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

Comparing libeio/eio.c (file contents):
Revision 1.66 by root, Sun Jun 5 19:58:37 2011 UTC vs.
Revision 1.101 by root, Wed Aug 3 15:25:38 2011 UTC

40#ifndef _WIN32 40#ifndef _WIN32
41# include "config.h" 41# include "config.h"
42#endif 42#endif
43 43
44#include "eio.h" 44#include "eio.h"
45#include "ecb.h"
45 46
46#ifdef EIO_STACKSIZE 47#ifdef EIO_STACKSIZE
47# define XTHREAD_STACKSIZE EIO_STACKSIZE 48# define XTHREAD_STACKSIZE EIO_STACKSIZE
48#endif 49#endif
49#include "xthread.h" 50#include "xthread.h"
53#include <stdlib.h> 54#include <stdlib.h>
54#include <string.h> 55#include <string.h>
55#include <errno.h> 56#include <errno.h>
56#include <sys/types.h> 57#include <sys/types.h>
57#include <sys/stat.h> 58#include <sys/stat.h>
58#include <sys/statvfs.h>
59#include <limits.h> 59#include <limits.h>
60#include <fcntl.h> 60#include <fcntl.h>
61#include <assert.h> 61#include <assert.h>
62 62
63/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */ 63/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
64/* intptr_t only comes form stdint.h, says idiot openbsd coder */ 64/* intptr_t only comes from stdint.h, says idiot openbsd coder */
65#if HAVE_STDINT_H 65#if HAVE_STDINT_H
66# include <stdint.h> 66# include <stdint.h>
67#endif 67#endif
68 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
69#ifndef EIO_FINISH 82#ifndef EIO_FINISH
70# 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
71#endif 84#endif
72 85
73#ifndef EIO_DESTROY 86#ifndef EIO_DESTROY
76 89
77#ifndef EIO_FEED 90#ifndef EIO_FEED
78# 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)
79#endif 92#endif
80 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
81#ifdef _WIN32 105#ifdef _WIN32
82 106
83 /*doh*/ 107 #undef PAGESIZE
108 #define PAGESIZE 4096 /* GetSystemInfo? */
109
110 /* TODO: look at how perl does stat (non-sloppy), unlink (ro-files), utime, link */
111
112 #ifdef EIO_STRUCT_STATI64
113 /* look at perl's non-sloppy stat */
114 #define stat(path,buf) _stati64 (path,buf)
115 #define fstat(fd,buf) _fstati64 (fd,buf)
116 #endif
117 #define lstat(path,buf) stat (path,buf)
118 #define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
119 #define mkdir(path,mode) _mkdir (path)
120 #define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
121
122 #define chmod(path,mode) _chmod (path, mode)
123 #define dup(fd) _dup (fd)
124 #define dup2(fd1,fd2) _dup2 (fd1, fd2)
125
126 #define fchmod(fd,mode) EIO_ENOSYS ()
127 #define chown(path,uid,gid) EIO_ENOSYS ()
128 #define fchown(fd,uid,gid) EIO_ENOSYS ()
129 #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
130 #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
131 #define mknod(path,mode,dev) EIO_ENOSYS ()
132 #define sync() EIO_ENOSYS ()
133 #define readlink(path,buf,s) EIO_ENOSYS ()
134 #define statvfs(path,buf) EIO_ENOSYS ()
135 #define fstatvfs(fd,buf) EIO_ENOSYS ()
136
137 /* rename() uses MoveFile, which fails to overwrite */
138 #define rename(old,neu) eio__rename (old, neu)
139
140 static int
141 eio__rename (const char *old, const char *neu)
142 {
143 if (MoveFileEx (old, neu, MOVEFILE_REPLACE_EXISTING))
144 return 0;
145
146 /* should steal _dosmaperr */
147 switch (GetLastError ())
148 {
149 case ERROR_FILE_NOT_FOUND:
150 case ERROR_PATH_NOT_FOUND:
151 case ERROR_INVALID_DRIVE:
152 case ERROR_NO_MORE_FILES:
153 case ERROR_BAD_NETPATH:
154 case ERROR_BAD_NET_NAME:
155 case ERROR_BAD_PATHNAME:
156 case ERROR_FILENAME_EXCED_RANGE:
157 errno = ENOENT;
158 break;
159
160 default:
161 errno = EACCES;
162 break;
163 }
164
165 return -1;
166 }
167
168 /* we could even stat and see if it exists */
169 static int
170 symlink (const char *old, const char *neu)
171 {
172 #if WINVER >= 0x0600
173 if (CreateSymbolicLink (neu, old, 1))
174 return 0;
175
176 if (CreateSymbolicLink (neu, old, 0))
177 return 0;
178 #endif
179
180 return EIO_ERRNO (ENOENT, -1);
181 }
182
183 /* POSIX API only */
184 #define CreateHardLink(neu,old,flags) 0
185 #define CreateSymbolicLink(neu,old,flags) 0
186
187 struct statvfs
188 {
189 int dummy;
190 };
191
192 #define DT_DIR EIO_DT_DIR
193 #define DT_REG EIO_DT_REG
194 #define D_NAME(entp) entp.cFileName
195 #define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
196
84#else 197#else
85 198
86# include <sys/time.h> 199 #include <sys/time.h>
87# include <sys/select.h> 200 #include <sys/select.h>
201 #include <sys/statvfs.h>
88# include <unistd.h> 202 #include <unistd.h>
203 #include <signal.h>
204 #include <dirent.h>
205
206 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
207 #include <sys/mman.h>
208 #endif
209
210 #define D_NAME(entp) entp->d_name
211
212 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
213 #if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
214 #define _DIRENT_HAVE_D_TYPE /* sigh */
215 #define D_INO(de) (de)->d_fileno
216 #define D_NAMLEN(de) (de)->d_namlen
217 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
218 #define D_INO(de) (de)->d_ino
219 #endif
220
221 #ifdef _D_EXACT_NAMLEN
222 #undef D_NAMLEN
223 #define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
224 #endif
225
226 #ifdef _DIRENT_HAVE_D_TYPE
227 #define D_TYPE(de) (de)->d_type
228 #endif
229
230 #ifndef EIO_STRUCT_DIRENT
231 #define EIO_STRUCT_DIRENT struct dirent
232 #endif
233
234#endif
235
236#if HAVE_UTIMES
89# include <utime.h> 237# include <utime.h>
90# include <signal.h>
91# include <dirent.h>
92
93#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
94# include <sys/mman.h>
95#endif
96
97/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
98# if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
99# define _DIRENT_HAVE_D_TYPE /* sigh */
100# define D_INO(de) (de)->d_fileno
101# define D_NAMLEN(de) (de)->d_namlen
102# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
103# define D_INO(de) (de)->d_ino
104# endif 238#endif
105 239
106#ifdef _D_EXACT_NAMLEN 240#if HAVE_SYS_SYSCALL_H
107# undef D_NAMLEN 241# include <sys/syscall.h>
108# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
109#endif
110
111# ifdef _DIRENT_HAVE_D_TYPE
112# define D_TYPE(de) (de)->d_type
113# endif 242#endif
114 243
115# ifndef EIO_STRUCT_DIRENT 244#if HAVE_SYS_PRCTL_H
116# define EIO_STRUCT_DIRENT struct dirent 245# include <sys/prctl.h>
117# endif
118
119#endif 246#endif
120 247
121#if HAVE_SENDFILE 248#if HAVE_SENDFILE
122# if __linux 249# if __linux
123# include <sys/sendfile.h> 250# include <sys/sendfile.h>
138#endif 265#endif
139#ifndef D_INO 266#ifndef D_INO
140# define D_INO(de) 0 267# define D_INO(de) 0
141#endif 268#endif
142#ifndef D_NAMLEN 269#ifndef D_NAMLEN
143# define D_NAMLEN(de) strlen ((de)->d_name) 270# define D_NAMLEN(entp) strlen (D_NAME (entp))
144#endif 271#endif
145 272
146/* used for struct dirent, AIX doesn't provide it */ 273/* used for struct dirent, AIX doesn't provide it */
147#ifndef NAME_MAX 274#ifndef NAME_MAX
148# define NAME_MAX 4096 275# define NAME_MAX 4096
155 282
156/* buffer size for various temporary buffers */ 283/* buffer size for various temporary buffers */
157#define EIO_BUFSIZE 65536 284#define EIO_BUFSIZE 65536
158 285
159#define dBUF \ 286#define dBUF \
160 char *eio_buf; \
161 ETP_WORKER_LOCK (self); \
162 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 287 char *eio_buf = malloc (EIO_BUFSIZE); \
163 ETP_WORKER_UNLOCK (self); \
164 errno = ENOMEM; \ 288 errno = ENOMEM; \
165 if (!eio_buf) \ 289 if (!eio_buf) \
166 return -1; 290 return -1
291
292#define FUBd \
293 free (eio_buf)
167 294
168#define EIO_TICKS ((1000000 + 1023) >> 10) 295#define EIO_TICKS ((1000000 + 1023) >> 10)
169
170/*****************************************************************************/
171
172#if __GNUC__ >= 3
173# define expect(expr,value) __builtin_expect ((expr),(value))
174#else
175# define expect(expr,value) (expr)
176#endif
177
178#define expect_false(expr) expect ((expr) != 0, 0)
179#define expect_true(expr) expect ((expr) != 0, 1)
180
181/*****************************************************************************/
182 296
183#define ETP_PRI_MIN EIO_PRI_MIN 297#define ETP_PRI_MIN EIO_PRI_MIN
184#define ETP_PRI_MAX EIO_PRI_MAX 298#define ETP_PRI_MAX EIO_PRI_MAX
185 299
186struct etp_worker; 300struct etp_worker;
190static int eio_finish (eio_req *req); 304static int eio_finish (eio_req *req);
191#define ETP_FINISH(req) eio_finish (req) 305#define ETP_FINISH(req) eio_finish (req)
192static void eio_execute (struct etp_worker *self, eio_req *req); 306static void eio_execute (struct etp_worker *self, eio_req *req);
193#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) 307#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
194 308
195#define ETP_WORKER_CLEAR(req) \
196 if (wrk->dbuf) \
197 { \
198 free (wrk->dbuf); \
199 wrk->dbuf = 0; \
200 } \
201 \
202 if (wrk->dirp) \
203 { \
204 closedir (wrk->dirp); \
205 wrk->dirp = 0; \
206 }
207
208#define ETP_WORKER_COMMON \
209 void *dbuf; \
210 DIR *dirp;
211
212/*****************************************************************************/ 309/*****************************************************************************/
213 310
214#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) 311#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
215 312
216/* calculate time difference in ~1/EIO_TICKS of a second */ 313/* calculate time difference in ~1/EIO_TICKS of a second */
314ecb_inline int
217static int tvdiff (struct timeval *tv1, struct timeval *tv2) 315tvdiff (struct timeval *tv1, struct timeval *tv2)
218{ 316{
219 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 317 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
220 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 318 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
221} 319}
222 320
226static void (*done_poll_cb) (void); 324static void (*done_poll_cb) (void);
227 325
228static unsigned int max_poll_time; /* reslock */ 326static unsigned int max_poll_time; /* reslock */
229static unsigned int max_poll_reqs; /* reslock */ 327static unsigned int max_poll_reqs; /* reslock */
230 328
231static volatile unsigned int nreqs; /* reqlock */ 329static unsigned int nreqs; /* reqlock */
232static volatile unsigned int nready; /* reqlock */ 330static unsigned int nready; /* reqlock */
233static volatile unsigned int npending; /* reqlock */ 331static unsigned int npending; /* reqlock */
234static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */ 332static unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
235static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */ 333static unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
236 334
237static xmutex_t wrklock; 335static xmutex_t wrklock;
238static xmutex_t reslock; 336static xmutex_t reslock;
239static xmutex_t reqlock; 337static xmutex_t reqlock;
240static xcond_t reqwait; 338static xcond_t reqwait;
243/* 341/*
244 * make our pread/pwrite emulation safe against themselves, but not against 342 * make our pread/pwrite emulation safe against themselves, but not against
245 * normal read/write by using a mutex. slows down execution a lot, 343 * normal read/write by using a mutex. slows down execution a lot,
246 * but that's your problem, not mine. 344 * but that's your problem, not mine.
247 */ 345 */
248static xmutex_t preadwritelock = X_MUTEX_INIT; 346static xmutex_t preadwritelock;
249#endif 347#endif
250 348
251typedef struct etp_worker 349typedef struct etp_worker
252{ 350{
253 /* locked by wrklock */ 351 /* locked by wrklock */
256 xthread_t tid; 354 xthread_t tid;
257 355
258 /* locked by reslock, reqlock or wrklock */ 356 /* locked by reslock, reqlock or wrklock */
259 ETP_REQ *req; /* currently processed request */ 357 ETP_REQ *req; /* currently processed request */
260 358
359#ifdef ETP_WORKER_COMMON
261 ETP_WORKER_COMMON 360 ETP_WORKER_COMMON
361#endif
262} etp_worker; 362} etp_worker;
263 363
264static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */ 364static etp_worker wrk_first; /* NOT etp */
265 365
266#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock) 366#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
267#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock) 367#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
268 368
269/* worker threads management */ 369/* worker threads management */
270 370
371static void ecb_cold
271static void etp_worker_clear (etp_worker *wrk) 372etp_worker_clear (etp_worker *wrk)
272{ 373{
273 ETP_WORKER_CLEAR (wrk);
274} 374}
275 375
376static void ecb_cold
276static void etp_worker_free (etp_worker *wrk) 377etp_worker_free (etp_worker *wrk)
277{ 378{
278 wrk->next->prev = wrk->prev; 379 wrk->next->prev = wrk->prev;
279 wrk->prev->next = wrk->next; 380 wrk->prev->next = wrk->next;
280 381
281 free (wrk); 382 free (wrk);
282} 383}
283 384
284static unsigned int etp_nreqs (void) 385static unsigned int
386etp_nreqs (void)
285{ 387{
286 int retval; 388 int retval;
287 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 389 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
288 retval = nreqs; 390 retval = nreqs;
289 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 391 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
290 return retval; 392 return retval;
291} 393}
292 394
293static unsigned int etp_nready (void) 395static unsigned int
396etp_nready (void)
294{ 397{
295 unsigned int retval; 398 unsigned int retval;
296 399
297 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 400 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
298 retval = nready; 401 retval = nready;
299 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 402 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
300 403
301 return retval; 404 return retval;
302} 405}
303 406
304static unsigned int etp_npending (void) 407static unsigned int
408etp_npending (void)
305{ 409{
306 unsigned int retval; 410 unsigned int retval;
307 411
308 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 412 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
309 retval = npending; 413 retval = npending;
310 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 414 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
311 415
312 return retval; 416 return retval;
313} 417}
314 418
315static unsigned int etp_nthreads (void) 419static unsigned int
420etp_nthreads (void)
316{ 421{
317 unsigned int retval; 422 unsigned int retval;
318 423
319 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 424 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
320 retval = started; 425 retval = started;
334} etp_reqq; 439} etp_reqq;
335 440
336static etp_reqq req_queue; 441static etp_reqq req_queue;
337static etp_reqq res_queue; 442static etp_reqq res_queue;
338 443
444static void ecb_noinline ecb_cold
445reqq_init (etp_reqq *q)
446{
447 int pri;
448
449 for (pri = 0; pri < ETP_NUM_PRI; ++pri)
450 q->qs[pri] = q->qe[pri] = 0;
451
452 q->size = 0;
453}
454
455static int ecb_noinline
339static int reqq_push (etp_reqq *q, ETP_REQ *req) 456reqq_push (etp_reqq *q, ETP_REQ *req)
340{ 457{
341 int pri = req->pri; 458 int pri = req->pri;
342 req->next = 0; 459 req->next = 0;
343 460
344 if (q->qe[pri]) 461 if (q->qe[pri])
350 q->qe[pri] = q->qs[pri] = req; 467 q->qe[pri] = q->qs[pri] = req;
351 468
352 return q->size++; 469 return q->size++;
353} 470}
354 471
472static ETP_REQ * ecb_noinline
355static ETP_REQ *reqq_shift (etp_reqq *q) 473reqq_shift (etp_reqq *q)
356{ 474{
357 int pri; 475 int pri;
358 476
359 if (!q->size) 477 if (!q->size)
360 return 0; 478 return 0;
375 } 493 }
376 494
377 abort (); 495 abort ();
378} 496}
379 497
380static void etp_thread_init (void) 498static int ecb_cold
499etp_init (void (*want_poll)(void), void (*done_poll)(void))
381{ 500{
382 X_MUTEX_CREATE (wrklock); 501 X_MUTEX_CREATE (wrklock);
383 X_MUTEX_CREATE (reslock); 502 X_MUTEX_CREATE (reslock);
384 X_MUTEX_CREATE (reqlock); 503 X_MUTEX_CREATE (reqlock);
385 X_COND_CREATE (reqwait); 504 X_COND_CREATE (reqwait);
386}
387 505
388static void etp_atfork_prepare (void) 506 reqq_init (&req_queue);
389{ 507 reqq_init (&res_queue);
390 X_LOCK (wrklock);
391 X_LOCK (reqlock);
392 X_LOCK (reslock);
393#if !HAVE_PREADWRITE
394 X_LOCK (preadwritelock);
395#endif
396}
397 508
398static void etp_atfork_parent (void) 509 wrk_first.next =
399{ 510 wrk_first.prev = &wrk_first;
400#if !HAVE_PREADWRITE
401 X_UNLOCK (preadwritelock);
402#endif
403 X_UNLOCK (reslock);
404 X_UNLOCK (reqlock);
405 X_UNLOCK (wrklock);
406}
407
408static void etp_atfork_child (void)
409{
410 ETP_REQ *prv;
411
412 while ((prv = reqq_shift (&req_queue)))
413 ETP_DESTROY (prv);
414
415 while ((prv = reqq_shift (&res_queue)))
416 ETP_DESTROY (prv);
417
418 while (wrk_first.next != &wrk_first)
419 {
420 etp_worker *wrk = wrk_first.next;
421
422 if (wrk->req)
423 ETP_DESTROY (wrk->req);
424
425 etp_worker_clear (wrk);
426 etp_worker_free (wrk);
427 }
428 511
429 started = 0; 512 started = 0;
430 idle = 0; 513 idle = 0;
431 nreqs = 0; 514 nreqs = 0;
432 nready = 0; 515 nready = 0;
433 npending = 0; 516 npending = 0;
434 517
435 etp_thread_init ();
436}
437
438static void
439etp_once_init (void)
440{
441 etp_thread_init ();
442 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
443}
444
445static int
446etp_init (void (*want_poll)(void), void (*done_poll)(void))
447{
448 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
449
450 pthread_once (&doinit, etp_once_init);
451
452 want_poll_cb = want_poll; 518 want_poll_cb = want_poll;
453 done_poll_cb = done_poll; 519 done_poll_cb = done_poll;
454 520
455 return 0; 521 return 0;
456} 522}
457 523
458X_THREAD_PROC (etp_proc); 524X_THREAD_PROC (etp_proc);
459 525
526static void ecb_cold
460static void etp_start_thread (void) 527etp_start_thread (void)
461{ 528{
462 etp_worker *wrk = calloc (1, sizeof (etp_worker)); 529 etp_worker *wrk = calloc (1, sizeof (etp_worker));
463 530
464 /*TODO*/ 531 /*TODO*/
465 assert (("unable to allocate worker thread data", wrk)); 532 assert (("unable to allocate worker thread data", wrk));
478 free (wrk); 545 free (wrk);
479 546
480 X_UNLOCK (wrklock); 547 X_UNLOCK (wrklock);
481} 548}
482 549
550static void
483static void etp_maybe_start_thread (void) 551etp_maybe_start_thread (void)
484{ 552{
485 if (expect_true (etp_nthreads () >= wanted)) 553 if (ecb_expect_true (etp_nthreads () >= wanted))
486 return; 554 return;
487 555
488 /* todo: maybe use idle here, but might be less exact */ 556 /* todo: maybe use idle here, but might be less exact */
489 if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ())) 557 if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
490 return; 558 return;
491 559
492 etp_start_thread (); 560 etp_start_thread ();
493} 561}
494 562
563static void ecb_cold
495static void etp_end_thread (void) 564etp_end_thread (void)
496{ 565{
497 eio_req *req = calloc (1, sizeof (eio_req)); 566 eio_req *req = calloc (1, sizeof (eio_req));
498 567
499 req->type = -1; 568 req->type = -1;
500 req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 569 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
507 X_LOCK (wrklock); 576 X_LOCK (wrklock);
508 --started; 577 --started;
509 X_UNLOCK (wrklock); 578 X_UNLOCK (wrklock);
510} 579}
511 580
512static int etp_poll (void) 581static int
582etp_poll (void)
513{ 583{
514 unsigned int maxreqs; 584 unsigned int maxreqs;
515 unsigned int maxtime; 585 unsigned int maxtime;
516 struct timeval tv_start, tv_now; 586 struct timeval tv_start, tv_now;
517 587
547 617
548 X_LOCK (reqlock); 618 X_LOCK (reqlock);
549 --nreqs; 619 --nreqs;
550 X_UNLOCK (reqlock); 620 X_UNLOCK (reqlock);
551 621
552 if (expect_false (req->type == EIO_GROUP && req->size)) 622 if (ecb_expect_false (req->type == EIO_GROUP && req->size))
553 { 623 {
554 req->int1 = 1; /* mark request as delayed */ 624 req->int1 = 1; /* mark request as delayed */
555 continue; 625 continue;
556 } 626 }
557 else 627 else
558 { 628 {
559 int res = ETP_FINISH (req); 629 int res = ETP_FINISH (req);
560 if (expect_false (res)) 630 if (ecb_expect_false (res))
561 return res; 631 return res;
562 } 632 }
563 633
564 if (expect_false (maxreqs && !--maxreqs)) 634 if (ecb_expect_false (maxreqs && !--maxreqs))
565 break; 635 break;
566 636
567 if (maxtime) 637 if (maxtime)
568 { 638 {
569 gettimeofday (&tv_now, 0); 639 gettimeofday (&tv_now, 0);
575 645
576 errno = EAGAIN; 646 errno = EAGAIN;
577 return -1; 647 return -1;
578} 648}
579 649
650static void
580static void etp_cancel (ETP_REQ *req) 651etp_cancel (ETP_REQ *req)
581{ 652{
582 X_LOCK (wrklock); 653 req->cancelled = 1;
583 req->flags |= EIO_FLAG_CANCELLED;
584 X_UNLOCK (wrklock);
585 654
586 eio_grp_cancel (req); 655 eio_grp_cancel (req);
587} 656}
588 657
658static void
589static void etp_submit (ETP_REQ *req) 659etp_submit (ETP_REQ *req)
590{ 660{
591 req->pri -= ETP_PRI_MIN; 661 req->pri -= ETP_PRI_MIN;
592 662
593 if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN; 663 if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
594 if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 664 if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
595 665
596 if (expect_false (req->type == EIO_GROUP)) 666 if (ecb_expect_false (req->type == EIO_GROUP))
597 { 667 {
598 /* I hope this is worth it :/ */ 668 /* I hope this is worth it :/ */
599 X_LOCK (reqlock); 669 X_LOCK (reqlock);
600 ++nreqs; 670 ++nreqs;
601 X_UNLOCK (reqlock); 671 X_UNLOCK (reqlock);
620 690
621 etp_maybe_start_thread (); 691 etp_maybe_start_thread ();
622 } 692 }
623} 693}
624 694
695static void ecb_cold
625static void etp_set_max_poll_time (double nseconds) 696etp_set_max_poll_time (double nseconds)
626{ 697{
627 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 698 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
628 max_poll_time = nseconds * EIO_TICKS; 699 max_poll_time = nseconds * EIO_TICKS;
629 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 700 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
630} 701}
631 702
703static void ecb_cold
632static void etp_set_max_poll_reqs (unsigned int maxreqs) 704etp_set_max_poll_reqs (unsigned int maxreqs)
633{ 705{
634 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 706 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
635 max_poll_reqs = maxreqs; 707 max_poll_reqs = maxreqs;
636 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 708 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
637} 709}
638 710
711static void ecb_cold
639static void etp_set_max_idle (unsigned int nthreads) 712etp_set_max_idle (unsigned int nthreads)
640{ 713{
641 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 714 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
642 max_idle = nthreads; 715 max_idle = nthreads;
643 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 716 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
644} 717}
645 718
719static void ecb_cold
646static void etp_set_idle_timeout (unsigned int seconds) 720etp_set_idle_timeout (unsigned int seconds)
647{ 721{
648 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 722 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
649 idle_timeout = seconds; 723 idle_timeout = seconds;
650 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 724 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
651} 725}
652 726
727static void ecb_cold
653static void etp_set_min_parallel (unsigned int nthreads) 728etp_set_min_parallel (unsigned int nthreads)
654{ 729{
655 if (wanted < nthreads) 730 if (wanted < nthreads)
656 wanted = nthreads; 731 wanted = nthreads;
657} 732}
658 733
734static void ecb_cold
659static void etp_set_max_parallel (unsigned int nthreads) 735etp_set_max_parallel (unsigned int nthreads)
660{ 736{
661 if (wanted > nthreads) 737 if (wanted > nthreads)
662 wanted = nthreads; 738 wanted = nthreads;
663 739
664 while (started > wanted) 740 while (started > wanted)
665 etp_end_thread (); 741 etp_end_thread ();
666} 742}
667 743
668/*****************************************************************************/ 744/*****************************************************************************/
669 745
746static void
670static void grp_try_feed (eio_req *grp) 747grp_try_feed (eio_req *grp)
671{ 748{
672 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 749 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
673 { 750 {
674 grp->flags &= ~EIO_FLAG_GROUPADD; 751 grp->flags &= ~EIO_FLAG_GROUPADD;
675 752
682 break; 759 break;
683 } 760 }
684 } 761 }
685} 762}
686 763
764static int
687static int grp_dec (eio_req *grp) 765grp_dec (eio_req *grp)
688{ 766{
689 --grp->size; 767 --grp->size;
690 768
691 /* call feeder, if applicable */ 769 /* call feeder, if applicable */
692 grp_try_feed (grp); 770 grp_try_feed (grp);
696 return eio_finish (grp); 774 return eio_finish (grp);
697 else 775 else
698 return 0; 776 return 0;
699} 777}
700 778
779static void
701void eio_destroy (eio_req *req) 780eio_destroy (eio_req *req)
702{ 781{
