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

Comparing libeio/eio.c (file contents):
Revision 1.65 by root, Mon May 30 12:56:50 2011 UTC vs.
Revision 1.106 by root, Mon Sep 26 20:19:08 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 X_STACKSIZE EIO_STACKSIZE
48#endif 49#endif
49#include "xthread.h" 50#include "xthread.h"
50 51
51#include <errno.h> 52#include <errno.h>
52#include <stddef.h> 53#include <stddef.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#if _POSIX_VERSION >= 200809L
64# define HAVE_AT 1
65#else
66# define HAVE_AT 0
67#endif
68
69/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
70/* intptr_t only comes from stdint.h, says idiot openbsd coder */
71#if HAVE_STDINT_H
72# include <stdint.h>
73#endif
74
75#ifndef ECANCELED
76# define ECANCELED EDOM
77#endif
78#ifndef ELOOP
79# define ELOOP EDOM
80#endif
81
82#if !defined(ENOTSOCK) && defined(WSAENOTSOCK)
83# define ENOTSOCK WSAENOTSOCK
84#endif
85
86static void eio_destroy (eio_req *req);
87
63#ifndef EIO_FINISH 88#ifndef EIO_FINISH
64# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0 89# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
65#endif 90#endif
66 91
67#ifndef EIO_DESTROY 92#ifndef EIO_DESTROY
… …
70 95
71#ifndef EIO_FEED 96#ifndef EIO_FEED
72# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0) 97# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)
73#endif 98#endif
74 99
100#ifndef EIO_FD_TO_WIN32_HANDLE
101# define EIO_FD_TO_WIN32_HANDLE(fd) _get_osfhandle (fd)
102#endif
103#ifndef EIO_WIN32_HANDLE_TO_FD
104# define EIO_WIN32_HANDLE_TO_FD(handle) _open_osfhandle (handle, 0)
105#endif
106
107#define EIO_ERRNO(errval,retval) ((errno = errval), retval)
108
109#define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1)
110
75#ifdef _WIN32 111#ifdef _WIN32
76 112
77 /*doh*/ 113 #undef PAGESIZE
114 #define PAGESIZE 4096 /* GetSystemInfo? */
115
116 /* TODO: look at how perl does stat (non-sloppy), unlink (ro-files), utime, link */
117
118 #ifdef EIO_STRUCT_STATI64
119 /* look at perl's non-sloppy stat */
120 #define stat(path,buf) _stati64 (path,buf)
121 #define fstat(fd,buf) _fstati64 (fd,buf)
122 #endif
123 #define lstat(path,buf) stat (path,buf)
124 #define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
125 #define mkdir(path,mode) _mkdir (path)
126 #define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
127
128 #define chmod(path,mode) _chmod (path, mode)
129 #define dup(fd) _dup (fd)
130 #define dup2(fd1,fd2) _dup2 (fd1, fd2)
131
132 #define fchmod(fd,mode) EIO_ENOSYS ()
133 #define chown(path,uid,gid) EIO_ENOSYS ()
134 #define fchown(fd,uid,gid) EIO_ENOSYS ()
135 #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
136 #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
137 #define mknod(path,mode,dev) EIO_ENOSYS ()
138 #define sync() EIO_ENOSYS ()
139 #define readlink(path,buf,s) EIO_ENOSYS ()
140 #define statvfs(path,buf) EIO_ENOSYS ()
141 #define fstatvfs(fd,buf) EIO_ENOSYS ()
142
143 /* rename() uses MoveFile, which fails to overwrite */
144 #define rename(old,neu) eio__rename (old, neu)
145
146 static int
147 eio__rename (const char *old, const char *neu)
148 {
149 if (MoveFileEx (old, neu, MOVEFILE_REPLACE_EXISTING))
150 return 0;
151
152 /* should steal _dosmaperr */
153 switch (GetLastError ())
154 {
155 case ERROR_FILE_NOT_FOUND:
156 case ERROR_PATH_NOT_FOUND:
157 case ERROR_INVALID_DRIVE:
158 case ERROR_NO_MORE_FILES:
159 case ERROR_BAD_NETPATH:
160 case ERROR_BAD_NET_NAME:
161 case ERROR_BAD_PATHNAME:
162 case ERROR_FILENAME_EXCED_RANGE:
163 errno = ENOENT;
164 break;
165
166 default:
167 errno = EACCES;
168 break;
169 }
170
171 return -1;
172 }
173
174 /* we could even stat and see if it exists */
175 static int
176 symlink (const char *old, const char *neu)
177 {
178 #if WINVER >= 0x0600
179 if (CreateSymbolicLink (neu, old, 1))
180 return 0;
181
182 if (CreateSymbolicLink (neu, old, 0))
183 return 0;
184 #endif
185
186 return EIO_ERRNO (ENOENT, -1);
187 }
188
189 /* POSIX API only */
190 #define CreateHardLink(neu,old,flags) 0
191 #define CreateSymbolicLink(neu,old,flags) 0
192
193 struct statvfs
194 {
195 int dummy;
196 };
197
198 #define DT_DIR EIO_DT_DIR
199 #define DT_REG EIO_DT_REG
200 #define D_NAME(entp) entp.cFileName
201 #define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
202
78#else 203#else
79 204
80# include <sys/time.h> 205 #include <sys/time.h>
81# include <sys/select.h> 206 #include <sys/select.h>
207 #include <sys/statvfs.h>
82# include <unistd.h> 208 #include <unistd.h>
209 #include <signal.h>
210 #include <dirent.h>
211
212 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
213 #include <sys/mman.h>
214 #endif
215
216 #define D_NAME(entp) entp->d_name
217
218 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
219 #if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
220 #define _DIRENT_HAVE_D_TYPE /* sigh */
221 #define D_INO(de) (de)->d_fileno
222 #define D_NAMLEN(de) (de)->d_namlen
223 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
224 #define D_INO(de) (de)->d_ino
225 #endif
226
227 #ifdef _D_EXACT_NAMLEN
228 #undef D_NAMLEN
229 #define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
230 #endif
231
232 #ifdef _DIRENT_HAVE_D_TYPE
233 #define D_TYPE(de) (de)->d_type
234 #endif
235
236 #ifndef EIO_STRUCT_DIRENT
237 #define EIO_STRUCT_DIRENT struct dirent
238 #endif
239
240#endif
241
242#if HAVE_UTIMES
83# include <utime.h> 243# include <utime.h>
84# include <signal.h>
85# include <dirent.h>
86
87#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
88# include <sys/mman.h>
89#endif
90
91/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
92# if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
93# define _DIRENT_HAVE_D_TYPE /* sigh */
94# define D_INO(de) (de)->d_fileno
95# define D_NAMLEN(de) (de)->d_namlen
96# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
97# define D_INO(de) (de)->d_ino
98# endif 244#endif
99 245
100#ifdef _D_EXACT_NAMLEN 246#if HAVE_SYS_SYSCALL_H
101# undef D_NAMLEN 247# include <sys/syscall.h>
102# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
103#endif
104
105# ifdef _DIRENT_HAVE_D_TYPE
106# define D_TYPE(de) (de)->d_type
107# endif 248#endif
108 249
109# ifndef EIO_STRUCT_DIRENT 250#if HAVE_SYS_PRCTL_H
110# define EIO_STRUCT_DIRENT struct dirent 251# include <sys/prctl.h>
111# endif
112
113#endif 252#endif
114 253
115#if HAVE_SENDFILE 254#if HAVE_SENDFILE
116# if __linux 255# if __linux
117# include <sys/sendfile.h> 256# include <sys/sendfile.h>
… …
132#endif 271#endif
133#ifndef D_INO 272#ifndef D_INO
134# define D_INO(de) 0 273# define D_INO(de) 0
135#endif 274#endif
136#ifndef D_NAMLEN 275#ifndef D_NAMLEN
137# define D_NAMLEN(de) strlen ((de)->d_name) 276# define D_NAMLEN(entp) strlen (D_NAME (entp))
138#endif 277#endif
139 278
140/* used for struct dirent, AIX doesn't provide it */ 279/* used for struct dirent, AIX doesn't provide it */
141#ifndef NAME_MAX 280#ifndef NAME_MAX
142# define NAME_MAX 4096 281# define NAME_MAX 4096
… …
149 288
150/* buffer size for various temporary buffers */ 289/* buffer size for various temporary buffers */
151#define EIO_BUFSIZE 65536 290#define EIO_BUFSIZE 65536
152 291
153#define dBUF \ 292#define dBUF \
154 char *eio_buf; \
155 ETP_WORKER_LOCK (self); \
156 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 293 char *eio_buf = malloc (EIO_BUFSIZE); \
157 ETP_WORKER_UNLOCK (self); \
158 errno = ENOMEM; \ 294 errno = ENOMEM; \
159 if (!eio_buf) \ 295 if (!eio_buf) \
160 return -1; 296 return -1
297
298#define FUBd \
299 free (eio_buf)
161 300
162#define EIO_TICKS ((1000000 + 1023) >> 10) 301#define EIO_TICKS ((1000000 + 1023) >> 10)
163 302
164/*****************************************************************************/ 303/*****************************************************************************/
165 304
166#if __GNUC__ >= 3 305struct tmpbuf
167# define expect(expr,value) __builtin_expect ((expr),(value)) 306{
168#else 307 void *ptr;
169# define expect(expr,value) (expr) 308 int len;
170#endif 309};
171 310
172#define expect_false(expr) expect ((expr) != 0, 0) 311static void *
173#define expect_true(expr) expect ((expr) != 0, 1) 312tmpbuf_get (struct tmpbuf *buf, int len)
313{
314 if (buf->len < len)
315 {
316 free (buf->ptr);
317 buf->ptr = malloc (buf->len = len);
318 }
319
320 return buf->ptr;
321}
174 322
175/*****************************************************************************/ 323/*****************************************************************************/
176 324
177#define ETP_PRI_MIN EIO_PRI_MIN 325#define ETP_PRI_MIN EIO_PRI_MIN
178#define ETP_PRI_MAX EIO_PRI_MAX 326#define ETP_PRI_MAX EIO_PRI_MAX
… …
184static int eio_finish (eio_req *req); 332static int eio_finish (eio_req *req);
185#define ETP_FINISH(req) eio_finish (req) 333#define ETP_FINISH(req) eio_finish (req)
186static void eio_execute (struct etp_worker *self, eio_req *req); 334static void eio_execute (struct etp_worker *self, eio_req *req);
187#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) 335#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
188 336
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/*****************************************************************************/ 337/*****************************************************************************/
207 338
208#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) 339#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
209 340
210/* calculate time difference in ~1/EIO_TICKS of a second */ 341/* calculate time difference in ~1/EIO_TICKS of a second */
342ecb_inline int
211static int tvdiff (struct timeval *tv1, struct timeval *tv2) 343tvdiff (struct timeval *tv1, struct timeval *tv2)
212{ 344{
213 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 345 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
214 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 346 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
215} 347}
216 348
… …
220static void (*done_poll_cb) (void); 352static void (*done_poll_cb) (void);
221 353
222static unsigned int max_poll_time; /* reslock */ 354static unsigned int max_poll_time; /* reslock */
223static unsigned int max_poll_reqs; /* reslock */ 355static unsigned int max_poll_reqs; /* reslock */
224 356
225static volatile unsigned int nreqs; /* reqlock */ 357static unsigned int nreqs; /* reqlock */
226static volatile unsigned int nready; /* reqlock */ 358static unsigned int nready; /* reqlock */
227static volatile unsigned int npending; /* reqlock */ 359static unsigned int npending; /* reqlock */
228static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */ 360static unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
229static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */ 361static unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
230 362
231static xmutex_t wrklock; 363static xmutex_t wrklock;
232static xmutex_t reslock; 364static xmutex_t reslock;
233static xmutex_t reqlock; 365static xmutex_t reqlock;
234static xcond_t reqwait; 366static xcond_t reqwait;
… …
237/* 369/*
238 * make our pread/pwrite emulation safe against themselves, but not against 370 * make our pread/pwrite emulation safe against themselves, but not against
239 * normal read/write by using a mutex. slows down execution a lot, 371 * normal read/write by using a mutex. slows down execution a lot,
240 * but that's your problem, not mine. 372 * but that's your problem, not mine.
241 */ 373 */
242static xmutex_t preadwritelock = X_MUTEX_INIT; 374static xmutex_t preadwritelock;
243#endif 375#endif
244 376
245typedef struct etp_worker 377typedef struct etp_worker
246{ 378{
379 struct tmpbuf tmpbuf;
380
247 /* locked by wrklock */ 381 /* locked by wrklock */
248 struct etp_worker *prev, *next; 382 struct etp_worker *prev, *next;
249 383
250 xthread_t tid; 384 xthread_t tid;
251 385
252 /* locked by reslock, reqlock or wrklock */ 386#ifdef ETP_WORKER_COMMON
253 ETP_REQ *req; /* currently processed request */
254
255 ETP_WORKER_COMMON 387 ETP_WORKER_COMMON
388#endif
256} etp_worker; 389} etp_worker;
257 390
258static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */ 391static etp_worker wrk_first; /* NOT etp */
259 392
260#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock) 393#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
261#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock) 394#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
262 395
263/* worker threads management */ 396/* worker threads management */
264 397
398static void ecb_cold
265static void etp_worker_clear (etp_worker *wrk) 399etp_worker_clear (etp_worker *wrk)
266{ 400{
267 ETP_WORKER_CLEAR (wrk);
268} 401}
269 402
403static void ecb_cold
270static void etp_worker_free (etp_worker *wrk) 404etp_worker_free (etp_worker *wrk)
271{ 405{
406 free (wrk->tmpbuf.ptr);
407
272 wrk->next->prev = wrk->prev; 408 wrk->next->prev = wrk->prev;
273 wrk->prev->next = wrk->next; 409 wrk->prev->next = wrk->next;
274 410
275 free (wrk); 411 free (wrk);
276} 412}
277 413
278static unsigned int etp_nreqs (void) 414static unsigned int
415etp_nreqs (void)
279{ 416{
280 int retval; 417 int retval;
281 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 418 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
282 retval = nreqs; 419 retval = nreqs;
283 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 420 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
284 return retval; 421 return retval;
285} 422}
286 423
287static unsigned int etp_nready (void) 424static unsigned int
425etp_nready (void)
288{ 426{
289 unsigned int retval; 427 unsigned int retval;
290 428
291 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 429 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
292 retval = nready; 430 retval = nready;
293 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 431 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
294 432
295 return retval; 433 return retval;
296} 434}
297 435
298static unsigned int etp_npending (void) 436static unsigned int
437etp_npending (void)
299{ 438{
300 unsigned int retval; 439 unsigned int retval;
301 440
302 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 441 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
303 retval = npending; 442 retval = npending;
304 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 443 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
305 444
306 return retval; 445 return retval;
307} 446}
308 447
309static unsigned int etp_nthreads (void) 448static unsigned int
449etp_nthreads (void)
310{ 450{
311 unsigned int retval; 451 unsigned int retval;
312 452
313 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 453 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
314 retval = started; 454 retval = started;
… …
328} etp_reqq; 468} etp_reqq;
329 469
330static etp_reqq req_queue; 470static etp_reqq req_queue;
331static etp_reqq res_queue; 471static etp_reqq res_queue;
332 472
473static void ecb_noinline ecb_cold
474reqq_init (etp_reqq *q)
475{
476 int pri;
477
478 for (pri = 0; pri < ETP_NUM_PRI; ++pri)
479 q->qs[pri] = q->qe[pri] = 0;
480
481 q->size = 0;
482}
483
484static int ecb_noinline
333static int reqq_push (etp_reqq *q, ETP_REQ *req) 485reqq_push (etp_reqq *q, ETP_REQ *req)
334{ 486{
335 int pri = req->pri; 487 int pri = req->pri;
336 req->next = 0; 488 req->next = 0;
337 489
338 if (q->qe[pri]) 490 if (q->qe[pri])
… …
344 q->qe[pri] = q->qs[pri] = req; 496 q->qe[pri] = q->qs[pri] = req;
345 497
346 return q->size++; 498 return q->size++;
347} 499}
348 500
501static ETP_REQ * ecb_noinline
349static ETP_REQ *reqq_shift (etp_reqq *q) 502reqq_shift (etp_reqq *q)
350{ 503{
351 int pri; 504 int pri;
352 505
353 if (!q->size) 506 if (!q->size)
354 return 0; 507 return 0;
… …
369 } 522 }
370 523
371 abort (); 524 abort ();
372} 525}
373 526
374static void etp_thread_init (void) 527static int ecb_cold
528etp_init (void (*want_poll)(void), void (*done_poll)(void))
375{ 529{
376 X_MUTEX_CREATE (wrklock); 530 X_MUTEX_CREATE (wrklock);
377 X_MUTEX_CREATE (reslock); 531 X_MUTEX_CREATE (reslock);
378 X_MUTEX_CREATE (reqlock); 532 X_MUTEX_CREATE (reqlock);
379 X_COND_CREATE (reqwait); 533 X_COND_CREATE (reqwait);
380}
381 534
382static void etp_atfork_prepare (void) 535 reqq_init (&req_queue);
383{ 536 reqq_init (&res_queue);
384 X_LOCK (wrklock);
385 X_LOCK (reqlock);
386 X_LOCK (reslock);
387#if !HAVE_PREADWRITE
388 X_LOCK (preadwritelock);
389#endif
390}
391 537
392static void etp_atfork_parent (void) 538 wrk_first.next =
393{ 539 wrk_first.prev = &wrk_first;
394#if !HAVE_PREADWRITE
395 X_UNLOCK (preadwritelock);
396#endif
397 X_UNLOCK (reslock);
398 X_UNLOCK (reqlock);
399 X_UNLOCK (wrklock);
400}
401
402static void etp_atfork_child (void)
403{
404 ETP_REQ *prv;
405
406 while ((prv = reqq_shift (&req_queue)))
407 ETP_DESTROY (prv);
408
409 while ((prv = reqq_shift (&res_queue)))
410 ETP_DESTROY (prv);
411
412 while (wrk_first.next != &wrk_first)
413 {
414 etp_worker *wrk = wrk_first.next;
415
416 if (wrk->req)
417 ETP_DESTROY (wrk->req);
418
419 etp_worker_clear (wrk);
420 etp_worker_free (wrk);
421 }
422 540
423 started = 0; 541 started = 0;
424 idle = 0; 542 idle = 0;
425 nreqs = 0; 543 nreqs = 0;
426 nready = 0; 544 nready = 0;
427 npending = 0; 545 npending = 0;
428 546
429 etp_thread_init ();
430}
431
432static void
433etp_once_init (void)
434{
435 etp_thread_init ();
436 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
437}
438
439static int
440etp_init (void (*want_poll)(void), void (*done_poll)(void))
441{
442 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
443
444 pthread_once (&doinit, etp_once_init);
445
446 want_poll_cb = want_poll; 547 want_poll_cb = want_poll;
447 done_poll_cb = done_poll; 548 done_poll_cb = done_poll;
448 549
449 return 0; 550 return 0;
450} 551}
451 552
452X_THREAD_PROC (etp_proc); 553X_THREAD_PROC (etp_proc);
453 554
555static void ecb_cold
454static void etp_start_thread (void) 556etp_start_thread (void)
455{ 557{
456 etp_worker *wrk = calloc (1, sizeof (etp_worker)); 558 etp_worker *wrk = calloc (1, sizeof (etp_worker));
457 559
458 /*TODO*/ 560 /*TODO*/
459 assert (("unable to allocate worker thread data", wrk)); 561 assert (("unable to allocate worker thread data", wrk));
… …
472 free (wrk); 574 free (wrk);
473 575
474 X_UNLOCK (wrklock); 576 X_UNLOCK (wrklock);
475} 577}
476 578
579static void
477static void etp_maybe_start_thread (void) 580etp_maybe_start_thread (void)
478{ 581{
479 if (expect_true (etp_nthreads () >= wanted)) 582 if (ecb_expect_true (etp_nthreads () >= wanted))
480 return; 583 return;
481 584
482 /* todo: maybe use idle here, but might be less exact */ 585 /* todo: maybe use idle here, but might be less exact */
483 if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ())) 586 if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
484 return; 587 return;
485 588
486 etp_start_thread (); 589 etp_start_thread ();
487} 590}
488 591
592static void ecb_cold
489static void etp_end_thread (void) 593etp_end_thread (void)
490{ 594{
491 eio_req *req = calloc (1, sizeof (eio_req)); 595 eio_req *req = calloc (1, sizeof (eio_req)); /* will be freed by worker */
492 596
493 req->type = -1; 597 req->type = -1;
494 req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 598 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
495 599
496 X_LOCK (reqlock); 600 X_LOCK (reqlock);
… …
501 X_LOCK (wrklock); 605 X_LOCK (wrklock);
502 --started; 606 --started;
503 X_UNLOCK (wrklock); 607 X_UNLOCK (wrklock);
504} 608}
505 609
506static int etp_poll (void) 610static int
611etp_poll (void)
507{ 612{
508 unsigned int maxreqs; 613 unsigned int maxreqs;
509 unsigned int maxtime; 614 unsigned int maxtime;
510 struct timeval tv_start, tv_now; 615 struct timeval tv_start, tv_now;
511 616
… …
541 646
542 X_LOCK (reqlock); 647 X_LOCK (reqlock);
543 --nreqs; 648 --nreqs;
544 X_UNLOCK (reqlock); 649 X_UNLOCK (reqlock);
545 650
546 if (expect_false (req->type == EIO_GROUP && req->size)) 651 if (ecb_expect_false (req->type == EIO_GROUP && req->size))
547 { 652 {
548 req->int1 = 1; /* mark request as delayed */ 653 req->int1 = 1; /* mark request as delayed */
549 continue; 654 continue;
550 } 655 }
551 else 656 else
552 { 657 {
553 int res = ETP_FINISH (req); 658 int res = ETP_FINISH (req);
554 if (expect_false (res)) 659 if (ecb_expect_false (res))
555 return res; 660 return res;
556 } 661 }
557 662
558 if (expect_false (maxreqs && !--maxreqs)) 663 if (ecb_expect_false (maxreqs && !--maxreqs))
559 break; 664 break;
560 665
561 if (maxtime) 666 if (maxtime)
562 { 667 {
563 gettimeofday (&tv_now, 0); 668 gettimeofday (&tv_now, 0);
… …
569 674
570 errno = EAGAIN; 675 errno = EAGAIN;
571 return -1; 676 return -1;
572} 677}
573 678
679static void
574static void etp_cancel (ETP_REQ *req) 680etp_cancel (ETP_REQ *req)
575{ 681{
576 X_LOCK (wrklock); 682 req->cancelled = 1;
577 req->flags |= EIO_FLAG_CANCELLED;
578 X_UNLOCK (wrklock);
579 683
580 eio_grp_cancel (req); 684 eio_grp_cancel (req);
581} 685}
582 686
687static void
583static void etp_submit (ETP_REQ *req) 688etp_submit (ETP_REQ *req)
584{ 689{
585 req->pri -= ETP_PRI_MIN; 690 req->pri -= ETP_PRI_MIN;
586 691
587 if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN; 692 if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
588 if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 693 if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
589 694
590 if (expect_false (req->type == EIO_GROUP)) 695 if (ecb_expect_false (req->type == EIO_GROUP))
591 { 696 {
592 /* I hope this is worth it :/ */ 697 /* I hope this is worth it :/ */
593 X_LOCK (reqlock); 698 X_LOCK (reqlock);
594 ++nreqs; 699 ++nreqs;
595 X_UNLOCK (reqlock); 700 X_UNLOCK (reqlock);
… …
614 719
615 etp_maybe_start_thread (); 720 etp_maybe_start_thread ();
616 } 721 }
617} 722}
618 723
724static void ecb_cold
619static void etp_set_max_poll_time (double nseconds) 725etp_set_max_poll_time (double nseconds)
620{ 726{
621 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 727 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
622 max_poll_time = nseconds * EIO_TICKS; 728 max_poll_time = nseconds * EIO_TICKS;
623 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 729 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
624} 730}
625 731
732static void ecb_cold
626static void etp_set_max_poll_reqs (unsigned int maxreqs) 733etp_set_max_poll_reqs (unsigned int maxreqs)
627{ 734{
628 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 735 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
629 max_poll_reqs = maxreqs; 736 max_poll_reqs = maxreqs;
630 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 737 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
631} 738}
632 739
740static void ecb_cold
633static void etp_set_max_idle (unsigned int nthreads) 741etp_set_max_idle (unsigned int nthreads)
634{ 742{
635 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 743 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
636 max_idle = nthreads; 744 max_idle = nthreads;
637 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 745 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
638} 746}
639 747
748static void ecb_cold
640static void etp_set_idle_timeout (unsigned int seconds) 749etp_set_idle_timeout (unsigned int seconds)
641{ 750{
642 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 751 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
643 idle_timeout = seconds; 752 idle_timeout = seconds;
644 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 753 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
645} 754}
646 755
756static void ecb_cold
647static void etp_set_min_parallel (unsigned int nthreads) 757etp_set_min_parallel (unsigned int nthreads)
648{ 758{
649 if (wanted < nthreads) 759 if (wanted < nthreads)
650 wanted = nthreads; 760 wanted = nthreads;
651} 761}
652 762
763static void ecb_cold
653static void etp_set_max_parallel (unsigned int nthreads) 764etp_set_max_parallel (unsigned int nthreads)
654{ 765{
655 if (wanted > nthreads) 766 if (wanted > nthreads)
656 wanted = nthreads; 767 wanted = nthreads;
657 768
658 while (started > wanted) 769 while (started > wanted)
659 etp_end_thread (); 770 etp_end_thread ();
660} 771}
661 772
662/*****************************************************************************/ 773/*****************************************************************************/
663 774
775static void
664static void grp_try_feed (eio_req *grp) 776grp_try_feed (eio_req *grp)
665{ 777{
666 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 778 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
667 { 779 {
668 grp->flags &= ~EIO_FLAG_GROUPADD; 780 grp->flags &= ~EIO_FLAG_GROUPADD;
669 781
… …
676 break; 788 break;
677 } 789 }
678 } 790 }
679} 791}
680 792
793static int
681static int grp_dec (eio_req *grp) 794grp_dec (eio_req *grp)
682{ 795{
683 --grp->size; 796 --grp->size;
684 797
685 /* call feeder, if applicable */ 798 /* call feeder, if applicable */
686 grp_try_feed (grp); 799 grp_try_feed (grp);
… …
690 return eio_finish (grp); 803 return eio_finish (grp);
691 else 804 else
692 return 0; 805 return 0;
693} 806}
694 807
808static void
695void eio_destroy (eio_req *req) 809eio_destroy (eio_req *req)
696{ 810{
697 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); 811 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
698 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); 812 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
699 813
700 EIO_DESTROY (req); 814 EIO_DESTROY (req);
701} 815}
702 816
817static int
703static int eio_finish (eio_req *req) 818eio_finish (eio_req *req)
704{ 819{
705 int res = EIO_FINISH (req); 820 int res = EIO_FINISH (req);
706 821
707 if (req->grp) 822 if (req->grp)
708 { 823 {
… …
716 if (grp->grp_first == req) 831 if (grp->grp_first == req)
717 grp->grp_first = req->grp_next; 832 grp->grp_first = req->grp_next;
718 833
719 res2 = grp_dec (grp); 834 res2 = grp_dec (grp);
720 835
721 if (!res && res2) 836 if (!res)
722 res = res2; 837 res = res2;
723 } 838 }
724 839
725 eio_destroy (req); 840 eio_destroy (req);
726 841
727 return res; 842 return res;
728} 843}
729 844
845void
730void eio_grp_cancel (eio_req *grp) 846eio_grp_cancel (eio_req *grp)
731{ 847{
732 for (grp = grp->grp_first; grp; grp = grp->grp_next) 848 for (grp = grp->grp_first; grp; grp = grp->grp_next)
733 eio_cancel (grp); 849 eio_cancel (grp);
734} 850}
735 851
852void
736void eio_cancel (eio_req *req) 853eio_cancel (eio_req *req)
737{ 854{
738 etp_cancel (req); 855 etp_cancel (req);
739} 856}
740 857
858void
741void eio_submit (eio_req *req) 859eio_submit (eio_req *req)
742{ 860{
743 etp_submit (req); 861 etp_submit (req);
744} 862}
745 863
746unsigned int eio_nreqs (void) 864unsigned int
865eio_nreqs (void)
747{ 866{
748 return etp_nreqs (); 867 return etp_nreqs ();
749} 868}
750 869
751unsigned int eio_nready (void) 870unsigned int
871eio_nready (void)
752{ 872{
753 return etp_nready (); 873 return etp_nready ();
754} 874}
755 875
756unsigned int eio_npending (void) 876unsigned int
877eio_npending (void)
757{ 878{
758 return etp_npending (); 879 return etp_npending ();
759} 880}
760 881
761unsigned int eio_nthreads (void) 882unsigned int ecb_cold
883eio_nthreads (void)
762{ 884{
763 return etp_nthreads (); 885 return etp_nthreads ();
764} 886}
765 887
888void ecb_cold
766void eio_set_max_poll_time (double nseconds) 889eio_set_max_poll_time (double nseconds)
767{ 890{
768 etp_set_max_poll_time (nseconds); 891 etp_set_max_poll_time (nseconds);
769} 892}
770 893
894void ecb_cold
771void eio_set_max_poll_reqs (unsigned int maxreqs) 895eio_set_max_poll_reqs (unsigned int maxreqs)
772{ 896{
773 etp_set_max_poll_reqs (maxreqs); 897 etp_set_max_poll_reqs (maxreqs);
774} 898}
775 899
900void ecb_cold
776void eio_set_max_idle (unsigned int nthreads) 901eio_set_max_idle (unsigned int nthreads)
777{ 902{
778 etp_set_max_idle (nthreads); 903 etp_set_max_idle (nthreads);
779} 904}
780 905
906void ecb_cold
781void eio_set_idle_timeout (unsigned int seconds) 907eio_set_idle_timeout (unsigned int seconds)
782{ 908{
783 etp_set_idle_timeout (seconds); 909 etp_set_idle_timeout (seconds);
784} 910}
785 911
912void ecb_cold
786void eio_set_min_parallel (unsigned int nthreads) 913eio_set_min_parallel (unsigned int nthreads)
787{ 914{
788 etp_set_min_parallel (nthreads); 915 etp_set_min_parallel (nthreads);
789} 916}
790 917
918void ecb_cold
791void eio_set_max_parallel (unsigned int nthreads) 919eio_set_max_parallel (unsigned int nthreads)
792{ 920{
793 etp_set_max_parallel (nthreads); 921 etp_set_max_parallel (nthreads);
794} 922}
795 923
796int eio_poll (void) 924int eio_poll (void)
… …
805# undef pread 933# undef pread
806# undef pwrite 934# undef pwrite
807# define pread eio__pread 935# define pread eio__pread
808# define pwrite eio__pwrite 936# define pwrite eio__pwrite
809 937
810static ssize_t 938static eio_ssize_t
811eio__pread (int fd, void *buf, size_t count, off_t offset) 939eio__pread (int fd, void *buf, size_t count, off_t offset)
812{ 940{
813 ssize_t res; 941 eio_ssize_t res;
814 off_t ooffset; 942 off_t ooffset;
815 943
816 X_LOCK (preadwritelock); 944 X_LOCK (preadwritelock);
817 ooffset = lseek (fd, 0, SEEK_CUR); 945 ooffset = lseek (fd, 0, SEEK_CUR);
818 lseek (fd, offset, SEEK_SET); 946 lseek (fd, offset, SEEK_SET);
… …
821 X_UNLOCK (preadwritelock); 949 X_UNLOCK (preadwritelock);
822 950
823 return res; 951 return res;
824} 952}
825 953
826static ssize_t 954static eio_ssize_t
827eio__pwrite (int fd, void *buf, size_t count, off_t offset) 955eio__pwrite (int fd, void *buf, size_t count, off_t offset)
828{ 956{
829 ssize_t res; 957 eio_ssize_t res;
830 off_t ooffset; 958 off_t ooffset;
831 959
832 X_LOCK (preadwritelock); 960 X_LOCK (preadwritelock);
833 ooffset = lseek (fd, 0, SEEK_CUR); 961 ooffset = lseek (fd, 0, SEEK_CUR);
834 lseek (fd, offset, SEEK_SET); 962 lseek (fd, offset, SEEK_SET);
… …
866#ifndef HAVE_FUTIMES 994#ifndef HAVE_FUTIMES
867 995
868# undef futimes 996# undef futimes
869# define futimes(fd,times) eio__futimes (fd, times) 997# define futimes(fd,times) eio__futimes (fd, times)
870 998
999static int
871static int eio__futimes (int fd, const struct timeval tv[2]) 1000eio__futimes (int fd, const struct timeval tv[2])
872{ 1001{
873 errno = ENOSYS; 1002 errno = ENOSYS;
874 return -1; 1003 return -1;
875} 1004}
876 1005
… …
879#if !HAVE_FDATASYNC 1008#if !HAVE_FDATASYNC
880# undef fdatasync 1009# undef fdatasync
881# define fdatasync(fd) fsync (fd) 1010# define fdatasync(fd) fsync (fd)
882#endif 1011#endif
883 1012
1013static int
1014eio__syncfs (int fd)
1015{
1016 int res;
1017
1018#if HAVE_SYS_SYNCFS
1019 res = (int)syscall (__NR_syncfs, (int)(fd));
1020#else
1021 res = -1;
1022 errno = ENOSYS;
1023#endif
1024
1025 if (res < 0 && errno == ENOSYS && fd >= 0)
1026 sync ();
1027
1028 return res;
1029}
1030
884/* sync_file_range always needs emulation */ 1031/* sync_file_range always needs emulation */
885int 1032static int
886eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags) 1033eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
887{ 1034{
888#if HAVE_SYNC_FILE_RANGE 1035#if HAVE_SYNC_FILE_RANGE
889 int res; 1036 int res;
890 1037
… …
907 /* even though we could play tricks with the flags, it's better to always 1054 /* even though we could play tricks with the flags, it's better to always
908 * call fdatasync, as that matches the expectation of its users best */ 1055 * call fdatasync, as that matches the expectation of its users best */
909 return fdatasync (fd); 1056 return fdatasync (fd);
910} 1057}
911 1058
1059static int
1060eio__fallocate (int fd, int mode, off_t offset, size_t len)
1061{
1062#if HAVE_FALLOCATE
1063 return fallocate (fd, mode, offset, len);
1064#else
1065 errno = ENOSYS;
1066 return -1;
1067#endif
1068}
1069
912#if !HAVE_READAHEAD 1070#if !HAVE_READAHEAD
913# undef readahead 1071# undef readahead
914# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 1072# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
915 1073
916static ssize_t 1074static eio_ssize_t
917eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) 1075eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
918{ 1076{
919 size_t todo = count; 1077 size_t todo = count;
920 dBUF; 1078 dBUF;
921 1079
… …
926 pread (fd, eio_buf, len, offset); 1084 pread (fd, eio_buf, len, offset);
927 offset += len; 1085 offset += len;
928 todo -= len; 1086 todo -= len;
929 } 1087 }
930 1088
1089 FUBd;
1090
931 errno = 0; 1091 errno = 0;
932 return count; 1092 return count;
933} 1093}
934 1094
935#endif 1095#endif
936 1096
937/* sendfile always needs emulation */ 1097/* sendfile always needs emulation */
938static ssize_t 1098static eio_ssize_t
939eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) 1099eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
940{ 1100{
1101 eio_ssize_t written = 0;
941 ssize_t res; 1102 eio_ssize_t res;
942 1103
943 if (!count) 1104 if (!count)
944 return 0; 1105 return 0;
945 1106
1107 for (;;)
1108 {
1109#ifdef __APPLE__
1110# undef HAVE_SENDFILE /* broken, as everything on os x */
1111#endif
946#if HAVE_SENDFILE 1112#if HAVE_SENDFILE
947# if __linux 1113# if __linux
1114 off_t soffset = offset;
948 res = sendfile (ofd, ifd, &offset, count); 1115 res = sendfile (ofd, ifd, &soffset, count);
949 1116
950# elif __FreeBSD__ 1117# elif __FreeBSD__
951 /* 1118 /*
952 * Of course, the freebsd sendfile is a dire hack with no thoughts 1119 * Of course, the freebsd sendfile is a dire hack with no thoughts
953 * wasted on making it similar to other I/O functions. 1120 * wasted on making it similar to other I/O functions.
954 */ 1121 */
955 {
956 off_t sbytes; 1122 off_t sbytes;
957 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 1123 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
958 1124
959 #if 0 /* according to the manpage, this is correct, but broken behaviour */ 1125 #if 0 /* according to the manpage, this is correct, but broken behaviour */
960 /* freebsd' sendfile will return 0 on success */ 1126 /* freebsd' sendfile will return 0 on success */
961 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */ 1127 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
962 /* not on e.g. EIO or EPIPE - sounds broken */ 1128 /* not on e.g. EIO or EPIPE - sounds broken */
963 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0) 1129 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
964 res = sbytes; 1130 res = sbytes;
965 #endif 1131 #endif
966 1132
967 /* according to source inspection, this is correct, and useful behaviour */ 1133 /* according to source inspection, this is correct, and useful behaviour */
968 if (sbytes) 1134 if (sbytes)
969 res = sbytes; 1135 res = sbytes;
970 }
971 1136
972# elif defined (__APPLE__) 1137# elif defined (__APPLE__)
973
974 {
975 off_t sbytes = count; 1138 off_t sbytes = count;
976 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0); 1139 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
977 1140
978 /* according to the manpage, sbytes is always valid */ 1141 /* according to the manpage, sbytes is always valid */
979 if (sbytes) 1142 if (sbytes)
980 res = sbytes; 1143 res = sbytes;
981 }
982 1144
983# elif __hpux 1145# elif __hpux
984 res = sendfile (ofd, ifd, offset, count, 0, 0); 1146 res = sendfile (ofd, ifd, offset, count, 0, 0);
985 1147
986# elif __solaris 1148# elif __solaris
987 {
988 struct sendfilevec vec; 1149 struct sendfilevec vec;
989 size_t sbytes; 1150 size_t sbytes;
990 1151
991 vec.sfv_fd = ifd; 1152 vec.sfv_fd = ifd;
992 vec.sfv_flag = 0; 1153 vec.sfv_flag = 0;
993 vec.sfv_off = offset; 1154 vec.sfv_off = offset;
994 vec.sfv_len = count; 1155 vec.sfv_len = count;
995 1156
996 res = sendfilev (ofd, &vec, 1, &sbytes); 1157 res = sendfilev (ofd, &vec, 1, &sbytes);
997 1158
998 if (res < 0 && sbytes) 1159 if (res < 0 && sbytes)
999 res = sbytes; 1160 res = sbytes;
1000 }
1001 1161
1002# endif 1162# endif
1003 1163
1004#elif defined (_WIN32) 1164#elif defined (_WIN32) && 0
1005
1006 /* does not work, just for documentation of what would need to be done */ 1165 /* does not work, just for documentation of what would need to be done */
1007 { 1166 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
1167 /* libeio guarantees that the file offset does not change, and windows */
1168 /* has no way to get an independent handle to the same file description */
1008 HANDLE h = TO_SOCKET (ifd); 1169 HANDLE h = TO_SOCKET (ifd);
1009 SetFilePointer (h, offset, 0, FILE_BEGIN); 1170 SetFilePointer (h, offset, 0, FILE_BEGIN);
1010 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0); 1171 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1011 }
1012 1172
1013#else 1173#else
1014 res = -1; 1174 res = -1;
1015 errno = ENOSYS; 1175 errno = ENOSYS;
1016#endif 1176#endif
1017 1177
1178 /* we assume sendfile can copy at least 128mb in one go */
1179 if (res <= 128 * 1024 * 1024)
1180 {
1181 if (res > 0)
1182 written += res;
1183
1184 if (written)
1185 return written;
1186
1187 break;
1188 }
1189 else
1190 {
1191 /* if we requested more, then probably the kernel was lazy */
1192 written += res;
1193 offset += res;
1194 count -= res;
1195
1196 if (!count)
1197 return written;
1198 }
1199 }
1200
1018 if (res < 0 1201 if (res < 0
1019 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1202 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1020 /* BSDs */ 1203 /* BSDs */
1021#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */ 1204#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1022 || errno == ENOTSUP 1205 || errno == ENOTSUP
1023#endif 1206#endif
1207#ifdef EOPNOTSUPP /* windows */
1024 || errno == EOPNOTSUPP /* BSDs */ 1208 || errno == EOPNOTSUPP /* BSDs */
1209#endif
1025#if __solaris 1210#if __solaris
1026 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1211 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
1027#endif 1212#endif
1028 ) 1213 )
1029 ) 1214 )
… …
1033 1218
1034 res = 0; 1219 res = 0;
1035 1220
1036 while (count) 1221 while (count)
1037 { 1222 {
1038 ssize_t cnt; 1223 eio_ssize_t cnt;
1039 1224
1040 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 1225 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
1041 1226
1042 if (cnt <= 0) 1227 if (cnt <= 0)
1043 { 1228 {
… …
1055 1240
1056 offset += cnt; 1241 offset += cnt;
1057 res += cnt; 1242 res += cnt;
1058 count -= cnt; 1243 count -= cnt;
1059 } 1244 }
1245
1246 FUBd;
1060 } 1247 }
1061 1248
1062 return res; 1249 return res;
1250}
1251
1252#ifdef PAGESIZE
1253# define eio_pagesize() PAGESIZE
1254#else
1255static intptr_t
1256eio_pagesize (void)
1257{
1258 static intptr_t page;
1259
1260 if (!page)
1261 page = sysconf (_SC_PAGESIZE);
1262
1263 return page;
1264}
1265#endif
1266
1267static void
1268eio_page_align (void **addr, size_t *length)
1269{
1270 intptr_t mask = eio_pagesize () - 1;
1271
1272 /* round down addr */
1273 intptr_t adj = mask & (intptr_t)*addr;
1274
1275 *addr = (void *)((intptr_t)*addr - adj);
1276 *length += adj;
1277
1278 /* round up length */
1279 *length = (*length + mask) & ~mask;
1280}
1281
1282#if !_POSIX_MEMLOCK
1283# define eio__mlockall(a) EIO_ENOSYS ()
1284#else
1285
1286static int
1287eio__mlockall (int flags)
1288{
1289 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1290 extern int mallopt (int, int);
1291 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1292 #endif
1293
1294 if (EIO_MCL_CURRENT != MCL_CURRENT
1295 || EIO_MCL_FUTURE != MCL_FUTURE)
1296 {
1297 flags = 0
1298 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1299 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1300 }
1301
1302 return mlockall (flags);
1303}
1304#endif
1305
1306#if !_POSIX_MEMLOCK_RANGE
1307# define eio__mlock(a,b) EIO_ENOSYS ()
1308#else
1309
1310static int
1311eio__mlock (void *addr, size_t length)
1312{
1313 eio_page_align (&addr, &length);
1314
1315 return mlock (addr, length);
1316}
1317
1318#endif
1319
1320#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1321# define eio__msync(a,b,c) EIO_ENOSYS ()
1322#else
1323
1324static int
1325eio__msync (void *mem, size_t len, int flags)
1326{
1327 eio_page_align (&mem, &len);
1328
1329 if (EIO_MS_ASYNC != MS_SYNC
1330 || EIO_MS_INVALIDATE != MS_INVALIDATE
1331 || EIO_MS_SYNC != MS_SYNC)
1332 {
1333 flags = 0
1334 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1335 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1336 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1337 }
1338
1339 return msync (mem, len, flags);
1340}
1341
1342#endif
1343
1344static int
1345eio__mtouch (eio_req *req)
1346{
1347 void *mem = req->ptr2;
1348 size_t len = req->size;
1349 int flags = req->int1;
1350
1351 eio_page_align (&mem, &len);
1352
1353 {
1354 intptr_t addr = (intptr_t)mem;
1355 intptr_t end = addr + len;
1356 intptr_t page = eio_pagesize ();
1357
1358 if (addr < end)
1359 if (flags & EIO_MT_MODIFY) /* modify */
1360 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1361 else
1362 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1363 }
1364
1365 return 0;
1366}
1367
1368/*****************************************************************************/
1369/* requests implemented outside eio_execute, because they are so large */
1370
1371/* copies some absolute path to tmpbuf */
1372static char *
1373eio__getwd (struct tmpbuf *tmpbuf, eio_wd wd)
1374{
1375 if (wd == EIO_CWD)
1376 return getcwd (tmpbuf->ptr, PATH_MAX);
1377
1378#if HAVE_AT
1379 abort (); /*TODO*/
1380#else
1381 strcpy (tmpbuf->ptr, wd);
1382#endif
1383 return tmpbuf->ptr;
1384}
1385
1386/* result will always end up in tmpbuf, there is always space for adding a 0-byte */
1387static int
1388eio__realpath (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
1389{
1390 const char *rel = path;
1391 char *res;
1392 char *tmp1, *tmp2;
1393#if SYMLOOP_MAX > 32
1394 int symlinks = SYMLOOP_MAX;
1395#else
1396 int symlinks = 32;
1397#endif
1398
1399 /*D*/ /*TODO: wd ignored */
1400
1401 errno = EINVAL;
1402 if (!rel)
1403 return -1;
1404
1405 errno = ENOENT;
1406 if (!*rel)
1407 return -1;
1408
1409 res = tmpbuf_get (tmpbuf, PATH_MAX * 3);
1410 tmp1 = res + PATH_MAX;
1411 tmp2 = tmp1 + PATH_MAX;
1412
1413#if 0 /* disabled, the musl way to do things is just too racy */
1414#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1415 /* on linux we may be able to ask the kernel */
1416 {
1417 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1418
1419 if (fd >= 0)
1420 {
1421 sprintf (tmp1, "/proc/self/fd/%d", fd);
1422 req->result = readlink (tmp1, res, PATH_MAX);
1423 close (fd);
1424
1425 /* here we should probably stat the open file and the disk file, to make sure they still match */
1426
1427 if (req->result > 0)
1428 goto done;
1429 }
1430 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1431 return;
1432 }
1433#endif
1434#endif
1435
1436 if (*rel != '/')
1437 {
1438 if (!eio__getwd (tmpbuf, wd))
1439 return -1;
1440
1441 if (res [1]) /* only use if not / */
1442 res += strlen (res);
1443 }
1444
1445 while (*rel)
1446 {
1447 eio_ssize_t len, linklen;
1448 const char *beg = rel;
1449
1450 while (*rel && *rel != '/')
1451 ++rel;
1452
1453 len = rel - beg;
1454
1455 if (!len) /* skip slashes */
1456 {
1457 ++rel;
1458 continue;
1459 }
1460
1461 if (beg [0] == '.')
1462 {
1463 if (len == 1)
1464 continue; /* . - nop */
1465
1466 if (beg [1] == '.' && len == 2)
1467 {
1468 /* .. - back up one component, if possible */
1469
1470 while (res != tmpbuf->ptr)
1471 if (*--res == '/')
1472 break;
1473
1474 continue;
1475 }
1476 }
1477
1478 errno = ENAMETOOLONG;
1479 if (res + 1 + len + 1 >= tmp1)
1480 return;
1481
1482 /* copy one component */
1483 *res = '/';
1484 memcpy (res + 1, beg, len);
1485
1486 /* zero-terminate, for readlink */
1487 res [len + 1] = 0;
1488
1489 /* now check if it's a symlink */
1490 linklen = readlink (tmpbuf->ptr, tmp1, PATH_MAX);
1491
1492 if (linklen < 0)
1493 {
1494 if (errno != EINVAL)
1495 return -1;
1496
1497 /* it's a normal directory. hopefully */
1498 res += len + 1;
1499 }
1500 else
1501 {
1502 /* yay, it was a symlink - build new path in tmp2 */
1503 int rellen = strlen (rel);
1504
1505 errno = ENAMETOOLONG;
1506 if (linklen + 1 + rellen >= PATH_MAX)
1507 return -1;
1508
1509 errno = ELOOP;
1510 if (!--symlinks)
1511 return -1;
1512
1513 if (*tmp1 == '/')
1514 res = tmpbuf->ptr; /* symlink resolves to an absolute path */
1515
1516 /* we need to be careful, as rel might point into tmp2 already */
1517 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1518 tmp2 [linklen] = '/';
1519 memcpy (tmp2, tmp1, linklen);
1520
1521 rel = tmp2;
1522 }
1523 }
1524
1525 /* special case for the lone root path */
1526 if (res == tmpbuf->ptr)
1527 *res++ = '/';
1528
1529 return res - (char *)tmpbuf->ptr;
1063} 1530}
1064 1531
1065static signed char 1532static signed char
1066eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) 1533eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1067{ 1534{
… …
1075 1542
1076#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */ 1543#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1077#define EIO_SORT_FAST 60 /* when to only use insertion sort */ 1544#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1078 1545
1079static void 1546static void
1080eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1547eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1081{ 1548{
1082 unsigned char bits [9 + sizeof (ino_t) * 8]; 1549 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1083 unsigned char *bit = bits; 1550 unsigned char *bit = bits;
1084 1551
1085 assert (CHAR_BIT == 8); 1552 assert (CHAR_BIT == 8);
1086 assert (sizeof (eio_dirent) * 8 < 256); 1553 assert (sizeof (eio_dirent) * 8 < 256);
1087 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */ 1554 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
… …
1089 1556
1090 if (size <= EIO_SORT_FAST) 1557 if (size <= EIO_SORT_FAST)
1091 return; 1558 return;
1092 1559
1093 /* first prepare an array of bits to test in our radix sort */ 1560 /* first prepare an array of bits to test in our radix sort */
1094 /* try to take endianness into account, as well as differences in ino_t sizes */ 1561 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1095 /* inode_bits must contain all inodes ORed together */ 1562 /* inode_bits must contain all inodes ORed together */
1096 /* which is used to skip bits that are 0 everywhere, which is very common */ 1563 /* which is used to skip bits that are 0 everywhere, which is very common */
1097 { 1564 {
1098 ino_t endianness; 1565 eio_ino_t endianness;
1099 int i, j; 1566 int i, j;
1100 1567
1101 /* we store the byte offset of byte n into byte n of "endianness" */ 1568 /* we store the byte offset of byte n into byte n of "endianness" */
1102 for (i = 0; i < sizeof (ino_t); ++i) 1569 for (i = 0; i < sizeof (eio_ino_t); ++i)
1103 ((unsigned char *)&endianness)[i] = i; 1570 ((unsigned char *)&endianness)[i] = i;
1104 1571
1105 *bit++ = 0; 1572 *bit++ = 0;
1106 1573
1107 for (i = 0; i < sizeof (ino_t); ++i) 1574 for (i = 0; i < sizeof (eio_ino_t); ++i)
1108 { 1575 {
1109 /* shifting off the byte offsets out of "endianness" */ 1576 /* shifting off the byte offsets out of "endianness" */
1110 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8; 1577 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1111 endianness >>= 8; 1578 endianness >>= 8;
1112 1579
1113 for (j = 0; j < 8; ++j) 1580 for (j = 0; j < 8; ++j)
1114 if (inode_bits & (((ino_t)1) << (i * 8 + j))) 1581 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1115 *bit++ = offs + j; 1582 *bit++ = offs + j;
1116 } 1583 }
1117 1584
1118 for (j = 0; j < 8; ++j) 1585 for (j = 0; j < 8; ++j)
1119 if (score_bits & (1 << j)) 1586 if (score_bits & (1 << j))
1120 *bit++ = offsetof (eio_dirent, score) * 8 + j; 1587 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1121 } 1588 }
1122 1589
1123 /* now actually do the sorting (a variant of MSD radix sort) */ 1590 /* now actually do the sorting (a variant of MSD radix sort) */
1124 { 1591 {
1125 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base; 1592 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1126 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end; 1593 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1127 unsigned char *bit_stk [9 + sizeof (ino_t) * 8]; 1594 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1128 int stk_idx = 0; 1595 int stk_idx = 0;
1129 1596
1130 base_stk [stk_idx] = dents; 1597 base_stk [stk_idx] = dents;
1131 end_stk [stk_idx] = dents + size; 1598 end_stk [stk_idx] = dents + size;
1132 bit_stk [stk_idx] = bit - 1; 1599 bit_stk [stk_idx] = bit - 1;
… …
1211 } 1678 }
1212 } 1679 }
1213} 1680}
1214 1681
1215static void 1682static void
1216eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1683eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1217{ 1684{
1218 if (size <= 1) 1685 if (size <= 1)
1219 return; /* our insertion sort relies on size > 0 */ 1686 return; /* our insertion sort relies on size > 0 */
1220 1687
1221 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */ 1688 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
… …
1227 eio_dent_insertion_sort (dents, size); 1694 eio_dent_insertion_sort (dents, size);
1228} 1695}
1229 1696
1230/* read a full directory */ 1697/* read a full directory */
1231static void 1698static void
1232eio__scandir (eio_req *req, etp_worker *self) 1699eio__scandir (eio_req *req)
1233{ 1700{
1234 DIR *dirp;
1235 EIO_STRUCT_DIRENT *entp;
1236 char *name, *names; 1701 char *name, *names;
1237 int namesalloc = 4096; 1702 int namesalloc = 4096 - sizeof (void *) * 4;
1238 int namesoffs = 0; 1703 int namesoffs = 0;
1239 int flags = req->int1; 1704 int flags = req->int1;
1240 eio_dirent *dents = 0; 1705 eio_dirent *dents = 0;
1241 int dentalloc = 128; 1706 int dentalloc = 128;
1242 int dentoffs = 0; 1707 int dentoffs = 0;
1243 ino_t inode_bits = 0; 1708 eio_ino_t inode_bits = 0;
1709#ifdef _WIN32
1710 HANDLE dirp;
1711 WIN32_FIND_DATA entp;
1712#else
1713 DIR *dirp;
1714 EIO_STRUCT_DIRENT *entp;
1715#endif
1244 1716
1245 req->result = -1; 1717 req->result = -1;
1246 1718
1247 if (!(flags & EIO_READDIR_DENTS)) 1719 if (!(flags & EIO_READDIR_DENTS))
1248 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); 1720 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1249 1721
1250 X_LOCK (wrklock); 1722#ifdef _WIN32
1251 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 1723 {
1724 int len = strlen ((const char *)req->ptr1);
1725 char *path = malloc (MAX_PATH);
1726 const char *fmt;
1727
1728 if (!len)
1729 fmt = "./*";
1730 else if (((const char *)req->ptr1)[len - 1] == '/' || ((const char *)req->ptr1)[len - 1] == '\\')
1731 fmt = "%s*";
1732 else
1733 fmt = "%s/*";
1734
1735 _snprintf (path, MAX_PATH, fmt, (const char *)req->ptr1);
1736 dirp = FindFirstFile (path, &entp);
1737 free (path);
1738
1739 if (dirp == INVALID_HANDLE_VALUE)
1740 {
1741 /* should steal _dosmaperr */
1742 switch (GetLastError ())
1743 {
1744 case ERROR_FILE_NOT_FOUND:
1745 req->result = 0;
1746 break;
1747
1748 case ERROR_INVALID_NAME:
1749 case ERROR_PATH_NOT_FOUND:
1750 case ERROR_NO_MORE_FILES:
1751 errno = ENOENT;
1752 break;
1753
1754 case ERROR_NOT_ENOUGH_MEMORY:
1755 errno = ENOMEM;
1756 break;
1757
1758 default:
1759 errno = EINVAL;
1760 break;
1761 }
1762
1763 return;
1764 }
1765 }
1766#else
1252 self->dirp = dirp = opendir (req->ptr1); 1767 dirp = opendir (req->ptr1);
1768
1769 if (!dirp)
1770 return;
1771#endif
1772
1773 if (req->flags & EIO_FLAG_PTR1_FREE)
1774 free (req->ptr1);
1253 1775
1254 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1776 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1255 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; 1777 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1256 req->ptr2 = names = malloc (namesalloc); 1778 req->ptr2 = names = malloc (namesalloc);
1257 X_UNLOCK (wrklock);
1258 1779
1259 if (dirp && names && (!flags || dents)) 1780 if (!names || (flags && !dents))
1781 return;
1782
1260 for (;;) 1783 for (;;)
1261 { 1784 {
1785 int done;
1786
1787#ifdef _WIN32
1788 done = !dirp;
1789#else
1262 errno = 0; 1790 errno = 0;
1263 entp = readdir (dirp); 1791 entp = readdir (dirp);
1792 done = !entp;
1793#endif
1264 1794
1265 if (!entp) 1795 if (done)
1266 { 1796 {
1797#ifndef _WIN32
1798 int old_errno = errno;
1799 closedir (dirp);
1800 errno = old_errno;
1801
1267 if (errno) 1802 if (errno)
1268 break; 1803 break;
1804#endif
1269 1805
1270 /* sort etc. */ 1806 /* sort etc. */
1271 req->int1 = flags; 1807 req->int1 = flags;
1272 req->result = dentoffs; 1808 req->result = dentoffs;
1273 1809
1274 if (flags & EIO_READDIR_STAT_ORDER) 1810 if (flags & EIO_READDIR_STAT_ORDER)
1275 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits); 1811 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1276 else if (flags & EIO_READDIR_DIRS_FIRST) 1812 else if (flags & EIO_READDIR_DIRS_FIRST)
1277 if (flags & EIO_READDIR_FOUND_UNKNOWN) 1813 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1278 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */ 1814 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1279 else 1815 else
1816 {
1817 /* in this case, all is known, and we just put dirs first and sort them */
1818 eio_dirent *oth = dents + dentoffs;
1819 eio_dirent *dir = dents;
1820
1821 /* now partition dirs to the front, and non-dirs to the back */
1822 /* by walking from both sides and swapping if necessary */
1823 while (oth > dir)
1824 {
1825 if (dir->type == EIO_DT_DIR)
1826 ++dir;
1827 else if ((--oth)->type == EIO_DT_DIR)
1828 {
1829 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1830
1831 ++dir;
1832 }
1833 }
1834
1835 /* now sort the dirs only (dirs all have the same score) */
1836 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1837 }
1838
1839 break;
1840 }
1841
1842 /* now add the entry to our list(s) */
1843 name = D_NAME (entp);
1844
1845 /* skip . and .. entries */
1846 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1847 {
1848 int len = D_NAMLEN (entp) + 1;
1849
1850 while (ecb_expect_false (namesoffs + len > namesalloc))
1851 {
1852 namesalloc *= 2;
1853 req->ptr2 = names = realloc (names, namesalloc);
1854
1855 if (!names)
1856 break;
1857 }
1858
1859 memcpy (names + namesoffs, name, len);
1860
1861 if (dents)
1862 {
1863 struct eio_dirent *ent;
1864
1865 if (ecb_expect_false (dentoffs == dentalloc))
1280 { 1866 {
1281 /* in this case, all is known, and we just put dirs first and sort them */ 1867 dentalloc *= 2;
1868 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1869
1870 if (!dents)
1871 break;
1872 }
1873
1282 eio_dirent *oth = dents + dentoffs; 1874 ent = dents + dentoffs;
1283 eio_dirent *dir = dents;
1284 1875
1285 /* now partition dirs to the front, and non-dirs to the back */ 1876 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1286 /* by walking from both sides and swapping if necessary */ 1877 ent->namelen = len - 1;
1287 while (oth > dir) 1878 ent->inode = D_INO (entp);
1879
1880 inode_bits |= ent->inode;
1881
1882 switch (D_TYPE (entp))
1883 {
1884 default:
1885 ent->type = EIO_DT_UNKNOWN;
1886 flags |= EIO_READDIR_FOUND_UNKNOWN;
1887 break;
1888
1889 #ifdef DT_FIFO
1890 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1891 #endif
1892 #ifdef DT_CHR
1893 case DT_CHR: ent->type = EIO_DT_CHR; break;
1894 #endif
1895 #ifdef DT_MPC
1896 case DT_MPC: ent->type = EIO_DT_MPC; break;
1897 #endif
1898 #ifdef DT_DIR
1899 case DT_DIR: ent->type = EIO_DT_DIR; break;
1900 #endif
1901 #ifdef DT_NAM
1902 case DT_NAM: ent->type = EIO_DT_NAM; break;
1903 #endif
1904 #ifdef DT_BLK
1905 case DT_BLK: ent->type = EIO_DT_BLK; break;
1906 #endif
1907 #ifdef DT_MPB
1908 case DT_MPB: ent->type = EIO_DT_MPB; break;
1909 #endif
1910 #ifdef DT_REG
1911 case DT_REG: ent->type = EIO_DT_REG; break;
1912 #endif
1913 #ifdef DT_NWK
1914 case DT_NWK: ent->type = EIO_DT_NWK; break;
1915 #endif
1916 #ifdef DT_CMP
1917 case DT_CMP: ent->type = EIO_DT_CMP; break;
1918 #endif
1919 #ifdef DT_LNK
1920 case DT_LNK: ent->type = EIO_DT_LNK; break;
1921 #endif
1922 #ifdef DT_SOCK
1923 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1924 #endif
1925 #ifdef DT_DOOR
1926 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1927 #endif
1928 #ifdef DT_WHT
1929 case DT_WHT: ent->type = EIO_DT_WHT; break;
1930 #endif
1931 }
1932
1933 ent->score = 7;
1934
1935 if (flags & EIO_READDIR_DIRS_FIRST)
1936 {
1937 if (ent->type == EIO_DT_UNKNOWN)
1288 { 1938 {
1289 if (dir->type == EIO_DT_DIR) 1939 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1290 ++dir; 1940 ent->score = 1;
1291 else if ((--oth)->type == EIO_DT_DIR) 1941 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1292 { 1942 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1293 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1294
1295 ++dir;
1296 }
1297 } 1943 }
1298 1944 else if (ent->type == EIO_DT_DIR)
1299 /* now sort the dirs only (dirs all have the same score) */ 1945 ent->score = 0;
1300 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1301 } 1946 }
1302
1303 break;
1304 }
1305
1306 /* now add the entry to our list(s) */
1307 name = entp->d_name;
1308
1309 /* skip . and .. entries */
1310 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1311 {
1312 int len = D_NAMLEN (entp) + 1;
1313
1314 while (expect_false (namesoffs + len > namesalloc))
1315 {
1316 namesalloc *= 2;
1317 X_LOCK (wrklock);
1318 req->ptr2 = names = realloc (names, namesalloc);
1319 X_UNLOCK (wrklock);
1320
1321 if (!names)
1322 break;
1323 } 1947 }
1324 1948
1325 memcpy (names + namesoffs, name, len);
1326
1327 if (dents)
1328 {
1329 struct eio_dirent *ent;
1330
1331 if (expect_false (dentoffs == dentalloc))
1332 {
1333 dentalloc *= 2;
1334 X_LOCK (wrklock);
1335 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1336 X_UNLOCK (wrklock);
1337
1338 if (!dents)
1339 break;
1340 }
1341
1342 ent = dents + dentoffs;
1343
1344 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1345 ent->namelen = len - 1;
1346 ent->inode = D_INO (entp);
1347
1348 inode_bits |= ent->inode;
1349
1350 switch (D_TYPE (entp))
1351 {
1352 default:
1353 ent->type = EIO_DT_UNKNOWN;
1354 flags |= EIO_READDIR_FOUND_UNKNOWN;
1355 break;
1356
1357 #ifdef DT_FIFO
1358 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1359 #endif
1360 #ifdef DT_CHR
1361 case DT_CHR: ent->type = EIO_DT_CHR; break;
1362 #endif
1363 #ifdef DT_MPC
1364 case DT_MPC: ent->type = EIO_DT_MPC; break;
1365 #endif
1366 #ifdef DT_DIR
1367 case DT_DIR: ent->type = EIO_DT_DIR; break;
1368 #endif
1369 #ifdef DT_NAM
1370 case DT_NAM: ent->type = EIO_DT_NAM; break;
1371 #endif
1372 #ifdef DT_BLK
1373 case DT_BLK: ent->type = EIO_DT_BLK; break;
1374 #endif
1375 #ifdef DT_MPB
1376 case DT_MPB: ent->type = EIO_DT_MPB; break;
1377 #endif
1378 #ifdef DT_REG
1379 case DT_REG: ent->type = EIO_DT_REG; break;
1380 #endif
1381 #ifdef DT_NWK
1382 case DT_NWK: ent->type = EIO_DT_NWK; break;
1383 #endif
1384 #ifdef DT_CMP
1385 case DT_CMP: ent->type = EIO_DT_CMP; break;
1386 #endif
1387 #ifdef DT_LNK
1388 case DT_LNK: ent->type = EIO_DT_LNK; break;
1389 #endif
1390 #ifdef DT_SOCK
1391 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1392 #endif
1393 #ifdef DT_DOOR
1394 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1395 #endif
1396 #ifdef DT_WHT
1397 case DT_WHT: ent->type = EIO_DT_WHT; break;
1398 #endif
1399 }
1400
1401 ent->score = 7;
1402
1403 if (flags & EIO_READDIR_DIRS_FIRST)
1404 {
1405 if (ent->type == EIO_DT_UNKNOWN)
1406 {
1407 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1408 ent->score = 1;
1409 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1410 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1411 }
1412 else if (ent->type == EIO_DT_DIR)
1413 ent->score = 0;
1414 }
1415 }
1416
1417 namesoffs += len; 1949 namesoffs += len;
1418 ++dentoffs; 1950 ++dentoffs;
1419 } 1951 }
1420 1952
1421 if (EIO_CANCELLED (req)) 1953 if (EIO_CANCELLED (req))
1422 { 1954 {
1423 errno = ECANCELED; 1955 errno = ECANCELED;
1424 break; 1956 break;
1425 } 1957 }
1958
1959#ifdef _WIN32
1960 if (!FindNextFile (dirp, &entp))
1961 {
1962 FindClose (dirp);
1963 dirp = 0;
1426 } 1964 }
1965#endif
1966 }
1427} 1967}
1428 1968
1429#ifdef PAGESIZE 1969/*****************************************************************************/
1430# define eio_pagesize() PAGESIZE 1970/* working directory stuff */
1971
1972#if HAVE_AT
1973
1974#define WD2FD(wd) ((wd) ? ((int)(long)(wd)) - 1 : AT_FDCWD)
1975
1976#ifndef O_SEARCH
1977# define O_SEARCH O_RDONLY
1978#endif
1979
1980eio_wd
1981eio_wd_open_sync (eio_wd wd, const char *path)
1982{
1983 int fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1984
1985 return fd >= 0 ? (eio_wd)(long)(fd + 1) : EIO_INVALID_WD;
1986}
1987
1988static eio_wd
1989eio__wd_open_sync (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
1990{
1991 return eio_wd_open_sync (wd, path);
1992}
1993
1994void
1995eio_wd_close_sync (eio_wd wd)
1996{
1997 int fd = WD2FD (wd);
1998
1999 if (fd >= 0)
2000 close (fd);
2001}
2002
2003static int
2004eio__truncateat (int dirfd, const char *path, off_t length)
2005{
2006 int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
2007 int res;
2008
2009 if (fd < 0)
2010 return fd;
2011
2012 res = ftruncate (fd, length);
2013 close (fd);
2014 return res;
2015}
2016
2017static int
2018eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
2019{
2020 int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
2021 int res;
2022
2023 if (fd < 0)
2024 return fd;
2025
2026 res = fstatvfs (fd, buf);
2027 close (fd);
2028 return res;
2029
2030}
2031
1431#else 2032#else
1432static intptr_t
1433eio_pagesize (void)
1434{
1435 static intptr_t page;
1436 2033
1437 if (!page) 2034/* on legacy systems, we represent the working directories simply by their path strings */
1438 page = sysconf (_SC_PAGESIZE);
1439 2035
2036static const char *
2037wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
2038{
2039 if (!wd || *path == '/')
1440 return page; 2040 return path;
1441}
1442#endif
1443 2041
2042 {
2043 int l1 = strlen ((const char *)wd);
2044 int l2 = strlen (path);
2045
2046 char *res = tmpbuf_get (tmpbuf, l1 + l2 + 2);
2047
2048 memcpy (res, wd, l1);
2049 res [l1] = '/';
2050 memcpy (res + l1 + 1, path, l2 + 1);
2051
2052 return res;
2053 }
2054}
2055
1444static void 2056static eio_wd
1445eio_page_align (void **addr, size_t *length) 2057eio__wd_open_sync (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
1446{ 2058{
1447 intptr_t mask = eio_pagesize () - 1; 2059 if (*path == '/') /* absolute paths ignore wd */
1448 2060 path = strdup (path);
1449 /* round down addr */ 2061 else if (path [0] == '.' && !path [1]) /* special case '.', as it is common */
1450 intptr_t adj = mask & (intptr_t)*addr; 2062 return wd;
1451 2063 else
1452 *addr = (void *)((intptr_t)*addr - adj);
1453 *length += adj;
1454
1455 /* round up length */
1456 *length = (*length + mask) & ~mask;
1457}
1458
1459#if !_POSIX_MEMLOCK
1460# define eio__mlockall(a) ((errno = ENOSYS), -1)
1461#else
1462
1463static int
1464eio__mlockall (int flags)
1465{
1466 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1467 extern int mallopt (int, int);
1468 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1469 #endif
1470
1471 if (EIO_MCL_CURRENT != MCL_CURRENT
1472 || EIO_MCL_FUTURE != MCL_FUTURE)
1473 { 2064 {
1474 flags = 0 2065 int len = eio__realpath (tmpbuf, wd, path);
1475 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1476 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1477 }
1478 2066
1479 return mlockall (flags); 2067 path = EIO_INVALID_WD;
1480}
1481#endif
1482 2068
1483#if !_POSIX_MEMLOCK_RANGE 2069 if (len >= 0)
1484# define eio__mlock(a,b) ((errno = ENOSYS), -1) 2070 {
1485#else 2071 ((char *)tmpbuf->ptr)[len] = 0;
1486 2072 path = strdup (tmpbuf->ptr);
1487static int 2073 }
1488eio__mlock (void *addr, size_t length)
1489{
1490 eio_page_align (&addr, &length);
1491
1492 return mlock (addr, length);
1493}
1494
1495#endif
1496
1497#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1498# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
1499#else
1500
1501int
1502eio__msync (void *mem, size_t len, int flags)
1503{
1504 eio_page_align (&mem, &len);
1505
1506 if (EIO_MS_ASYNC != MS_SYNC
1507 || EIO_MS_INVALIDATE != MS_INVALIDATE
1508 || EIO_MS_SYNC != MS_SYNC)
1509 { 2074 }
1510 flags = 0
1511 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1512 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1513 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1514 }
1515 2075
1516 return msync (mem, len, flags); 2076 if (!path)
1517} 2077 path = EIO_INVALID_WD;
1518 2078
1519#endif 2079 return (eio_wd)path;
2080}
1520 2081
1521int 2082eio_wd
1522eio__mtouch (void *mem, size_t len, int flags) 2083eio_wd_open_sync (eio_wd wd, const char *path)
1523{ 2084{
1524 eio_page_align (&mem, &len); 2085 struct tmpbuf tmpbuf = { 0 };
2086 wd = eio__wd_open_sync (&tmpbuf, wd, path);
2087 free (tmpbuf.ptr);
1525 2088
1526 {
1527 intptr_t addr = (intptr_t)mem;
1528 intptr_t end = addr + len;
1529 intptr_t page = eio_pagesize ();
1530
1531 if (addr < end)
1532 if (flags & EIO_MT_MODIFY) /* modify */
1533 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
1534 else
1535 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
1536 }
1537
1538 return 0; 2089 return wd;
1539} 2090}
2091
2092void
2093eio_wd_close_sync (eio_wd wd)
2094{
2095 if (wd != EIO_INVALID_WD)
2096 free (wd);
2097}
2098
2099#endif
1540 2100
1541/*****************************************************************************/ 2101/*****************************************************************************/
1542 2102
1543#define ALLOC(len) \ 2103#define ALLOC(len) \
1544 if (!req->ptr2) \ 2104 if (!req->ptr2) \
… …
1559{ 2119{
1560 ETP_REQ *req; 2120 ETP_REQ *req;
1561 struct timespec ts; 2121 struct timespec ts;
1562 etp_worker *self = (etp_worker *)thr_arg; 2122 etp_worker *self = (etp_worker *)thr_arg;
1563 2123
2124#if HAVE_PRCTL_SET_NAME
2125 prctl (PR_SET_NAME, (unsigned long)"eio_thread", 0, 0, 0);
2126#endif
2127
1564 /* try to distribute timeouts somewhat randomly */ 2128 /* try to distribute timeouts somewhat evenly */
1565 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 2129 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1566 2130
1567 for (;;) 2131 for (;;)
1568 { 2132 {
2133 ts.tv_sec = 0;
2134
1569 X_LOCK (reqlock); 2135 X_LOCK (reqlock);
1570 2136
1571 for (;;) 2137 for (;;)
1572 { 2138 {
1573 self->req = req = reqq_shift (&req_queue); 2139 req = reqq_shift (&req_queue);
1574 2140
1575 if (req) 2141 if (req)
1576 break; 2142 break;
1577 2143
2144 if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
2145 {
2146 X_UNLOCK (reqlock);
2147 X_LOCK (wrklock);
2148 --started;
2149 X_UNLOCK (wrklock);
2150 goto quit;
2151 }
2152
1578 ++idle; 2153 ++idle;
1579 2154
1580 ts.tv_sec = time (0) + idle_timeout; 2155 if (idle <= max_idle)
1581 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 2156 /* we are allowed to idle, so do so without any timeout */
2157 X_COND_WAIT (reqwait, reqlock);
2158 else
1582 { 2159 {
1583 if (idle > max_idle) 2160 /* initialise timeout once */
1584 { 2161 if (!ts.tv_sec)
1585 --idle; 2162 ts.tv_sec = time (0) + idle_timeout;
1586 X_UNLOCK (reqlock);
1587 X_LOCK (wrklock);
1588 --started;
1589 X_UNLOCK (wrklock);
1590 goto quit;
1591 }
1592 2163
1593 /* we are allowed to idle, so do so without any timeout */
1594 X_COND_WAIT (reqwait, reqlock); 2164 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
2165 ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1595 } 2166 }
1596 2167
1597 --idle; 2168 --idle;
1598 } 2169 }
1599 2170
… …
1602 X_UNLOCK (reqlock); 2173 X_UNLOCK (reqlock);
1603 2174
1604 if (req->type < 0) 2175 if (req->type < 0)
1605 goto quit; 2176 goto quit;
1606 2177
1607 if (!EIO_CANCELLED (req))
1608 ETP_EXECUTE (self, req); 2178 ETP_EXECUTE (self, req);
1609 2179
1610 X_LOCK (reslock); 2180 X_LOCK (reslock);
1611 2181
1612 ++npending; 2182 ++npending;
1613 2183
1614 if (!reqq_push (&res_queue, req) && want_poll_cb) 2184 if (!reqq_push (&res_queue, req) && want_poll_cb)
1615 want_poll_cb (); 2185 want_poll_cb ();
1616 2186
1617 self->req = 0;
1618 etp_worker_clear (self); 2187 etp_worker_clear (self);
1619 2188
1620 X_UNLOCK (reslock); 2189 X_UNLOCK (reslock);
1621 } 2190 }
1622 2191
1623quit: 2192quit:
2193 free (req);
2194
1624 X_LOCK (wrklock); 2195 X_LOCK (wrklock);
1625 etp_worker_free (self); 2196 etp_worker_free (self);
1626 X_UNLOCK (wrklock); 2197 X_UNLOCK (wrklock);
1627 2198
1628 return 0; 2199 return 0;
1629} 2200}
1630 2201
1631/*****************************************************************************/ 2202/*****************************************************************************/
1632 2203
2204int ecb_cold
1633int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 2205eio_init (void (*want_poll)(void), void (*done_poll)(void))
1634{ 2206{
2207#if !HAVE_PREADWRITE
2208 X_MUTEX_CREATE (preadwritelock);
2209#endif
2210
1635 return etp_init (want_poll, done_poll); 2211 return etp_init (want_poll, done_poll);
1636} 2212}
1637 2213
2214ecb_inline void
1638static void eio_api_destroy (eio_req *req) 2215eio_api_destroy (eio_req *req)
1639{ 2216{
1640 free (req); 2217 free (req);
1641} 2218}
1642 2219
1643#define REQ(rtype) \ 2220#define REQ(rtype) \
… …
1662 { \ 2239 { \
1663 eio_api_destroy (req); \ 2240 eio_api_destroy (req); \
1664 return 0; \ 2241 return 0; \
1665 } 2242 }
1666 2243
2244static void
1667static void eio_execute (etp_worker *self, eio_req *req) 2245eio_execute (etp_worker *self, eio_req *req)
1668{ 2246{
2247#if HAVE_AT
2248 int dirfd;
2249#else
2250 const char *path;
2251#endif
2252
2253 if (ecb_expect_false (EIO_CANCELLED (req)))
2254 {
2255 req->result = -1;
2256 req->errorno = ECANCELED;
2257 return;
2258 }
2259
2260 if (ecb_expect_false (req->wd == EIO_INVALID_WD))
2261 {
2262 req->result = -1;
2263 req->errorno = ENOENT;
2264 return;
2265 }
2266
2267 if (req->type >= EIO_OPEN)
2268 {
2269 #if HAVE_AT
2270 dirfd = WD2FD (req->wd);
2271 #else
2272 path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
2273 #endif
2274 }
2275
1669 switch (req->type) 2276 switch (req->type)
1670 { 2277 {
2278 case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
2279 req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
2280 break;
2281 case EIO_WD_CLOSE: req->result = 0;
2282 eio_wd_close_sync (req->wd); break;
2283
1671 case EIO_READ: ALLOC (req->size); 2284 case EIO_READ: ALLOC (req->size);
1672 req->result = req->offs >= 0 2285 req->result = req->offs >= 0
1673 ? pread (req->int1, req->ptr2, req->size, req->offs) 2286 ? pread (req->int1, req->ptr2, req->size, req->offs)
1674 : read (req->int1, req->ptr2, req->size); break; 2287 : read (req->int1, req->ptr2, req->size); break;
1675 case EIO_WRITE: req->result = req->offs >= 0 2288 case EIO_WRITE: req->result = req->offs >= 0
1676 ? pwrite (req->int1, req->ptr2, req->size, req->offs) 2289 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1677 : write (req->int1, req->ptr2, req->size); break; 2290 : write (req->int1, req->ptr2, req->size); break;
1678 2291
1679 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break; 2292 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1680 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break; 2293 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
2294
2295#if HAVE_AT
1681 2296
1682 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2297 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1683 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2298 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break;
1684 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2299 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2300 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, AT_SYMLINK_NOFOLLOW); break;
2301 case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break;
2302 case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break;
2303 case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break;
2304 case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break;
2305
2306 case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
2307 case EIO_RMDIR: req->result = unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break;
2308 case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); break;
2309 case EIO_RENAME: req->result = renameat (dirfd, req->ptr1, WD2FD (req->int3), req->ptr2); break;
2310 case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD (req->int3), req->ptr2, 0); break;
2311 case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
2312 case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
2313 case EIO_READLINK: ALLOC (PATH_MAX);
2314 req->result = readlinkat (dirfd, req->ptr1, req->ptr2, PATH_MAX); break;
2315 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2316 req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2317 case EIO_UTIME:
2318 case EIO_FUTIME:
2319 {
2320 struct timespec ts[2];
2321 struct timespec *times;
2322
2323 if (req->nv1 != -1. || req->nv2 != -1.)
2324 {
2325 ts[0].tv_sec = req->nv1;
2326 ts[0].tv_nsec = (req->nv1 - ts[0].tv_sec) * 1e9;
2327 ts[1].tv_sec = req->nv2;
2328 ts[1].tv_nsec = (req->nv2 - ts[1].tv_sec) * 1e9;
2329
2330 times = ts;
2331 }
2332 else
2333 times = 0;
2334
2335 req->result = req->type == EIO_FUTIME
2336 ? futimens (req->int1, times)
2337 : utimensat (dirfd, req->ptr1, times, 0);
2338 }
2339 break;
2340
2341#else
2342
2343 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2344 req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
2345 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1685 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2346 req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
2347 case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
2348 case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
2349 case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
2350 case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
2351
2352 case EIO_UNLINK: req->result = unlink (path ); break;
2353 case EIO_RMDIR: req->result = rmdir (path ); break;
2354 case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
2355 case EIO_RENAME: req->result = rename (path , req->ptr2); break;
2356 case EIO_LINK: req->result = link (path , req->ptr2); break;
2357 case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
2358 case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
2359 case EIO_READLINK: ALLOC (PATH_MAX);
2360 req->result = readlink (path, req->ptr2, PATH_MAX); break;
2361 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2362 req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
2363
2364 case EIO_UTIME:
2365 case EIO_FUTIME:
2366 {
2367 struct timeval tv[2];
2368 struct timeval *times;
2369
2370 if (req->nv1 != -1. || req->nv2 != -1.)
2371 {
2372 tv[0].tv_sec = req->nv1;
2373 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
2374 tv[1].tv_sec = req->nv2;
2375 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
2376
2377 times = tv;
2378 }
2379 else
2380 times = 0;
2381
2382 req->result = req->type == EIO_FUTIME
2383 ? futimes (req->int1, times)
2384 : utimes (req->ptr1, times);
2385 }
2386 break;
2387
2388#endif
2389
2390 case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
2391 {
2392 ALLOC (req->result);
2393 memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
2394 }
2395 break;
2396
1686 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2397 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1687 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; 2398 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1688 2399
1689 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1690 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1691 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS)); 2400 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1692 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break; 2401 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1693 2402
1694 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1695 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; 2403 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1696 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1697 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break; 2404 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1698 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1699 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break; 2405 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1700 2406
1701 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
1702 case EIO_CLOSE: req->result = close (req->int1); break; 2407 case EIO_CLOSE: req->result = close (req->int1); break;
1703 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break; 2408 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1704 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
1705 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1706 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1707 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1708 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1709 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1710 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1711
1712 case EIO_READLINK: ALLOC (PATH_MAX);
1713 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1714
1715 case EIO_SYNC: req->result = 0; sync (); break; 2409 case EIO_SYNC: req->result = 0; sync (); break;
1716 case EIO_FSYNC: req->result = fsync (req->int1); break; 2410 case EIO_FSYNC: req->result = fsync (req->int1); break;
1717 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 2411 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
2412 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
2413 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1718 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break; 2414 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1719 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break; 2415 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1720 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break; 2416 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1721 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break; 2417 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1722 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break; 2418 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
1723 2419
1724 case EIO_READDIR: eio__scandir (req, self); break; 2420 case EIO_READDIR: eio__scandir (req); break;
1725 2421
1726 case EIO_BUSY: 2422 case EIO_BUSY:
1727#ifdef _WIN32 2423#ifdef _WIN32
1728 Sleep (req->nv1 * 1e3); 2424 Sleep (req->nv1 * 1e3);
1729#else 2425#else
… …
1736 req->result = select (0, 0, 0, 0, &tv); 2432 req->result = select (0, 0, 0, 0, &tv);
1737 } 2433 }
1738#endif 2434#endif
1739 break; 2435 break;
1740 2436
1741 case EIO_UTIME:
1742 case EIO_FUTIME:
1743 {
1744 struct timeval tv[2];
1745 struct timeval *times;
1746
1747 if (req->nv1 != -1. || req->nv2 != -1.)
1748 {
1749 tv[0].tv_sec = req->nv1;
1750 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
1751 tv[1].tv_sec = req->nv2;
1752 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
1753
1754 times = tv;
1755 }
1756 else
1757 times = 0;
1758
1759 req->result = req->type == EIO_FUTIME
1760 ? futimes (req->int1, times)
1761 : utimes (req->ptr1, times);
1762 }
1763 break;
1764
1765 case EIO_GROUP: 2437 case EIO_GROUP:
1766 abort (); /* handled in eio_request */ 2438 abort (); /* handled in eio_request */
1767 2439
1768 case EIO_NOP: 2440 case EIO_NOP:
1769 req->result = 0; 2441 req->result = 0;
1770 break; 2442 break;
1771 2443
1772 case EIO_CUSTOM: 2444 case EIO_CUSTOM:
1773 ((void (*)(eio_req *))req->feed) (req); 2445 req->feed (req);
1774 break; 2446 break;
1775 2447
1776 default: 2448 default:
1777 errno = ENOSYS; 2449 errno = ENOSYS;
1778 req->result = -1; 2450 req->result = -1;
… …
1807eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2479eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1808{ 2480{
1809 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2481 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1810} 2482}
1811 2483
2484eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2485{
2486 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2487}
2488
2489eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
2490{
2491 REQ (EIO_SYNCFS); req->int1 = fd; SEND;
2492}
2493
2494eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2495{
2496 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2497}
2498
1812eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2499eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1813{ 2500{
1814 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2501 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1815} 2502}
1816 2503
… …
1822eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data) 2509eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
1823{ 2510{
1824 REQ (EIO_MLOCKALL); req->int1 = flags; SEND; 2511 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
1825} 2512}
1826 2513
1827eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) 2514eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
1828{ 2515{
1829 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; 2516 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
1830}
1831
1832eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1833{
1834 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1835} 2517}
1836 2518
1837eio_req *eio_close (int fd, int pri, eio_cb cb, void *data) 2519eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
1838{ 2520{
1839 REQ (EIO_CLOSE); req->int1 = fd; SEND; 2521 REQ (EIO_CLOSE); req->int1 = fd; SEND;
… …
1877eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) 2559eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1878{ 2560{
1879 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2561 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1880} 2562}
1881 2563
1882eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2564eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1883{ 2565{
1884 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2566 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1885} 2567}
1886 2568
1887eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) 2569eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
… …
1907eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data) 2589eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1908{ 2590{
1909 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2591 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1910} 2592}
1911 2593
1912eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2594eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1913{ 2595{
1914 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2596 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1915} 2597}
1916 2598
1917eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) 2599eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
… …
1931} 2613}
1932 2614
1933eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data) 2615eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1934{ 2616{
1935 return eio__1path (EIO_READLINK, path, pri, cb, data); 2617 return eio__1path (EIO_READLINK, path, pri, cb, data);
2618}
2619
2620eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2621{
2622 return eio__1path (EIO_REALPATH, path, pri, cb, data);
1936} 2623}
1937 2624
1938eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) 2625eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1939{ 2626{
1940 return eio__1path (EIO_STAT, path, pri, cb, data); 2627 return eio__1path (EIO_STAT, path, pri, cb, data);
… …
1999eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2686eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2000{ 2687{
2001 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2688 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
2002} 2689}
2003 2690
2004eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) 2691eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
2005{ 2692{
2006 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND; 2693 REQ (EIO_CUSTOM); req->feed = execute; SEND;
2007} 2694}
2008 2695
2009#endif 2696#endif
2010 2697
2011eio_req *eio_grp (eio_cb cb, void *data) 2698eio_req *eio_grp (eio_cb cb, void *data)
… …
2020#undef SEND 2707#undef SEND
2021 2708
2022/*****************************************************************************/ 2709/*****************************************************************************/
2023/* grp functions */ 2710/* grp functions */
2024 2711
2712void
2025void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2713eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
2026{ 2714{
2027 grp->int2 = limit; 2715 grp->int2 = limit;
2028 grp->feed = feed; 2716 grp->feed = feed;
2029 2717
2030 grp_try_feed (grp); 2718 grp_try_feed (grp);
2031} 2719}
2032 2720
2721void
2033void eio_grp_limit (eio_req *grp, int limit) 2722eio_grp_limit (eio_req *grp, int limit)
2034{ 2723{
2035 grp->int2 = limit; 2724 grp->int2 = limit;
2036 2725
2037 grp_try_feed (grp); 2726 grp_try_feed (grp);
2038} 2727}
2039 2728
2729void
2040void eio_grp_add (eio_req *grp, eio_req *req) 2730eio_grp_add (eio_req *grp, eio_req *req)
2041{ 2731{
2042 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2732 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
2043 2733
2044 grp->flags |= EIO_FLAG_GROUPADD; 2734 grp->flags |= EIO_FLAG_GROUPADD;
2045 2735
… …
2056} 2746}
2057 2747
2058/*****************************************************************************/ 2748/*****************************************************************************/
2059/* misc garbage */ 2749/* misc garbage */
2060 2750
2751eio_ssize_t
2061ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2752eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2062{ 2753{
2063 etp_worker wrk;
2064 ssize_t ret;
2065
2066 wrk.dbuf = 0;
2067
2068 ret = eio__sendfile (ofd, ifd, offset, count, &wrk); 2754 return eio__sendfile (ofd, ifd, offset, count);
2069
2070 if (wrk.dbuf)
2071 free (wrk.dbuf);
2072
2073 return ret;
2074} 2755}
2075 2756

Diff Legend

– Removed lines
+ Added lines
< Changed lines
> Changed lines