1, times) 2057 ? futimes (req->int1, times)
1206 : utimes (req->ptr1, times); 2058 : utimes (req->ptr1, times);
1207 } 2059 }
2060 break;
1208 2061
1209 case EIO_GROUP: 2062 case EIO_GROUP:
2063 abort (); /* handled in eio_request */
2064
1210 case EIO_NOP: 2065 case EIO_NOP:
1211 req->result = 0; 2066 req->result = 0;
1212 break; 2067 break;
1213 2068
1214 case EIO_CUSTOM: 2069 case EIO_CUSTOM:
1215 req->feed (req); 2070 req->feed (req);
1216 break; 2071 break;
1217 2072
1218 default: 2073 default:
2074 errno = ENOSYS;
1219 req->result = -1; 2075 req->result = -1;
1220 break; 2076 break;
1221 } 2077 }
1222 2078
1223 req->errorno = errno; 2079 req->errorno = errno;
1243eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data) 2099eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
1244{ 2100{
1245 REQ (EIO_FSYNC); req->int1 = fd; SEND; 2101 REQ (EIO_FSYNC); req->int1 = fd; SEND;
1246} 2102}
1247 2103
2104eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2105{
2106 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2107}
2108
2109eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2110{
2111 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2112}
2113
2114eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
2115{
2116 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
2117}
2118
2119eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
2120{
2121 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2122}
2123
2124eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2125{
2126 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2127}
2128
2129eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2130{
2131 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
2132}
2133
1248eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) 2134eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1249{ 2135{
1250 REQ (EIO_FDATASYNC); req->int1 = fd; SEND; 2136 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1251} 2137}
1252 2138
1273eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) 2159eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1274{ 2160{
1275 REQ (EIO_FSTAT); req->int1 = fd; SEND; 2161 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1276} 2162}
1277 2163
2164eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
2165{
2166 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
2167}
2168
1278eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) 2169eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1279{ 2170{
1280 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; 2171 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1281} 2172}
1282 2173
1288eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) 2179eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1289{ 2180{
1290 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2181 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1291} 2182}
1292 2183
1293eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2184eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1294{ 2185{
1295 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2186 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1296} 2187}
1297 2188
1298eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) 2189eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
1318eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data) 2209eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1319{ 2210{
1320 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2211 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1321} 2212}
1322 2213
1323eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2214eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1324{ 2215{
1325 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2216 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1326} 2217}
1327 2218
1328eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) 2219eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
1344eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data) 2235eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1345{ 2236{
1346 return eio__1path (EIO_READLINK, path, pri, cb, data); 2237 return eio__1path (EIO_READLINK, path, pri, cb, data);
1347} 2238}
1348 2239
2240eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2241{
2242 return eio__1path (EIO_REALPATH, path, pri, cb, data);
2243}
2244
1349eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) 2245eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1350{ 2246{
1351 return eio__1path (EIO_STAT, path, pri, cb, data); 2247 return eio__1path (EIO_STAT, path, pri, cb, data);
1352} 2248}
1353 2249
1354eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) 2250eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1355{ 2251{
1356 return eio__1path (EIO_LSTAT, path, pri, cb, data); 2252 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1357} 2253}
1358 2254
2255eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
2256{
2257 return eio__1path (EIO_STATVFS, path, pri, cb, data);
2258}
2259
1359eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) 2260eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1360{ 2261{
1361 return eio__1path (EIO_UNLINK, path, pri, cb, data); 2262 return eio__1path (EIO_UNLINK, path, pri, cb, data);
1362} 2263}
1363 2264
1364eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data) 2265eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1365{ 2266{
1366 return eio__1path (EIO_RMDIR, path, pri, cb, data); 2267 return eio__1path (EIO_RMDIR, path, pri, cb, data);
1367} 2268}
1368 2269
1369eio_req *eio_readdir (const char *path, int pri, eio_cb cb, void *data) 2270eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1370{ 2271{
1371 return eio__1path (EIO_READDIR, path, pri, cb, data); 2272 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1372} 2273}
1373 2274
1374eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) 2275eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1375{ 2276{
1376 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int2 = (long)dev; SEND; 2277 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1377} 2278}
1378 2279
1379static eio_req * 2280static eio_req *
1380eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2281eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1381{ 2282{
1405eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2306eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1406{ 2307{
1407 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2308 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1408} 2309}
1409 2310
1410eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) 2311eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
1411{ 2312{
1412 REQ (EIO_CUSTOM); req->feed = execute; SEND; 2313 REQ (EIO_CUSTOM); req->feed = execute; SEND;
1413} 2314}
1414 2315
1415#endif 2316#endif
1426#undef SEND 2327#undef SEND
1427 2328
1428/*****************************************************************************/ 2329/*****************************************************************************/
1429/* grp functions */ 2330/* grp functions */
1430 2331
2332void
1431void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2333eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1432{ 2334{
1433 grp->int2 = limit; 2335 grp->int2 = limit;
1434 grp->feed = feed; 2336 grp->feed = feed;
1435 2337
1436 grp_try_feed (grp); 2338 grp_try_feed (grp);
1437} 2339}
1438 2340
2341void
1439void eio_grp_limit (eio_req *grp, int limit) 2342eio_grp_limit (eio_req *grp, int limit)
1440{ 2343{
1441 grp->int2 = limit; 2344 grp->int2 = limit;
1442 2345
1443 grp_try_feed (grp); 2346 grp_try_feed (grp);
1444} 2347}
1445 2348
2349void
1446void eio_grp_add (eio_req *grp, eio_req *req) 2350eio_grp_add (eio_req *grp, eio_req *req)
1447{ 2351{
1448 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2352 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
2353
2354 grp->flags |= EIO_FLAG_GROUPADD;
1449 2355
1450 ++grp->size; 2356 ++grp->size;
1451 req->grp = grp; 2357 req->grp = grp;
1452 2358
1453 req->grp_prev = 0; 2359 req->grp_prev = 0;
1460} 2366}
1461 2367
1462/*****************************************************************************/ 2368/*****************************************************************************/
1463/* misc garbage */ 2369/* misc garbage */
1464 2370
2371eio_ssize_t
1465ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2372eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1466{ 2373{
1467 etp_worker wrk; 2374 etp_worker wrk;
2375 eio_ssize_t ret;
1468 2376
1469 wrk.dbuf = 0; 2377 wrk.dbuf = 0;
1470 2378
1471 eio__sendfile (ofd, ifd, offset, count, &wrk); 2379 ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
1472 2380
1473 if (wrk.dbuf) 2381 if (wrk.dbuf)
1474 free (wrk.dbuf); 2382 free (wrk.dbuf);
1475}
1476 2383
2384 return ret;
2385}
2386

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines