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

Comparing cvsroot/libeio/eio.c (file contents):
Revision 1.66 by root, Sun Jun 5 19:58:37 2011 UTC vs.
Revision 1.113 by root, Sun Oct 9 22:25:00 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/* 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)
169 296
170/*****************************************************************************/ 297/*****************************************************************************/
171 298
172#if __GNUC__ >= 3 299struct tmpbuf
173# define expect(expr,value) __builtin_expect ((expr),(value)) 300{
301 void *ptr;
302 int len;
303};
304
305static void *
306tmpbuf_get (struct tmpbuf *buf, int len)
307{
308 if (buf->len < len)
309 {
310 free (buf->ptr);
311 buf->ptr = malloc (buf->len = len);
312 }
313
314 return buf->ptr;
315}
316
317struct tmpbuf;
318
319#if _POSIX_VERSION >= 200809L
320 #define HAVE_AT 1
321 #define WD2FD(wd) ((wd) ? (wd)->fd : AT_FDCWD)
322 #ifndef O_SEARCH
323 #define O_SEARCH O_RDONLY
324 #endif
174#else 325#else
175# define expect(expr,value) (expr) 326 #define HAVE_AT 0
327 static const char *wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path);
176#endif 328#endif
177 329
178#define expect_false(expr) expect ((expr) != 0, 0) 330struct eio_pwd
179#define expect_true(expr) expect ((expr) != 0, 1) 331{
332#if HAVE_AT
333 int fd;
334#endif
335 int len;
336 char str[1]; /* actually, a 0-terminated canonical path */
337};
180 338
181/*****************************************************************************/ 339/*****************************************************************************/
182 340
183#define ETP_PRI_MIN EIO_PRI_MIN 341#define ETP_PRI_MIN EIO_PRI_MIN
184#define ETP_PRI_MAX EIO_PRI_MAX 342#define ETP_PRI_MAX EIO_PRI_MAX
… …
190static int eio_finish (eio_req *req); 348static int eio_finish (eio_req *req);
191#define ETP_FINISH(req) eio_finish (req) 349#define ETP_FINISH(req) eio_finish (req)
192static void eio_execute (struct etp_worker *self, eio_req *req); 350static void eio_execute (struct etp_worker *self, eio_req *req);
193#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) 351#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
194 352
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/*****************************************************************************/ 353/*****************************************************************************/
213 354
214#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) 355#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
215 356
216/* calculate time difference in ~1/EIO_TICKS of a second */ 357/* calculate time difference in ~1/EIO_TICKS of a second */
358ecb_inline int
217static int tvdiff (struct timeval *tv1, struct timeval *tv2) 359tvdiff (struct timeval *tv1, struct timeval *tv2)
218{ 360{
219 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 361 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
220 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 362 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
221} 363}
222 364
… …
226static void (*done_poll_cb) (void); 368static void (*done_poll_cb) (void);
227 369
228static unsigned int max_poll_time; /* reslock */ 370static unsigned int max_poll_time; /* reslock */
229static unsigned int max_poll_reqs; /* reslock */ 371static unsigned int max_poll_reqs; /* reslock */
230 372
231static volatile unsigned int nreqs; /* reqlock */ 373static unsigned int nreqs; /* reqlock */
232static volatile unsigned int nready; /* reqlock */ 374static unsigned int nready; /* reqlock */
233static volatile unsigned int npending; /* reqlock */ 375static unsigned int npending; /* reqlock */
234static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */ 376static 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 */ 377static unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
236 378
237static xmutex_t wrklock; 379static xmutex_t wrklock;
238static xmutex_t reslock; 380static xmutex_t reslock;
239static xmutex_t reqlock; 381static xmutex_t reqlock;
240static xcond_t reqwait; 382static xcond_t reqwait;
… …
243/* 385/*
244 * make our pread/pwrite emulation safe against themselves, but not against 386 * make our pread/pwrite emulation safe against themselves, but not against
245 * normal read/write by using a mutex. slows down execution a lot, 387 * normal read/write by using a mutex. slows down execution a lot,
246 * but that's your problem, not mine. 388 * but that's your problem, not mine.
247 */ 389 */
248static xmutex_t preadwritelock = X_MUTEX_INIT; 390static xmutex_t preadwritelock;
249#endif 391#endif
250 392
251typedef struct etp_worker 393typedef struct etp_worker
252{ 394{
395 struct tmpbuf tmpbuf;
396
253 /* locked by wrklock */ 397 /* locked by wrklock */
254 struct etp_worker *prev, *next; 398 struct etp_worker *prev, *next;
255 399
256 xthread_t tid; 400 xthread_t tid;
257 401
258 /* locked by reslock, reqlock or wrklock */ 402#ifdef ETP_WORKER_COMMON
259 ETP_REQ *req; /* currently processed request */
260
261 ETP_WORKER_COMMON 403 ETP_WORKER_COMMON
404#endif
262} etp_worker; 405} etp_worker;
263 406
264static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */ 407static etp_worker wrk_first; /* NOT etp */
265 408
266#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock) 409#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
267#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock) 410#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
268 411
269/* worker threads management */ 412/* worker threads management */
270 413
414static void
271static void etp_worker_clear (etp_worker *wrk) 415etp_worker_clear (etp_worker *wrk)
272{ 416{
273 ETP_WORKER_CLEAR (wrk);
274} 417}
275 418
419static void ecb_cold
276static void etp_worker_free (etp_worker *wrk) 420etp_worker_free (etp_worker *wrk)
277{ 421{
422 free (wrk->tmpbuf.ptr);
423
278 wrk->next->prev = wrk->prev; 424 wrk->next->prev = wrk->prev;
279 wrk->prev->next = wrk->next; 425 wrk->prev->next = wrk->next;
280 426
281 free (wrk); 427 free (wrk);
282} 428}
283 429
284static unsigned int etp_nreqs (void) 430static unsigned int
431etp_nreqs (void)
285{ 432{
286 int retval; 433 int retval;
287 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 434 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
288 retval = nreqs; 435 retval = nreqs;
289 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 436 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
290 return retval; 437 return retval;
291} 438}
292 439
293static unsigned int etp_nready (void) 440static unsigned int
441etp_nready (void)
294{ 442{
295 unsigned int retval; 443 unsigned int retval;
296 444
297 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 445 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
298 retval = nready; 446 retval = nready;
299 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 447 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
300 448
301 return retval; 449 return retval;
302} 450}
303 451
304static unsigned int etp_npending (void) 452static unsigned int
453etp_npending (void)
305{ 454{
306 unsigned int retval; 455 unsigned int retval;
307 456
308 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 457 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
309 retval = npending; 458 retval = npending;
310 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 459 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
311 460
312 return retval; 461 return retval;
313} 462}
314 463
315static unsigned int etp_nthreads (void) 464static unsigned int
465etp_nthreads (void)
316{ 466{
317 unsigned int retval; 467 unsigned int retval;
318 468
319 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 469 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
320 retval = started; 470 retval = started;
… …
334} etp_reqq; 484} etp_reqq;
335 485
336static etp_reqq req_queue; 486static etp_reqq req_queue;
337static etp_reqq res_queue; 487static etp_reqq res_queue;
338 488
489static void ecb_noinline ecb_cold
490reqq_init (etp_reqq *q)
491{
492 int pri;
493
494 for (pri = 0; pri < ETP_NUM_PRI; ++pri)
495 q->qs[pri] = q->qe[pri] = 0;
496
497 q->size = 0;
498}
499
500static int ecb_noinline
339static int reqq_push (etp_reqq *q, ETP_REQ *req) 501reqq_push (etp_reqq *q, ETP_REQ *req)
340{ 502{
341 int pri = req->pri; 503 int pri = req->pri;
342 req->next = 0; 504 req->next = 0;
343 505
344 if (q->qe[pri]) 506 if (q->qe[pri])
… …
350 q->qe[pri] = q->qs[pri] = req; 512 q->qe[pri] = q->qs[pri] = req;
351 513
352 return q->size++; 514 return q->size++;
353} 515}
354 516
517static ETP_REQ * ecb_noinline
355static ETP_REQ *reqq_shift (etp_reqq *q) 518reqq_shift (etp_reqq *q)
356{ 519{
357 int pri; 520 int pri;
358 521
359 if (!q->size) 522 if (!q->size)
360 return 0; 523 return 0;
… …
375 } 538 }
376 539
377 abort (); 540 abort ();
378} 541}
379 542
380static void etp_thread_init (void) 543static int ecb_cold
544etp_init (void (*want_poll)(void), void (*done_poll)(void))
381{ 545{
382 X_MUTEX_CREATE (wrklock); 546 X_MUTEX_CREATE (wrklock);
383 X_MUTEX_CREATE (reslock); 547 X_MUTEX_CREATE (reslock);
384 X_MUTEX_CREATE (reqlock); 548 X_MUTEX_CREATE (reqlock);
385 X_COND_CREATE (reqwait); 549 X_COND_CREATE (reqwait);
386}
387 550
388static void etp_atfork_prepare (void) 551 reqq_init (&req_queue);
389{ 552 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}
397 553
398static void etp_atfork_parent (void) 554 wrk_first.next =
399{ 555 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 }
428 556
429 started = 0; 557 started = 0;
430 idle = 0; 558 idle = 0;
431 nreqs = 0; 559 nreqs = 0;
432 nready = 0; 560 nready = 0;
433 npending = 0; 561 npending = 0;
434 562
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; 563 want_poll_cb = want_poll;
453 done_poll_cb = done_poll; 564 done_poll_cb = done_poll;
454 565
455 return 0; 566 return 0;
456} 567}
457 568
458X_THREAD_PROC (etp_proc); 569X_THREAD_PROC (etp_proc);
459 570
571static void ecb_cold
460static void etp_start_thread (void) 572etp_start_thread (void)
461{ 573{
462 etp_worker *wrk = calloc (1, sizeof (etp_worker)); 574 etp_worker *wrk = calloc (1, sizeof (etp_worker));
463 575
464 /*TODO*/ 576 /*TODO*/
465 assert (("unable to allocate worker thread data", wrk)); 577 assert (("unable to allocate worker thread data", wrk));
… …
478 free (wrk); 590 free (wrk);
479 591
480 X_UNLOCK (wrklock); 592 X_UNLOCK (wrklock);
481} 593}
482 594
595static void
483static void etp_maybe_start_thread (void) 596etp_maybe_start_thread (void)
484{ 597{
485 if (expect_true (etp_nthreads () >= wanted)) 598 if (ecb_expect_true (etp_nthreads () >= wanted))
486 return; 599 return;
487 600
488 /* todo: maybe use idle here, but might be less exact */ 601 /* todo: maybe use idle here, but might be less exact */
489 if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ())) 602 if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
490 return; 603 return;
491 604
492 etp_start_thread (); 605 etp_start_thread ();
493} 606}
494 607
608static void ecb_cold
495static void etp_end_thread (void) 609etp_end_thread (void)
496{ 610{
497 eio_req *req = calloc (1, sizeof (eio_req)); 611 eio_req *req = calloc (1, sizeof (eio_req)); /* will be freed by worker */
498 612
499 req->type = -1; 613 req->type = -1;
500 req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 614 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
501 615
502 X_LOCK (reqlock); 616 X_LOCK (reqlock);
… …
507 X_LOCK (wrklock); 621 X_LOCK (wrklock);
508 --started; 622 --started;
509 X_UNLOCK (wrklock); 623 X_UNLOCK (wrklock);
510} 624}
511 625
512static int etp_poll (void) 626static int
627etp_poll (void)
513{ 628{
514 unsigned int maxreqs; 629 unsigned int maxreqs;
515 unsigned int maxtime; 630 unsigned int maxtime;
516 struct timeval tv_start, tv_now; 631 struct timeval tv_start, tv_now;
517 632
… …
547 662
548 X_LOCK (reqlock); 663 X_LOCK (reqlock);
549 --nreqs; 664 --nreqs;
550 X_UNLOCK (reqlock); 665 X_UNLOCK (reqlock);
551 666
552 if (expect_false (req->type == EIO_GROUP && req->size)) 667 if (ecb_expect_false (req->type == EIO_GROUP && req->size))
553 { 668 {
554 req->int1 = 1; /* mark request as delayed */ 669 req->int1 = 1; /* mark request as delayed */
555 continue; 670 continue;
556 } 671 }
557 else 672 else
558 { 673 {
559 int res = ETP_FINISH (req); 674 int res = ETP_FINISH (req);
560 if (expect_false (res)) 675 if (ecb_expect_false (res))
561 return res; 676 return res;
562 } 677 }
563 678
564 if (expect_false (maxreqs && !--maxreqs)) 679 if (ecb_expect_false (maxreqs && !--maxreqs))
565 break; 680 break;
566 681
567 if (maxtime) 682 if (maxtime)
568 { 683 {
569 gettimeofday (&tv_now, 0); 684 gettimeofday (&tv_now, 0);
… …
575 690
576 errno = EAGAIN; 691 errno = EAGAIN;
577 return -1; 692 return -1;
578} 693}
579 694
695static void
580static void etp_cancel (ETP_REQ *req) 696etp_cancel (ETP_REQ *req)
581{ 697{
582 X_LOCK (wrklock); 698 req->cancelled = 1;
583 req->flags |= EIO_FLAG_CANCELLED;
584 X_UNLOCK (wrklock);
585 699
586 eio_grp_cancel (req); 700 eio_grp_cancel (req);
587} 701}
588 702
703static void
589static void etp_submit (ETP_REQ *req) 704etp_submit (ETP_REQ *req)
590{ 705{
591 req->pri -= ETP_PRI_MIN; 706 req->pri -= ETP_PRI_MIN;
592 707
593 if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN; 708 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; 709 if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
595 710
596 if (expect_false (req->type == EIO_GROUP)) 711 if (ecb_expect_false (req->type == EIO_GROUP))
597 { 712 {
598 /* I hope this is worth it :/ */ 713 /* I hope this is worth it :/ */
599 X_LOCK (reqlock); 714 X_LOCK (reqlock);
600 ++nreqs; 715 ++nreqs;
601 X_UNLOCK (reqlock); 716 X_UNLOCK (reqlock);
… …
620 735
621 etp_maybe_start_thread (); 736 etp_maybe_start_thread ();
622 } 737 }
623} 738}
624 739
740static void ecb_cold
625static void etp_set_max_poll_time (double nseconds) 741etp_set_max_poll_time (double nseconds)
626{ 742{
627 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 743 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
628 max_poll_time = nseconds * EIO_TICKS; 744 max_poll_time = nseconds * EIO_TICKS;
629 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 745 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
630} 746}
631 747
748static void ecb_cold
632static void etp_set_max_poll_reqs (unsigned int maxreqs) 749etp_set_max_poll_reqs (unsigned int maxreqs)
633{ 750{
634 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 751 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
635 max_poll_reqs = maxreqs; 752 max_poll_reqs = maxreqs;
636 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 753 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
637} 754}
638 755
756static void ecb_cold
639static void etp_set_max_idle (unsigned int nthreads) 757etp_set_max_idle (unsigned int nthreads)
640{ 758{
641 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 759 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
642 max_idle = nthreads; 760 max_idle = nthreads;
643 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 761 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
644} 762}
645 763
764static void ecb_cold
646static void etp_set_idle_timeout (unsigned int seconds) 765etp_set_idle_timeout (unsigned int seconds)
647{ 766{
648 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 767 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
649 idle_timeout = seconds; 768 idle_timeout = seconds;
650 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 769 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
651} 770}
652 771
772static void ecb_cold
653static void etp_set_min_parallel (unsigned int nthreads) 773etp_set_min_parallel (unsigned int nthreads)
654{ 774{
655 if (wanted < nthreads) 775 if (wanted < nthreads)
656 wanted = nthreads; 776 wanted = nthreads;
657} 777}
658 778
779static void ecb_cold
659static void etp_set_max_parallel (unsigned int nthreads) 780etp_set_max_parallel (unsigned int nthreads)
660{ 781{
661 if (wanted > nthreads) 782 if (wanted > nthreads)
662 wanted = nthreads; 783 wanted = nthreads;
663 784
664 while (started > wanted) 785 while (started > wanted)
665 etp_end_thread (); 786 etp_end_thread ();
666} 787}
667 788
668/*****************************************************************************/ 789/*****************************************************************************/
669 790
791static void
670static void grp_try_feed (eio_req *grp) 792grp_try_feed (eio_req *grp)
671{ 793{
672 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 794 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
673 { 795 {
674 grp->flags &= ~EIO_FLAG_GROUPADD; 796 grp->flags &= ~EIO_FLAG_GROUPADD;
675 797
… …
682 break; 804 break;
683 } 805 }
684 } 806 }
685} 807}
686 808
809static int
687static int grp_dec (eio_req *grp) 810grp_dec (eio_req *grp)
688{ 811{
689 --grp->size; 812 --grp->size;
690 813
691 /* call feeder, if applicable */ 814 /* call feeder, if applicable */
692 grp_try_feed (grp); 815 grp_try_feed (grp);
… …
696 return eio_finish (grp); 819 return eio_finish (grp);
697 else 820 else
698 return 0; 821 return 0;
699} 822}
700 823
824static void
701void eio_destroy (eio_req *req) 825eio_destroy (eio_req *req)
702{ 826{
703 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); 827 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
704 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); 828 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
705 829
706 EIO_DESTROY (req); 830 EIO_DESTROY (req);
707} 831}
708 832
833static int
709static int eio_finish (eio_req *req) 834eio_finish (eio_req *req)
710{ 835{
711 int res = EIO_FINISH (req); 836 int res = EIO_FINISH (req);
712 837
713 if (req->grp) 838 if (req->grp)
714 { 839 {
… …
722 if (grp->grp_first == req) 847 if (grp->grp_first == req)
723 grp->grp_first = req->grp_next; 848 grp->grp_first = req->grp_next;
724 849
725 res2 = grp_dec (grp); 850 res2 = grp_dec (grp);
726 851
727 if (!res && res2) 852 if (!res)
728 res = res2; 853 res = res2;
729 } 854 }
730 855
731 eio_destroy (req); 856 eio_destroy (req);
732 857
733 return res; 858 return res;
734} 859}
735 860
861void
736void eio_grp_cancel (eio_req *grp) 862eio_grp_cancel (eio_req *grp)
737{ 863{
738 for (grp = grp->grp_first; grp; grp = grp->grp_next) 864 for (grp = grp->grp_first; grp; grp = grp->grp_next)
739 eio_cancel (grp); 865 eio_cancel (grp);
740} 866}
741 867
868void
742void eio_cancel (eio_req *req) 869eio_cancel (eio_req *req)
743{ 870{
744 etp_cancel (req); 871 etp_cancel (req);
745} 872}
746 873
874void
747void eio_submit (eio_req *req) 875eio_submit (eio_req *req)
748{ 876{
749 etp_submit (req); 877 etp_submit (req);
750} 878}
751 879
752unsigned int eio_nreqs (void) 880unsigned int
881eio_nreqs (void)
753{ 882{
754 return etp_nreqs (); 883 return etp_nreqs ();
755} 884}
756 885
757unsigned int eio_nready (void) 886unsigned int
887eio_nready (void)
758{ 888{
759 return etp_nready (); 889 return etp_nready ();
760} 890}
761 891
762unsigned int eio_npending (void) 892unsigned int
893eio_npending (void)
763{ 894{
764 return etp_npending (); 895 return etp_npending ();
765} 896}
766 897
767unsigned int eio_nthreads (void) 898unsigned int ecb_cold
899eio_nthreads (void)
768{ 900{
769 return etp_nthreads (); 901 return etp_nthreads ();
770} 902}
771 903
904void ecb_cold
772void eio_set_max_poll_time (double nseconds) 905eio_set_max_poll_time (double nseconds)
773{ 906{
774 etp_set_max_poll_time (nseconds); 907 etp_set_max_poll_time (nseconds);
775} 908}
776 909
910void ecb_cold
777void eio_set_max_poll_reqs (unsigned int maxreqs) 911eio_set_max_poll_reqs (unsigned int maxreqs)
778{ 912{
779 etp_set_max_poll_reqs (maxreqs); 913 etp_set_max_poll_reqs (maxreqs);
780} 914}
781 915
916void ecb_cold
782void eio_set_max_idle (unsigned int nthreads) 917eio_set_max_idle (unsigned int nthreads)
783{ 918{
784 etp_set_max_idle (nthreads); 919 etp_set_max_idle (nthreads);
785} 920}
786 921
922void ecb_cold
787void eio_set_idle_timeout (unsigned int seconds) 923eio_set_idle_timeout (unsigned int seconds)
788{ 924{
789 etp_set_idle_timeout (seconds); 925 etp_set_idle_timeout (seconds);
790} 926}
791 927
928void ecb_cold
792void eio_set_min_parallel (unsigned int nthreads) 929eio_set_min_parallel (unsigned int nthreads)
793{ 930{
794 etp_set_min_parallel (nthreads); 931 etp_set_min_parallel (nthreads);
795} 932}
796 933
934void ecb_cold
797void eio_set_max_parallel (unsigned int nthreads) 935eio_set_max_parallel (unsigned int nthreads)
798{ 936{
799 etp_set_max_parallel (nthreads); 937 etp_set_max_parallel (nthreads);
800} 938}
801 939
802int eio_poll (void) 940int eio_poll (void)
… …
811# undef pread 949# undef pread
812# undef pwrite 950# undef pwrite
813# define pread eio__pread 951# define pread eio__pread
814# define pwrite eio__pwrite 952# define pwrite eio__pwrite
815 953
816static ssize_t 954static eio_ssize_t
817eio__pread (int fd, void *buf, size_t count, off_t offset) 955eio__pread (int fd, void *buf, size_t count, off_t offset)
818{ 956{
819 ssize_t res; 957 eio_ssize_t res;
820 off_t ooffset; 958 off_t ooffset;
821 959
822 X_LOCK (preadwritelock); 960 X_LOCK (preadwritelock);
823 ooffset = lseek (fd, 0, SEEK_CUR); 961 ooffset = lseek (fd, 0, SEEK_CUR);
824 lseek (fd, offset, SEEK_SET); 962 lseek (fd, offset, SEEK_SET);
… …
827 X_UNLOCK (preadwritelock); 965 X_UNLOCK (preadwritelock);
828 966
829 return res; 967 return res;
830} 968}
831 969
832static ssize_t 970static eio_ssize_t
833eio__pwrite (int fd, void *buf, size_t count, off_t offset) 971eio__pwrite (int fd, void *buf, size_t count, off_t offset)
834{ 972{
835 ssize_t res; 973 eio_ssize_t res;
836 off_t ooffset; 974 off_t ooffset;
837 975
838 X_LOCK (preadwritelock); 976 X_LOCK (preadwritelock);
839 ooffset = lseek (fd, 0, SEEK_CUR); 977 ooffset = lseek (fd, 0, SEEK_CUR);
840 lseek (fd, offset, SEEK_SET); 978 lseek (fd, offset, SEEK_SET);
… …
872#ifndef HAVE_FUTIMES 1010#ifndef HAVE_FUTIMES
873 1011
874# undef futimes 1012# undef futimes
875# define futimes(fd,times) eio__futimes (fd, times) 1013# define futimes(fd,times) eio__futimes (fd, times)
876 1014
1015static int
877static int eio__futimes (int fd, const struct timeval tv[2]) 1016eio__futimes (int fd, const struct timeval tv[2])
878{ 1017{
879 errno = ENOSYS; 1018 errno = ENOSYS;
880 return -1; 1019 return -1;
881} 1020}
882 1021
… …
885#if !HAVE_FDATASYNC 1024#if !HAVE_FDATASYNC
886# undef fdatasync 1025# undef fdatasync
887# define fdatasync(fd) fsync (fd) 1026# define fdatasync(fd) fsync (fd)
888#endif 1027#endif
889 1028
1029static int
1030eio__syncfs (int fd)
1031{
1032 int res;
1033
1034#if HAVE_SYS_SYNCFS
1035 res = (int)syscall (__NR_syncfs, (int)(fd));
1036#else
1037 res = -1;
1038 errno = ENOSYS;
1039#endif
1040
1041 if (res < 0 && errno == ENOSYS && fd >= 0)
1042 sync ();
1043
1044 return res;
1045}
1046
890/* sync_file_range always needs emulation */ 1047/* sync_file_range always needs emulation */
891int 1048static int
892eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags) 1049eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
893{ 1050{
894#if HAVE_SYNC_FILE_RANGE 1051#if HAVE_SYNC_FILE_RANGE
895 int res; 1052 int res;
896 1053
… …
913 /* even though we could play tricks with the flags, it's better to always 1070 /* 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 */ 1071 * call fdatasync, as that matches the expectation of its users best */
915 return fdatasync (fd); 1072 return fdatasync (fd);
916} 1073}
917 1074
1075static int
1076eio__fallocate (int fd, int mode, off_t offset, size_t len)
1077{
1078#if HAVE_FALLOCATE
1079 return fallocate (fd, mode, offset, len);
1080#else
1081 errno = ENOSYS;
1082 return -1;
1083#endif
1084}
1085
918#if !HAVE_READAHEAD 1086#if !HAVE_READAHEAD
919# undef readahead 1087# undef readahead
920# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 1088# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
921 1089
922static ssize_t 1090static eio_ssize_t
923eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) 1091eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
924{ 1092{
925 size_t todo = count; 1093 size_t todo = count;
926 dBUF; 1094 dBUF;
927 1095
… …
932 pread (fd, eio_buf, len, offset); 1100 pread (fd, eio_buf, len, offset);
933 offset += len; 1101 offset += len;
934 todo -= len; 1102 todo -= len;
935 } 1103 }
936 1104
1105 FUBd;
1106
937 errno = 0; 1107 errno = 0;
938 return count; 1108 return count;
939} 1109}
940 1110
941#endif 1111#endif
942 1112
943/* sendfile always needs emulation */ 1113/* sendfile always needs emulation */
944static ssize_t 1114static eio_ssize_t
945eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) 1115eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
946{ 1116{
1117 eio_ssize_t written = 0;
947 ssize_t res; 1118 eio_ssize_t res;
948 1119
949 if (!count) 1120 if (!count)
950 return 0; 1121 return 0;
951 1122
1123 for (;;)
1124 {
1125#ifdef __APPLE__
1126# undef HAVE_SENDFILE /* broken, as everything on os x */
1127#endif
952#if HAVE_SENDFILE 1128#if HAVE_SENDFILE
953# if __linux 1129# if __linux
1130 off_t soffset = offset;
954 res = sendfile (ofd, ifd, &offset, count); 1131 res = sendfile (ofd, ifd, &soffset, count);
955 1132
956# elif __FreeBSD__ 1133# elif __FreeBSD__
957 /* 1134 /*
958 * Of course, the freebsd sendfile is a dire hack with no thoughts 1135 * Of course, the freebsd sendfile is a dire hack with no thoughts
959 * wasted on making it similar to other I/O functions. 1136 * wasted on making it similar to other I/O functions.
960 */ 1137 */
961 {
962 off_t sbytes; 1138 off_t sbytes;
963 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 1139 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
964 1140
965 #if 0 /* according to the manpage, this is correct, but broken behaviour */ 1141 #if 0 /* according to the manpage, this is correct, but broken behaviour */
966 /* freebsd' sendfile will return 0 on success */ 1142 /* freebsd' sendfile will return 0 on success */
967 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */ 1143 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
968 /* not on e.g. EIO or EPIPE - sounds broken */ 1144 /* not on e.g. EIO or EPIPE - sounds broken */
969 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0) 1145 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
970 res = sbytes; 1146 res = sbytes;
971 #endif 1147 #endif
972 1148
973 /* according to source inspection, this is correct, and useful behaviour */ 1149 /* according to source inspection, this is correct, and useful behaviour */
974 if (sbytes) 1150 if (sbytes)
975 res = sbytes; 1151 res = sbytes;
976 }
977 1152
978# elif defined (__APPLE__) 1153# elif defined (__APPLE__)
979
980 {
981 off_t sbytes = count; 1154 off_t sbytes = count;
982 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0); 1155 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
983 1156
984 /* according to the manpage, sbytes is always valid */ 1157 /* according to the manpage, sbytes is always valid */
985 if (sbytes) 1158 if (sbytes)
986 res = sbytes; 1159 res = sbytes;
987 }
988 1160
989# elif __hpux 1161# elif __hpux
990 res = sendfile (ofd, ifd, offset, count, 0, 0); 1162 res = sendfile (ofd, ifd, offset, count, 0, 0);
991 1163
992# elif __solaris 1164# elif __solaris
993 {
994 struct sendfilevec vec; 1165 struct sendfilevec vec;
995 size_t sbytes; 1166 size_t sbytes;
996 1167
997 vec.sfv_fd = ifd; 1168 vec.sfv_fd = ifd;
998 vec.sfv_flag = 0; 1169 vec.sfv_flag = 0;
999 vec.sfv_off = offset; 1170 vec.sfv_off = offset;
1000 vec.sfv_len = count; 1171 vec.sfv_len = count;
1001 1172
1002 res = sendfilev (ofd, &vec, 1, &sbytes); 1173 res = sendfilev (ofd, &vec, 1, &sbytes);
1003 1174
1004 if (res < 0 && sbytes) 1175 if (res < 0 && sbytes)
1005 res = sbytes; 1176 res = sbytes;
1006 }
1007 1177
1008# endif 1178# endif
1009 1179
1010#elif defined (_WIN32) 1180#elif defined (_WIN32) && 0
1011
1012 /* does not work, just for documentation of what would need to be done */ 1181 /* does not work, just for documentation of what would need to be done */
1013 { 1182 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
1183 /* libeio guarantees that the file offset does not change, and windows */
1184 /* has no way to get an independent handle to the same file description */
1014 HANDLE h = TO_SOCKET (ifd); 1185 HANDLE h = TO_SOCKET (ifd);
1015 SetFilePointer (h, offset, 0, FILE_BEGIN); 1186 SetFilePointer (h, offset, 0, FILE_BEGIN);
1016 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0); 1187 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1017 }
1018 1188
1019#else 1189#else
1020 res = -1; 1190 res = -1;
1021 errno = ENOSYS; 1191 errno = ENOSYS;
1022#endif 1192#endif
1023 1193
1194 /* we assume sendfile can copy at least 128mb in one go */
1195 if (res <= 128 * 1024 * 1024)
1196 {
1197 if (res > 0)
1198 written += res;
1199
1200 if (written)
1201 return written;
1202
1203 break;
1204 }
1205 else
1206 {
1207 /* if we requested more, then probably the kernel was lazy */
1208 written += res;
1209 offset += res;
1210 count -= res;
1211
1212 if (!count)
1213 return written;
1214 }
1215 }
1216
1024 if (res < 0 1217 if (res < 0
1025 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1218 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1026 /* BSDs */ 1219 /* BSDs */
1027#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */ 1220#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1028 || errno == ENOTSUP 1221 || errno == ENOTSUP
1029#endif 1222#endif
1223#ifdef EOPNOTSUPP /* windows */
1030 || errno == EOPNOTSUPP /* BSDs */ 1224 || errno == EOPNOTSUPP /* BSDs */
1225#endif
1031#if __solaris 1226#if __solaris
1032 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1227 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
1033#endif 1228#endif
1034 ) 1229 )
1035 ) 1230 )
… …
1039 1234
1040 res = 0; 1235 res = 0;
1041 1236
1042 while (count) 1237 while (count)
1043 { 1238 {
1044 ssize_t cnt; 1239 eio_ssize_t cnt;
1045 1240
1046 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 1241 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
1047 1242
1048 if (cnt <= 0) 1243 if (cnt <= 0)
1049 { 1244 {
… …
1061 1256
1062 offset += cnt; 1257 offset += cnt;
1063 res += cnt; 1258 res += cnt;
1064 count -= cnt; 1259 count -= cnt;
1065 } 1260 }
1261
1262 FUBd;
1066 } 1263 }
1067 1264
1068 return res; 1265 return res;
1266}
1267
1268#ifdef PAGESIZE
1269# define eio_pagesize() PAGESIZE
1270#else
1271static intptr_t
1272eio_pagesize (void)
1273{
1274 static intptr_t page;
1275
1276 if (!page)
1277 page = sysconf (_SC_PAGESIZE);
1278
1279 return page;
1280}
1281#endif
1282
1283static void
1284eio_page_align (void **addr, size_t *length)
1285{
1286 intptr_t mask = eio_pagesize () - 1;
1287
1288 /* round down addr */
1289 intptr_t adj = mask & (intptr_t)*addr;
1290
1291 *addr = (void *)((intptr_t)*addr - adj);
1292 *length += adj;
1293
1294 /* round up length */
1295 *length = (*length + mask) & ~mask;
1296}
1297
1298#if !_POSIX_MEMLOCK
1299# define eio__mlockall(a) EIO_ENOSYS ()
1300#else
1301
1302static int
1303eio__mlockall (int flags)
1304{
1305 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1306 extern int mallopt (int, int);
1307 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1308 #endif
1309
1310 if (EIO_MCL_CURRENT != MCL_CURRENT
1311 || EIO_MCL_FUTURE != MCL_FUTURE)
1312 {
1313 flags = 0
1314 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1315 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1316 }
1317
1318 return mlockall (flags);
1319}
1320#endif
1321
1322#if !_POSIX_MEMLOCK_RANGE
1323# define eio__mlock(a,b) EIO_ENOSYS ()
1324#else
1325
1326static int
1327eio__mlock (void *addr, size_t length)
1328{
1329 eio_page_align (&addr, &length);
1330
1331 return mlock (addr, length);
1332}
1333
1334#endif
1335
1336#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1337# define eio__msync(a,b,c) EIO_ENOSYS ()
1338#else
1339
1340static int
1341eio__msync (void *mem, size_t len, int flags)
1342{
1343 eio_page_align (&mem, &len);
1344
1345 if (EIO_MS_ASYNC != MS_SYNC
1346 || EIO_MS_INVALIDATE != MS_INVALIDATE
1347 || EIO_MS_SYNC != MS_SYNC)
1348 {
1349 flags = 0
1350 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1351 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1352 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1353 }
1354
1355 return msync (mem, len, flags);
1356}
1357
1358#endif
1359
1360static int
1361eio__mtouch (eio_req *req)
1362{
1363 void *mem = req->ptr2;
1364 size_t len = req->size;
1365 int flags = req->int1;
1366
1367 eio_page_align (&mem, &len);
1368
1369 {
1370 intptr_t addr = (intptr_t)mem;
1371 intptr_t end = addr + len;
1372 intptr_t page = eio_pagesize ();
1373
1374 if (addr < end)
1375 if (flags & EIO_MT_MODIFY) /* modify */
1376 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1377 else
1378 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1379 }
1380
1381 return 0;
1382}
1383
1384/*****************************************************************************/
1385/* requests implemented outside eio_execute, because they are so large */
1386
1387/* result will always end up in tmpbuf, there is always space for adding a 0-byte */
1388static int
1389eio__realpath (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
1390{
1391 const char *rel = path;
1392 char *res;
1393 char *tmp1, *tmp2;
1394#if SYMLOOP_MAX > 32
1395 int symlinks = SYMLOOP_MAX;
1396#else
1397 int symlinks = 32;
1398#endif
1399
1400 errno = EINVAL;
1401 if (!rel)
1402 return -1;
1403
1404 errno = ENOENT;
1405 if (!*rel)
1406 return -1;
1407
1408 res = tmpbuf_get (tmpbuf, PATH_MAX * 3);
1409 tmp1 = res + PATH_MAX;
1410 tmp2 = tmp1 + PATH_MAX;
1411
1412#if 0 /* disabled, the musl way to do things is just too racy */
1413#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1414 /* on linux we may be able to ask the kernel */
1415 {
1416 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1417
1418 if (fd >= 0)
1419 {
1420 sprintf (tmp1, "/proc/self/fd/%d", fd);
1421 req->result = readlink (tmp1, res, PATH_MAX);
1422 close (fd);
1423
1424 /* here we should probably stat the open file and the disk file, to make sure they still match */
1425
1426 if (req->result > 0)
1427 goto done;
1428 }
1429 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1430 return;
1431 }
1432#endif
1433#endif
1434
1435 if (*rel != '/')
1436 {
1437 int len;
1438
1439 errno = ENOENT;
1440 if (wd == EIO_INVALID_WD)
1441 return -1;
1442
1443 if (wd == EIO_CWD)
1444 {
1445 if (!getcwd (res, PATH_MAX))
1446 return -1;
1447
1448 len = strlen (res);
1449 }
1450 else
1451 memcpy (res, wd->str, len = wd->len);
1452
1453 if (res [1]) /* only use if not / */
1454 res += len;
1455 }
1456
1457 while (*rel)
1458 {
1459 eio_ssize_t len, linklen;
1460 const char *beg = rel;
1461
1462 while (*rel && *rel != '/')
1463 ++rel;
1464
1465 len = rel - beg;
1466
1467 if (!len) /* skip slashes */
1468 {
1469 ++rel;
1470 continue;
1471 }
1472
1473 if (beg [0] == '.')
1474 {
1475 if (len == 1)
1476 continue; /* . - nop */
1477
1478 if (beg [1] == '.' && len == 2)
1479 {
1480 /* .. - back up one component, if possible */
1481
1482 while (res != tmpbuf->ptr)
1483 if (*--res == '/')
1484 break;
1485
1486 continue;
1487 }
1488 }
1489
1490 errno = ENAMETOOLONG;
1491 if (res + 1 + len + 1 >= tmp1)
1492 return;
1493
1494 /* copy one component */
1495 *res = '/';
1496 memcpy (res + 1, beg, len);
1497
1498 /* zero-terminate, for readlink */
1499 res [len + 1] = 0;
1500
1501 /* now check if it's a symlink */
1502 linklen = readlink (tmpbuf->ptr, tmp1, PATH_MAX);
1503
1504 if (linklen < 0)
1505 {
1506 if (errno != EINVAL)
1507 return -1;
1508
1509 /* it's a normal directory. hopefully */
1510 res += len + 1;
1511 }
1512 else
1513 {
1514 /* yay, it was a symlink - build new path in tmp2 */
1515 int rellen = strlen (rel);
1516
1517 errno = ENAMETOOLONG;
1518 if (linklen + 1 + rellen >= PATH_MAX)
1519 return -1;
1520
1521 errno = ELOOP;
1522 if (!--symlinks)
1523 return -1;
1524
1525 if (*tmp1 == '/')
1526 res = tmpbuf->ptr; /* symlink resolves to an absolute path */
1527
1528 /* we need to be careful, as rel might point into tmp2 already */
1529 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1530 tmp2 [linklen] = '/';
1531 memcpy (tmp2, tmp1, linklen);
1532
1533 rel = tmp2;
1534 }
1535 }
1536
1537 /* special case for the lone root path */
1538 if (res == tmpbuf->ptr)
1539 *res++ = '/';
1540
1541 return res - (char *)tmpbuf->ptr;
1069} 1542}
1070 1543
1071static signed char 1544static signed char
1072eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) 1545eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1073{ 1546{
… …
1081 1554
1082#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */ 1555#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 */ 1556#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1084 1557
1085static void 1558static void
1086eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1559eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1087{ 1560{
1088 unsigned char bits [9 + sizeof (ino_t) * 8]; 1561 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1089 unsigned char *bit = bits; 1562 unsigned char *bit = bits;
1090 1563
1091 assert (CHAR_BIT == 8); 1564 assert (CHAR_BIT == 8);
1092 assert (sizeof (eio_dirent) * 8 < 256); 1565 assert (sizeof (eio_dirent) * 8 < 256);
1093 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */ 1566 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
… …
1095 1568
1096 if (size <= EIO_SORT_FAST) 1569 if (size <= EIO_SORT_FAST)
1097 return; 1570 return;
1098 1571
1099 /* first prepare an array of bits to test in our radix sort */ 1572 /* 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 */ 1573 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1101 /* inode_bits must contain all inodes ORed together */ 1574 /* inode_bits must contain all inodes ORed together */
1102 /* which is used to skip bits that are 0 everywhere, which is very common */ 1575 /* which is used to skip bits that are 0 everywhere, which is very common */
1103 { 1576 {
1104 ino_t endianness; 1577 eio_ino_t endianness;
1105 int i, j; 1578 int i, j;
1106 1579
1107 /* we store the byte offset of byte n into byte n of "endianness" */ 1580 /* we store the byte offset of byte n into byte n of "endianness" */
1108 for (i = 0; i < sizeof (ino_t); ++i) 1581 for (i = 0; i < sizeof (eio_ino_t); ++i)
1109 ((unsigned char *)&endianness)[i] = i; 1582 ((unsigned char *)&endianness)[i] = i;
1110 1583
1111 *bit++ = 0; 1584 *bit++ = 0;
1112 1585
1113 for (i = 0; i < sizeof (ino_t); ++i) 1586 for (i = 0; i < sizeof (eio_ino_t); ++i)
1114 { 1587 {
1115 /* shifting off the byte offsets out of "endianness" */ 1588 /* shifting off the byte offsets out of "endianness" */
1116 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8; 1589 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1117 endianness >>= 8; 1590 endianness >>= 8;
1118 1591
1119 for (j = 0; j < 8; ++j) 1592 for (j = 0; j < 8; ++j)
1120 if (inode_bits & (((ino_t)1) << (i * 8 + j))) 1593 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1121 *bit++ = offs + j; 1594 *bit++ = offs + j;
1122 } 1595 }
1123 1596
1124 for (j = 0; j < 8; ++j) 1597 for (j = 0; j < 8; ++j)
1125 if (score_bits & (1 << j)) 1598 if (score_bits & (1 << j))
1126 *bit++ = offsetof (eio_dirent, score) * 8 + j; 1599 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1127 } 1600 }
1128 1601
1129 /* now actually do the sorting (a variant of MSD radix sort) */ 1602 /* now actually do the sorting (a variant of MSD radix sort) */
1130 { 1603 {
1131 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base; 1604 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1132 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end; 1605 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1133 unsigned char *bit_stk [9 + sizeof (ino_t) * 8]; 1606 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1134 int stk_idx = 0; 1607 int stk_idx = 0;
1135 1608
1136 base_stk [stk_idx] = dents; 1609 base_stk [stk_idx] = dents;
1137 end_stk [stk_idx] = dents + size; 1610 end_stk [stk_idx] = dents + size;
1138 bit_stk [stk_idx] = bit - 1; 1611 bit_stk [stk_idx] = bit - 1;
… …
1217 } 1690 }
1218 } 1691 }
1219} 1692}
1220 1693
1221static void 1694static void
1222eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1695eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1223{ 1696{
1224 if (size <= 1) 1697 if (size <= 1)
1225 return; /* our insertion sort relies on size > 0 */ 1698 return; /* our insertion sort relies on size > 0 */
1226 1699
1227 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */ 1700 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
… …
1235 1708
1236/* read a full directory */ 1709/* read a full directory */
1237static void 1710static void
1238eio__scandir (eio_req *req, etp_worker *self) 1711eio__scandir (eio_req *req, etp_worker *self)
1239{ 1712{
1240 DIR *dirp;
1241 EIO_STRUCT_DIRENT *entp;
1242 char *name, *names; 1713 char *name, *names;
1243 int namesalloc = 4096; 1714 int namesalloc = 4096 - sizeof (void *) * 4;
1244 int namesoffs = 0; 1715 int namesoffs = 0;
1245 int flags = req->int1; 1716 int flags = req->int1;
1246 eio_dirent *dents = 0; 1717 eio_dirent *dents = 0;
1247 int dentalloc = 128; 1718 int dentalloc = 128;
1248 int dentoffs = 0; 1719 int dentoffs = 0;
1249 ino_t inode_bits = 0; 1720 eio_ino_t inode_bits = 0;
1721#ifdef _WIN32
1722 HANDLE dirp;
1723 WIN32_FIND_DATA entp;
1724#else
1725 DIR *dirp;
1726 EIO_STRUCT_DIRENT *entp;
1727#endif
1250 1728
1251 req->result = -1; 1729 req->result = -1;
1252 1730
1253 if (!(flags & EIO_READDIR_DENTS)) 1731 if (!(flags & EIO_READDIR_DENTS))
1254 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); 1732 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1255 1733
1256 X_LOCK (wrklock); 1734#ifdef _WIN32
1257 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 1735 {
1736 int len = strlen ((const char *)req->ptr1);
1737 char *path = malloc (MAX_PATH);
1738 const char *fmt;
1739 const char *reqpath = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1740
1741 if (!len)
1742 fmt = "./*";
1743 else if (reqpath[len - 1] == '/' || reqpath[len - 1] == '\\')
1744 fmt = "%s*";
1745 else
1746 fmt = "%s/*";
1747
1748 _snprintf (path, MAX_PATH, fmt, reqpath);
1749 dirp = FindFirstFile (path, &entp);
1750 free (path);
1751
1752 if (dirp == INVALID_HANDLE_VALUE)
1753 {
1754 /* should steal _dosmaperr */
1755 switch (GetLastError ())
1756 {
1757 case ERROR_FILE_NOT_FOUND:
1758 req->result = 0;
1759 break;
1760
1761 case ERROR_INVALID_NAME:
1762 case ERROR_PATH_NOT_FOUND:
1763 case ERROR_NO_MORE_FILES:
1764 errno = ENOENT;
1765 break;
1766
1767 case ERROR_NOT_ENOUGH_MEMORY:
1768 errno = ENOMEM;
1769 break;
1770
1771 default:
1772 errno = EINVAL;
1773 break;
1774 }
1775
1776 return;
1777 }
1778 }
1779#else
1780 #if HAVE_AT
1781 if (req->wd)
1782 {
1783 int fd = openat (WD2FD (req->wd), req->ptr1, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1784
1785 if (fd < 0)
1786 return;
1787
1788 dirp = fdopendir (fd);
1789
1790 if (!dirp)
1791 close (fd);
1792 }
1793 else
1258 self->dirp = dirp = opendir (req->ptr1); 1794 dirp = opendir (req->ptr1);
1795 #else
1796 dirp = opendir (wd_expand (&self->tmpbuf, req->wd, req->ptr1));
1797 #endif
1798
1799 if (!dirp)
1800 return;
1801#endif
1802
1803 if (req->flags & EIO_FLAG_PTR1_FREE)
1804 free (req->ptr1);
1259 1805
1260 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1806 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1261 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; 1807 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1262 req->ptr2 = names = malloc (namesalloc); 1808 req->ptr2 = names = malloc (namesalloc);
1263 X_UNLOCK (wrklock);
1264 1809
1265 if (dirp && names && (!flags || dents)) 1810 if (!names || (flags && !dents))
1811 return;
1812
1266 for (;;) 1813 for (;;)
1267 { 1814 {
1815 int done;
1816
1817#ifdef _WIN32
1818 done = !dirp;
1819#else
1268 errno = 0; 1820 errno = 0;
1269 entp = readdir (dirp); 1821 entp = readdir (dirp);
1822 done = !entp;
1823#endif
1270 1824
1271 if (!entp) 1825 if (done)
1272 { 1826 {
1827#ifndef _WIN32
1828 int old_errno = errno;
1829 closedir (dirp);
1830 errno = old_errno;
1831
1273 if (errno) 1832 if (errno)
1274 break; 1833 break;
1834#endif
1275 1835
1276 /* sort etc. */ 1836 /* sort etc. */
1277 req->int1 = flags; 1837 req->int1 = flags;
1278 req->result = dentoffs; 1838 req->result = dentoffs;
1279 1839
1280 if (flags & EIO_READDIR_STAT_ORDER) 1840 if (flags & EIO_READDIR_STAT_ORDER)
1281 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits); 1841 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1282 else if (flags & EIO_READDIR_DIRS_FIRST) 1842 else if (flags & EIO_READDIR_DIRS_FIRST)
1283 if (flags & EIO_READDIR_FOUND_UNKNOWN) 1843 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1284 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */ 1844 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1285 else 1845 else
1846 {
1847 /* in this case, all is known, and we just put dirs first and sort them */
1848 eio_dirent *oth = dents + dentoffs;
1849 eio_dirent *dir = dents;
1850
1851 /* now partition dirs to the front, and non-dirs to the back */
1852 /* by walking from both sides and swapping if necessary */
1853 while (oth > dir)
1854 {
1855 if (dir->type == EIO_DT_DIR)
1856 ++dir;
1857 else if ((--oth)->type == EIO_DT_DIR)
1858 {
1859 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1860
1861 ++dir;
1862 }
1863 }
1864
1865 /* now sort the dirs only (dirs all have the same score) */
1866 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1867 }
1868
1869 break;
1870 }
1871
1872 /* now add the entry to our list(s) */
1873 name = D_NAME (entp);
1874
1875 /* skip . and .. entries */
1876 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1877 {
1878 int len = D_NAMLEN (entp) + 1;
1879
1880 while (ecb_expect_false (namesoffs + len > namesalloc))
1881 {
1882 namesalloc *= 2;
1883 req->ptr2 = names = realloc (names, namesalloc);
1884
1885 if (!names)
1886 break;
1887 }
1888
1889 memcpy (names + namesoffs, name, len);
1890
1891 if (dents)
1892 {
1893 struct eio_dirent *ent;
1894
1895 if (ecb_expect_false (dentoffs == dentalloc))
1286 { 1896 {
1287 /* in this case, all is known, and we just put dirs first and sort them */ 1897 dentalloc *= 2;
1898 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1899
1900 if (!dents)
1901 break;
1902 }
1903
1288 eio_dirent *oth = dents + dentoffs; 1904 ent = dents + dentoffs;
1289 eio_dirent *dir = dents;
1290 1905
1291 /* now partition dirs to the front, and non-dirs to the back */ 1906 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1292 /* by walking from both sides and swapping if necessary */ 1907 ent->namelen = len - 1;
1293 while (oth > dir) 1908 ent->inode = D_INO (entp);
1909
1910 inode_bits |= ent->inode;
1911
1912 switch (D_TYPE (entp))
1913 {
1914 default:
1915 ent->type = EIO_DT_UNKNOWN;
1916 flags |= EIO_READDIR_FOUND_UNKNOWN;
1917 break;
1918
1919 #ifdef DT_FIFO
1920 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1921 #endif
1922 #ifdef DT_CHR
1923 case DT_CHR: ent->type = EIO_DT_CHR; break;
1924 #endif
1925 #ifdef DT_MPC
1926 case DT_MPC: ent->type = EIO_DT_MPC; break;
1927 #endif
1928 #ifdef DT_DIR
1929 case DT_DIR: ent->type = EIO_DT_DIR; break;
1930 #endif
1931 #ifdef DT_NAM
1932 case DT_NAM: ent->type = EIO_DT_NAM; break;
1933 #endif
1934 #ifdef DT_BLK
1935 case DT_BLK: ent->type = EIO_DT_BLK; break;
1936 #endif
1937 #ifdef DT_MPB
1938 case DT_MPB: ent->type = EIO_DT_MPB; break;
1939 #endif
1940 #ifdef DT_REG
1941 case DT_REG: ent->type = EIO_DT_REG; break;
1942 #endif
1943 #ifdef DT_NWK
1944 case DT_NWK: ent->type = EIO_DT_NWK; break;
1945 #endif
1946 #ifdef DT_CMP
1947 case DT_CMP: ent->type = EIO_DT_CMP; break;
1948 #endif
1949 #ifdef DT_LNK
1950 case DT_LNK: ent->type = EIO_DT_LNK; break;
1951 #endif
1952 #ifdef DT_SOCK
1953 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1954 #endif
1955 #ifdef DT_DOOR
1956 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1957 #endif
1958 #ifdef DT_WHT
1959 case DT_WHT: ent->type = EIO_DT_WHT; break;
1960 #endif
1961 }
1962
1963 ent->score = 7;
1964
1965 if (flags & EIO_READDIR_DIRS_FIRST)
1966 {
1967 if (ent->type == EIO_DT_UNKNOWN)
1294 { 1968 {
1295 if (dir->type == EIO_DT_DIR) 1969 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1296 ++dir; 1970 ent->score = 1;
1297 else if ((--oth)->type == EIO_DT_DIR) 1971 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1298 { 1972 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1299 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1300
1301 ++dir;
1302 }
1303 } 1973 }
1304 1974 else if (ent->type == EIO_DT_DIR)
1305 /* now sort the dirs only (dirs all have the same score) */ 1975 ent->score = 0;
1306 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1307 } 1976 }
1308
1309 break;
1310 }
1311
1312 /* now add the entry to our list(s) */
1313 name = entp->d_name;
1314
1315 /* skip . and .. entries */
1316 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1317 {
1318 int len = D_NAMLEN (entp) + 1;
1319
1320 while (expect_false (namesoffs + len > namesalloc))
1321 {
1322 namesalloc *= 2;
1323 X_LOCK (wrklock);
1324 req->ptr2 = names = realloc (names, namesalloc);
1325 X_UNLOCK (wrklock);
1326
1327 if (!names)
1328 break;
1329 } 1977 }
1330 1978
1331 memcpy (names + namesoffs, name, len);
1332
1333 if (dents)
1334 {
1335 struct eio_dirent *ent;
1336
1337 if (expect_false (dentoffs == dentalloc))
1338 {
1339 dentalloc *= 2;
1340 X_LOCK (wrklock);
1341 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1342 X_UNLOCK (wrklock);
1343
1344 if (!dents)
1345 break;
1346 }
1347
1348 ent = dents + dentoffs;
1349
1350 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1351 ent->namelen = len - 1;
1352 ent->inode = D_INO (entp);
1353
1354 inode_bits |= ent->inode;
1355
1356 switch (D_TYPE (entp))
1357 {
1358 default:
1359 ent->type = EIO_DT_UNKNOWN;
1360 flags |= EIO_READDIR_FOUND_UNKNOWN;
1361 break;
1362
1363 #ifdef DT_FIFO
1364 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1365 #endif
1366 #ifdef DT_CHR
1367 case DT_CHR: ent->type = EIO_DT_CHR; break;
1368 #endif
1369 #ifdef DT_MPC
1370 case DT_MPC: ent->type = EIO_DT_MPC; break;
1371 #endif
1372 #ifdef DT_DIR
1373 case DT_DIR: ent->type = EIO_DT_DIR; break;
1374 #endif
1375 #ifdef DT_NAM
1376 case DT_NAM: ent->type = EIO_DT_NAM; break;
1377 #endif
1378 #ifdef DT_BLK
1379 case DT_BLK: ent->type = EIO_DT_BLK; break;
1380 #endif
1381 #ifdef DT_MPB
1382 case DT_MPB: ent->type = EIO_DT_MPB; break;
1383 #endif
1384 #ifdef DT_REG
1385 case DT_REG: ent->type = EIO_DT_REG; break;
1386 #endif
1387 #ifdef DT_NWK
1388 case DT_NWK: ent->type = EIO_DT_NWK; break;
1389 #endif
1390 #ifdef DT_CMP
1391 case DT_CMP: ent->type = EIO_DT_CMP; break;
1392 #endif
1393 #ifdef DT_LNK
1394 case DT_LNK: ent->type = EIO_DT_LNK; break;
1395 #endif
1396 #ifdef DT_SOCK
1397 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1398 #endif
1399 #ifdef DT_DOOR
1400 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1401 #endif
1402 #ifdef DT_WHT
1403 case DT_WHT: ent->type = EIO_DT_WHT; break;
1404 #endif
1405 }
1406
1407 ent->score = 7;
1408
1409 if (flags & EIO_READDIR_DIRS_FIRST)
1410 {
1411 if (ent->type == EIO_DT_UNKNOWN)
1412 {
1413 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1414 ent->score = 1;
1415 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1416 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1417 }
1418 else if (ent->type == EIO_DT_DIR)
1419 ent->score = 0;
1420 }
1421 }
1422
1423 namesoffs += len; 1979 namesoffs += len;
1424 ++dentoffs; 1980 ++dentoffs;
1425 } 1981 }
1426 1982
1427 if (EIO_CANCELLED (req)) 1983 if (EIO_CANCELLED (req))
1428 { 1984 {
1429 errno = ECANCELED; 1985 errno = ECANCELED;
1430 break; 1986 break;
1431 } 1987 }
1988
1989#ifdef _WIN32
1990 if (!FindNextFile (dirp, &entp))
1991 {
1992 FindClose (dirp);
1993 dirp = 0;
1432 } 1994 }
1995#endif
1996 }
1433} 1997}
1434 1998
1435#ifdef PAGESIZE 1999/*****************************************************************************/
1436# define eio_pagesize() PAGESIZE 2000/* working directory stuff */
1437#else 2001/* various deficiencies in the posix 2008 api force us to */
1438static intptr_t 2002/* keep the absolute path in string form at all times */
1439eio_pagesize (void) 2003/* fuck yeah. */
1440{
1441 static intptr_t page;
1442 2004
1443 if (!page) 2005#if !HAVE_AT
1444 page = sysconf (_SC_PAGESIZE);
1445 2006
2007/* a bit like realpath, but usually faster because it doesn'T have to return */
2008/* an absolute or canonical path */
2009static const char *
2010wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
2011{
2012 if (!wd || *path == '/')
2013 return path;
2014
2015 if (path [0] == '.' && !path [1])
2016 return wd->str;
2017
2018 {
2019 int l1 = wd->len;
2020 int l2 = strlen (path);
2021
2022 char *res = tmpbuf_get (tmpbuf, l1 + l2 + 2);
2023
2024 memcpy (res, wd->str, l1);
2025 res [l1] = '/';
2026 memcpy (res + l1 + 1, path, l2 + 1);
2027
2028 return res;
2029 }
2030}
2031
2032#endif
2033
2034static eio_wd
2035eio__wd_open_sync (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
2036{
2037 int fd;
2038 eio_wd res;
2039 int len = eio__realpath (tmpbuf, wd, path);
2040
2041 if (len < 0)
2042 return EIO_INVALID_WD;
2043
2044#if HAVE_AT
2045 fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
2046
2047 if (fd < 0)
2048 return EIO_INVALID_WD;
2049#endif
2050
2051 res = malloc (sizeof (*res) + len); /* one extra 0-byte */
2052
2053#if HAVE_AT
2054 res->fd = fd;
2055#endif
2056
2057 res->len = len;
2058 memcpy (res->str, tmpbuf->ptr, len);
2059 res->str [len] = 0;
2060
1446 return page; 2061 return res;
1447} 2062}
1448#endif
1449 2063
1450static void 2064eio_wd
1451eio_page_align (void **addr, size_t *length) 2065eio_wd_open_sync (eio_wd wd, const char *path)
1452{ 2066{
1453 intptr_t mask = eio_pagesize () - 1; 2067 struct tmpbuf tmpbuf = { 0 };
2068 wd = eio__wd_open_sync (&tmpbuf, wd, path);
2069 free (tmpbuf.ptr);
1454 2070
1455 /* round down addr */ 2071 return wd;
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} 2072}
1464 2073
1465#if !_POSIX_MEMLOCK 2074void
1466# define eio__mlockall(a) ((errno = ENOSYS), -1) 2075eio_wd_close_sync (eio_wd wd)
1467#else 2076{
2077 if (wd != EIO_INVALID_WD && wd != EIO_CWD)
2078 {
2079 #if HAVE_AT
2080 close (wd->fd);
2081 #endif
2082 free (wd);
2083 }
2084}
2085
2086#if HAVE_AT
2087
2088/* they forgot these */
1468 2089
1469static int 2090static int
1470eio__mlockall (int flags) 2091eio__truncateat (int dirfd, const char *path, off_t length)
1471{ 2092{
1472 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7 2093 int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
1473 extern int mallopt (int, int); 2094 int res;
1474 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1475 #endif
1476 2095
1477 if (EIO_MCL_CURRENT != MCL_CURRENT 2096 if (fd < 0)
1478 || EIO_MCL_FUTURE != MCL_FUTURE) 2097 return fd;
1479 {
1480 flags = 0
1481 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1482 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1483 }
1484 2098
1485 return mlockall (flags); 2099 res = ftruncate (fd, length);
2100 close (fd);
2101 return res;
1486} 2102}
1487#endif
1488
1489#if !_POSIX_MEMLOCK_RANGE
1490# define eio__mlock(a,b) ((errno = ENOSYS), -1)
1491#else
1492 2103
1493static int 2104static int
1494eio__mlock (void *addr, size_t length) 2105eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
1495{ 2106{
1496 eio_page_align (&addr, &length); 2107 int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
2108 int res;
1497 2109
1498 return mlock (addr, length); 2110 if (fd < 0)
1499} 2111 return fd;
1500 2112
1501#endif 2113 res = fstatvfs (fd, buf);
1502 2114 close (fd);
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; 2115 return res;
2116
1545} 2117}
2118
2119#endif
1546 2120
1547/*****************************************************************************/ 2121/*****************************************************************************/
1548 2122
1549#define ALLOC(len) \ 2123#define ALLOC(len) \
1550 if (!req->ptr2) \ 2124 if (!req->ptr2) \
… …
1559 req->result = -1; \ 2133 req->result = -1; \
1560 break; \ 2134 break; \
1561 } \ 2135 } \
1562 } 2136 }
1563 2137
2138static void ecb_noinline ecb_cold
2139etp_proc_init (void)
2140{
2141#if HAVE_PRCTL_SET_NAME
2142 /* provide a more sensible "thread name" */
2143 char name[16 + 1];
2144 const int namelen = sizeof (name) - 1;
2145 int len;
2146
2147 prctl (PR_GET_NAME, (unsigned long)name, 0, 0, 0);
2148 name [namelen] = 0;
2149 len = strlen (name);
2150 strcpy (name + (len <= namelen - 4 ? len : namelen - 4), "/eio");
2151 prctl (PR_SET_NAME, (unsigned long)name, 0, 0, 0);
2152#endif
2153}
2154
1564X_THREAD_PROC (etp_proc) 2155X_THREAD_PROC (etp_proc)
1565{ 2156{
1566 ETP_REQ *req; 2157 ETP_REQ *req;
1567 struct timespec ts; 2158 struct timespec ts;
1568 etp_worker *self = (etp_worker *)thr_arg; 2159 etp_worker *self = (etp_worker *)thr_arg;
1569 2160
2161 etp_proc_init ();
2162
1570 /* try to distribute timeouts somewhat randomly */ 2163 /* try to distribute timeouts somewhat evenly */
1571 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 2164 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1572 2165
1573 for (;;) 2166 for (;;)
1574 { 2167 {
2168 ts.tv_sec = 0;
2169
1575 X_LOCK (reqlock); 2170 X_LOCK (reqlock);
1576 2171
1577 for (;;) 2172 for (;;)
1578 { 2173 {
1579 self->req = req = reqq_shift (&req_queue); 2174 req = reqq_shift (&req_queue);
1580 2175
1581 if (req) 2176 if (req)
1582 break; 2177 break;
1583 2178
2179 if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
2180 {
2181 X_UNLOCK (reqlock);
2182 X_LOCK (wrklock);
2183 --started;
2184 X_UNLOCK (wrklock);
2185 goto quit;
2186 }
2187
1584 ++idle; 2188 ++idle;
1585 2189
1586 ts.tv_sec = time (0) + idle_timeout; 2190 if (idle <= max_idle)
1587 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 2191 /* we are allowed to idle, so do so without any timeout */
2192 X_COND_WAIT (reqwait, reqlock);
2193 else
1588 { 2194 {
1589 if (idle > max_idle) 2195 /* initialise timeout once */
1590 { 2196 if (!ts.tv_sec)
1591 --idle; 2197 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 2198
1599 /* we are allowed to idle, so do so without any timeout */
1600 X_COND_WAIT (reqwait, reqlock); 2199 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
2200 ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1601 } 2201 }
1602 2202
1603 --idle; 2203 --idle;
1604 } 2204 }
1605 2205
… …
1608 X_UNLOCK (reqlock); 2208 X_UNLOCK (reqlock);
1609 2209
1610 if (req->type < 0) 2210 if (req->type < 0)
1611 goto quit; 2211 goto quit;
1612 2212
1613 if (!EIO_CANCELLED (req))
1614 ETP_EXECUTE (self, req); 2213 ETP_EXECUTE (self, req);
1615 2214
1616 X_LOCK (reslock); 2215 X_LOCK (reslock);
1617 2216
1618 ++npending; 2217 ++npending;
1619 2218
1620 if (!reqq_push (&res_queue, req) && want_poll_cb) 2219 if (!reqq_push (&res_queue, req) && want_poll_cb)
1621 want_poll_cb (); 2220 want_poll_cb ();
1622 2221
1623 self->req = 0;
1624 etp_worker_clear (self); 2222 etp_worker_clear (self);
1625 2223
1626 X_UNLOCK (reslock); 2224 X_UNLOCK (reslock);
1627 } 2225 }
1628 2226
1629quit: 2227quit:
2228 free (req);
2229
1630 X_LOCK (wrklock); 2230 X_LOCK (wrklock);
1631 etp_worker_free (self); 2231 etp_worker_free (self);
1632 X_UNLOCK (wrklock); 2232 X_UNLOCK (wrklock);
1633 2233
1634 return 0; 2234 return 0;
1635} 2235}
1636 2236
1637/*****************************************************************************/ 2237/*****************************************************************************/
1638 2238
2239int ecb_cold
1639int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 2240eio_init (void (*want_poll)(void), void (*done_poll)(void))
1640{ 2241{
2242#if !HAVE_PREADWRITE
2243 X_MUTEX_CREATE (preadwritelock);
2244#endif
2245
1641 return etp_init (want_poll, done_poll); 2246 return etp_init (want_poll, done_poll);
1642} 2247}
1643 2248
2249ecb_inline void
1644static void eio_api_destroy (eio_req *req) 2250eio_api_destroy (eio_req *req)
1645{ 2251{
1646 free (req); 2252 free (req);
1647} 2253}
1648 2254
1649#define REQ(rtype) \ 2255#define REQ(rtype) \
… …
1668 { \ 2274 { \
1669 eio_api_destroy (req); \ 2275 eio_api_destroy (req); \
1670 return 0; \ 2276 return 0; \
1671 } 2277 }
1672 2278
2279static void
1673static void eio_execute (etp_worker *self, eio_req *req) 2280eio_execute (etp_worker *self, eio_req *req)
1674{ 2281{
2282#if HAVE_AT
2283 int dirfd;
2284#else
2285 const char *path;
2286#endif
2287
2288 if (ecb_expect_false (EIO_CANCELLED (req)))
2289 {
2290 req->result = -1;
2291 req->errorno = ECANCELED;
2292 return;
2293 }
2294
2295 if (ecb_expect_false (req->wd == EIO_INVALID_WD))
2296 {
2297 req->result = -1;
2298 req->errorno = ENOENT;
2299 return;
2300 }
2301
2302 if (req->type >= EIO_OPEN)
2303 {
2304 #if HAVE_AT
2305 dirfd = WD2FD (req->wd);
2306 #else
2307 path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
2308 #endif
2309 }
2310
1675 switch (req->type) 2311 switch (req->type)
1676 { 2312 {
2313 case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
2314 req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
2315 break;
2316 case EIO_WD_CLOSE: req->result = 0;
2317 eio_wd_close_sync (req->wd); break;
2318
1677 case EIO_READ: ALLOC (req->size); 2319 case EIO_READ: ALLOC (req->size);
1678 req->result = req->offs >= 0 2320 req->result = req->offs >= 0
1679 ? pread (req->int1, req->ptr2, req->size, req->offs) 2321 ? pread (req->int1, req->ptr2, req->size, req->offs)
1680 : read (req->int1, req->ptr2, req->size); break; 2322 : read (req->int1, req->ptr2, req->size); break;
1681 case EIO_WRITE: req->result = req->offs >= 0 2323 case EIO_WRITE: req->result = req->offs >= 0
1682 ? pwrite (req->int1, req->ptr2, req->size, req->offs) 2324 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1683 : write (req->int1, req->ptr2, req->size); break; 2325 : write (req->int1, req->ptr2, req->size); break;
1684 2326
1685 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break; 2327 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; 2328 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
2329
2330#if HAVE_AT
1687 2331
1688 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2332 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1689 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2333 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break;
1690 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2334 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2335 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, AT_SYMLINK_NOFOLLOW); break;
2336 case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break;
2337 case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break;
2338 case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break;
2339 case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break;
2340
2341 case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
2342 case EIO_RMDIR: req->result = unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break;
2343 case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); break;
2344 case EIO_RENAME: req->result = renameat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2); break;
2345 case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2, 0); break;
2346 case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
2347 case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
2348 case EIO_READLINK: ALLOC (PATH_MAX);
2349 req->result = readlinkat (dirfd, req->ptr1, req->ptr2, PATH_MAX); break;
2350 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2351 req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2352 case EIO_UTIME:
2353 case EIO_FUTIME:
2354 {
2355 struct timespec ts[2];
2356 struct timespec *times;
2357
2358 if (req->nv1 != -1. || req->nv2 != -1.)
2359 {
2360 ts[0].tv_sec = req->nv1;
2361 ts[0].tv_nsec = (req->nv1 - ts[0].tv_sec) * 1e9;
2362 ts[1].tv_sec = req->nv2;
2363 ts[1].tv_nsec = (req->nv2 - ts[1].tv_sec) * 1e9;
2364
2365 times = ts;
2366 }
2367 else
2368 times = 0;
2369
2370 req->result = req->type == EIO_FUTIME
2371 ? futimens (req->int1, times)
2372 : utimensat (dirfd, req->ptr1, times, 0);
2373 }
2374 break;
2375
2376#else
2377
2378 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2379 req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
2380 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1691 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2381 req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
2382 case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
2383 case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
2384 case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
2385 case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
2386
2387 case EIO_UNLINK: req->result = unlink (path ); break;
2388 case EIO_RMDIR: req->result = rmdir (path ); break;
2389 case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
2390 case EIO_RENAME: req->result = rename (path , req->ptr2); break;
2391 case EIO_LINK: req->result = link (path , req->ptr2); break;
2392 case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
2393 case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
2394 case EIO_READLINK: ALLOC (PATH_MAX);
2395 req->result = readlink (path, req->ptr2, PATH_MAX); break;
2396 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2397 req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
2398
2399 case EIO_UTIME:
2400 case EIO_FUTIME:
2401 {
2402 struct timeval tv[2];
2403 struct timeval *times;
2404
2405 if (req->nv1 != -1. || req->nv2 != -1.)
2406 {
2407 tv[0].tv_sec = req->nv1;
2408 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
2409 tv[1].tv_sec = req->nv2;
2410 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
2411
2412 times = tv;
2413 }
2414 else
2415 times = 0;
2416
2417 req->result = req->type == EIO_FUTIME
2418 ? futimes (req->int1, times)
2419 : utimes (req->ptr1, times);
2420 }
2421 break;
2422
2423#endif
2424
2425 case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
2426 {
2427 ALLOC (req->result);
2428 memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
2429 }
2430 break;
2431
1692 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2432 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1693 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; 2433 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1694 2434
1695 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1696 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1697 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS)); 2435 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1698 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break; 2436 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1699 2437
1700 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1701 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; 2438 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1702 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1703 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break; 2439 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1704 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1705 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break; 2440 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1706 2441
1707 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
1708 case EIO_CLOSE: req->result = close (req->int1); break; 2442 case EIO_CLOSE: req->result = close (req->int1); break;
1709 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break; 2443 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1710 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
1711 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1712 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1713 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1714 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1715 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;
1717
1718 case EIO_READLINK: ALLOC (PATH_MAX);
1719 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1720
1721 case EIO_SYNC: req->result = 0; sync (); break; 2444 case EIO_SYNC: req->result = 0; sync (); break;
1722 case EIO_FSYNC: req->result = fsync (req->int1); break; 2445 case EIO_FSYNC: req->result = fsync (req->int1); break;
1723 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 2446 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
2447 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
2448 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; 2449 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; 2450 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1726 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break; 2451 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1727 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break; 2452 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; 2453 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
1729 2454
1730 case EIO_READDIR: eio__scandir (req, self); break; 2455 case EIO_READDIR: eio__scandir (req, self); break;
1731 2456
1732 case EIO_BUSY: 2457 case EIO_BUSY:
1733#ifdef _WIN32 2458#ifdef _WIN32
… …
1742 req->result = select (0, 0, 0, 0, &tv); 2467 req->result = select (0, 0, 0, 0, &tv);
1743 } 2468 }
1744#endif 2469#endif
1745 break; 2470 break;
1746 2471
1747 case EIO_UTIME:
1748 case EIO_FUTIME:
1749 {
1750 struct timeval tv[2];
1751 struct timeval *times;
1752
1753 if (req->nv1 != -1. || req->nv2 != -1.)
1754 {
1755 tv[0].tv_sec = req->nv1;
1756 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
1757 tv[1].tv_sec = req->nv2;
1758 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
1759
1760 times = tv;
1761 }
1762 else
1763 times = 0;
1764
1765 req->result = req->type == EIO_FUTIME
1766 ? futimes (req->int1, times)
1767 : utimes (req->ptr1, times);
1768 }
1769 break;
1770
1771 case EIO_GROUP: 2472 case EIO_GROUP:
1772 abort (); /* handled in eio_request */ 2473 abort (); /* handled in eio_request */
1773 2474
1774 case EIO_NOP: 2475 case EIO_NOP:
1775 req->result = 0; 2476 req->result = 0;
1776 break; 2477 break;
1777 2478
1778 case EIO_CUSTOM: 2479 case EIO_CUSTOM:
1779 ((void (*)(eio_req *))req->feed) (req); 2480 req->feed (req);
1780 break; 2481 break;
1781 2482
1782 default: 2483 default:
1783 errno = ENOSYS; 2484 errno = ENOSYS;
1784 req->result = -1; 2485 req->result = -1;
… …
1788 req->errorno = errno; 2489 req->errorno = errno;
1789} 2490}
1790 2491
1791#ifndef EIO_NO_WRAPPERS 2492#ifndef EIO_NO_WRAPPERS
1792 2493
2494eio_req *eio_wd_open (const char *path, int pri, eio_cb cb, void *data)
2495{
2496 REQ (EIO_WD_OPEN); PATH; SEND;
2497}
2498
2499eio_req *eio_wd_close (eio_wd wd, int pri, eio_cb cb, void *data)
2500{
2501 REQ (EIO_WD_CLOSE); req->wd = wd; SEND;
2502}
2503
1793eio_req *eio_nop (int pri, eio_cb cb, void *data) 2504eio_req *eio_nop (int pri, eio_cb cb, void *data)
1794{ 2505{
1795 REQ (EIO_NOP); SEND; 2506 REQ (EIO_NOP); SEND;
1796} 2507}
1797 2508
… …
1813eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2524eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1814{ 2525{
1815 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2526 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1816} 2527}
1817 2528
2529eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2530{
2531 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2532}
2533
2534eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
2535{
2536 REQ (EIO_SYNCFS); req->int1 = fd; SEND;
2537}
2538
2539eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2540{
2541 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2542}
2543
1818eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2544eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1819{ 2545{
1820 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2546 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1821} 2547}
1822 2548
… …
1828eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data) 2554eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
1829{ 2555{
1830 REQ (EIO_MLOCKALL); req->int1 = flags; SEND; 2556 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
1831} 2557}
1832 2558
1833eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) 2559eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
1834{ 2560{
1835 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; 2561 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} 2562}
1842 2563
1843eio_req *eio_close (int fd, int pri, eio_cb cb, void *data) 2564eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
1844{ 2565{
1845 REQ (EIO_CLOSE); req->int1 = fd; SEND; 2566 REQ (EIO_CLOSE); req->int1 = fd; SEND;
… …
1883eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) 2604eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1884{ 2605{
1885 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2606 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1886} 2607}
1887 2608
1888eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2609eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1889{ 2610{
1890 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2611 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1891} 2612}
1892 2613
1893eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) 2614eio_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) 2634eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1914{ 2635{
1915 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2636 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1916} 2637}
1917 2638
1918eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2639eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1919{ 2640{
1920 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2641 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1921} 2642}
1922 2643
1923eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) 2644eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
… …
1937} 2658}
1938 2659
1939eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data) 2660eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1940{ 2661{
1941 return eio__1path (EIO_READLINK, path, pri, cb, data); 2662 return eio__1path (EIO_READLINK, path, pri, cb, data);
2663}
2664
2665eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2666{
2667 return eio__1path (EIO_REALPATH, path, pri, cb, data);
1942} 2668}
1943 2669
1944eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) 2670eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1945{ 2671{
1946 return eio__1path (EIO_STAT, path, pri, cb, data); 2672 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) 2731eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2006{ 2732{
2007 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2733 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
2008} 2734}
2009 2735
2010eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) 2736eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
2011{ 2737{
2012 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND; 2738 REQ (EIO_CUSTOM); req->feed = execute; SEND;
2013} 2739}
2014 2740
2015#endif 2741#endif
2016 2742
2017eio_req *eio_grp (eio_cb cb, void *data) 2743eio_req *eio_grp (eio_cb cb, void *data)
… …
2026#undef SEND 2752#undef SEND
2027 2753
2028/*****************************************************************************/ 2754/*****************************************************************************/
2029/* grp functions */ 2755/* grp functions */
2030 2756
2757void
2031void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2758eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
2032{ 2759{
2033 grp->int2 = limit; 2760 grp->int2 = limit;
2034 grp->feed = feed; 2761 grp->feed = feed;
2035 2762
2036 grp_try_feed (grp); 2763 grp_try_feed (grp);
2037} 2764}
2038 2765
2766void
2039void eio_grp_limit (eio_req *grp, int limit) 2767eio_grp_limit (eio_req *grp, int limit)
2040{ 2768{
2041 grp->int2 = limit; 2769 grp->int2 = limit;
2042 2770
2043 grp_try_feed (grp); 2771 grp_try_feed (grp);
2044} 2772}
2045 2773
2774void
2046void eio_grp_add (eio_req *grp, eio_req *req) 2775eio_grp_add (eio_req *grp, eio_req *req)
2047{ 2776{
2048 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2777 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
2049 2778
2050 grp->flags |= EIO_FLAG_GROUPADD; 2779 grp->flags |= EIO_FLAG_GROUPADD;
2051 2780
… …
2062} 2791}
2063 2792
2064/*****************************************************************************/ 2793/*****************************************************************************/
2065/* misc garbage */ 2794/* misc garbage */
2066 2795
2796eio_ssize_t
2067ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2797eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2068{ 2798{
2069 etp_worker wrk;
2070 ssize_t ret;
2071
2072 wrk.dbuf = 0;
2073
2074 ret = eio__sendfile (ofd, ifd, offset, count, &wrk); 2799 return eio__sendfile (ofd, ifd, offset, count);
2075
2076 if (wrk.dbuf)
2077 free (wrk.dbuf);
2078
2079 return ret;
2080} 2800}
2081 2801

Diff Legend

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