703 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); 782 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
704 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); 783 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
705 784
706 EIO_DESTROY (req); 785 EIO_DESTROY (req);
707} 786}
708 787
788static int
709static int eio_finish (eio_req *req) 789eio_finish (eio_req *req)
710{ 790{
711 int res = EIO_FINISH (req); 791 int res = EIO_FINISH (req);
712 792
713 if (req->grp) 793 if (req->grp)
714 { 794 {
722 if (grp->grp_first == req) 802 if (grp->grp_first == req)
723 grp->grp_first = req->grp_next; 803 grp->grp_first = req->grp_next;
724 804
725 res2 = grp_dec (grp); 805 res2 = grp_dec (grp);
726 806
727 if (!res && res2) 807 if (!res)
728 res = res2; 808 res = res2;
729 } 809 }
730 810
731 eio_destroy (req); 811 eio_destroy (req);
732 812
733 return res; 813 return res;
734} 814}
735 815
816void
736void eio_grp_cancel (eio_req *grp) 817eio_grp_cancel (eio_req *grp)
737{ 818{
738 for (grp = grp->grp_first; grp; grp = grp->grp_next) 819 for (grp = grp->grp_first; grp; grp = grp->grp_next)
739 eio_cancel (grp); 820 eio_cancel (grp);
740} 821}
741 822
823void
742void eio_cancel (eio_req *req) 824eio_cancel (eio_req *req)
743{ 825{
744 etp_cancel (req); 826 etp_cancel (req);
745} 827}
746 828
829void
747void eio_submit (eio_req *req) 830eio_submit (eio_req *req)
748{ 831{
749 etp_submit (req); 832 etp_submit (req);
750} 833}
751 834
752unsigned int eio_nreqs (void) 835unsigned int
836eio_nreqs (void)
753{ 837{
754 return etp_nreqs (); 838 return etp_nreqs ();
755} 839}
756 840
757unsigned int eio_nready (void) 841unsigned int
842eio_nready (void)
758{ 843{
759 return etp_nready (); 844 return etp_nready ();
760} 845}
761 846
762unsigned int eio_npending (void) 847unsigned int
848eio_npending (void)
763{ 849{
764 return etp_npending (); 850 return etp_npending ();
765} 851}
766 852
767unsigned int eio_nthreads (void) 853unsigned int ecb_cold
854eio_nthreads (void)
768{ 855{
769 return etp_nthreads (); 856 return etp_nthreads ();
770} 857}
771 858
859void ecb_cold
772void eio_set_max_poll_time (double nseconds) 860eio_set_max_poll_time (double nseconds)
773{ 861{
774 etp_set_max_poll_time (nseconds); 862 etp_set_max_poll_time (nseconds);
775} 863}
776 864
865void ecb_cold
777void eio_set_max_poll_reqs (unsigned int maxreqs) 866eio_set_max_poll_reqs (unsigned int maxreqs)
778{ 867{
779 etp_set_max_poll_reqs (maxreqs); 868 etp_set_max_poll_reqs (maxreqs);
780} 869}
781 870
871void ecb_cold
782void eio_set_max_idle (unsigned int nthreads) 872eio_set_max_idle (unsigned int nthreads)
783{ 873{
784 etp_set_max_idle (nthreads); 874 etp_set_max_idle (nthreads);
785} 875}
786 876
877void ecb_cold
787void eio_set_idle_timeout (unsigned int seconds) 878eio_set_idle_timeout (unsigned int seconds)
788{ 879{
789 etp_set_idle_timeout (seconds); 880 etp_set_idle_timeout (seconds);
790} 881}
791 882
883void ecb_cold
792void eio_set_min_parallel (unsigned int nthreads) 884eio_set_min_parallel (unsigned int nthreads)
793{ 885{
794 etp_set_min_parallel (nthreads); 886 etp_set_min_parallel (nthreads);
795} 887}
796 888
889void ecb_cold
797void eio_set_max_parallel (unsigned int nthreads) 890eio_set_max_parallel (unsigned int nthreads)
798{ 891{
799 etp_set_max_parallel (nthreads); 892 etp_set_max_parallel (nthreads);
800} 893}
801 894
802int eio_poll (void) 895int eio_poll (void)
811# undef pread 904# undef pread
812# undef pwrite 905# undef pwrite
813# define pread eio__pread 906# define pread eio__pread
814# define pwrite eio__pwrite 907# define pwrite eio__pwrite
815 908
816static ssize_t 909static eio_ssize_t
817eio__pread (int fd, void *buf, size_t count, off_t offset) 910eio__pread (int fd, void *buf, size_t count, off_t offset)
818{ 911{
819 ssize_t res; 912 eio_ssize_t res;
820 off_t ooffset; 913 off_t ooffset;
821 914
822 X_LOCK (preadwritelock); 915 X_LOCK (preadwritelock);
823 ooffset = lseek (fd, 0, SEEK_CUR); 916 ooffset = lseek (fd, 0, SEEK_CUR);
824 lseek (fd, offset, SEEK_SET); 917 lseek (fd, offset, SEEK_SET);
827 X_UNLOCK (preadwritelock); 920 X_UNLOCK (preadwritelock);
828 921
829 return res; 922 return res;
830} 923}
831 924
832static ssize_t 925static eio_ssize_t
833eio__pwrite (int fd, void *buf, size_t count, off_t offset) 926eio__pwrite (int fd, void *buf, size_t count, off_t offset)
834{ 927{
835 ssize_t res; 928 eio_ssize_t res;
836 off_t ooffset; 929 off_t ooffset;
837 930
838 X_LOCK (preadwritelock); 931 X_LOCK (preadwritelock);
839 ooffset = lseek (fd, 0, SEEK_CUR); 932 ooffset = lseek (fd, 0, SEEK_CUR);
840 lseek (fd, offset, SEEK_SET); 933 lseek (fd, offset, SEEK_SET);
872#ifndef HAVE_FUTIMES 965#ifndef HAVE_FUTIMES
873 966
874# undef futimes 967# undef futimes
875# define futimes(fd,times) eio__futimes (fd, times) 968# define futimes(fd,times) eio__futimes (fd, times)
876 969
970static int
877static int eio__futimes (int fd, const struct timeval tv[2]) 971eio__futimes (int fd, const struct timeval tv[2])
878{ 972{
879 errno = ENOSYS; 973 errno = ENOSYS;
880 return -1; 974 return -1;
881} 975}
882 976
885#if !HAVE_FDATASYNC 979#if !HAVE_FDATASYNC
886# undef fdatasync 980# undef fdatasync
887# define fdatasync(fd) fsync (fd) 981# define fdatasync(fd) fsync (fd)
888#endif 982#endif
889 983
984static int
985eio__syncfs (int fd)
986{
987 int res;
988
989#if HAVE_SYS_SYNCFS
990 res = (int)syscall (__NR_syncfs, (int)(fd));
991#else
992 res = -1;
993 errno = ENOSYS;
994#endif
995
996 if (res < 0 && errno == ENOSYS && fd >= 0)
997 sync ();
998
999 return res;
1000}
1001
890/* sync_file_range always needs emulation */ 1002/* sync_file_range always needs emulation */
891int 1003static int
892eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags) 1004eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
893{ 1005{
894#if HAVE_SYNC_FILE_RANGE 1006#if HAVE_SYNC_FILE_RANGE
895 int res; 1007 int res;
896 1008
913 /* even though we could play tricks with the flags, it's better to always 1025 /* even though we could play tricks with the flags, it's better to always
914 * call fdatasync, as that matches the expectation of its users best */ 1026 * call fdatasync, as that matches the expectation of its users best */
915 return fdatasync (fd); 1027 return fdatasync (fd);
916} 1028}
917 1029
1030static int
1031eio__fallocate (int fd, int mode, off_t offset, size_t len)
1032{
1033#if HAVE_FALLOCATE
1034 return fallocate (fd, mode, offset, len);
1035#else
1036 errno = ENOSYS;
1037 return -1;
1038#endif
1039}
1040
918#if !HAVE_READAHEAD 1041#if !HAVE_READAHEAD
919# undef readahead 1042# undef readahead
920# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 1043# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
921 1044
922static ssize_t 1045static eio_ssize_t
923eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) 1046eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
924{ 1047{
925 size_t todo = count; 1048 size_t todo = count;
926 dBUF; 1049 dBUF;
927 1050
932 pread (fd, eio_buf, len, offset); 1055 pread (fd, eio_buf, len, offset);
933 offset += len; 1056 offset += len;
934 todo -= len; 1057 todo -= len;
935 } 1058 }
936 1059
1060 FUBd;
1061
937 errno = 0; 1062 errno = 0;
938 return count; 1063 return count;
939} 1064}
940 1065
941#endif 1066#endif
942 1067
943/* sendfile always needs emulation */ 1068/* sendfile always needs emulation */
944static ssize_t 1069static eio_ssize_t
945eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) 1070eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
946{ 1071{
1072 eio_ssize_t written = 0;
947 ssize_t res; 1073 eio_ssize_t res;
948 1074
949 if (!count) 1075 if (!count)
950 return 0; 1076 return 0;
951 1077
1078 for (;;)
1079 {
1080#ifdef __APPLE__
1081# undef HAVE_SENDFILE /* broken, as everything on os x */
1082#endif
952#if HAVE_SENDFILE 1083#if HAVE_SENDFILE
953# if __linux 1084# if __linux
1085 off_t soffset = offset;
954 res = sendfile (ofd, ifd, &offset, count); 1086 res = sendfile (ofd, ifd, &soffset, count);
955 1087
956# elif __FreeBSD__ 1088# elif __FreeBSD__
957 /* 1089 /*
958 * Of course, the freebsd sendfile is a dire hack with no thoughts 1090 * Of course, the freebsd sendfile is a dire hack with no thoughts
959 * wasted on making it similar to other I/O functions. 1091 * wasted on making it similar to other I/O functions.
960 */ 1092 */
961 {
962 off_t sbytes; 1093 off_t sbytes;
963 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 1094 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
964 1095
965 #if 0 /* according to the manpage, this is correct, but broken behaviour */ 1096 #if 0 /* according to the manpage, this is correct, but broken behaviour */
966 /* freebsd' sendfile will return 0 on success */ 1097 /* freebsd' sendfile will return 0 on success */
967 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */ 1098 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
968 /* not on e.g. EIO or EPIPE - sounds broken */ 1099 /* not on e.g. EIO or EPIPE - sounds broken */
969 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0) 1100 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
970 res = sbytes; 1101 res = sbytes;
971 #endif 1102 #endif
972 1103
973 /* according to source inspection, this is correct, and useful behaviour */ 1104 /* according to source inspection, this is correct, and useful behaviour */
974 if (sbytes) 1105 if (sbytes)
975 res = sbytes; 1106 res = sbytes;
976 }
977 1107
978# elif defined (__APPLE__) 1108# elif defined (__APPLE__)
979
980 {
981 off_t sbytes = count; 1109 off_t sbytes = count;
982 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0); 1110 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
983 1111
984 /* according to the manpage, sbytes is always valid */ 1112 /* according to the manpage, sbytes is always valid */
985 if (sbytes) 1113 if (sbytes)
986 res = sbytes; 1114 res = sbytes;
987 }
988 1115
989# elif __hpux 1116# elif __hpux
990 res = sendfile (ofd, ifd, offset, count, 0, 0); 1117 res = sendfile (ofd, ifd, offset, count, 0, 0);
991 1118
992# elif __solaris 1119# elif __solaris
993 {
994 struct sendfilevec vec; 1120 struct sendfilevec vec;
995 size_t sbytes; 1121 size_t sbytes;
996 1122
997 vec.sfv_fd = ifd; 1123 vec.sfv_fd = ifd;
998 vec.sfv_flag = 0; 1124 vec.sfv_flag = 0;
999 vec.sfv_off = offset; 1125 vec.sfv_off = offset;
1000 vec.sfv_len = count; 1126 vec.sfv_len = count;
1001 1127
1002 res = sendfilev (ofd, &vec, 1, &sbytes); 1128 res = sendfilev (ofd, &vec, 1, &sbytes);
1003 1129
1004 if (res < 0 && sbytes) 1130 if (res < 0 && sbytes)
1005 res = sbytes; 1131 res = sbytes;
1006 }
1007 1132
1008# endif 1133# endif
1009 1134
1010#elif defined (_WIN32) 1135#elif defined (_WIN32) && 0
1011
1012 /* does not work, just for documentation of what would need to be done */ 1136 /* does not work, just for documentation of what would need to be done */
1013 { 1137 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
1138 /* libeio guarantees that the file offset does not change, and windows */
1139 /* has no way to get an independent handle to the same file description */
1014 HANDLE h = TO_SOCKET (ifd); 1140 HANDLE h = TO_SOCKET (ifd);
1015 SetFilePointer (h, offset, 0, FILE_BEGIN); 1141 SetFilePointer (h, offset, 0, FILE_BEGIN);
1016 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0); 1142 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1017 }
1018 1143
1019#else 1144#else
1020 res = -1; 1145 res = -1;
1021 errno = ENOSYS; 1146 errno = ENOSYS;
1022#endif 1147#endif
1023 1148
1149 /* we assume sendfile can copy at least 128mb in one go */
1150 if (res <= 128 * 1024 * 1024)
1151 {
1152 if (res > 0)
1153 written += res;
1154
1155 if (written)
1156 return written;
1157
1158 break;
1159 }
1160 else
1161 {
1162 /* if we requested more, then probably the kernel was lazy */
1163 written += res;
1164 offset += res;
1165 count -= res;
1166
1167 if (!count)
1168 return written;
1169 }
1170 }
1171
1024 if (res < 0 1172 if (res < 0
1025 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1173 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1026 /* BSDs */ 1174 /* BSDs */
1027#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */ 1175#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1028 || errno == ENOTSUP 1176 || errno == ENOTSUP
1029#endif 1177#endif
1178#ifdef EOPNOTSUPP /* windows */
1030 || errno == EOPNOTSUPP /* BSDs */ 1179 || errno == EOPNOTSUPP /* BSDs */
1180#endif
1031#if __solaris 1181#if __solaris
1032 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1182 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
1033#endif 1183#endif
1034 ) 1184 )
1035 ) 1185 )
1039 1189
1040 res = 0; 1190 res = 0;
1041 1191
1042 while (count) 1192 while (count)
1043 { 1193 {
1044 ssize_t cnt; 1194 eio_ssize_t cnt;
1045 1195
1046 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 1196 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
1047 1197
1048 if (cnt <= 0) 1198 if (cnt <= 0)
1049 { 1199 {
1061 1211
1062 offset += cnt; 1212 offset += cnt;
1063 res += cnt; 1213 res += cnt;
1064 count -= cnt; 1214 count -= cnt;
1065 } 1215 }
1216
1217 FUBd;
1066 } 1218 }
1067 1219
1068 return res; 1220 return res;
1221}
1222
1223#ifdef PAGESIZE
1224# define eio_pagesize() PAGESIZE
1225#else
1226static intptr_t
1227eio_pagesize (void)
1228{
1229 static intptr_t page;
1230
1231 if (!page)
1232 page = sysconf (_SC_PAGESIZE);
1233
1234 return page;
1235}
1236#endif
1237
1238static void
1239eio_page_align (void **addr, size_t *length)
1240{
1241 intptr_t mask = eio_pagesize () - 1;
1242
1243 /* round down addr */
1244 intptr_t adj = mask & (intptr_t)*addr;
1245
1246 *addr = (void *)((intptr_t)*addr - adj);
1247 *length += adj;
1248
1249 /* round up length */
1250 *length = (*length + mask) & ~mask;
1251}
1252
1253#if !_POSIX_MEMLOCK
1254# define eio__mlockall(a) EIO_ENOSYS ()
1255#else
1256
1257static int
1258eio__mlockall (int flags)
1259{
1260 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1261 extern int mallopt (int, int);
1262 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1263 #endif
1264
1265 if (EIO_MCL_CURRENT != MCL_CURRENT
1266 || EIO_MCL_FUTURE != MCL_FUTURE)
1267 {
1268 flags = 0
1269 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1270 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1271 }
1272
1273 return mlockall (flags);
1274}
1275#endif
1276
1277#if !_POSIX_MEMLOCK_RANGE
1278# define eio__mlock(a,b) EIO_ENOSYS ()
1279#else
1280
1281static int
1282eio__mlock (void *addr, size_t length)
1283{
1284 eio_page_align (&addr, &length);
1285
1286 return mlock (addr, length);
1287}
1288
1289#endif
1290
1291#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1292# define eio__msync(a,b,c) EIO_ENOSYS ()
1293#else
1294
1295static int
1296eio__msync (void *mem, size_t len, int flags)
1297{
1298 eio_page_align (&mem, &len);
1299
1300 if (EIO_MS_ASYNC != MS_SYNC
1301 || EIO_MS_INVALIDATE != MS_INVALIDATE
1302 || EIO_MS_SYNC != MS_SYNC)
1303 {
1304 flags = 0
1305 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1306 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1307 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1308 }
1309
1310 return msync (mem, len, flags);
1311}
1312
1313#endif
1314
1315static int
1316eio__mtouch (eio_req *req)
1317{
1318 void *mem = req->ptr2;
1319 size_t len = req->size;
1320 int flags = req->int1;
1321
1322 eio_page_align (&mem, &len);
1323
1324 {
1325 intptr_t addr = (intptr_t)mem;
1326 intptr_t end = addr + len;
1327 intptr_t page = eio_pagesize ();
1328
1329 if (addr < end)
1330 if (flags & EIO_MT_MODIFY) /* modify */
1331 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1332 else
1333 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1334 }
1335
1336 return 0;
1337}
1338
1339/*****************************************************************************/
1340/* requests implemented outside eio_execute, because they are so large */
1341
1342static void
1343eio__realpath (eio_req *req, etp_worker *self)
1344{
1345 char *rel = req->ptr1;
1346 char *res;
1347 char *tmp1, *tmp2;
1348#if SYMLOOP_MAX > 32
1349 int symlinks = SYMLOOP_MAX;
1350#else
1351 int symlinks = 32;
1352#endif
1353
1354 req->result = -1;
1355
1356 errno = EINVAL;
1357 if (!rel)
1358 return;
1359
1360 errno = ENOENT;
1361 if (!*rel)
1362 return;
1363
1364 if (!req->ptr2)
1365 {
1366 X_LOCK (wrklock);
1367 req->flags |= EIO_FLAG_PTR2_FREE;
1368 X_UNLOCK (wrklock);
1369 req->ptr2 = malloc (PATH_MAX * 3);
1370
1371 errno = ENOMEM;
1372 if (!req->ptr2)
1373 return;
1374 }
1375
1376 res = req->ptr2;
1377 tmp1 = res + PATH_MAX;
1378 tmp2 = tmp1 + PATH_MAX;
1379
1380#if 0 /* disabled, the musl way to do things is just too racy */
1381#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1382 /* on linux we may be able to ask the kernel */
1383 {
1384 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1385
1386 if (fd >= 0)
1387 {
1388 sprintf (tmp1, "/proc/self/fd/%d", fd);
1389 req->result = readlink (tmp1, res, PATH_MAX);
1390 close (fd);
1391
1392 /* here we should probably stat the open file and the disk file, to make sure they still match */
1393
1394 if (req->result > 0)
1395 goto done;
1396 }
1397 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1398 return;
1399 }
1400#endif
1401#endif
1402
1403 if (*rel != '/')
1404 {
1405 if (!getcwd (res, PATH_MAX))
1406 return;
1407
1408 if (res [1]) /* only use if not / */
1409 res += strlen (res);
1410 }
1411
1412 while (*rel)
1413 {
1414 eio_ssize_t len, linklen;
1415 char *beg = rel;
1416
1417 while (*rel && *rel != '/')
1418 ++rel;
1419
1420 len = rel - beg;
1421
1422 if (!len) /* skip slashes */
1423 {
1424 ++rel;
1425 continue;
1426 }
1427
1428 if (beg [0] == '.')
1429 {
1430 if (len == 1)
1431 continue; /* . - nop */
1432
1433 if (beg [1] == '.' && len == 2)
1434 {
1435 /* .. - back up one component, if possible */
1436
1437 while (res != req->ptr2)
1438 if (*--res == '/')
1439 break;
1440
1441 continue;
1442 }
1443 }
1444
1445 errno = ENAMETOOLONG;
1446 if (res + 1 + len + 1 >= tmp1)
1447 return;
1448
1449 /* copy one component */
1450 *res = '/';
1451 memcpy (res + 1, beg, len);
1452
1453 /* zero-terminate, for readlink */
1454 res [len + 1] = 0;
1455
1456 /* now check if it's a symlink */
1457 linklen = readlink (req->ptr2, tmp1, PATH_MAX);
1458
1459 if (linklen < 0)
1460 {
1461 if (errno != EINVAL)
1462 return;
1463
1464 /* it's a normal directory. hopefully */
1465 res += len + 1;
1466 }
1467 else
1468 {
1469 /* yay, it was a symlink - build new path in tmp2 */
1470 int rellen = strlen (rel);
1471
1472 errno = ENAMETOOLONG;
1473 if (linklen + 1 + rellen >= PATH_MAX)
1474 return;
1475
1476 errno = ELOOP;
1477 if (!--symlinks)
1478 return;
1479
1480 if (*tmp1 == '/')
1481 res = req->ptr2; /* symlink resolves to an absolute path */
1482
1483 /* we need to be careful, as rel might point into tmp2 already */
1484 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1485 tmp2 [linklen] = '/';
1486 memcpy (tmp2, tmp1, linklen);
1487
1488 rel = tmp2;
1489 }
1490 }
1491
1492 /* special case for the lone root path */
1493 if (res == req->ptr2)
1494 *res++ = '/';
1495
1496 req->result = res - (char *)req->ptr2;
1497
1498done:
1499 req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
1069} 1500}
1070 1501
1071static signed char 1502static signed char
1072eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) 1503eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1073{ 1504{
1081 1512
1082#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */ 1513#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1083#define EIO_SORT_FAST 60 /* when to only use insertion sort */ 1514#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1084 1515
1085static void 1516static void
1086eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1517eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1087{ 1518{
1088 unsigned char bits [9 + sizeof (ino_t) * 8]; 1519 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1089 unsigned char *bit = bits; 1520 unsigned char *bit = bits;
1090 1521
1091 assert (CHAR_BIT == 8); 1522 assert (CHAR_BIT == 8);
1092 assert (sizeof (eio_dirent) * 8 < 256); 1523 assert (sizeof (eio_dirent) * 8 < 256);
1093 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */ 1524 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1095 1526
1096 if (size <= EIO_SORT_FAST) 1527 if (size <= EIO_SORT_FAST)
1097 return; 1528 return;
1098 1529
1099 /* first prepare an array of bits to test in our radix sort */ 1530 /* first prepare an array of bits to test in our radix sort */
1100 /* try to take endianness into account, as well as differences in ino_t sizes */ 1531 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1101 /* inode_bits must contain all inodes ORed together */ 1532 /* inode_bits must contain all inodes ORed together */
1102 /* which is used to skip bits that are 0 everywhere, which is very common */ 1533 /* which is used to skip bits that are 0 everywhere, which is very common */
1103 { 1534 {
1104 ino_t endianness; 1535 eio_ino_t endianness;
1105 int i, j; 1536 int i, j;
1106 1537
1107 /* we store the byte offset of byte n into byte n of "endianness" */ 1538 /* we store the byte offset of byte n into byte n of "endianness" */
1108 for (i = 0; i < sizeof (ino_t); ++i) 1539 for (i = 0; i < sizeof (eio_ino_t); ++i)
1109 ((unsigned char *)&endianness)[i] = i; 1540 ((unsigned char *)&endianness)[i] = i;
1110 1541
1111 *bit++ = 0; 1542 *bit++ = 0;
1112 1543
1113 for (i = 0; i < sizeof (ino_t); ++i) 1544 for (i = 0; i < sizeof (eio_ino_t); ++i)
1114 { 1545 {
1115 /* shifting off the byte offsets out of "endianness" */ 1546 /* shifting off the byte offsets out of "endianness" */
1116 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8; 1547 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1117 endianness >>= 8; 1548 endianness >>= 8;
1118 1549
1119 for (j = 0; j < 8; ++j) 1550 for (j = 0; j < 8; ++j)
1120 if (inode_bits & (((ino_t)1) << (i * 8 + j))) 1551 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1121 *bit++ = offs + j; 1552 *bit++ = offs + j;
1122 } 1553 }
1123 1554
1124 for (j = 0; j < 8; ++j) 1555 for (j = 0; j < 8; ++j)
1125 if (score_bits & (1 << j)) 1556 if (score_bits & (1 << j))
1126 *bit++ = offsetof (eio_dirent, score) * 8 + j; 1557 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1127 } 1558 }
1128 1559
1129 /* now actually do the sorting (a variant of MSD radix sort) */ 1560 /* now actually do the sorting (a variant of MSD radix sort) */
1130 { 1561 {
1131 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base; 1562 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1132 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end; 1563 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1133 unsigned char *bit_stk [9 + sizeof (ino_t) * 8]; 1564 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1134 int stk_idx = 0; 1565 int stk_idx = 0;
1135 1566
1136 base_stk [stk_idx] = dents; 1567 base_stk [stk_idx] = dents;
1137 end_stk [stk_idx] = dents + size; 1568 end_stk [stk_idx] = dents + size;
1138 bit_stk [stk_idx] = bit - 1; 1569 bit_stk [stk_idx] = bit - 1;
1217 } 1648 }
1218 } 1649 }
1219} 1650}
1220 1651
1221static void 1652static void
1222eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1653eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1223{ 1654{
1224 if (size <= 1) 1655 if (size <= 1)
1225 return; /* our insertion sort relies on size > 0 */ 1656 return; /* our insertion sort relies on size > 0 */
1226 1657
1227 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */ 1658 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1235 1666
1236/* read a full directory */ 1667/* read a full directory */
1237static void 1668static void
1238eio__scandir (eio_req *req, etp_worker *self) 1669eio__scandir (eio_req *req, etp_worker *self)
1239{ 1670{
1240 DIR *dirp;
1241 EIO_STRUCT_DIRENT *entp;
1242 char *name, *names; 1671 char *name, *names;
1243 int namesalloc = 4096; 1672 int namesalloc = 4096 - sizeof (void *) * 4;
1244 int namesoffs = 0; 1673 int namesoffs = 0;
1245 int flags = req->int1; 1674 int flags = req->int1;
1246 eio_dirent *dents = 0; 1675 eio_dirent *dents = 0;
1247 int dentalloc = 128; 1676 int dentalloc = 128;
1248 int dentoffs = 0; 1677 int dentoffs = 0;
1249 ino_t inode_bits = 0; 1678 eio_ino_t inode_bits = 0;
1679#ifdef _WIN32
1680 HANDLE dirp;
1681 WIN32_FIND_DATA entp;
1682#else
1683 DIR *dirp;
1684 EIO_STRUCT_DIRENT *entp;
1685#endif
1250 1686
1251 req->result = -1; 1687 req->result = -1;
1252 1688
1253 if (!(flags & EIO_READDIR_DENTS)) 1689 if (!(flags & EIO_READDIR_DENTS))
1254 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); 1690 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1255 1691
1256 X_LOCK (wrklock); 1692#ifdef _WIN32
1257 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 1693 {
1694 int len = strlen ((const char *)req->ptr1);
1695 char *path = malloc (MAX_PATH);
1696 const char *fmt;
1697
1698 if (!len)
1699 fmt = "./*";
1700 else if (((const char *)req->ptr1)[len - 1] == '/' || ((const char *)req->ptr1)[len - 1] == '\\')
1701 fmt = "%s*";
1702 else
1703 fmt = "%s/*";
1704
1705 _snprintf (path, MAX_PATH, fmt, (const char *)req->ptr1);
1706 dirp = FindFirstFile (path, &entp);
1707 free (path);
1708
1709 if (dirp == INVALID_HANDLE_VALUE)
1710 {
1711 dirp = 0;
1712
1713 /* should steal _dosmaperr */
1714 switch (GetLastError ())
1715 {
1716 case ERROR_FILE_NOT_FOUND:
1717 req->result = 0;
1718 break;
1719
1720 case ERROR_INVALID_NAME:
1721 case ERROR_PATH_NOT_FOUND:
1722 case ERROR_NO_MORE_FILES:
1723 errno = ENOENT;
1724 break;
1725
1726 case ERROR_NOT_ENOUGH_MEMORY:
1727 errno = ENOMEM;
1728 break;
1729
1730 default:
1731 errno = EINVAL;
1732 break;
1733 }
1734 }
1735 }
1736#else
1258 self->dirp = dirp = opendir (req->ptr1); 1737 dirp = opendir (req->ptr1);
1738#endif
1739
1740 if (req->flags & EIO_FLAG_PTR1_FREE)
1741 free (req->ptr1);
1259 1742
1260 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1743 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1261 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; 1744 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1262 req->ptr2 = names = malloc (namesalloc); 1745 req->ptr2 = names = malloc (namesalloc);
1263 X_UNLOCK (wrklock);
1264 1746
1265 if (dirp && names && (!flags || dents)) 1747 if (dirp && names && (!flags || dents))
1266 for (;;) 1748 for (;;)
1267 { 1749 {
1750 int done;
1751
1752#ifdef _WIN32
1753 done = !dirp;
1754#else
1268 errno = 0; 1755 errno = 0;
1269 entp = readdir (dirp); 1756 entp = readdir (dirp);
1757 done = !entp;
1758#endif
1270 1759
1271 if (!entp) 1760 if (done)
1272 { 1761 {
1762#ifndef _WIN32
1763 int old_errno = errno;
1764 closedir (dirp);
1765 errno = old_errno;
1766
1273 if (errno) 1767 if (errno)
1274 break; 1768 break;
1769#endif
1275 1770
1276 /* sort etc. */ 1771 /* sort etc. */
1277 req->int1 = flags; 1772 req->int1 = flags;
1278 req->result = dentoffs; 1773 req->result = dentoffs;
1279 1774
1308 1803
1309 break; 1804 break;
1310 } 1805 }
1311 1806
1312 /* now add the entry to our list(s) */ 1807 /* now add the entry to our list(s) */
1313 name = entp->d_name; 1808 name = D_NAME (entp);
1314 1809
1315 /* skip . and .. entries */ 1810 /* skip . and .. entries */
1316 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1811 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1317 { 1812 {
1318 int len = D_NAMLEN (entp) + 1; 1813 int len = D_NAMLEN (entp) + 1;
1319 1814
1320 while (expect_false (namesoffs + len > namesalloc)) 1815 while (ecb_expect_false (namesoffs + len > namesalloc))
1321 { 1816 {
1322 namesalloc *= 2; 1817 namesalloc *= 2;
1323 X_LOCK (wrklock);
1324 req->ptr2 = names = realloc (names, namesalloc); 1818 req->ptr2 = names = realloc (names, namesalloc);
1325 X_UNLOCK (wrklock);
1326 1819
1327 if (!names) 1820 if (!names)
1328 break; 1821 break;
1329 } 1822 }
1330 1823
1332 1825
1333 if (dents) 1826 if (dents)
1334 { 1827 {
1335 struct eio_dirent *ent; 1828 struct eio_dirent *ent;
1336 1829
1337 if (expect_false (dentoffs == dentalloc)) 1830 if (ecb_expect_false (dentoffs == dentalloc))
1338 { 1831 {
1339 dentalloc *= 2; 1832 dentalloc *= 2;
1340 X_LOCK (wrklock);
1341 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent)); 1833 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1342 X_UNLOCK (wrklock);
1343 1834
1344 if (!dents) 1835 if (!dents)
1345 break; 1836 break;
1346 } 1837 }
1347 1838
1427 if (EIO_CANCELLED (req)) 1918 if (EIO_CANCELLED (req))
1428 { 1919 {
1429 errno = ECANCELED; 1920 errno = ECANCELED;
1430 break; 1921 break;
1431 } 1922 }
1923
1924#ifdef _WIN32
1925 if (!FindNextFile (dirp, &entp))
1926 {
1927 FindClose (dirp);
1928 dirp = 0;
1929 }
1930#endif
1432 } 1931 }
1433}
1434
1435#ifdef PAGESIZE
1436# define eio_pagesize() PAGESIZE
1437#else
1438static intptr_t
1439eio_pagesize (void)
1440{
1441 static intptr_t page;
1442
1443 if (!page)
1444 page = sysconf (_SC_PAGESIZE);
1445
1446 return page;
1447}
1448#endif
1449
1450static void
1451eio_page_align (void **addr, size_t *length)
1452{
1453 intptr_t mask = eio_pagesize () - 1;
1454
1455 /* round down addr */
1456 intptr_t adj = mask & (intptr_t)*addr;
1457
1458 *addr = (void *)((intptr_t)*addr - adj);
1459 *length += adj;
1460
1461 /* round up length */
1462 *length = (*length + mask) & ~mask;
1463}
1464
1465#if !_POSIX_MEMLOCK
1466# define eio__mlockall(a) ((errno = ENOSYS), -1)
1467#else
1468
1469static int
1470eio__mlockall (int flags)
1471{
1472 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1473 extern int mallopt (int, int);
1474 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1475 #endif
1476
1477 if (EIO_MCL_CURRENT != MCL_CURRENT
1478 || EIO_MCL_FUTURE != MCL_FUTURE)
1479 {
1480 flags = 0
1481 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1482 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1483 }
1484
1485 return mlockall (flags);
1486}
1487#endif
1488
1489#if !_POSIX_MEMLOCK_RANGE
1490# define eio__mlock(a,b) ((errno = ENOSYS), -1)
1491#else
1492
1493static int
1494eio__mlock (void *addr, size_t length)
1495{
1496 eio_page_align (&addr, &length);
1497
1498 return mlock (addr, length);
1499}
1500
1501#endif
1502
1503#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1504# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
1505#else
1506
1507int
1508eio__msync (void *mem, size_t len, int flags)
1509{
1510 eio_page_align (&mem, &len);
1511
1512 if (EIO_MS_ASYNC != MS_SYNC
1513 || EIO_MS_INVALIDATE != MS_INVALIDATE
1514 || EIO_MS_SYNC != MS_SYNC)
1515 {
1516 flags = 0
1517 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1518 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1519 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1520 }
1521
1522 return msync (mem, len, flags);
1523}
1524
1525#endif
1526
1527int
1528eio__mtouch (void *mem, size_t len, int flags)
1529{
1530 eio_page_align (&mem, &len);
1531
1532 {
1533 intptr_t addr = (intptr_t)mem;
1534 intptr_t end = addr + len;
1535 intptr_t page = eio_pagesize ();
1536
1537 if (addr < end)
1538 if (flags & EIO_MT_MODIFY) /* modify */
1539 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
1540 else
1541 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
1542 }
1543
1544 return 0;
1545} 1932}
1546 1933
1547/*****************************************************************************/ 1934/*****************************************************************************/
1548 1935
1549#define ALLOC(len) \ 1936#define ALLOC(len) \
1565{ 1952{
1566 ETP_REQ *req; 1953 ETP_REQ *req;
1567 struct timespec ts; 1954 struct timespec ts;
1568 etp_worker *self = (etp_worker *)thr_arg; 1955 etp_worker *self = (etp_worker *)thr_arg;
1569 1956
1957#if HAVE_PRCTL_SET_NAME
1958 prctl (PR_SET_NAME, (unsigned long)"eio_thread", 0, 0, 0);
1959#endif
1960
1570 /* try to distribute timeouts somewhat randomly */ 1961 /* try to distribute timeouts somewhat evenly */
1571 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 1962 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1572 1963
1573 for (;;) 1964 for (;;)
1574 { 1965 {
1966 ts.tv_sec = 0;
1967
1575 X_LOCK (reqlock); 1968 X_LOCK (reqlock);
1576 1969
1577 for (;;) 1970 for (;;)
1578 { 1971 {
1579 self->req = req = reqq_shift (&req_queue); 1972 self->req = req = reqq_shift (&req_queue);
1580 1973
1581 if (req) 1974 if (req)
1582 break; 1975 break;
1583 1976
1977 if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
1978 {
1979 X_UNLOCK (reqlock);
1980 X_LOCK (wrklock);
1981 --started;
1982 X_UNLOCK (wrklock);
1983 goto quit;
1984 }
1985
1584 ++idle; 1986 ++idle;
1585 1987
1586 ts.tv_sec = time (0) + idle_timeout; 1988 if (idle <= max_idle)
1587 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 1989 /* we are allowed to idle, so do so without any timeout */
1990 X_COND_WAIT (reqwait, reqlock);
1991 else
1588 { 1992 {
1589 if (idle > max_idle) 1993 /* initialise timeout once */
1590 { 1994 if (!ts.tv_sec)
1591 --idle; 1995 ts.tv_sec = time (0) + idle_timeout;
1592 X_UNLOCK (reqlock);
1593 X_LOCK (wrklock);
1594 --started;
1595 X_UNLOCK (wrklock);
1596 goto quit;
1597 }
1598 1996
1599 /* we are allowed to idle, so do so without any timeout */
1600 X_COND_WAIT (reqwait, reqlock); 1997 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
1998 ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1601 } 1999 }
1602 2000
1603 --idle; 2001 --idle;
1604 } 2002 }
1605 2003
1608 X_UNLOCK (reqlock); 2006 X_UNLOCK (reqlock);
1609 2007
1610 if (req->type < 0) 2008 if (req->type < 0)
1611 goto quit; 2009 goto quit;
1612 2010
1613 if (!EIO_CANCELLED (req))
1614 ETP_EXECUTE (self, req); 2011 ETP_EXECUTE (self, req);
1615 2012
1616 X_LOCK (reslock); 2013 X_LOCK (reslock);
1617 2014
1618 ++npending; 2015 ++npending;
1619 2016
1634 return 0; 2031 return 0;
1635} 2032}
1636 2033
1637/*****************************************************************************/ 2034/*****************************************************************************/
1638 2035
2036int ecb_cold
1639int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 2037eio_init (void (*want_poll)(void), void (*done_poll)(void))
1640{ 2038{
2039#if !HAVE_PREADWRITE
2040 X_MUTEX_CREATE (preadwritelock);
2041#endif
2042
1641 return etp_init (want_poll, done_poll); 2043 return etp_init (want_poll, done_poll);
1642} 2044}
1643 2045
2046ecb_inline void
1644static void eio_api_destroy (eio_req *req) 2047eio_api_destroy (eio_req *req)
1645{ 2048{
1646 free (req); 2049 free (req);
1647} 2050}
1648 2051
1649#define REQ(rtype) \ 2052#define REQ(rtype) \
1668 { \ 2071 { \
1669 eio_api_destroy (req); \ 2072 eio_api_destroy (req); \
1670 return 0; \ 2073 return 0; \
1671 } 2074 }
1672 2075
2076static void
1673static void eio_execute (etp_worker *self, eio_req *req) 2077eio_execute (etp_worker *self, eio_req *req)
1674{ 2078{
2079 if (ecb_expect_false (EIO_CANCELLED (req)))
2080 {
2081 req->result = -1;
2082 req->errorno = ECANCELED;
2083 return;
2084 }
2085
1675 switch (req->type) 2086 switch (req->type)
1676 { 2087 {
1677 case EIO_READ: ALLOC (req->size); 2088 case EIO_READ: ALLOC (req->size);
1678 req->result = req->offs >= 0 2089 req->result = req->offs >= 0
1679 ? pread (req->int1, req->ptr2, req->size, req->offs) 2090 ? pread (req->int1, req->ptr2, req->size, req->offs)
1681 case EIO_WRITE: req->result = req->offs >= 0 2092 case EIO_WRITE: req->result = req->offs >= 0
1682 ? pwrite (req->int1, req->ptr2, req->size, req->offs) 2093 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1683 : write (req->int1, req->ptr2, req->size); break; 2094 : write (req->int1, req->ptr2, req->size); break;
1684 2095
1685 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break; 2096 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1686 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break; 2097 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
1687 2098
1688 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2099 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1689 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2100 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1690 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2101 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1691 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2102 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1713 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; 2124 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1714 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; 2125 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1715 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; 2126 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1716 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break; 2127 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1717 2128
2129 case EIO_REALPATH: eio__realpath (req, self); break;
2130
1718 case EIO_READLINK: ALLOC (PATH_MAX); 2131 case EIO_READLINK: ALLOC (PATH_MAX);
1719 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break; 2132 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1720 2133
1721 case EIO_SYNC: req->result = 0; sync (); break; 2134 case EIO_SYNC: req->result = 0; sync (); break;
1722 case EIO_FSYNC: req->result = fsync (req->int1); break; 2135 case EIO_FSYNC: req->result = fsync (req->int1); break;
1723 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 2136 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
2137 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
2138 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1724 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break; 2139 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1725 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break; 2140 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1726 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break; 2141 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1727 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break; 2142 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1728 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break; 2143 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
1729 2144
1730 case EIO_READDIR: eio__scandir (req, self); break; 2145 case EIO_READDIR: eio__scandir (req, self); break;
1731 2146
1732 case EIO_BUSY: 2147 case EIO_BUSY:
1733#ifdef _WIN32 2148#ifdef _WIN32
1774 case EIO_NOP: 2189 case EIO_NOP:
1775 req->result = 0; 2190 req->result = 0;
1776 break; 2191 break;
1777 2192
1778 case EIO_CUSTOM: 2193 case EIO_CUSTOM:
1779 ((void (*)(eio_req *))req->feed) (req); 2194 req->feed (req);
1780 break; 2195 break;
1781 2196
1782 default: 2197 default:
1783 errno = ENOSYS; 2198 errno = ENOSYS;
1784 req->result = -1; 2199 req->result = -1;
1813eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2228eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1814{ 2229{
1815 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2230 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1816} 2231}
1817 2232
2233eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2234{
2235 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2236}
2237
2238eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
2239{
2240 REQ (EIO_SYNCFS); req->int1 = fd; SEND;
2241}
2242
2243eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2244{
2245 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2246}
2247
1818eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2248eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1819{ 2249{
1820 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2250 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1821} 2251}
1822 2252
1828eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data) 2258eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
1829{ 2259{
1830 REQ (EIO_MLOCKALL); req->int1 = flags; SEND; 2260 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
1831} 2261}
1832 2262
1833eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) 2263eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
1834{ 2264{
1835 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; 2265 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
1836}
1837
1838eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1839{
1840 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1841} 2266}
1842 2267
1843eio_req *eio_close (int fd, int pri, eio_cb cb, void *data) 2268eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
1844{ 2269{
1845 REQ (EIO_CLOSE); req->int1 = fd; SEND; 2270 REQ (EIO_CLOSE); req->int1 = fd; SEND;
1883eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) 2308eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1884{ 2309{
1885 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2310 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1886} 2311}
1887 2312
1888eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2313eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1889{ 2314{
1890 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2315 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1891} 2316}
1892 2317
1893eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) 2318eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
1913eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data) 2338eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1914{ 2339{
1915 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2340 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1916} 2341}
1917 2342
1918eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2343eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1919{ 2344{
1920 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2345 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1921} 2346}
1922 2347
1923eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) 2348eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
1937} 2362}
1938 2363
1939eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data) 2364eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1940{ 2365{
1941 return eio__1path (EIO_READLINK, path, pri, cb, data); 2366 return eio__1path (EIO_READLINK, path, pri, cb, data);
2367}
2368
2369eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2370{
2371 return eio__1path (EIO_REALPATH, path, pri, cb, data);
1942} 2372}
1943 2373
1944eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) 2374eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1945{ 2375{
1946 return eio__1path (EIO_STAT, path, pri, cb, data); 2376 return eio__1path (EIO_STAT, path, pri, cb, data);
2005eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2435eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2006{ 2436{
2007 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2437 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
2008} 2438}
2009 2439
2010eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) 2440eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
2011{ 2441{
2012 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND; 2442 REQ (EIO_CUSTOM); req->feed = execute; SEND;
2013} 2443}
2014 2444
2015#endif 2445#endif
2016 2446
2017eio_req *eio_grp (eio_cb cb, void *data) 2447eio_req *eio_grp (eio_cb cb, void *data)
2026#undef SEND 2456#undef SEND
2027 2457
2028/*****************************************************************************/ 2458/*****************************************************************************/
2029/* grp functions */ 2459/* grp functions */
2030 2460
2461void
2031void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2462eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
2032{ 2463{
2033 grp->int2 = limit; 2464 grp->int2 = limit;
2034 grp->feed = feed; 2465 grp->feed = feed;
2035 2466
2036 grp_try_feed (grp); 2467 grp_try_feed (grp);
2037} 2468}
2038 2469
2470void
2039void eio_grp_limit (eio_req *grp, int limit) 2471eio_grp_limit (eio_req *grp, int limit)
2040{ 2472{
2041 grp->int2 = limit; 2473 grp->int2 = limit;
2042 2474
2043 grp_try_feed (grp); 2475 grp_try_feed (grp);
2044} 2476}
2045 2477
2478void
2046void eio_grp_add (eio_req *grp, eio_req *req) 2479eio_grp_add (eio_req *grp, eio_req *req)
2047{ 2480{
2048 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2481 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
2049 2482
2050 grp->flags |= EIO_FLAG_GROUPADD; 2483 grp->flags |= EIO_FLAG_GROUPADD;
2051 2484
2062} 2495}
2063 2496
2064/*****************************************************************************/ 2497/*****************************************************************************/
2065/* misc garbage */ 2498/* misc garbage */
2066 2499
2500eio_ssize_t
2067ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2501eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2068{ 2502{
2069 etp_worker wrk;
2070 ssize_t ret;
2071
2072 wrk.dbuf = 0;
2073
2074 ret = eio__sendfile (ofd, ifd, offset, count, &wrk); 2503 return eio__sendfile (ofd, ifd, offset, count);
2075
2076 if (wrk.dbuf)
2077 free (wrk.dbuf);
2078
2079 return ret;
2080} 2504}
2081 2505

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines