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 1.61 by root, Fri Feb 11 00:53:24 2011 UTC vs.
Revision 1.96 by root, Tue Jul 19 05:20:36 2011 UTC

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"
41 46
42#ifdef EIO_STACKSIZE 47#ifdef EIO_STACKSIZE
43# define XTHREAD_STACKSIZE EIO_STACKSIZE 48# define XTHREAD_STACKSIZE EIO_STACKSIZE
44#endif 49#endif
45#include "xthread.h" 50#include "xthread.h"
49#include <stdlib.h> 54#include <stdlib.h>
50#include <string.h> 55#include <string.h>
51#include <errno.h> 56#include <errno.h>
52#include <sys/types.h> 57#include <sys/types.h>
53#include <sys/stat.h> 58#include <sys/stat.h>
54#include <sys/statvfs.h>
55#include <limits.h> 59#include <limits.h>
56#include <fcntl.h> 60#include <fcntl.h>
57#include <assert.h> 61#include <assert.h>
58 62
63/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
64/* intptr_t only comes from stdint.h, says idiot openbsd coder */
65#if HAVE_STDINT_H
66# include <stdint.h>
67#endif
68
69#ifndef ECANCELED
70# define ECANCELED EDOM
71#endif
72#ifndef ELOOP
73# define ELOOP EDOM
74#endif
75
76#if !defined(ENOTSOCK) && defined(WSAENOTSOCK)
77# define ENOTSOCK WSAENOTSOCK
78#endif
79
80static void eio_destroy (eio_req *req);
81
59#ifndef EIO_FINISH 82#ifndef EIO_FINISH
60# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0 83# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
61#endif 84#endif
62 85
63#ifndef EIO_DESTROY 86#ifndef EIO_DESTROY
66 89
67#ifndef EIO_FEED 90#ifndef EIO_FEED
68# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0) 91# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)
69#endif 92#endif
70 93
94#ifndef EIO_FD_TO_WIN32_HANDLE
95# define EIO_FD_TO_WIN32_HANDLE(fd) _get_osfhandle (fd)
96#endif
97#ifndef EIO_WIN32_HANDLE_TO_FD
98# define EIO_WIN32_HANDLE_TO_FD(handle) _open_osfhandle (handle, 0)
99#endif
100
101#define EIO_ERRNO(errval,retval) ((errno = errval), retval)
102
103#define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1)
104
71#ifdef _WIN32 105#ifdef _WIN32
72 106
73 /*doh*/ 107 #undef PAGESIZE
108 #define PAGESIZE 4096 /* GetSystemInfo? */
109
110 #ifdef EIO_STRUCT_STATI64
111 #define stat(path,buf) _stati64 (path,buf)
112 #define fstat(fd,buf) _fstati64 (fd,buf)
113 #endif
114 #define lstat(path,buf) stat (path,buf)
115 #define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
116 #define mkdir(path,mode) _mkdir (path)
117 #define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
118
119 #define chmod(path,mode) _chmod (path, mode)
120 #define dup(fd) _dup (fd)
121 #define dup2(fd1,fd2) _dup2 (fd1, fd2)
122
123 #define fchmod(fd,mode) EIO_ENOSYS ()
124 #define chown(path,uid,gid) EIO_ENOSYS ()
125 #define fchown(fd,uid,gid) EIO_ENOSYS ()
126 #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
127 #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
128 #define mknod(path,mode,dev) EIO_ENOSYS ()
129 #define sync() EIO_ENOSYS ()
130 #define readlink(path,buf,s) EIO_ENOSYS ()
131 #define statvfs(path,buf) EIO_ENOSYS ()
132 #define fstatvfs(fd,buf) EIO_ENOSYS ()
133
134 /* we could even stat and see if it exists */
135 static int
136 symlink (const char *old, const char *neu)
137 {
138 #if WINVER >= 0x0600
139 if (CreateSymbolicLink (neu, old, 1))
140 return 0;
141
142 if (CreateSymbolicLink (neu, old, 0))
143 return 0;
144 #endif
145
146 return EIO_ERRNO (ENOENT, -1);
147 }
148
149 /* POSIX API only */
150 #define CreateHardLink(neu,old,flags) 0
151 #define CreateSymbolicLink(neu,old,flags) 0
152
153 struct statvfs
154 {
155 int dummy;
156 };
157
158 #define DT_DIR EIO_DT_DIR
159 #define DT_REG EIO_DT_REG
160 #define D_NAME(entp) entp.cFileName
161 #define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
162
74#else 163#else
75 164
76# include "config.h"
77# include <sys/time.h> 165 #include <sys/time.h>
78# include <sys/select.h> 166 #include <sys/select.h>
167 #include <sys/statvfs.h>
79# include <unistd.h> 168 #include <unistd.h>
169 #include <signal.h>
170 #include <dirent.h>
171
172 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
173 #include <sys/mman.h>
174 #endif
175
176 #define D_NAME(entp) entp->d_name
177
178 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
179 #if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
180 #define _DIRENT_HAVE_D_TYPE /* sigh */
181 #define D_INO(de) (de)->d_fileno
182 #define D_NAMLEN(de) (de)->d_namlen
183 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
184 #define D_INO(de) (de)->d_ino
185 #endif
186
187 #ifdef _D_EXACT_NAMLEN
188 #undef D_NAMLEN
189 #define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
190 #endif
191
192 #ifdef _DIRENT_HAVE_D_TYPE
193 #define D_TYPE(de) (de)->d_type
194 #endif
195
196 #ifndef EIO_STRUCT_DIRENT
197 #define EIO_STRUCT_DIRENT struct dirent
198 #endif
199
200#endif
201
202#if HAVE_UTIMES
80# include <utime.h> 203# include <utime.h>
81# include <signal.h>
82# include <dirent.h>
83
84#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
85# include <sys/mman.h>
86#endif
87
88/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
89# if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
90# define _DIRENT_HAVE_D_TYPE /* sigh */
91# define D_INO(de) (de)->d_fileno
92# define D_NAMLEN(de) (de)->d_namlen
93# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
94# define D_INO(de) (de)->d_ino
95# endif
96
97#ifdef _D_EXACT_NAMLEN
98# undef D_NAMLEN
99# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
100#endif
101
102# ifdef _DIRENT_HAVE_D_TYPE
103# define D_TYPE(de) (de)->d_type
104# endif
105
106# ifndef EIO_STRUCT_DIRENT
107# define EIO_STRUCT_DIRENT struct dirent
108# endif
109
110#endif 204#endif
111 205
112#if HAVE_SENDFILE 206#if HAVE_SENDFILE
113# if __linux 207# if __linux
114# include <sys/sendfile.h> 208# include <sys/sendfile.h>
129#endif 223#endif
130#ifndef D_INO 224#ifndef D_INO
131# define D_INO(de) 0 225# define D_INO(de) 0
132#endif 226#endif
133#ifndef D_NAMLEN 227#ifndef D_NAMLEN
134# define D_NAMLEN(de) strlen ((de)->d_name) 228# define D_NAMLEN(entp) strlen (D_NAME (entp))
135#endif 229#endif
136
137/* number of seconds after which an idle threads exit */
138#define IDLE_TIMEOUT 10
139 230
140/* used for struct dirent, AIX doesn't provide it */ 231/* used for struct dirent, AIX doesn't provide it */
141#ifndef NAME_MAX 232#ifndef NAME_MAX
142# define NAME_MAX 4096 233# define NAME_MAX 4096
143#endif 234#endif
149 240
150/* buffer size for various temporary buffers */ 241/* buffer size for various temporary buffers */
151#define EIO_BUFSIZE 65536 242#define EIO_BUFSIZE 65536
152 243
153#define dBUF \ 244#define dBUF \
154 char *eio_buf; \
155 ETP_WORKER_LOCK (self); \
156 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 245 char *eio_buf = malloc (EIO_BUFSIZE); \
157 ETP_WORKER_UNLOCK (self); \
158 errno = ENOMEM; \ 246 errno = ENOMEM; \
159 if (!eio_buf) \ 247 if (!eio_buf) \
160 return -1; 248 return -1
249
250#define FUBd \
251 free (eio_buf)
161 252
162#define EIO_TICKS ((1000000 + 1023) >> 10) 253#define EIO_TICKS ((1000000 + 1023) >> 10)
163
164/*****************************************************************************/
165
166#if __GNUC__ >= 3
167# define expect(expr,value) __builtin_expect ((expr),(value))
168#else
169# define expect(expr,value) (expr)
170#endif
171
172#define expect_false(expr) expect ((expr) != 0, 0)
173#define expect_true(expr) expect ((expr) != 0, 1)
174
175/*****************************************************************************/
176 254
177#define ETP_PRI_MIN EIO_PRI_MIN 255#define ETP_PRI_MIN EIO_PRI_MIN
178#define ETP_PRI_MAX EIO_PRI_MAX 256#define ETP_PRI_MAX EIO_PRI_MAX
179 257
180struct etp_worker; 258struct etp_worker;
184static int eio_finish (eio_req *req); 262static int eio_finish (eio_req *req);
185#define ETP_FINISH(req) eio_finish (req) 263#define ETP_FINISH(req) eio_finish (req)
186static void eio_execute (struct etp_worker *self, eio_req *req); 264static void eio_execute (struct etp_worker *self, eio_req *req);
187#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) 265#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
188 266
189#define ETP_WORKER_CLEAR(req) \
190 if (wrk->dbuf) \
191 { \
192 free (wrk->dbuf); \
193 wrk->dbuf = 0; \
194 } \
195 \
196 if (wrk->dirp) \
197 { \
198 closedir (wrk->dirp); \
199 wrk->dirp = 0; \
200 }
201
202#define ETP_WORKER_COMMON \
203 void *dbuf; \
204 DIR *dirp;
205
206/*****************************************************************************/ 267/*****************************************************************************/
207 268
208#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) 269#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
209 270
210/* calculate time difference in ~1/EIO_TICKS of a second */ 271/* calculate time difference in ~1/EIO_TICKS of a second */
272ecb_inline int
211static int tvdiff (struct timeval *tv1, struct timeval *tv2) 273tvdiff (struct timeval *tv1, struct timeval *tv2)
212{ 274{
213 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 275 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
214 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 276 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
215} 277}
216 278
223static unsigned int max_poll_reqs; /* reslock */ 285static unsigned int max_poll_reqs; /* reslock */
224 286
225static volatile unsigned int nreqs; /* reqlock */ 287static volatile unsigned int nreqs; /* reqlock */
226static volatile unsigned int nready; /* reqlock */ 288static volatile unsigned int nready; /* reqlock */
227static volatile unsigned int npending; /* reqlock */ 289static volatile unsigned int npending; /* reqlock */
228static volatile unsigned int max_idle = 4; 290static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
291static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
229 292
230static xmutex_t wrklock = X_MUTEX_INIT; 293static xmutex_t wrklock;
231static xmutex_t reslock = X_MUTEX_INIT; 294static xmutex_t reslock;
232static xmutex_t reqlock = X_MUTEX_INIT; 295static xmutex_t reqlock;
233static xcond_t reqwait = X_COND_INIT; 296static xcond_t reqwait;
234 297
235#if !HAVE_PREADWRITE 298#if !HAVE_PREADWRITE
236/* 299/*
237 * make our pread/pwrite emulation safe against themselves, but not against 300 * make our pread/pwrite emulation safe against themselves, but not against
238 * normal read/write by using a mutex. slows down execution a lot, 301 * normal read/write by using a mutex. slows down execution a lot,
239 * but that's your problem, not mine. 302 * but that's your problem, not mine.
240 */ 303 */
241static xmutex_t preadwritelock = X_MUTEX_INIT; 304static xmutex_t preadwritelock;
242#endif 305#endif
243 306
244typedef struct etp_worker 307typedef struct etp_worker
245{ 308{
246 /* locked by wrklock */ 309 /* locked by wrklock */
249 xthread_t tid; 312 xthread_t tid;
250 313
251 /* locked by reslock, reqlock or wrklock */ 314 /* locked by reslock, reqlock or wrklock */
252 ETP_REQ *req; /* currently processed request */ 315 ETP_REQ *req; /* currently processed request */
253 316
317#ifdef ETP_WORKER_COMMON
254 ETP_WORKER_COMMON 318 ETP_WORKER_COMMON
319#endif
255} etp_worker; 320} etp_worker;
256 321
257static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */ 322static etp_worker wrk_first; /* NOT etp */
258 323
259#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock) 324#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
260#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock) 325#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
261 326
262/* worker threads management */ 327/* worker threads management */
263 328
329static void ecb_cold
264static void etp_worker_clear (etp_worker *wrk) 330etp_worker_clear (etp_worker *wrk)
265{ 331{
266 ETP_WORKER_CLEAR (wrk);
267} 332}
268 333
334static void ecb_cold
269static void etp_worker_free (etp_worker *wrk) 335etp_worker_free (etp_worker *wrk)
270{ 336{
271 wrk->next->prev = wrk->prev; 337 wrk->next->prev = wrk->prev;
272 wrk->prev->next = wrk->next; 338 wrk->prev->next = wrk->next;
273 339
274 free (wrk); 340 free (wrk);
275} 341}
276 342
277static unsigned int etp_nreqs (void) 343static unsigned int
344etp_nreqs (void)
278{ 345{
279 int retval; 346 int retval;
280 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 347 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
281 retval = nreqs; 348 retval = nreqs;
282 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 349 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
283 return retval; 350 return retval;
284} 351}
285 352
286static unsigned int etp_nready (void) 353static unsigned int
354etp_nready (void)
287{ 355{
288 unsigned int retval; 356 unsigned int retval;
289 357
290 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 358 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
291 retval = nready; 359 retval = nready;
292 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 360 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
293 361
294 return retval; 362 return retval;
295} 363}
296 364
297static unsigned int etp_npending (void) 365static unsigned int
366etp_npending (void)
298{ 367{
299 unsigned int retval; 368 unsigned int retval;
300 369
301 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 370 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
302 retval = npending; 371 retval = npending;
303 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 372 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
304 373
305 return retval; 374 return retval;
306} 375}
307 376
308static unsigned int etp_nthreads (void) 377static unsigned int
378etp_nthreads (void)
309{ 379{
310 unsigned int retval; 380 unsigned int retval;
311 381
312 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 382 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
313 retval = started; 383 retval = started;
327} etp_reqq; 397} etp_reqq;
328 398
329static etp_reqq req_queue; 399static etp_reqq req_queue;
330static etp_reqq res_queue; 400static etp_reqq res_queue;
331 401
402static void ecb_noinline ecb_cold
403reqq_init (etp_reqq *q)
404{
405 int pri;
406
407 for (pri = 0; pri < ETP_NUM_PRI; ++pri)
408 q->qs[pri] = q->qe[pri] = 0;
409
410 q->size = 0;
411}
412
413static int ecb_noinline
332static int reqq_push (etp_reqq *q, ETP_REQ *req) 414reqq_push (etp_reqq *q, ETP_REQ *req)
333{ 415{
334 int pri = req->pri; 416 int pri = req->pri;
335 req->next = 0; 417 req->next = 0;
336 418
337 if (q->qe[pri]) 419 if (q->qe[pri])
343 q->qe[pri] = q->qs[pri] = req; 425 q->qe[pri] = q->qs[pri] = req;
344 426
345 return q->size++; 427 return q->size++;
346} 428}
347 429
430static ETP_REQ * ecb_noinline
348static ETP_REQ *reqq_shift (etp_reqq *q) 431reqq_shift (etp_reqq *q)
349{ 432{
350 int pri; 433 int pri;
351 434
352 if (!q->size) 435 if (!q->size)
353 return 0; 436 return 0;
368 } 451 }
369 452
370 abort (); 453 abort ();
371} 454}
372 455
373static void etp_atfork_prepare (void) 456static int ecb_cold
457etp_init (void (*want_poll)(void), void (*done_poll)(void))
374{ 458{
375 X_LOCK (wrklock); 459 X_MUTEX_CREATE (wrklock);
376 X_LOCK (reqlock); 460 X_MUTEX_CREATE (reslock);
377 X_LOCK (reslock); 461 X_MUTEX_CREATE (reqlock);
378#if !HAVE_PREADWRITE 462 X_COND_CREATE (reqwait);
379 X_LOCK (preadwritelock);
380#endif
381}
382 463
383static void etp_atfork_parent (void) 464 reqq_init (&req_queue);
384{ 465 reqq_init (&res_queue);
385#if !HAVE_PREADWRITE
386 X_UNLOCK (preadwritelock);
387#endif
388 X_UNLOCK (reslock);
389 X_UNLOCK (reqlock);
390 X_UNLOCK (wrklock);
391}
392 466
393static void etp_atfork_child (void) 467 wrk_first.next =
394{ 468 wrk_first.prev = &wrk_first;
395 ETP_REQ *prv;
396
397 while ((prv = reqq_shift (&req_queue)))
398 ETP_DESTROY (prv);
399
400 while ((prv = reqq_shift (&res_queue)))
401 ETP_DESTROY (prv);
402
403 while (wrk_first.next != &wrk_first)
404 {
405 etp_worker *wrk = wrk_first.next;
406
407 if (wrk->req)
408 ETP_DESTROY (wrk->req);
409
410 etp_worker_clear (wrk);
411 etp_worker_free (wrk);
412 }
413 469
414 started = 0; 470 started = 0;
415 idle = 0; 471 idle = 0;
416 nreqs = 0; 472 nreqs = 0;
417 nready = 0; 473 nready = 0;
418 npending = 0; 474 npending = 0;
419 475
420 etp_atfork_parent ();
421}
422
423static void
424etp_once_init (void)
425{
426 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
427}
428
429static int
430etp_init (void (*want_poll)(void), void (*done_poll)(void))
431{
432 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
433
434 pthread_once (&doinit, etp_once_init);
435
436 want_poll_cb = want_poll; 476 want_poll_cb = want_poll;
437 done_poll_cb = done_poll; 477 done_poll_cb = done_poll;
438 478
439 return 0; 479 return 0;
440} 480}
441 481
442X_THREAD_PROC (etp_proc); 482X_THREAD_PROC (etp_proc);
443 483
484static void ecb_cold
444static void etp_start_thread (void) 485etp_start_thread (void)
445{ 486{
446 etp_worker *wrk = calloc (1, sizeof (etp_worker)); 487 etp_worker *wrk = calloc (1, sizeof (etp_worker));
447 488
448 /*TODO*/ 489 /*TODO*/
449 assert (("unable to allocate worker thread data", wrk)); 490 assert (("unable to allocate worker thread data", wrk));
462 free (wrk); 503 free (wrk);
463 504
464 X_UNLOCK (wrklock); 505 X_UNLOCK (wrklock);
465} 506}
466 507
508static void
467static void etp_maybe_start_thread (void) 509etp_maybe_start_thread (void)
468{ 510{
469 if (expect_true (etp_nthreads () >= wanted)) 511 if (ecb_expect_true (etp_nthreads () >= wanted))
470 return; 512 return;
471 513
472 /* todo: maybe use idle here, but might be less exact */ 514 /* todo: maybe use idle here, but might be less exact */
473 if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ())) 515 if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
474 return; 516 return;
475 517
476 etp_start_thread (); 518 etp_start_thread ();
477} 519}
478 520
521static void ecb_cold
479static void etp_end_thread (void) 522etp_end_thread (void)
480{ 523{
481 eio_req *req = calloc (1, sizeof (eio_req)); 524 eio_req *req = calloc (1, sizeof (eio_req));
482 525
483 req->type = -1; 526 req->type = -1;
484 req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 527 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
491 X_LOCK (wrklock); 534 X_LOCK (wrklock);
492 --started; 535 --started;
493 X_UNLOCK (wrklock); 536 X_UNLOCK (wrklock);
494} 537}
495 538
496static int etp_poll (void) 539static int
540etp_poll (void)
497{ 541{
498 unsigned int maxreqs; 542 unsigned int maxreqs;
499 unsigned int maxtime; 543 unsigned int maxtime;
500 struct timeval tv_start, tv_now; 544 struct timeval tv_start, tv_now;
501 545
531 575
532 X_LOCK (reqlock); 576 X_LOCK (reqlock);
533 --nreqs; 577 --nreqs;
534 X_UNLOCK (reqlock); 578 X_UNLOCK (reqlock);
535 579
536 if (expect_false (req->type == EIO_GROUP && req->size)) 580 if (ecb_expect_false (req->type == EIO_GROUP && req->size))
537 { 581 {
538 req->int1 = 1; /* mark request as delayed */ 582 req->int1 = 1; /* mark request as delayed */
539 continue; 583 continue;
540 } 584 }
541 else 585 else
542 { 586 {
543 int res = ETP_FINISH (req); 587 int res = ETP_FINISH (req);
544 if (expect_false (res)) 588 if (ecb_expect_false (res))
545 return res; 589 return res;
546 } 590 }
547 591
548 if (expect_false (maxreqs && !--maxreqs)) 592 if (ecb_expect_false (maxreqs && !--maxreqs))
549 break; 593 break;
550 594
551 if (maxtime) 595 if (maxtime)
552 { 596 {
553 gettimeofday (&tv_now, 0); 597 gettimeofday (&tv_now, 0);
559 603
560 errno = EAGAIN; 604 errno = EAGAIN;
561 return -1; 605 return -1;
562} 606}
563 607
608static void
564static void etp_cancel (ETP_REQ *req) 609etp_cancel (ETP_REQ *req)
565{ 610{
566 X_LOCK (wrklock); 611 req->cancelled = 1;
567 req->flags |= EIO_FLAG_CANCELLED;
568 X_UNLOCK (wrklock);
569 612
570 eio_grp_cancel (req); 613 eio_grp_cancel (req);
571} 614}
572 615
616static void
573static void etp_submit (ETP_REQ *req) 617etp_submit (ETP_REQ *req)
574{ 618{
575 req->pri -= ETP_PRI_MIN; 619 req->pri -= ETP_PRI_MIN;
576 620
577 if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN; 621 if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
578 if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 622 if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
579 623
580 if (expect_false (req->type == EIO_GROUP)) 624 if (ecb_expect_false (req->type == EIO_GROUP))
581 { 625 {
582 /* I hope this is worth it :/ */ 626 /* I hope this is worth it :/ */
583 X_LOCK (reqlock); 627 X_LOCK (reqlock);
584 ++nreqs; 628 ++nreqs;
585 X_UNLOCK (reqlock); 629 X_UNLOCK (reqlock);
604 648
605 etp_maybe_start_thread (); 649 etp_maybe_start_thread ();
606 } 650 }
607} 651}
608 652
653static void ecb_cold
609static void etp_set_max_poll_time (double nseconds) 654etp_set_max_poll_time (double nseconds)
610{ 655{
611 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 656 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
612 max_poll_time = nseconds * EIO_TICKS; 657 max_poll_time = nseconds * EIO_TICKS;
613 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 658 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
614} 659}
615 660
661static void ecb_cold
616static void etp_set_max_poll_reqs (unsigned int maxreqs) 662etp_set_max_poll_reqs (unsigned int maxreqs)
617{ 663{
618 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 664 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
619 max_poll_reqs = maxreqs; 665 max_poll_reqs = maxreqs;
620 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 666 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
621} 667}
622 668
669static void ecb_cold
623static void etp_set_max_idle (unsigned int nthreads) 670etp_set_max_idle (unsigned int nthreads)
624{ 671{
625 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 672 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
626 max_idle = nthreads <= 0 ? 1 : nthreads; 673 max_idle = nthreads;
627 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 674 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
628} 675}
629 676
677static void ecb_cold
678etp_set_idle_timeout (unsigned int seconds)
679{
680 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
681 idle_timeout = seconds;
682 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
683}
684
685static void ecb_cold
630static void etp_set_min_parallel (unsigned int nthreads) 686etp_set_min_parallel (unsigned int nthreads)
631{ 687{
632 if (wanted < nthreads) 688 if (wanted < nthreads)
633 wanted = nthreads; 689 wanted = nthreads;
634} 690}
635 691
692static void ecb_cold
636static void etp_set_max_parallel (unsigned int nthreads) 693etp_set_max_parallel (unsigned int nthreads)
637{ 694{
638 if (wanted > nthreads) 695 if (wanted > nthreads)
639 wanted = nthreads; 696 wanted = nthreads;
640 697
641 while (started > wanted) 698 while (started > wanted)
642 etp_end_thread (); 699 etp_end_thread ();
643} 700}
644 701
645/*****************************************************************************/ 702/*****************************************************************************/
646 703
704static void
647static void grp_try_feed (eio_req *grp) 705grp_try_feed (eio_req *grp)
648{ 706{
649 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 707 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
650 { 708 {
651 grp->flags &= ~EIO_FLAG_GROUPADD; 709 grp->flags &= ~EIO_FLAG_GROUPADD;
652 710
659 break; 717 break;
660 } 718 }
661 } 719 }
662} 720}
663 721
722static int
664static int grp_dec (eio_req *grp) 723grp_dec (eio_req *grp)
665{ 724{
666 --grp->size; 725 --grp->size;
667 726
668 /* call feeder, if applicable */ 727 /* call feeder, if applicable */
669 grp_try_feed (grp); 728 grp_try_feed (grp);
673 return eio_finish (grp); 732 return eio_finish (grp);
674 else 733 else
675 return 0; 734 return 0;
676} 735}
677 736
737static void
678void eio_destroy (eio_req *req) 738eio_destroy (eio_req *req)
679{ 739{
680 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); 740 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
681 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); 741 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
682 742
683 EIO_DESTROY (req); 743 EIO_DESTROY (req);
684} 744}
685 745
746static int
686static int eio_finish (eio_req *req) 747eio_finish (eio_req *req)
687{ 748{
688 int res = EIO_FINISH (req); 749 int res = EIO_FINISH (req);
689 750
690 if (req->grp) 751 if (req->grp)
691 { 752 {
699 if (grp->grp_first == req) 760 if (grp->grp_first == req)
700 grp->grp_first = req->grp_next; 761 grp->grp_first = req->grp_next;
701 762
702 res2 = grp_dec (grp); 763 res2 = grp_dec (grp);
703 764
704 if (!res && res2) 765 if (!res)
705 res = res2; 766 res = res2;
706 } 767 }
707 768
708 eio_destroy (req); 769 eio_destroy (req);
709 770
710 return res; 771 return res;
711} 772}
712 773
774void
713void eio_grp_cancel (eio_req *grp) 775eio_grp_cancel (eio_req *grp)
714{ 776{
715 for (grp = grp->grp_first; grp; grp = grp->grp_next) 777 for (grp = grp->grp_first; grp; grp = grp->grp_next)
716 eio_cancel (grp); 778 eio_cancel (grp);
717} 779}
718 780
781void
719void eio_cancel (eio_req *req) 782eio_cancel (eio_req *req)
720{ 783{
721 etp_cancel (req); 784 etp_cancel (req);
722} 785}
723 786
787void
724void eio_submit (eio_req *req) 788eio_submit (eio_req *req)
725{ 789{
726 etp_submit (req); 790 etp_submit (req);
727} 791}
728 792
729unsigned int eio_nreqs (void) 793unsigned int
794eio_nreqs (void)
730{ 795{
731 return etp_nreqs (); 796 return etp_nreqs ();
732} 797}
733 798
734unsigned int eio_nready (void) 799unsigned int
800eio_nready (void)
735{ 801{
736 return etp_nready (); 802 return etp_nready ();
737} 803}
738 804
739unsigned int eio_npending (void) 805unsigned int
806eio_npending (void)
740{ 807{
741 return etp_npending (); 808 return etp_npending ();
742} 809}
743 810
744unsigned int eio_nthreads (void) 811unsigned int ecb_cold
812eio_nthreads (void)
745{ 813{
746 return etp_nthreads (); 814 return etp_nthreads ();
747} 815}
748 816
817void ecb_cold
749void eio_set_max_poll_time (double nseconds) 818eio_set_max_poll_time (double nseconds)
750{ 819{
751 etp_set_max_poll_time (nseconds); 820 etp_set_max_poll_time (nseconds);
752} 821}
753 822
823void ecb_cold
754void eio_set_max_poll_reqs (unsigned int maxreqs) 824eio_set_max_poll_reqs (unsigned int maxreqs)
755{ 825{
756 etp_set_max_poll_reqs (maxreqs); 826 etp_set_max_poll_reqs (maxreqs);
757} 827}
758 828
829void ecb_cold
759void eio_set_max_idle (unsigned int nthreads) 830eio_set_max_idle (unsigned int nthreads)
760{ 831{
761 etp_set_max_idle (nthreads); 832 etp_set_max_idle (nthreads);
762} 833}
763 834
835void ecb_cold
836eio_set_idle_timeout (unsigned int seconds)
837{
838 etp_set_idle_timeout (seconds);
839}
840
841void ecb_cold
764void eio_set_min_parallel (unsigned int nthreads) 842eio_set_min_parallel (unsigned int nthreads)
765{ 843{
766 etp_set_min_parallel (nthreads); 844 etp_set_min_parallel (nthreads);
767} 845}
768 846
847void ecb_cold
769void eio_set_max_parallel (unsigned int nthreads) 848eio_set_max_parallel (unsigned int nthreads)
770{ 849{
771 etp_set_max_parallel (nthreads); 850 etp_set_max_parallel (nthreads);
772} 851}
773 852
774int eio_poll (void) 853int eio_poll (void)
783# undef pread 862# undef pread
784# undef pwrite 863# undef pwrite
785# define pread eio__pread 864# define pread eio__pread
786# define pwrite eio__pwrite 865# define pwrite eio__pwrite
787 866
788static ssize_t 867static eio_ssize_t
789eio__pread (int fd, void *buf, size_t count, off_t offset) 868eio__pread (int fd, void *buf, size_t count, off_t offset)
790{ 869{
791 ssize_t res; 870 eio_ssize_t res;
792 off_t ooffset; 871 off_t ooffset;
793 872
794 X_LOCK (preadwritelock); 873 X_LOCK (preadwritelock);
795 ooffset = lseek (fd, 0, SEEK_CUR); 874 ooffset = lseek (fd, 0, SEEK_CUR);
796 lseek (fd, offset, SEEK_SET); 875 lseek (fd, offset, SEEK_SET);
799 X_UNLOCK (preadwritelock); 878 X_UNLOCK (preadwritelock);
800 879
801 return res; 880 return res;
802} 881}
803 882
804static ssize_t 883static eio_ssize_t
805eio__pwrite (int fd, void *buf, size_t count, off_t offset) 884eio__pwrite (int fd, void *buf, size_t count, off_t offset)
806{ 885{
807 ssize_t res; 886 eio_ssize_t res;
808 off_t ooffset; 887 off_t ooffset;
809 888
810 X_LOCK (preadwritelock); 889 X_LOCK (preadwritelock);
811 ooffset = lseek (fd, 0, SEEK_CUR); 890 ooffset = lseek (fd, 0, SEEK_CUR);
812 lseek (fd, offset, SEEK_SET); 891 lseek (fd, offset, SEEK_SET);
844#ifndef HAVE_FUTIMES 923#ifndef HAVE_FUTIMES
845 924
846# undef futimes 925# undef futimes
847# define futimes(fd,times) eio__futimes (fd, times) 926# define futimes(fd,times) eio__futimes (fd, times)
848 927
928static int
849static int eio__futimes (int fd, const struct timeval tv[2]) 929eio__futimes (int fd, const struct timeval tv[2])
850{ 930{
851 errno = ENOSYS; 931 errno = ENOSYS;
852 return -1; 932 return -1;
853} 933}
854 934
858# undef fdatasync 938# undef fdatasync
859# define fdatasync(fd) fsync (fd) 939# define fdatasync(fd) fsync (fd)
860#endif 940#endif
861 941
862/* sync_file_range always needs emulation */ 942/* sync_file_range always needs emulation */
863int 943static int
864eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags) 944eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
865{ 945{
866#if HAVE_SYNC_FILE_RANGE 946#if HAVE_SYNC_FILE_RANGE
867 int res; 947 int res;
868 948
885 /* even though we could play tricks with the flags, it's better to always 965 /* even though we could play tricks with the flags, it's better to always
886 * call fdatasync, as that matches the expectation of its users best */ 966 * call fdatasync, as that matches the expectation of its users best */
887 return fdatasync (fd); 967 return fdatasync (fd);
888} 968}
889 969
970static int
971eio__fallocate (int fd, int mode, off_t offset, size_t len)
972{
973#if HAVE_FALLOCATE
974 return fallocate (fd, mode, offset, len);
975#else
976 errno = ENOSYS;
977 return -1;
978#endif
979}
980
890#if !HAVE_READAHEAD 981#if !HAVE_READAHEAD
891# undef readahead 982# undef readahead
892# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 983# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
893 984
894static ssize_t 985static eio_ssize_t
895eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) 986eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
896{ 987{
897 size_t todo = count; 988 size_t todo = count;
898 dBUF; 989 dBUF;
899 990
904 pread (fd, eio_buf, len, offset); 995 pread (fd, eio_buf, len, offset);
905 offset += len; 996 offset += len;
906 todo -= len; 997 todo -= len;
907 } 998 }
908 999
1000 FUBd;
1001
909 errno = 0; 1002 errno = 0;
910 return count; 1003 return count;
911} 1004}
912 1005
913#endif 1006#endif
914 1007
915/* sendfile always needs emulation */ 1008/* sendfile always needs emulation */
916static ssize_t 1009static eio_ssize_t
917eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) 1010eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
918{ 1011{
1012 eio_ssize_t written = 0;
919 ssize_t res; 1013 eio_ssize_t res;
920 1014
921 if (!count) 1015 if (!count)
922 return 0; 1016 return 0;
923 1017
1018 for (;;)
1019 {
1020#ifdef __APPLE__
1021# undef HAVE_SENDFILE /* broken, as everything on os x */
1022#endif
924#if HAVE_SENDFILE 1023#if HAVE_SENDFILE
925# if __linux 1024# if __linux
1025 off_t soffset = offset;
926 res = sendfile (ofd, ifd, &offset, count); 1026 res = sendfile (ofd, ifd, &soffset, count);
927 1027
928# elif __FreeBSD__ 1028# elif __FreeBSD__
929 /* 1029 /*
930 * Of course, the freebsd sendfile is a dire hack with no thoughts 1030 * Of course, the freebsd sendfile is a dire hack with no thoughts
931 * wasted on making it similar to other I/O functions. 1031 * wasted on making it similar to other I/O functions.
932 */ 1032 */
933 {
934 off_t sbytes; 1033 off_t sbytes;
935 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 1034 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
936 1035
937 #if 0 /* according to the manpage, this is correct, but broken behaviour */ 1036 #if 0 /* according to the manpage, this is correct, but broken behaviour */
938 /* freebsd' sendfile will return 0 on success */ 1037 /* freebsd' sendfile will return 0 on success */
939 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */ 1038 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
940 /* not on e.g. EIO or EPIPE - sounds broken */ 1039 /* not on e.g. EIO or EPIPE - sounds broken */
941 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0) 1040 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
942 res = sbytes; 1041 res = sbytes;
943 #endif 1042 #endif
944 1043
945 /* according to source inspection, this is correct, and useful behaviour */ 1044 /* according to source inspection, this is correct, and useful behaviour */
946 if (sbytes) 1045 if (sbytes)
947 res = sbytes; 1046 res = sbytes;
948 }
949 1047
950# elif defined (__APPLE__) 1048# elif defined (__APPLE__)
951
952 {
953 off_t sbytes = count; 1049 off_t sbytes = count;
954 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0); 1050 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
955 1051
956 /* according to the manpage, sbytes is always valid */ 1052 /* according to the manpage, sbytes is always valid */
957 if (sbytes) 1053 if (sbytes)
958 res = sbytes; 1054 res = sbytes;
959 }
960 1055
961# elif __hpux 1056# elif __hpux
962 res = sendfile (ofd, ifd, offset, count, 0, 0); 1057 res = sendfile (ofd, ifd, offset, count, 0, 0);
963 1058
964# elif __solaris 1059# elif __solaris
965 {
966 struct sendfilevec vec; 1060 struct sendfilevec vec;
967 size_t sbytes; 1061 size_t sbytes;
968 1062
969 vec.sfv_fd = ifd; 1063 vec.sfv_fd = ifd;
970 vec.sfv_flag = 0; 1064 vec.sfv_flag = 0;
971 vec.sfv_off = offset; 1065 vec.sfv_off = offset;
972 vec.sfv_len = count; 1066 vec.sfv_len = count;
973 1067
974 res = sendfilev (ofd, &vec, 1, &sbytes); 1068 res = sendfilev (ofd, &vec, 1, &sbytes);
975 1069
976 if (res < 0 && sbytes) 1070 if (res < 0 && sbytes)
977 res = sbytes; 1071 res = sbytes;
978 }
979 1072
980# endif 1073# endif
981 1074
982#elif defined (_WIN32) 1075#elif defined (_WIN32) && 0
983
984 /* does not work, just for documentation of what would need to be done */ 1076 /* does not work, just for documentation of what would need to be done */
985 { 1077 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
1078 /* libeio guarantees that the file offset does not change, and windows */
1079 /* has no way to get an independent handle to the same file description */
986 HANDLE h = TO_SOCKET (ifd); 1080 HANDLE h = TO_SOCKET (ifd);
987 SetFilePointer (h, offset, 0, FILE_BEGIN); 1081 SetFilePointer (h, offset, 0, FILE_BEGIN);
988 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0); 1082 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
989 }
990 1083
991#else 1084#else
992 res = -1; 1085 res = -1;
993 errno = ENOSYS; 1086 errno = ENOSYS;
994#endif 1087#endif
995 1088
1089 /* we assume sendfile can copy at least 128mb in one go */
1090 if (res <= 128 * 1024 * 1024)
1091 {
1092 if (res > 0)
1093 written += res;
1094
1095 if (written)
1096 return written;
1097
1098 break;
1099 }
1100 else
1101 {
1102 /* if we requested more, then probably the kernel was lazy */
1103 written += res;
1104 offset += res;
1105 count -= res;
1106
1107 if (!count)
1108 return written;
1109 }
1110 }
1111
996 if (res < 0 1112 if (res < 0
997 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1113 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
998 /* BSDs */ 1114 /* BSDs */
999#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */ 1115#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1000 || errno == ENOTSUP 1116 || errno == ENOTSUP
1001#endif 1117#endif
1118#ifdef EOPNOTSUPP /* windows */
1002 || errno == EOPNOTSUPP /* BSDs */ 1119 || errno == EOPNOTSUPP /* BSDs */
1120#endif
1003#if __solaris 1121#if __solaris
1004 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1122 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
1005#endif 1123#endif
1006 ) 1124 )
1007 ) 1125 )
1011 1129
1012 res = 0; 1130 res = 0;
1013 1131
1014 while (count) 1132 while (count)
1015 { 1133 {
1016 ssize_t cnt; 1134 eio_ssize_t cnt;
1017 1135
1018 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 1136 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
1019 1137
1020 if (cnt <= 0) 1138 if (cnt <= 0)
1021 { 1139 {
1033 1151
1034 offset += cnt; 1152 offset += cnt;
1035 res += cnt; 1153 res += cnt;
1036 count -= cnt; 1154 count -= cnt;
1037 } 1155 }
1156
1157 FUBd;
1038 } 1158 }
1039 1159
1040 return res; 1160 return res;
1161}
1162
1163#ifdef PAGESIZE
1164# define eio_pagesize() PAGESIZE
1165#else
1166static intptr_t
1167eio_pagesize (void)
1168{
1169 static intptr_t page;
1170
1171 if (!page)
1172 page = sysconf (_SC_PAGESIZE);
1173
1174 return page;
1175}
1176#endif
1177
1178static void
1179eio_page_align (void **addr, size_t *length)
1180{
1181 intptr_t mask = eio_pagesize () - 1;
1182
1183 /* round down addr */
1184 intptr_t adj = mask & (intptr_t)*addr;
1185
1186 *addr = (void *)((intptr_t)*addr - adj);
1187 *length += adj;
1188
1189 /* round up length */
1190 *length = (*length + mask) & ~mask;
1191}
1192
1193#if !_POSIX_MEMLOCK
1194# define eio__mlockall(a) EIO_ENOSYS ()
1195#else
1196
1197static int
1198eio__mlockall (int flags)
1199{
1200 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1201 extern int mallopt (int, int);
1202 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1203 #endif
1204
1205 if (EIO_MCL_CURRENT != MCL_CURRENT
1206 || EIO_MCL_FUTURE != MCL_FUTURE)
1207 {
1208 flags = 0
1209 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1210 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1211 }
1212
1213 return mlockall (flags);
1214}
1215#endif
1216
1217#if !_POSIX_MEMLOCK_RANGE
1218# define eio__mlock(a,b) EIO_ENOSYS ()
1219#else
1220
1221static int
1222eio__mlock (void *addr, size_t length)
1223{
1224 eio_page_align (&addr, &length);
1225
1226 return mlock (addr, length);
1227}
1228
1229#endif
1230
1231#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1232# define eio__msync(a,b,c) EIO_ENOSYS ()
1233#else
1234
1235static int
1236eio__msync (void *mem, size_t len, int flags)
1237{
1238 eio_page_align (&mem, &len);
1239
1240 if (EIO_MS_ASYNC != MS_SYNC
1241 || EIO_MS_INVALIDATE != MS_INVALIDATE
1242 || EIO_MS_SYNC != MS_SYNC)
1243 {
1244 flags = 0
1245 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1246 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1247 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1248 }
1249
1250 return msync (mem, len, flags);
1251}
1252
1253#endif
1254
1255static int
1256eio__mtouch (eio_req *req)
1257{
1258 void *mem = req->ptr2;
1259 size_t len = req->size;
1260 int flags = req->int1;
1261
1262 eio_page_align (&mem, &len);
1263
1264 {
1265 intptr_t addr = (intptr_t)mem;
1266 intptr_t end = addr + len;
1267 intptr_t page = eio_pagesize ();
1268
1269 if (addr < end)
1270 if (flags & EIO_MT_MODIFY) /* modify */
1271 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1272 else
1273 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1274 }
1275
1276 return 0;
1277}
1278
1279/*****************************************************************************/
1280/* requests implemented outside eio_execute, because they are so large */
1281
1282static void
1283eio__realpath (eio_req *req, etp_worker *self)
1284{
1285 char *rel = req->ptr1;
1286 char *res;
1287 char *tmp1, *tmp2;
1288#if SYMLOOP_MAX > 32
1289 int symlinks = SYMLOOP_MAX;
1290#else
1291 int symlinks = 32;
1292#endif
1293
1294 req->result = -1;
1295
1296 errno = EINVAL;
1297 if (!rel)
1298 return;
1299
1300 errno = ENOENT;
1301 if (!*rel)
1302 return;
1303
1304 if (!req->ptr2)
1305 {
1306 X_LOCK (wrklock);
1307 req->flags |= EIO_FLAG_PTR2_FREE;
1308 X_UNLOCK (wrklock);
1309 req->ptr2 = malloc (PATH_MAX * 3);
1310
1311 errno = ENOMEM;
1312 if (!req->ptr2)
1313 return;
1314 }
1315
1316 res = req->ptr2;
1317 tmp1 = res + PATH_MAX;
1318 tmp2 = tmp1 + PATH_MAX;
1319
1320#if 0 /* disabled, the musl way to do things is just too racy */
1321#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1322 /* on linux we may be able to ask the kernel */
1323 {
1324 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1325
1326 if (fd >= 0)
1327 {
1328 sprintf (tmp1, "/proc/self/fd/%d", fd);
1329 req->result = readlink (tmp1, res, PATH_MAX);
1330 close (fd);
1331
1332 /* here we should probably stat the open file and the disk file, to make sure they still match */
1333
1334 if (req->result > 0)
1335 goto done;
1336 }
1337 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1338 return;
1339 }
1340#endif
1341#endif
1342
1343 if (*rel != '/')
1344 {
1345 if (!getcwd (res, PATH_MAX))
1346 return;
1347
1348 if (res [1]) /* only use if not / */
1349 res += strlen (res);
1350 }
1351
1352 while (*rel)
1353 {
1354 eio_ssize_t len, linklen;
1355 char *beg = rel;
1356
1357 while (*rel && *rel != '/')
1358 ++rel;
1359
1360 len = rel - beg;
1361
1362 if (!len) /* skip slashes */
1363 {
1364 ++rel;
1365 continue;
1366 }
1367
1368 if (beg [0] == '.')
1369 {
1370 if (len == 1)
1371 continue; /* . - nop */
1372
1373 if (beg [1] == '.' && len == 2)
1374 {
1375 /* .. - back up one component, if possible */
1376
1377 while (res != req->ptr2)
1378 if (*--res == '/')
1379 break;
1380
1381 continue;
1382 }
1383 }
1384
1385 errno = ENAMETOOLONG;
1386 if (res + 1 + len + 1 >= tmp1)
1387 return;
1388
1389 /* copy one component */
1390 *res = '/';
1391 memcpy (res + 1, beg, len);
1392
1393 /* zero-terminate, for readlink */
1394 res [len + 1] = 0;
1395
1396 /* now check if it's a symlink */
1397 linklen = readlink (req->ptr2, tmp1, PATH_MAX);
1398
1399 if (linklen < 0)
1400 {
1401 if (errno != EINVAL)
1402 return;
1403
1404 /* it's a normal directory. hopefully */
1405 res += len + 1;
1406 }
1407 else
1408 {
1409 /* yay, it was a symlink - build new path in tmp2 */
1410 int rellen = strlen (rel);
1411
1412 errno = ENAMETOOLONG;
1413 if (linklen + 1 + rellen >= PATH_MAX)
1414 return;
1415
1416 errno = ELOOP;
1417 if (!--symlinks)
1418 return;
1419
1420 if (*tmp1 == '/')
1421 res = req->ptr2; /* symlink resolves to an absolute path */
1422
1423 /* we need to be careful, as rel might point into tmp2 already */
1424 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1425 tmp2 [linklen] = '/';
1426 memcpy (tmp2, tmp1, linklen);
1427
1428 rel = tmp2;
1429 }
1430 }
1431
1432 /* special case for the lone root path */
1433 if (res == req->ptr2)
1434 *res++ = '/';
1435
1436 req->result = res - (char *)req->ptr2;
1437
1438done:
1439 req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
1041} 1440}
1042 1441
1043static signed char 1442static signed char
1044eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) 1443eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1045{ 1444{
1046 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */ 1445 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1047 : a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0; 1446 : a->inode < b->inode ? -1
1447 : a->inode > b->inode ? 1
1448 : 0;
1048} 1449}
1049 1450
1050#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0 1451#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1051 1452
1052#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */ 1453#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1053#define EIO_SORT_FAST 60 /* when to only use insertion sort */ 1454#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1054 1455
1055static void 1456static void
1056eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1457eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1057{ 1458{
1058 unsigned char bits [9 + sizeof (ino_t) * 8]; 1459 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1059 unsigned char *bit = bits; 1460 unsigned char *bit = bits;
1060 1461
1061 assert (CHAR_BIT == 8); 1462 assert (CHAR_BIT == 8);
1062 assert (sizeof (eio_dirent) * 8 < 256); 1463 assert (sizeof (eio_dirent) * 8 < 256);
1063 assert (offsetof (eio_dirent, inode)); /* we use 0 as sentinel */ 1464 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1064 assert (offsetof (eio_dirent, score)); /* we use 0 as sentinel */ 1465 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1065 1466
1066 if (size <= EIO_SORT_FAST) 1467 if (size <= EIO_SORT_FAST)
1067 return; 1468 return;
1068 1469
1069 /* first prepare an array of bits to test in our radix sort */ 1470 /* first prepare an array of bits to test in our radix sort */
1070 /* try to take endianness into account, as well as differences in ino_t sizes */ 1471 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1071 /* inode_bits must contain all inodes ORed together */ 1472 /* inode_bits must contain all inodes ORed together */
1072 /* which is used to skip bits that are 0 everywhere, which is very common */ 1473 /* which is used to skip bits that are 0 everywhere, which is very common */
1073 { 1474 {
1074 ino_t endianness; 1475 eio_ino_t endianness;
1075 int i, j; 1476 int i, j;
1076 1477
1077 /* we store the byte offset of byte n into byte n of "endianness" */ 1478 /* we store the byte offset of byte n into byte n of "endianness" */
1078 for (i = 0; i < sizeof (ino_t); ++i) 1479 for (i = 0; i < sizeof (eio_ino_t); ++i)
1079 ((unsigned char *)&endianness)[i] = i; 1480 ((unsigned char *)&endianness)[i] = i;
1080 1481
1081 *bit++ = 0; 1482 *bit++ = 0;
1082 1483
1083 for (i = 0; i < sizeof (ino_t); ++i) 1484 for (i = 0; i < sizeof (eio_ino_t); ++i)
1084 { 1485 {
1085 /* shifting off the byte offsets out of "endianness" */ 1486 /* shifting off the byte offsets out of "endianness" */
1086 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8; 1487 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1087 endianness >>= 8; 1488 endianness >>= 8;
1088 1489
1089 for (j = 0; j < 8; ++j) 1490 for (j = 0; j < 8; ++j)
1090 if (inode_bits & (((ino_t)1) << (i * 8 + j))) 1491 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1091 *bit++ = offs + j; 1492 *bit++ = offs + j;
1092 } 1493 }
1093 1494
1094 for (j = 0; j < 8; ++j) 1495 for (j = 0; j < 8; ++j)
1095 if (score_bits & (1 << j)) 1496 if (score_bits & (1 << j))
1096 *bit++ = offsetof (eio_dirent, score) * 8 + j; 1497 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1097 } 1498 }
1098 1499
1099 /* now actually do the sorting (a variant of MSD radix sort) */ 1500 /* now actually do the sorting (a variant of MSD radix sort) */
1100 { 1501 {
1101 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base; 1502 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1102 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end; 1503 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1103 unsigned char *bit_stk [9 + sizeof (ino_t) * 8]; 1504 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1104 int stk_idx = 0; 1505 int stk_idx = 0;
1105 1506
1106 base_stk [stk_idx] = dents; 1507 base_stk [stk_idx] = dents;
1107 end_stk [stk_idx] = dents + size; 1508 end_stk [stk_idx] = dents + size;
1108 bit_stk [stk_idx] = bit - 1; 1509 bit_stk [stk_idx] = bit - 1;
1187 } 1588 }
1188 } 1589 }
1189} 1590}
1190 1591
1191static void 1592static void
1192eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1593eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1193{ 1594{
1194 if (size <= 1) 1595 if (size <= 1)
1195 return; /* our insertion sort relies on size > 0 */ 1596 return; /* our insertion sort relies on size > 0 */
1196 1597
1197 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */ 1598 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1205 1606
1206/* read a full directory */ 1607/* read a full directory */
1207static void 1608static void
1208eio__scandir (eio_req *req, etp_worker *self) 1609eio__scandir (eio_req *req, etp_worker *self)
1209{ 1610{
1210 DIR *dirp;
1211 EIO_STRUCT_DIRENT *entp;
1212 char *name, *names; 1611 char *name, *names;
1213 int namesalloc = 4096; 1612 int namesalloc = 4096 - sizeof (void *) * 4;
1214 int namesoffs = 0; 1613 int namesoffs = 0;
1215 int flags = req->int1; 1614 int flags = req->int1;
1216 eio_dirent *dents = 0; 1615 eio_dirent *dents = 0;
1217 int dentalloc = 128; 1616 int dentalloc = 128;
1218 int dentoffs = 0; 1617 int dentoffs = 0;
1219 ino_t inode_bits = 0; 1618 eio_ino_t inode_bits = 0;
1619#ifdef _WIN32
1620 HANDLE dirp;
1621 WIN32_FIND_DATA entp;
1622#else
1623 DIR *dirp;
1624 EIO_STRUCT_DIRENT *entp;
1625#endif
1220 1626
1221 req->result = -1; 1627 req->result = -1;
1222 1628
1223 if (!(flags & EIO_READDIR_DENTS)) 1629 if (!(flags & EIO_READDIR_DENTS))
1224 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); 1630 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1225 1631
1226 X_LOCK (wrklock); 1632#ifdef _WIN32
1227 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 1633 {
1634 int len = strlen ((const char *)req->ptr1);
1635 char *path = malloc (MAX_PATH);
1636 const char *fmt;
1637
1638 if (!len)
1639 fmt = "./*";
1640 else if (((const char *)req->ptr1)[len - 1] == '/' || ((const char *)req->ptr1)[len - 1] == '\\')
1641 fmt = "%s*";
1642 else
1643 fmt = "%s/*";
1644
1645 _snprintf (path, MAX_PATH, fmt, (const char *)req->ptr1);
1646 dirp = FindFirstFile (path, &entp);
1647 free (path);
1648
1649 if (dirp == INVALID_HANDLE_VALUE)
1650 {
1651 dirp = 0;
1652
1653 switch (GetLastError ())
1654 {
1655 case ERROR_FILE_NOT_FOUND:
1656 req->result = 0;
1657 break;
1658
1659 case ERROR_INVALID_NAME:
1660 case ERROR_PATH_NOT_FOUND:
1661 case ERROR_NO_MORE_FILES:
1662 errno = ENOENT;
1663 break;
1664
1665 case ERROR_NOT_ENOUGH_MEMORY:
1666 errno = ENOMEM;
1667 break;
1668
1669 default:
1670 errno = EINVAL;
1671 break;
1672 }
1673 }
1674 }
1675#else
1228 self->dirp = dirp = opendir (req->ptr1); 1676 dirp = opendir (req->ptr1);
1677#endif
1678
1679 if (req->flags & EIO_FLAG_PTR1_FREE)
1680 free (req->ptr1);
1681
1229 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1682 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1230 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; 1683 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1231 req->ptr2 = names = malloc (namesalloc); 1684 req->ptr2 = names = malloc (namesalloc);
1232 X_UNLOCK (wrklock);
1233 1685
1234 if (dirp && names && (!flags || dents)) 1686 if (dirp && names && (!flags || dents))
1235 for (;;) 1687 for (;;)
1236 { 1688 {
1689 int done;
1690
1691#ifdef _WIN32
1692 done = !dirp;
1693#else
1237 errno = 0; 1694 errno = 0;
1238 entp = readdir (dirp); 1695 entp = readdir (dirp);
1696 done = !entp;
1697#endif
1239 1698
1240 if (!entp) 1699 if (done)
1241 { 1700 {
1701#ifndef _WIN32
1702 int old_errno = errno;
1703 closedir (dirp);
1704 errno = old_errno;
1705
1242 if (errno) 1706 if (errno)
1243 break; 1707 break;
1708#endif
1244 1709
1245 /* sort etc. */ 1710 /* sort etc. */
1246 req->int1 = flags; 1711 req->int1 = flags;
1247 req->result = dentoffs; 1712 req->result = dentoffs;
1248 1713
1249 if (flags & EIO_READDIR_STAT_ORDER) 1714 if (flags & EIO_READDIR_STAT_ORDER)
1250 eio_dent_sort (dents, dentoffs, 0, inode_bits); /* sort by inode exclusively */ 1715 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1251 else if (flags & EIO_READDIR_DIRS_FIRST) 1716 else if (flags & EIO_READDIR_DIRS_FIRST)
1252 if (flags & EIO_READDIR_FOUND_UNKNOWN) 1717 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1253 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */ 1718 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1254 else 1719 else
1255 { 1720 {
1257 eio_dirent *oth = dents + dentoffs; 1722 eio_dirent *oth = dents + dentoffs;
1258 eio_dirent *dir = dents; 1723 eio_dirent *dir = dents;
1259 1724
1260 /* now partition dirs to the front, and non-dirs to the back */ 1725 /* now partition dirs to the front, and non-dirs to the back */
1261 /* by walking from both sides and swapping if necessary */ 1726 /* by walking from both sides and swapping if necessary */
1262 /* also clear score, so it doesn't influence sorting */
1263 while (oth > dir) 1727 while (oth > dir)
1264 { 1728 {
1265 if (dir->type == EIO_DT_DIR) 1729 if (dir->type == EIO_DT_DIR)
1266 ++dir; 1730 ++dir;
1267 else if ((--oth)->type == EIO_DT_DIR) 1731 else if ((--oth)->type == EIO_DT_DIR)
1270 1734
1271 ++dir; 1735 ++dir;
1272 } 1736 }
1273 } 1737 }
1274 1738
1275 /* now sort the dirs only */ 1739 /* now sort the dirs only (dirs all have the same score) */
1276 eio_dent_sort (dents, dir - dents, 0, inode_bits); 1740 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1277 } 1741 }
1278 1742
1279 break; 1743 break;
1280 } 1744 }
1281 1745
1282 /* now add the entry to our list(s) */ 1746 /* now add the entry to our list(s) */
1283 name = entp->d_name; 1747 name = D_NAME (entp);
1284 1748
1285 /* skip . and .. entries */ 1749 /* skip . and .. entries */
1286 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1750 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1287 { 1751 {
1288 int len = D_NAMLEN (entp) + 1; 1752 int len = D_NAMLEN (entp) + 1;
1289 1753
1290 while (expect_false (namesoffs + len > namesalloc)) 1754 while (ecb_expect_false (namesoffs + len > namesalloc))
1291 { 1755 {
1292 namesalloc *= 2; 1756 namesalloc *= 2;
1293 X_LOCK (wrklock);
1294 req->ptr2 = names = realloc (names, namesalloc); 1757 req->ptr2 = names = realloc (names, namesalloc);
1295 X_UNLOCK (wrklock);
1296 1758
1297 if (!names) 1759 if (!names)
1298 break; 1760 break;
1299 } 1761 }
1300 1762
1302 1764
1303 if (dents) 1765 if (dents)
1304 { 1766 {
1305 struct eio_dirent *ent; 1767 struct eio_dirent *ent;
1306 1768
1307 if (expect_false (dentoffs == dentalloc)) 1769 if (ecb_expect_false (dentoffs == dentalloc))
1308 { 1770 {
1309 dentalloc *= 2; 1771 dentalloc *= 2;
1310 X_LOCK (wrklock);
1311 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent)); 1772 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1312 X_UNLOCK (wrklock);
1313 1773
1314 if (!dents) 1774 if (!dents)
1315 break; 1775 break;
1316 } 1776 }
1317 1777
1397 if (EIO_CANCELLED (req)) 1857 if (EIO_CANCELLED (req))
1398 { 1858 {
1399 errno = ECANCELED; 1859 errno = ECANCELED;
1400 break; 1860 break;
1401 } 1861 }
1862
1863#ifdef _WIN32
1864 if (!FindNextFile (dirp, &entp))
1865 {
1866 FindClose (dirp);
1867 dirp = 0;
1868 }
1869#endif
1402 } 1870 }
1403}
1404
1405#ifdef PAGESIZE
1406# define eio_pagesize() PAGESIZE
1407#else
1408static intptr_t
1409eio_pagesize (void)
1410{
1411 static intptr_t page;
1412
1413 if (!page)
1414 page = sysconf (_SC_PAGESIZE);
1415
1416 return page;
1417}
1418#endif
1419
1420static void
1421eio_page_align (void **addr, size_t *length)
1422{
1423 intptr_t mask = eio_pagesize () - 1;
1424
1425 /* round down addr */
1426 intptr_t adj = mask & (intptr_t)*addr;
1427
1428 *addr = (void *)((intptr_t)*addr - adj);
1429 *length += adj;
1430
1431 /* round up length */
1432 *length = (*length + mask) & ~mask;
1433}
1434
1435#if !_POSIX_MEMLOCK
1436# define eio__mlockall(a) ((errno = ENOSYS), -1)
1437#else
1438
1439static int
1440eio__mlockall (int flags)
1441{
1442 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1443 extern int mallopt (int, int);
1444 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1445 #endif
1446
1447 if (EIO_MCL_CURRENT != MCL_CURRENT
1448 || EIO_MCL_FUTURE != MCL_FUTURE)
1449 {
1450 flags = 0
1451 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1452 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1453 }
1454
1455 return mlockall (flags);
1456}
1457#endif
1458
1459#if !_POSIX_MEMLOCK_RANGE
1460# define eio__mlock(a,b) ((errno = ENOSYS), -1)
1461#else
1462
1463static int
1464eio__mlock (void *addr, size_t length)
1465{
1466 eio_page_align (&addr, &length);
1467
1468 return mlock (addr, length);
1469}
1470
1471#endif
1472
1473#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1474# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
1475#else
1476
1477int
1478eio__msync (void *mem, size_t len, int flags)
1479{
1480 eio_page_align (&mem, &len);
1481
1482 if (EIO_MS_ASYNC != MS_SYNC
1483 || EIO_MS_INVALIDATE != MS_INVALIDATE
1484 || EIO_MS_SYNC != MS_SYNC)
1485 {
1486 flags = 0
1487 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1488 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1489 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1490 }
1491
1492 return msync (mem, len, flags);
1493}
1494
1495#endif
1496
1497int
1498eio__mtouch (void *mem, size_t len, int flags)
1499{
1500 eio_page_align (&mem, &len);
1501
1502 {
1503 intptr_t addr = (intptr_t)mem;
1504 intptr_t end = addr + len;
1505 intptr_t page = eio_pagesize ();
1506
1507 if (addr < end)
1508 if (flags & EIO_MT_MODIFY) /* modify */
1509 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
1510 else
1511 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
1512 }
1513
1514 return 0;
1515} 1871}
1516 1872
1517/*****************************************************************************/ 1873/*****************************************************************************/
1518 1874
1519#define ALLOC(len) \ 1875#define ALLOC(len) \
1535{ 1891{
1536 ETP_REQ *req; 1892 ETP_REQ *req;
1537 struct timespec ts; 1893 struct timespec ts;
1538 etp_worker *self = (etp_worker *)thr_arg; 1894 etp_worker *self = (etp_worker *)thr_arg;
1539 1895
1540 /* try to distribute timeouts somewhat randomly */ 1896 /* try to distribute timeouts somewhat evenly */
1541 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 1897 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1542 1898
1543 for (;;) 1899 for (;;)
1544 { 1900 {
1901 ts.tv_sec = 0;
1902
1545 X_LOCK (reqlock); 1903 X_LOCK (reqlock);
1546 1904
1547 for (;;) 1905 for (;;)
1548 { 1906 {
1549 self->req = req = reqq_shift (&req_queue); 1907 self->req = req = reqq_shift (&req_queue);
1550 1908
1551 if (req) 1909 if (req)
1552 break; 1910 break;
1553 1911
1912 if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
1913 {
1914 X_UNLOCK (reqlock);
1915 X_LOCK (wrklock);
1916 --started;
1917 X_UNLOCK (wrklock);
1918 goto quit;
1919 }
1920
1554 ++idle; 1921 ++idle;
1555 1922
1556 ts.tv_sec = time (0) + IDLE_TIMEOUT; 1923 if (idle <= max_idle)
1557 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 1924 /* we are allowed to idle, so do so without any timeout */
1925 X_COND_WAIT (reqwait, reqlock);
1926 else
1558 { 1927 {
1559 if (idle > max_idle) 1928 /* initialise timeout once */
1560 { 1929 if (!ts.tv_sec)
1561 --idle; 1930 ts.tv_sec = time (0) + idle_timeout;
1562 X_UNLOCK (reqlock);
1563 X_LOCK (wrklock);
1564 --started;
1565 X_UNLOCK (wrklock);
1566 goto quit;
1567 }
1568 1931
1569 /* we are allowed to idle, so do so without any timeout */
1570 X_COND_WAIT (reqwait, reqlock); 1932 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
1933 ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1571 } 1934 }
1572 1935
1573 --idle; 1936 --idle;
1574 } 1937 }
1575 1938
1578 X_UNLOCK (reqlock); 1941 X_UNLOCK (reqlock);
1579 1942
1580 if (req->type < 0) 1943 if (req->type < 0)
1581 goto quit; 1944 goto quit;
1582 1945
1583 if (!EIO_CANCELLED (req))
1584 ETP_EXECUTE (self, req); 1946 ETP_EXECUTE (self, req);
1585 1947
1586 X_LOCK (reslock); 1948 X_LOCK (reslock);
1587 1949
1588 ++npending; 1950 ++npending;
1589 1951
1604 return 0; 1966 return 0;
1605} 1967}
1606 1968
1607/*****************************************************************************/ 1969/*****************************************************************************/
1608 1970
1971int ecb_cold
1609int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1972eio_init (void (*want_poll)(void), void (*done_poll)(void))
1610{ 1973{
1974#if !HAVE_PREADWRITE
1975 X_MUTEX_CREATE (preadwritelock);
1976#endif
1977
1611 return etp_init (want_poll, done_poll); 1978 return etp_init (want_poll, done_poll);
1612} 1979}
1613 1980
1981ecb_inline void
1614static void eio_api_destroy (eio_req *req) 1982eio_api_destroy (eio_req *req)
1615{ 1983{
1616 free (req); 1984 free (req);
1617} 1985}
1618 1986
1619#define REQ(rtype) \ 1987#define REQ(rtype) \
1638 { \ 2006 { \
1639 eio_api_destroy (req); \ 2007 eio_api_destroy (req); \
1640 return 0; \ 2008 return 0; \
1641 } 2009 }
1642 2010
2011static void
1643static void eio_execute (etp_worker *self, eio_req *req) 2012eio_execute (etp_worker *self, eio_req *req)
1644{ 2013{
2014 if (ecb_expect_false (EIO_CANCELLED (req)))
2015 {
2016 req->result = -1;
2017 req->errorno = ECANCELED;
2018 return;
2019 }
2020
1645 switch (req->type) 2021 switch (req->type)
1646 { 2022 {
1647 case EIO_READ: ALLOC (req->size); 2023 case EIO_READ: ALLOC (req->size);
1648 req->result = req->offs >= 0 2024 req->result = req->offs >= 0
1649 ? pread (req->int1, req->ptr2, req->size, req->offs) 2025 ? pread (req->int1, req->ptr2, req->size, req->offs)
1651 case EIO_WRITE: req->result = req->offs >= 0 2027 case EIO_WRITE: req->result = req->offs >= 0
1652 ? pwrite (req->int1, req->ptr2, req->size, req->offs) 2028 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1653 : write (req->int1, req->ptr2, req->size); break; 2029 : write (req->int1, req->ptr2, req->size); break;
1654 2030
1655 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break; 2031 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1656 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break; 2032 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
1657 2033
1658 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2034 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1659 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2035 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1660 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2036 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1661 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2037 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1683 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; 2059 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1684 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; 2060 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1685 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; 2061 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1686 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break; 2062 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1687 2063
2064 case EIO_REALPATH: eio__realpath (req, self); break;
2065
1688 case EIO_READLINK: ALLOC (PATH_MAX); 2066 case EIO_READLINK: ALLOC (PATH_MAX);
1689 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break; 2067 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1690 2068
1691 case EIO_SYNC: req->result = 0; sync (); break; 2069 case EIO_SYNC: req->result = 0; sync (); break;
1692 case EIO_FSYNC: req->result = fsync (req->int1); break; 2070 case EIO_FSYNC: req->result = fsync (req->int1); break;
1693 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 2071 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1694 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break; 2072 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1695 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break; 2073 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1696 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break; 2074 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1697 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break; 2075 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1698 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break; 2076 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
2077 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
1699 2078
1700 case EIO_READDIR: eio__scandir (req, self); break; 2079 case EIO_READDIR: eio__scandir (req, self); break;
1701 2080
1702 case EIO_BUSY: 2081 case EIO_BUSY:
1703#ifdef _WIN32 2082#ifdef _WIN32
1744 case EIO_NOP: 2123 case EIO_NOP:
1745 req->result = 0; 2124 req->result = 0;
1746 break; 2125 break;
1747 2126
1748 case EIO_CUSTOM: 2127 case EIO_CUSTOM:
1749 ((void (*)(eio_req *))req->feed) (req); 2128 req->feed (req);
1750 break; 2129 break;
1751 2130
1752 default: 2131 default:
1753 errno = ENOSYS; 2132 errno = ENOSYS;
1754 req->result = -1; 2133 req->result = -1;
1803eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) 2182eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
1804{ 2183{
1805 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; 2184 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
1806} 2185}
1807 2186
2187eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2188{
2189 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
2190}
2191
1808eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) 2192eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1809{ 2193{
1810 REQ (EIO_FDATASYNC); req->int1 = fd; SEND; 2194 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1811} 2195}
1812 2196
1853eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) 2237eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1854{ 2238{
1855 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2239 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1856} 2240}
1857 2241
1858eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2242eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1859{ 2243{
1860 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2244 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1861} 2245}
1862 2246
1863eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) 2247eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
1883eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data) 2267eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1884{ 2268{
1885 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2269 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1886} 2270}
1887 2271
1888eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2272eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1889{ 2273{
1890 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2274 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1891} 2275}
1892 2276
1893eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) 2277eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
1907} 2291}
1908 2292
1909eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data) 2293eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1910{ 2294{
1911 return eio__1path (EIO_READLINK, path, pri, cb, data); 2295 return eio__1path (EIO_READLINK, path, pri, cb, data);
2296}
2297
2298eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2299{
2300 return eio__1path (EIO_REALPATH, path, pri, cb, data);
1912} 2301}
1913 2302
1914eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) 2303eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1915{ 2304{
1916 return eio__1path (EIO_STAT, path, pri, cb, data); 2305 return eio__1path (EIO_STAT, path, pri, cb, data);
1975eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2364eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1976{ 2365{
1977 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2366 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1978} 2367}
1979 2368
1980eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) 2369eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
1981{ 2370{
1982 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND; 2371 REQ (EIO_CUSTOM); req->feed = execute; SEND;
1983} 2372}
1984 2373
1985#endif 2374#endif
1986 2375
1987eio_req *eio_grp (eio_cb cb, void *data) 2376eio_req *eio_grp (eio_cb cb, void *data)
1996#undef SEND 2385#undef SEND
1997 2386
1998/*****************************************************************************/ 2387/*****************************************************************************/
1999/* grp functions */ 2388/* grp functions */
2000 2389
2390void
2001void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2391eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
2002{ 2392{
2003 grp->int2 = limit; 2393 grp->int2 = limit;
2004 grp->feed = feed; 2394 grp->feed = feed;
2005 2395
2006 grp_try_feed (grp); 2396 grp_try_feed (grp);
2007} 2397}
2008 2398
2399void
2009void eio_grp_limit (eio_req *grp, int limit) 2400eio_grp_limit (eio_req *grp, int limit)
2010{ 2401{
2011 grp->int2 = limit; 2402 grp->int2 = limit;
2012 2403
2013 grp_try_feed (grp); 2404 grp_try_feed (grp);
2014} 2405}
2015 2406
2407void
2016void eio_grp_add (eio_req *grp, eio_req *req) 2408eio_grp_add (eio_req *grp, eio_req *req)
2017{ 2409{
2018 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2410 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
2019 2411
2020 grp->flags |= EIO_FLAG_GROUPADD; 2412 grp->flags |= EIO_FLAG_GROUPADD;
2021 2413
2032} 2424}
2033 2425
2034/*****************************************************************************/ 2426/*****************************************************************************/
2035/* misc garbage */ 2427/* misc garbage */
2036 2428
2429eio_ssize_t
2037ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2430eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2038{ 2431{
2039 etp_worker wrk;
2040 ssize_t ret;
2041
2042 wrk.dbuf = 0;
2043
2044 ret = eio__sendfile (ofd, ifd, offset, count, &wrk); 2432 return eio__sendfile (ofd, ifd, offset, count);
2045
2046 if (wrk.dbuf)
2047 free (wrk.dbuf);
2048
2049 return ret;
2050} 2433}
2051 2434

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines