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 1.70 by root, Fri Jun 10 06:55:10 2011 UTC vs.
Revision 1.118 by root, Mon Apr 2 17:53:27 2012 UTC

1/* 1/*
2 * libeio implementation 2 * libeio implementation
3 * 3 *
4 * Copyright (c) 2007,2008,2009,2010,2011 Marc Alexander Lehmann <libeio@schmorp.de> 4 * Copyright (c) 2007,2008,2009,2010,2011,2012 Marc Alexander Lehmann <libeio@schmorp.de>
5 * All rights reserved. 5 * All rights reserved.
6 * 6 *
7 * Redistribution and use in source and binary forms, with or without modifica- 7 * Redistribution and use in source and binary forms, with or without modifica-
8 * tion, are permitted provided that the following conditions are met: 8 * tion, are permitted provided that the following conditions are met:
9 * 9 *
… …
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__ || __NetBSD__ || __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
937 errno = 0; 1105 FUBd;
1106
1107 /* linux's readahead basically only fails for EBADF or EINVAL (not mmappable) */
1108 /* but not for e.g. EIO or eof, so we also never fail */
938 return count; 1109 return 0;
939} 1110}
940 1111
941#endif 1112#endif
942 1113
943/* sendfile always needs emulation */ 1114/* sendfile always needs emulation */
944static ssize_t 1115static eio_ssize_t
945eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) 1116eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
946{ 1117{
947 ssize_t written = 0; 1118 eio_ssize_t written = 0;
948 ssize_t res; 1119 eio_ssize_t res;
949 1120
950 if (!count) 1121 if (!count)
951 return 0; 1122 return 0;
952 1123
953 for (;;) 1124 for (;;)
954 { 1125 {
1126#ifdef __APPLE__
1127# undef HAVE_SENDFILE /* broken, as everything on os x */
1128#endif
955#if HAVE_SENDFILE 1129#if HAVE_SENDFILE
956# if __linux 1130# if __linux
957 off_t soffset = offset; 1131 off_t soffset = offset;
958 res = sendfile (ofd, ifd, &soffset, count); 1132 res = sendfile (ofd, ifd, &soffset, count);
959 1133
… …
975 1149
976 /* according to source inspection, this is correct, and useful behaviour */ 1150 /* according to source inspection, this is correct, and useful behaviour */
977 if (sbytes) 1151 if (sbytes)
978 res = sbytes; 1152 res = sbytes;
979 1153
980# elif defined (__APPLE__) 1154# elif defined __APPLE__
981 off_t sbytes = count; 1155 off_t sbytes = count;
982 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0); 1156 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
983 1157
984 /* according to the manpage, sbytes is always valid */ 1158 /* according to the manpage, sbytes is always valid */
985 if (sbytes) 1159 if (sbytes)
… …
1002 if (res < 0 && sbytes) 1176 if (res < 0 && sbytes)
1003 res = sbytes; 1177 res = sbytes;
1004 1178
1005# endif 1179# endif
1006 1180
1007#elif defined (_WIN32) 1181#elif defined (_WIN32) && 0
1008 /* does not work, just for documentation of what would need to be done */ 1182 /* does not work, just for documentation of what would need to be done */
1009 /* actually, cannot be done like this, as TransmitFile changes the file offset, */ 1183 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
1010 /* libeio guarantees that the file offset does not change, and windows */ 1184 /* libeio guarantees that the file offset does not change, and windows */
1011 /* has no way to get an independent handle to the same file description */ 1185 /* has no way to get an independent handle to the same file description */
1012 HANDLE h = TO_SOCKET (ifd); 1186 HANDLE h = TO_SOCKET (ifd);
… …
1045 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1219 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1046 /* BSDs */ 1220 /* BSDs */
1047#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */ 1221#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1048 || errno == ENOTSUP 1222 || errno == ENOTSUP
1049#endif 1223#endif
1224#ifdef EOPNOTSUPP /* windows */
1050 || errno == EOPNOTSUPP /* BSDs */ 1225 || errno == EOPNOTSUPP /* BSDs */
1226#endif
1051#if __solaris 1227#if __solaris
1052 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1228 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
1053#endif 1229#endif
1054 ) 1230 )
1055 ) 1231 )
… …
1059 1235
1060 res = 0; 1236 res = 0;
1061 1237
1062 while (count) 1238 while (count)
1063 { 1239 {
1064 ssize_t cnt; 1240 eio_ssize_t cnt;
1065 1241
1066 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 1242 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
1067 1243
1068 if (cnt <= 0) 1244 if (cnt <= 0)
1069 { 1245 {
… …
1081 1257
1082 offset += cnt; 1258 offset += cnt;
1083 res += cnt; 1259 res += cnt;
1084 count -= cnt; 1260 count -= cnt;
1085 } 1261 }
1262
1263 FUBd;
1086 } 1264 }
1087 1265
1088 return res; 1266 return res;
1267}
1268
1269#ifdef PAGESIZE
1270# define eio_pagesize() PAGESIZE
1271#else
1272static intptr_t
1273eio_pagesize (void)
1274{
1275 static intptr_t page;
1276
1277 if (!page)
1278 page = sysconf (_SC_PAGESIZE);
1279
1280 return page;
1281}
1282#endif
1283
1284static void
1285eio_page_align (void **addr, size_t *length)
1286{
1287 intptr_t mask = eio_pagesize () - 1;
1288
1289 /* round down addr */
1290 intptr_t adj = mask & (intptr_t)*addr;
1291
1292 *addr = (void *)((intptr_t)*addr - adj);
1293 *length += adj;
1294
1295 /* round up length */
1296 *length = (*length + mask) & ~mask;
1297}
1298
1299#if !_POSIX_MEMLOCK
1300# define eio__mlockall(a) EIO_ENOSYS ()
1301#else
1302
1303static int
1304eio__mlockall (int flags)
1305{
1306 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1307 extern int mallopt (int, int);
1308 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1309 #endif
1310
1311 if (EIO_MCL_CURRENT != MCL_CURRENT
1312 || EIO_MCL_FUTURE != MCL_FUTURE)
1313 {
1314 flags = 0
1315 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1316 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1317 }
1318
1319 return mlockall (flags);
1320}
1321#endif
1322
1323#if !_POSIX_MEMLOCK_RANGE
1324# define eio__mlock(a,b) EIO_ENOSYS ()
1325#else
1326
1327static int
1328eio__mlock (void *addr, size_t length)
1329{
1330 eio_page_align (&addr, &length);
1331
1332 return mlock (addr, length);
1333}
1334
1335#endif
1336
1337#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1338# define eio__msync(a,b,c) EIO_ENOSYS ()
1339#else
1340
1341static int
1342eio__msync (void *mem, size_t len, int flags)
1343{
1344 eio_page_align (&mem, &len);
1345
1346 if (EIO_MS_ASYNC != MS_SYNC
1347 || EIO_MS_INVALIDATE != MS_INVALIDATE
1348 || EIO_MS_SYNC != MS_SYNC)
1349 {
1350 flags = 0
1351 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1352 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1353 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1354 }
1355
1356 return msync (mem, len, flags);
1357}
1358
1359#endif
1360
1361static int
1362eio__mtouch (eio_req *req)
1363{
1364 void *mem = req->ptr2;
1365 size_t len = req->size;
1366 int flags = req->int1;
1367
1368 eio_page_align (&mem, &len);
1369
1370 {
1371 intptr_t addr = (intptr_t)mem;
1372 intptr_t end = addr + len;
1373 intptr_t page = eio_pagesize ();
1374
1375 if (addr < end)
1376 if (flags & EIO_MT_MODIFY) /* modify */
1377 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1378 else
1379 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1380 }
1381
1382 return 0;
1383}
1384
1385/*****************************************************************************/
1386/* requests implemented outside eio_execute, because they are so large */
1387
1388static void
1389eio__lseek (eio_req *req)
1390{
1391 /* this usually gets optimised away completely, or your compiler sucks, */
1392 /* or the whence constants really are not 0, 1, 2 */
1393 int whence = req->int2 == EIO_SEEK_SET ? SEEK_SET
1394 : req->int2 == EIO_SEEK_CUR ? SEEK_CUR
1395 : req->int2 == EIO_SEEK_END ? SEEK_END
1396 : req->int2;
1397
1398 req->offs = lseek (req->int1, req->offs, whence);
1399 req->result = req->offs == (off_t)-1 ? -1 : 0;
1400}
1401
1402/* result will always end up in tmpbuf, there is always space for adding a 0-byte */
1403static int
1404eio__realpath (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
1405{
1406 const char *rel = path;
1407 char *res;
1408 char *tmp1, *tmp2;
1409#if SYMLOOP_MAX > 32
1410 int symlinks = SYMLOOP_MAX;
1411#else
1412 int symlinks = 32;
1413#endif
1414
1415 errno = EINVAL;
1416 if (!rel)
1417 return -1;
1418
1419 errno = ENOENT;
1420 if (!*rel)
1421 return -1;
1422
1423 res = tmpbuf_get (tmpbuf, PATH_MAX * 3);
1424 tmp1 = res + PATH_MAX;
1425 tmp2 = tmp1 + PATH_MAX;
1426
1427#if 0 /* disabled, the musl way to do things is just too racy */
1428#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1429 /* on linux we may be able to ask the kernel */
1430 {
1431 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1432
1433 if (fd >= 0)
1434 {
1435 sprintf (tmp1, "/proc/self/fd/%d", fd);
1436 req->result = readlink (tmp1, res, PATH_MAX);
1437 close (fd);
1438
1439 /* here we should probably stat the open file and the disk file, to make sure they still match */
1440
1441 if (req->result > 0)
1442 goto done;
1443 }
1444 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1445 return;
1446 }
1447#endif
1448#endif
1449
1450 if (*rel != '/')
1451 {
1452 int len;
1453
1454 errno = ENOENT;
1455 if (wd == EIO_INVALID_WD)
1456 return -1;
1457
1458 if (wd == EIO_CWD)
1459 {
1460 if (!getcwd (res, PATH_MAX))
1461 return -1;
1462
1463 len = strlen (res);
1464 }
1465 else
1466 memcpy (res, wd->str, len = wd->len);
1467
1468 if (res [1]) /* only use if not / */
1469 res += len;
1470 }
1471
1472 while (*rel)
1473 {
1474 eio_ssize_t len, linklen;
1475 const char *beg = rel;
1476
1477 while (*rel && *rel != '/')
1478 ++rel;
1479
1480 len = rel - beg;
1481
1482 if (!len) /* skip slashes */
1483 {
1484 ++rel;
1485 continue;
1486 }
1487
1488 if (beg [0] == '.')
1489 {
1490 if (len == 1)
1491 continue; /* . - nop */
1492
1493 if (beg [1] == '.' && len == 2)
1494 {
1495 /* .. - back up one component, if possible */
1496
1497 while (res != tmpbuf->ptr)
1498 if (*--res == '/')
1499 break;
1500
1501 continue;
1502 }
1503 }
1504
1505 errno = ENAMETOOLONG;
1506 if (res + 1 + len + 1 >= tmp1)
1507 return -1;
1508
1509 /* copy one component */
1510 *res = '/';
1511 memcpy (res + 1, beg, len);
1512
1513 /* zero-terminate, for readlink */
1514 res [len + 1] = 0;
1515
1516 /* now check if it's a symlink */
1517 linklen = readlink (tmpbuf->ptr, tmp1, PATH_MAX);
1518
1519 if (linklen < 0)
1520 {
1521 if (errno != EINVAL)
1522 return -1;
1523
1524 /* it's a normal directory. hopefully */
1525 res += len + 1;
1526 }
1527 else
1528 {
1529 /* yay, it was a symlink - build new path in tmp2 */
1530 int rellen = strlen (rel);
1531
1532 errno = ENAMETOOLONG;
1533 if (linklen + 1 + rellen >= PATH_MAX)
1534 return -1;
1535
1536 errno = ELOOP;
1537 if (!--symlinks)
1538 return -1;
1539
1540 if (*tmp1 == '/')
1541 res = tmpbuf->ptr; /* symlink resolves to an absolute path */
1542
1543 /* we need to be careful, as rel might point into tmp2 already */
1544 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1545 tmp2 [linklen] = '/';
1546 memcpy (tmp2, tmp1, linklen);
1547
1548 rel = tmp2;
1549 }
1550 }
1551
1552 /* special case for the lone root path */
1553 if (res == tmpbuf->ptr)
1554 *res++ = '/';
1555
1556 return res - (char *)tmpbuf->ptr;
1089} 1557}
1090 1558
1091static signed char 1559static signed char
1092eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) 1560eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1093{ 1561{
… …
1101 1569
1102#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */ 1570#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1103#define EIO_SORT_FAST 60 /* when to only use insertion sort */ 1571#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1104 1572
1105static void 1573static void
1106eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1574eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1107{ 1575{
1108 unsigned char bits [9 + sizeof (ino_t) * 8]; 1576 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1109 unsigned char *bit = bits; 1577 unsigned char *bit = bits;
1110 1578
1111 assert (CHAR_BIT == 8); 1579 assert (CHAR_BIT == 8);
1112 assert (sizeof (eio_dirent) * 8 < 256); 1580 assert (sizeof (eio_dirent) * 8 < 256);
1113 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */ 1581 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
… …
1115 1583
1116 if (size <= EIO_SORT_FAST) 1584 if (size <= EIO_SORT_FAST)
1117 return; 1585 return;
1118 1586
1119 /* first prepare an array of bits to test in our radix sort */ 1587 /* first prepare an array of bits to test in our radix sort */
1120 /* try to take endianness into account, as well as differences in ino_t sizes */ 1588 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1121 /* inode_bits must contain all inodes ORed together */ 1589 /* inode_bits must contain all inodes ORed together */
1122 /* which is used to skip bits that are 0 everywhere, which is very common */ 1590 /* which is used to skip bits that are 0 everywhere, which is very common */
1123 { 1591 {
1124 ino_t endianness; 1592 eio_ino_t endianness;
1125 int i, j; 1593 int i, j;
1126 1594
1127 /* we store the byte offset of byte n into byte n of "endianness" */ 1595 /* we store the byte offset of byte n into byte n of "endianness" */
1128 for (i = 0; i < sizeof (ino_t); ++i) 1596 for (i = 0; i < sizeof (eio_ino_t); ++i)
1129 ((unsigned char *)&endianness)[i] = i; 1597 ((unsigned char *)&endianness)[i] = i;
1130 1598
1131 *bit++ = 0; 1599 *bit++ = 0;
1132 1600
1133 for (i = 0; i < sizeof (ino_t); ++i) 1601 for (i = 0; i < sizeof (eio_ino_t); ++i)
1134 { 1602 {
1135 /* shifting off the byte offsets out of "endianness" */ 1603 /* shifting off the byte offsets out of "endianness" */
1136 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8; 1604 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1137 endianness >>= 8; 1605 endianness >>= 8;
1138 1606
1139 for (j = 0; j < 8; ++j) 1607 for (j = 0; j < 8; ++j)
1140 if (inode_bits & (((ino_t)1) << (i * 8 + j))) 1608 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1141 *bit++ = offs + j; 1609 *bit++ = offs + j;
1142 } 1610 }
1143 1611
1144 for (j = 0; j < 8; ++j) 1612 for (j = 0; j < 8; ++j)
1145 if (score_bits & (1 << j)) 1613 if (score_bits & (1 << j))
1146 *bit++ = offsetof (eio_dirent, score) * 8 + j; 1614 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1147 } 1615 }
1148 1616
1149 /* now actually do the sorting (a variant of MSD radix sort) */ 1617 /* now actually do the sorting (a variant of MSD radix sort) */
1150 { 1618 {
1151 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base; 1619 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1152 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end; 1620 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1153 unsigned char *bit_stk [9 + sizeof (ino_t) * 8]; 1621 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1154 int stk_idx = 0; 1622 int stk_idx = 0;
1155 1623
1156 base_stk [stk_idx] = dents; 1624 base_stk [stk_idx] = dents;
1157 end_stk [stk_idx] = dents + size; 1625 end_stk [stk_idx] = dents + size;
1158 bit_stk [stk_idx] = bit - 1; 1626 bit_stk [stk_idx] = bit - 1;
… …
1237 } 1705 }
1238 } 1706 }
1239} 1707}
1240 1708
1241static void 1709static void
1242eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1710eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1243{ 1711{
1244 if (size <= 1) 1712 if (size <= 1)
1245 return; /* our insertion sort relies on size > 0 */ 1713 return; /* our insertion sort relies on size > 0 */
1246 1714
1247 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */ 1715 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
… …
1255 1723
1256/* read a full directory */ 1724/* read a full directory */
1257static void 1725static void
1258eio__scandir (eio_req *req, etp_worker *self) 1726eio__scandir (eio_req *req, etp_worker *self)
1259{ 1727{
1260 DIR *dirp;
1261 EIO_STRUCT_DIRENT *entp;
1262 char *name, *names; 1728 char *name, *names;
1263 int namesalloc = 4096; 1729 int namesalloc = 4096 - sizeof (void *) * 4;
1264 int namesoffs = 0; 1730 int namesoffs = 0;
1265 int flags = req->int1; 1731 int flags = req->int1;
1266 eio_dirent *dents = 0; 1732 eio_dirent *dents = 0;
1267 int dentalloc = 128; 1733 int dentalloc = 128;
1268 int dentoffs = 0; 1734 int dentoffs = 0;
1269 ino_t inode_bits = 0; 1735 eio_ino_t inode_bits = 0;
1736#ifdef _WIN32
1737 HANDLE dirp;
1738 WIN32_FIND_DATA entp;
1739#else
1740 DIR *dirp;
1741 EIO_STRUCT_DIRENT *entp;
1742#endif
1270 1743
1271 req->result = -1; 1744 req->result = -1;
1272 1745
1273 if (!(flags & EIO_READDIR_DENTS)) 1746 if (!(flags & EIO_READDIR_DENTS))
1274 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); 1747 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1275 1748
1276 X_LOCK (wrklock); 1749#ifdef _WIN32
1277 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 1750 {
1751 int len = strlen ((const char *)req->ptr1);
1752 char *path = malloc (MAX_PATH);
1753 const char *fmt;
1754 const char *reqpath = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1755
1756 if (!len)
1757 fmt = "./*";
1758 else if (reqpath[len - 1] == '/' || reqpath[len - 1] == '\\')
1759 fmt = "%s*";
1760 else
1761 fmt = "%s/*";
1762
1763 _snprintf (path, MAX_PATH, fmt, reqpath);
1764 dirp = FindFirstFile (path, &entp);
1765 free (path);
1766
1767 if (dirp == INVALID_HANDLE_VALUE)
1768 {
1769 /* should steal _dosmaperr */
1770 switch (GetLastError ())
1771 {
1772 case ERROR_FILE_NOT_FOUND:
1773 req->result = 0;
1774 break;
1775
1776 case ERROR_INVALID_NAME:
1777 case ERROR_PATH_NOT_FOUND:
1778 case ERROR_NO_MORE_FILES:
1779 errno = ENOENT;
1780 break;
1781
1782 case ERROR_NOT_ENOUGH_MEMORY:
1783 errno = ENOMEM;
1784 break;
1785
1786 default:
1787 errno = EINVAL;
1788 break;
1789 }
1790
1791 return;
1792 }
1793 }
1794#else
1795 #if HAVE_AT
1796 if (req->wd)
1797 {
1798 int fd = openat (WD2FD (req->wd), req->ptr1, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1799
1800 if (fd < 0)
1801 return;
1802
1803 dirp = fdopendir (fd);
1804
1805 if (!dirp)
1806 close (fd);
1807 }
1808 else
1278 self->dirp = dirp = opendir (req->ptr1); 1809 dirp = opendir (req->ptr1);
1810 #else
1811 dirp = opendir (wd_expand (&self->tmpbuf, req->wd, req->ptr1));
1812 #endif
1813
1814 if (!dirp)
1815 return;
1816#endif
1817
1818 if (req->flags & EIO_FLAG_PTR1_FREE)
1819 free (req->ptr1);
1279 1820
1280 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1821 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1281 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; 1822 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1282 req->ptr2 = names = malloc (namesalloc); 1823 req->ptr2 = names = malloc (namesalloc);
1283 X_UNLOCK (wrklock);
1284 1824
1285 if (dirp && names && (!flags || dents)) 1825 if (!names || (flags && !dents))
1826 return;
1827
1286 for (;;) 1828 for (;;)
1287 { 1829 {
1830 int done;
1831
1832#ifdef _WIN32
1833 done = !dirp;
1834#else
1288 errno = 0; 1835 errno = 0;
1289 entp = readdir (dirp); 1836 entp = readdir (dirp);
1837 done = !entp;
1838#endif
1290 1839
1291 if (!entp) 1840 if (done)
1292 { 1841 {
1842#ifndef _WIN32
1843 int old_errno = errno;
1844 closedir (dirp);
1845 errno = old_errno;
1846
1293 if (errno) 1847 if (errno)
1294 break; 1848 break;
1849#endif
1295 1850
1296 /* sort etc. */ 1851 /* sort etc. */
1297 req->int1 = flags; 1852 req->int1 = flags;
1298 req->result = dentoffs; 1853 req->result = dentoffs;
1299 1854
1300 if (flags & EIO_READDIR_STAT_ORDER) 1855 if (flags & EIO_READDIR_STAT_ORDER)
1301 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits); 1856 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1302 else if (flags & EIO_READDIR_DIRS_FIRST) 1857 else if (flags & EIO_READDIR_DIRS_FIRST)
1303 if (flags & EIO_READDIR_FOUND_UNKNOWN) 1858 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1304 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */ 1859 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1305 else 1860 else
1861 {
1862 /* in this case, all is known, and we just put dirs first and sort them */
1863 eio_dirent *oth = dents + dentoffs;
1864 eio_dirent *dir = dents;
1865
1866 /* now partition dirs to the front, and non-dirs to the back */
1867 /* by walking from both sides and swapping if necessary */
1868 while (oth > dir)
1869 {
1870 if (dir->type == EIO_DT_DIR)
1871 ++dir;
1872 else if ((--oth)->type == EIO_DT_DIR)
1873 {
1874 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1875
1876 ++dir;
1877 }
1878 }
1879
1880 /* now sort the dirs only (dirs all have the same score) */
1881 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1882 }
1883
1884 break;
1885 }
1886
1887 /* now add the entry to our list(s) */
1888 name = D_NAME (entp);
1889
1890 /* skip . and .. entries */
1891 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1892 {
1893 int len = D_NAMLEN (entp) + 1;
1894
1895 while (ecb_expect_false (namesoffs + len > namesalloc))
1896 {
1897 namesalloc *= 2;
1898 req->ptr2 = names = realloc (names, namesalloc);
1899
1900 if (!names)
1901 break;
1902 }
1903
1904 memcpy (names + namesoffs, name, len);
1905
1906 if (dents)
1907 {
1908 struct eio_dirent *ent;
1909
1910 if (ecb_expect_false (dentoffs == dentalloc))
1306 { 1911 {
1307 /* in this case, all is known, and we just put dirs first and sort them */ 1912 dentalloc *= 2;
1913 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1914
1915 if (!dents)
1916 break;
1917 }
1918
1308 eio_dirent *oth = dents + dentoffs; 1919 ent = dents + dentoffs;
1309 eio_dirent *dir = dents;
1310 1920
1311 /* now partition dirs to the front, and non-dirs to the back */ 1921 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1312 /* by walking from both sides and swapping if necessary */ 1922 ent->namelen = len - 1;
1313 while (oth > dir) 1923 ent->inode = D_INO (entp);
1924
1925 inode_bits |= ent->inode;
1926
1927 switch (D_TYPE (entp))
1928 {
1929 default:
1930 ent->type = EIO_DT_UNKNOWN;
1931 flags |= EIO_READDIR_FOUND_UNKNOWN;
1932 break;
1933
1934 #ifdef DT_FIFO
1935 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1936 #endif
1937 #ifdef DT_CHR
1938 case DT_CHR: ent->type = EIO_DT_CHR; break;
1939 #endif
1940 #ifdef DT_MPC
1941 case DT_MPC: ent->type = EIO_DT_MPC; break;
1942 #endif
1943 #ifdef DT_DIR
1944 case DT_DIR: ent->type = EIO_DT_DIR; break;
1945 #endif
1946 #ifdef DT_NAM
1947 case DT_NAM: ent->type = EIO_DT_NAM; break;
1948 #endif
1949 #ifdef DT_BLK
1950 case DT_BLK: ent->type = EIO_DT_BLK; break;
1951 #endif
1952 #ifdef DT_MPB
1953 case DT_MPB: ent->type = EIO_DT_MPB; break;
1954 #endif
1955 #ifdef DT_REG
1956 case DT_REG: ent->type = EIO_DT_REG; break;
1957 #endif
1958 #ifdef DT_NWK
1959 case DT_NWK: ent->type = EIO_DT_NWK; break;
1960 #endif
1961 #ifdef DT_CMP
1962 case DT_CMP: ent->type = EIO_DT_CMP; break;
1963 #endif
1964 #ifdef DT_LNK
1965 case DT_LNK: ent->type = EIO_DT_LNK; break;
1966 #endif
1967 #ifdef DT_SOCK
1968 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1969 #endif
1970 #ifdef DT_DOOR
1971 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1972 #endif
1973 #ifdef DT_WHT
1974 case DT_WHT: ent->type = EIO_DT_WHT; break;
1975 #endif
1976 }
1977
1978 ent->score = 7;
1979
1980 if (flags & EIO_READDIR_DIRS_FIRST)
1981 {
1982 if (ent->type == EIO_DT_UNKNOWN)
1314 { 1983 {
1315 if (dir->type == EIO_DT_DIR) 1984 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1316 ++dir; 1985 ent->score = 1;
1317 else if ((--oth)->type == EIO_DT_DIR) 1986 else if (!strchr (name, '.')) /* absence of dots indicate likely dirs */
1318 { 1987 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1319 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1320
1321 ++dir;
1322 }
1323 } 1988 }
1324 1989 else if (ent->type == EIO_DT_DIR)
1325 /* now sort the dirs only (dirs all have the same score) */ 1990 ent->score = 0;
1326 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1327 } 1991 }
1328
1329 break;
1330 }
1331
1332 /* now add the entry to our list(s) */
1333 name = entp->d_name;
1334
1335 /* skip . and .. entries */
1336 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1337 {
1338 int len = D_NAMLEN (entp) + 1;
1339
1340 while (expect_false (namesoffs + len > namesalloc))
1341 {
1342 namesalloc *= 2;
1343 X_LOCK (wrklock);
1344 req->ptr2 = names = realloc (names, namesalloc);
1345 X_UNLOCK (wrklock);
1346
1347 if (!names)
1348 break;
1349 } 1992 }
1350 1993
1351 memcpy (names + namesoffs, name, len);
1352
1353 if (dents)
1354 {
1355 struct eio_dirent *ent;
1356
1357 if (expect_false (dentoffs == dentalloc))
1358 {
1359 dentalloc *= 2;
1360 X_LOCK (wrklock);
1361 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1362 X_UNLOCK (wrklock);
1363
1364 if (!dents)
1365 break;
1366 }
1367
1368 ent = dents + dentoffs;
1369
1370 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1371 ent->namelen = len - 1;
1372 ent->inode = D_INO (entp);
1373
1374 inode_bits |= ent->inode;
1375
1376 switch (D_TYPE (entp))
1377 {
1378 default:
1379 ent->type = EIO_DT_UNKNOWN;
1380 flags |= EIO_READDIR_FOUND_UNKNOWN;
1381 break;
1382
1383 #ifdef DT_FIFO
1384 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1385 #endif
1386 #ifdef DT_CHR
1387 case DT_CHR: ent->type = EIO_DT_CHR; break;
1388 #endif
1389 #ifdef DT_MPC
1390 case DT_MPC: ent->type = EIO_DT_MPC; break;
1391 #endif
1392 #ifdef DT_DIR
1393 case DT_DIR: ent->type = EIO_DT_DIR; break;
1394 #endif
1395 #ifdef DT_NAM
1396 case DT_NAM: ent->type = EIO_DT_NAM; break;
1397 #endif
1398 #ifdef DT_BLK
1399 case DT_BLK: ent->type = EIO_DT_BLK; break;
1400 #endif
1401 #ifdef DT_MPB
1402 case DT_MPB: ent->type = EIO_DT_MPB; break;
1403 #endif
1404 #ifdef DT_REG
1405 case DT_REG: ent->type = EIO_DT_REG; break;
1406 #endif
1407 #ifdef DT_NWK
1408 case DT_NWK: ent->type = EIO_DT_NWK; break;
1409 #endif
1410 #ifdef DT_CMP
1411 case DT_CMP: ent->type = EIO_DT_CMP; break;
1412 #endif
1413 #ifdef DT_LNK
1414 case DT_LNK: ent->type = EIO_DT_LNK; break;
1415 #endif
1416 #ifdef DT_SOCK
1417 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1418 #endif
1419 #ifdef DT_DOOR
1420 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1421 #endif
1422 #ifdef DT_WHT
1423 case DT_WHT: ent->type = EIO_DT_WHT; break;
1424 #endif
1425 }
1426
1427 ent->score = 7;
1428
1429 if (flags & EIO_READDIR_DIRS_FIRST)
1430 {
1431 if (ent->type == EIO_DT_UNKNOWN)
1432 {
1433 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1434 ent->score = 1;
1435 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1436 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1437 }
1438 else if (ent->type == EIO_DT_DIR)
1439 ent->score = 0;
1440 }
1441 }
1442
1443 namesoffs += len; 1994 namesoffs += len;
1444 ++dentoffs; 1995 ++dentoffs;
1445 } 1996 }
1446 1997
1447 if (EIO_CANCELLED (req)) 1998 if (EIO_CANCELLED (req))
1448 { 1999 {
1449 errno = ECANCELED; 2000 errno = ECANCELED;
1450 break; 2001 break;
1451 } 2002 }
2003
2004#ifdef _WIN32
2005 if (!FindNextFile (dirp, &entp))
2006 {
2007 FindClose (dirp);
2008 dirp = 0;
1452 } 2009 }
2010#endif
2011 }
1453} 2012}
1454 2013
1455#ifdef PAGESIZE 2014/*****************************************************************************/
1456# define eio_pagesize() PAGESIZE 2015/* working directory stuff */
1457#else 2016/* various deficiencies in the posix 2008 api force us to */
1458static intptr_t 2017/* keep the absolute path in string form at all times */
1459eio_pagesize (void) 2018/* fuck yeah. */
1460{
1461 static intptr_t page;
1462 2019
1463 if (!page) 2020#if !HAVE_AT
1464 page = sysconf (_SC_PAGESIZE);
1465 2021
2022/* a bit like realpath, but usually faster because it doesn'T have to return */
2023/* an absolute or canonical path */
2024static const char *
2025wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
2026{
2027 if (!wd || *path == '/')
2028 return path;
2029
2030 if (path [0] == '.' && !path [1])
2031 return wd->str;
2032
2033 {
2034 int l1 = wd->len;
2035 int l2 = strlen (path);
2036
2037 char *res = tmpbuf_get (tmpbuf, l1 + l2 + 2);
2038
2039 memcpy (res, wd->str, l1);
2040 res [l1] = '/';
2041 memcpy (res + l1 + 1, path, l2 + 1);
2042
2043 return res;
2044 }
2045}
2046
2047#endif
2048
2049static eio_wd
2050eio__wd_open_sync (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
2051{
2052 int fd;
2053 eio_wd res;
2054 int len = eio__realpath (tmpbuf, wd, path);
2055
2056 if (len < 0)
2057 return EIO_INVALID_WD;
2058
2059#if HAVE_AT
2060 fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
2061
2062 if (fd < 0)
2063 return EIO_INVALID_WD;
2064#endif
2065
2066 res = malloc (sizeof (*res) + len); /* one extra 0-byte */
2067
2068#if HAVE_AT
2069 res->fd = fd;
2070#endif
2071
2072 res->len = len;
2073 memcpy (res->str, tmpbuf->ptr, len);
2074 res->str [len] = 0;
2075
1466 return page; 2076 return res;
1467} 2077}
1468#endif
1469 2078
1470static void 2079eio_wd
1471eio_page_align (void **addr, size_t *length) 2080eio_wd_open_sync (eio_wd wd, const char *path)
1472{ 2081{
1473 intptr_t mask = eio_pagesize () - 1; 2082 struct tmpbuf tmpbuf = { 0 };
2083 wd = eio__wd_open_sync (&tmpbuf, wd, path);
2084 free (tmpbuf.ptr);
1474 2085
1475 /* round down addr */ 2086 return wd;
1476 intptr_t adj = mask & (intptr_t)*addr;
1477
1478 *addr = (void *)((intptr_t)*addr - adj);
1479 *length += adj;
1480
1481 /* round up length */
1482 *length = (*length + mask) & ~mask;
1483} 2087}
1484 2088
1485#if !_POSIX_MEMLOCK 2089void
1486# define eio__mlockall(a) ((errno = ENOSYS), -1) 2090eio_wd_close_sync (eio_wd wd)
1487#else 2091{
2092 if (wd != EIO_INVALID_WD && wd != EIO_CWD)
2093 {
2094 #if HAVE_AT
2095 close (wd->fd);
2096 #endif
2097 free (wd);
2098 }
2099}
2100
2101#if HAVE_AT
2102
2103/* they forgot these */
1488 2104
1489static int 2105static int
1490eio__mlockall (int flags) 2106eio__truncateat (int dirfd, const char *path, off_t length)
1491{ 2107{
1492 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7 2108 int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
1493 extern int mallopt (int, int); 2109 int res;
1494 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1495 #endif
1496 2110
1497 if (EIO_MCL_CURRENT != MCL_CURRENT 2111 if (fd < 0)
1498 || EIO_MCL_FUTURE != MCL_FUTURE) 2112 return fd;
1499 {
1500 flags = 0
1501 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1502 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1503 }
1504 2113
1505 return mlockall (flags); 2114 res = ftruncate (fd, length);
2115 close (fd);
2116 return res;
1506} 2117}
1507#endif
1508
1509#if !_POSIX_MEMLOCK_RANGE
1510# define eio__mlock(a,b) ((errno = ENOSYS), -1)
1511#else
1512 2118
1513static int 2119static int
1514eio__mlock (void *addr, size_t length) 2120eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
1515{ 2121{
1516 eio_page_align (&addr, &length); 2122 int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
2123 int res;
1517 2124
1518 return mlock (addr, length); 2125 if (fd < 0)
1519} 2126 return fd;
1520 2127
1521#endif 2128 res = fstatvfs (fd, buf);
1522 2129 close (fd);
1523#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1524# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
1525#else
1526
1527int
1528eio__msync (void *mem, size_t len, int flags)
1529{
1530 eio_page_align (&mem, &len);
1531
1532 if (EIO_MS_ASYNC != MS_SYNC
1533 || EIO_MS_INVALIDATE != MS_INVALIDATE
1534 || EIO_MS_SYNC != MS_SYNC)
1535 {
1536 flags = 0
1537 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1538 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1539 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1540 }
1541
1542 return msync (mem, len, flags);
1543}
1544
1545#endif
1546
1547int
1548eio__mtouch (eio_req *req)
1549{
1550 void *mem = req->ptr2;
1551 size_t len = req->size;
1552 int flags = req->int1;
1553
1554 eio_page_align (&mem, &len);
1555
1556 {
1557 intptr_t addr = (intptr_t)mem;
1558 intptr_t end = addr + len;
1559 intptr_t page = eio_pagesize ();
1560
1561 if (addr < end)
1562 if (flags & EIO_MT_MODIFY) /* modify */
1563 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1564 else
1565 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1566 }
1567
1568 return 0; 2130 return res;
2131
1569} 2132}
2133
2134#endif
1570 2135
1571/*****************************************************************************/ 2136/*****************************************************************************/
1572 2137
1573#define ALLOC(len) \ 2138#define ALLOC(len) \
1574 if (!req->ptr2) \ 2139 if (!req->ptr2) \
… …
1583 req->result = -1; \ 2148 req->result = -1; \
1584 break; \ 2149 break; \
1585 } \ 2150 } \
1586 } 2151 }
1587 2152
2153static void ecb_noinline ecb_cold
2154etp_proc_init (void)
2155{
2156#if HAVE_PRCTL_SET_NAME
2157 /* provide a more sensible "thread name" */
2158 char name[16 + 1];
2159 const int namelen = sizeof (name) - 1;
2160 int len;
2161
2162 prctl (PR_GET_NAME, (unsigned long)name, 0, 0, 0);
2163 name [namelen] = 0;
2164 len = strlen (name);
2165 strcpy (name + (len <= namelen - 4 ? len : namelen - 4), "/eio");
2166 prctl (PR_SET_NAME, (unsigned long)name, 0, 0, 0);
2167#endif
2168}
2169
1588X_THREAD_PROC (etp_proc) 2170X_THREAD_PROC (etp_proc)
1589{ 2171{
1590 ETP_REQ *req; 2172 ETP_REQ *req;
1591 struct timespec ts; 2173 struct timespec ts;
1592 etp_worker *self = (etp_worker *)thr_arg; 2174 etp_worker *self = (etp_worker *)thr_arg;
1593 2175
2176 etp_proc_init ();
2177
1594 /* try to distribute timeouts somewhat randomly */ 2178 /* try to distribute timeouts somewhat evenly */
1595 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 2179 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1596 2180
1597 for (;;) 2181 for (;;)
1598 { 2182 {
2183 ts.tv_sec = 0;
2184
1599 X_LOCK (reqlock); 2185 X_LOCK (reqlock);
1600 2186
1601 for (;;) 2187 for (;;)
1602 { 2188 {
1603 self->req = req = reqq_shift (&req_queue); 2189 req = reqq_shift (&req_queue);
1604 2190
1605 if (req) 2191 if (req)
1606 break; 2192 break;
1607 2193
2194 if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
2195 {
2196 X_UNLOCK (reqlock);
2197 X_LOCK (wrklock);
2198 --started;
2199 X_UNLOCK (wrklock);
2200 goto quit;
2201 }
2202
1608 ++idle; 2203 ++idle;
1609 2204
1610 ts.tv_sec = time (0) + idle_timeout; 2205 if (idle <= max_idle)
1611 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 2206 /* we are allowed to idle, so do so without any timeout */
2207 X_COND_WAIT (reqwait, reqlock);
2208 else
1612 { 2209 {
1613 if (idle > max_idle) 2210 /* initialise timeout once */
1614 { 2211 if (!ts.tv_sec)
1615 --idle; 2212 ts.tv_sec = time (0) + idle_timeout;
1616 X_UNLOCK (reqlock);
1617 X_LOCK (wrklock);
1618 --started;
1619 X_UNLOCK (wrklock);
1620 goto quit;
1621 }
1622 2213
1623 /* we are allowed to idle, so do so without any timeout */
1624 X_COND_WAIT (reqwait, reqlock); 2214 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
2215 ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1625 } 2216 }
1626 2217
1627 --idle; 2218 --idle;
1628 } 2219 }
1629 2220
… …
1632 X_UNLOCK (reqlock); 2223 X_UNLOCK (reqlock);
1633 2224
1634 if (req->type < 0) 2225 if (req->type < 0)
1635 goto quit; 2226 goto quit;
1636 2227
1637 if (!EIO_CANCELLED (req))
1638 ETP_EXECUTE (self, req); 2228 ETP_EXECUTE (self, req);
1639 2229
1640 X_LOCK (reslock); 2230 X_LOCK (reslock);
1641 2231
1642 ++npending; 2232 ++npending;
1643 2233
1644 if (!reqq_push (&res_queue, req) && want_poll_cb) 2234 if (!reqq_push (&res_queue, req) && want_poll_cb)
1645 want_poll_cb (); 2235 want_poll_cb ();
1646 2236
1647 self->req = 0;
1648 etp_worker_clear (self); 2237 etp_worker_clear (self);
1649 2238
1650 X_UNLOCK (reslock); 2239 X_UNLOCK (reslock);
1651 } 2240 }
1652 2241
1653quit: 2242quit:
2243 free (req);
2244
1654 X_LOCK (wrklock); 2245 X_LOCK (wrklock);
1655 etp_worker_free (self); 2246 etp_worker_free (self);
1656 X_UNLOCK (wrklock); 2247 X_UNLOCK (wrklock);
1657 2248
1658 return 0; 2249 return 0;
1659} 2250}
1660 2251
1661/*****************************************************************************/ 2252/*****************************************************************************/
1662 2253
2254int ecb_cold
1663int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 2255eio_init (void (*want_poll)(void), void (*done_poll)(void))
1664{ 2256{
2257#if !HAVE_PREADWRITE
2258 X_MUTEX_CREATE (preadwritelock);
2259#endif
2260
1665 return etp_init (want_poll, done_poll); 2261 return etp_init (want_poll, done_poll);
1666} 2262}
1667 2263
2264ecb_inline void
1668static void eio_api_destroy (eio_req *req) 2265eio_api_destroy (eio_req *req)
1669{ 2266{
1670 free (req); 2267 free (req);
1671} 2268}
1672 2269
1673#define REQ(rtype) \ 2270#define REQ(rtype) \
… …
1692 { \ 2289 { \
1693 eio_api_destroy (req); \ 2290 eio_api_destroy (req); \
1694 return 0; \ 2291 return 0; \
1695 } 2292 }
1696 2293
2294static void
1697static void eio_execute (etp_worker *self, eio_req *req) 2295eio_execute (etp_worker *self, eio_req *req)
1698{ 2296{
2297#if HAVE_AT
2298 int dirfd;
2299#else
2300 const char *path;
2301#endif
2302
2303 if (ecb_expect_false (EIO_CANCELLED (req)))
2304 {
2305 req->result = -1;
2306 req->errorno = ECANCELED;
2307 return;
2308 }
2309
2310 if (ecb_expect_false (req->wd == EIO_INVALID_WD))
2311 {
2312 req->result = -1;
2313 req->errorno = ENOENT;
2314 return;
2315 }
2316
2317 if (req->type >= EIO_OPEN)
2318 {
2319 #if HAVE_AT
2320 dirfd = WD2FD (req->wd);
2321 #else
2322 path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
2323 #endif
2324 }
2325
1699 switch (req->type) 2326 switch (req->type)
1700 { 2327 {
2328 case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
2329 req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
2330 break;
2331 case EIO_WD_CLOSE: req->result = 0;
2332 eio_wd_close_sync (req->wd); break;
2333
2334 case EIO_SEEK: eio__lseek (req); break;
1701 case EIO_READ: ALLOC (req->size); 2335 case EIO_READ: ALLOC (req->size);
1702 req->result = req->offs >= 0 2336 req->result = req->offs >= 0
1703 ? pread (req->int1, req->ptr2, req->size, req->offs) 2337 ? pread (req->int1, req->ptr2, req->size, req->offs)
1704 : read (req->int1, req->ptr2, req->size); break; 2338 : read (req->int1, req->ptr2, req->size); break;
1705 case EIO_WRITE: req->result = req->offs >= 0 2339 case EIO_WRITE: req->result = req->offs >= 0
1706 ? pwrite (req->int1, req->ptr2, req->size, req->offs) 2340 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1707 : write (req->int1, req->ptr2, req->size); break; 2341 : write (req->int1, req->ptr2, req->size); break;
1708 2342
1709 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break; 2343 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1710 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break; 2344 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
2345
2346#if HAVE_AT
1711 2347
1712 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2348 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1713 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2349 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break;
1714 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2350 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2351 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, AT_SYMLINK_NOFOLLOW); break;
2352 case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break;
2353 case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break;
2354 case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break;
2355 case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break;
2356
2357 case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
2358 case EIO_RMDIR: req->result = unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break;
2359 case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); break;
2360 case EIO_RENAME: req->result = renameat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2); break;
2361 case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2, 0); break;
2362 case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
2363 case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
2364 case EIO_READLINK: ALLOC (PATH_MAX);
2365 req->result = readlinkat (dirfd, req->ptr1, req->ptr2, PATH_MAX); break;
2366 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2367 req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2368 case EIO_UTIME:
2369 case EIO_FUTIME:
2370 {
2371 struct timespec ts[2];
2372 struct timespec *times;
2373
2374 if (req->nv1 != -1. || req->nv2 != -1.)
2375 {
2376 ts[0].tv_sec = req->nv1;
2377 ts[0].tv_nsec = (req->nv1 - ts[0].tv_sec) * 1e9;
2378 ts[1].tv_sec = req->nv2;
2379 ts[1].tv_nsec = (req->nv2 - ts[1].tv_sec) * 1e9;
2380
2381 times = ts;
2382 }
2383 else
2384 times = 0;
2385
2386 req->result = req->type == EIO_FUTIME
2387 ? futimens (req->int1, times)
2388 : utimensat (dirfd, req->ptr1, times, 0);
2389 }
2390 break;
2391
2392#else
2393
2394 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2395 req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
2396 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1715 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2397 req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
2398 case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
2399 case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
2400 case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
2401 case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
2402
2403 case EIO_UNLINK: req->result = unlink (path ); break;
2404 case EIO_RMDIR: req->result = rmdir (path ); break;
2405 case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
2406 case EIO_RENAME: req->result = rename (path , req->ptr2); break;
2407 case EIO_LINK: req->result = link (path , req->ptr2); break;
2408 case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
2409 case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
2410 case EIO_READLINK: ALLOC (PATH_MAX);
2411 req->result = readlink (path, req->ptr2, PATH_MAX); break;
2412 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2413 req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
2414
2415 case EIO_UTIME:
2416 case EIO_FUTIME:
2417 {
2418 struct timeval tv[2];
2419 struct timeval *times;
2420
2421 if (req->nv1 != -1. || req->nv2 != -1.)
2422 {
2423 tv[0].tv_sec = req->nv1;
2424 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
2425 tv[1].tv_sec = req->nv2;
2426 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
2427
2428 times = tv;
2429 }
2430 else
2431 times = 0;
2432
2433 req->result = req->type == EIO_FUTIME
2434 ? futimes (req->int1, times)
2435 : utimes (req->ptr1, times);
2436 }
2437 break;
2438
2439#endif
2440
2441 case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
2442 {
2443 ALLOC (req->result);
2444 memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
2445 }
2446 break;
2447
1716 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2448 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1717 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; 2449 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1718 2450
1719 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1720 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1721 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS)); 2451 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1722 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break; 2452 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1723 2453
1724 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1725 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; 2454 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1726 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1727 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break; 2455 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1728 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1729 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break; 2456 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1730 2457
1731 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
1732 case EIO_CLOSE: req->result = close (req->int1); break; 2458 case EIO_CLOSE: req->result = close (req->int1); break;
1733 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break; 2459 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1734 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
1735 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1736 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1737 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1738 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1739 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1740 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1741
1742 case EIO_READLINK: ALLOC (PATH_MAX);
1743 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1744
1745 case EIO_SYNC: req->result = 0; sync (); break; 2460 case EIO_SYNC: req->result = 0; sync (); break;
1746 case EIO_FSYNC: req->result = fsync (req->int1); break; 2461 case EIO_FSYNC: req->result = fsync (req->int1); break;
1747 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 2462 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
2463 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
2464 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1748 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break; 2465 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1749 case EIO_MTOUCH: req->result = eio__mtouch (req); break; 2466 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1750 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break; 2467 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1751 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break; 2468 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1752 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break; 2469 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
1753 2470
1754 case EIO_READDIR: eio__scandir (req, self); break; 2471 case EIO_READDIR: eio__scandir (req, self); break;
1755 2472
1756 case EIO_BUSY: 2473 case EIO_BUSY:
1757#ifdef _WIN32 2474#ifdef _WIN32
… …
1766 req->result = select (0, 0, 0, 0, &tv); 2483 req->result = select (0, 0, 0, 0, &tv);
1767 } 2484 }
1768#endif 2485#endif
1769 break; 2486 break;
1770 2487
1771 case EIO_UTIME:
1772 case EIO_FUTIME:
1773 {
1774 struct timeval tv[2];
1775 struct timeval *times;
1776
1777 if (req->nv1 != -1. || req->nv2 != -1.)
1778 {
1779 tv[0].tv_sec = req->nv1;
1780 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
1781 tv[1].tv_sec = req->nv2;
1782 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
1783
1784 times = tv;
1785 }
1786 else
1787 times = 0;
1788
1789 req->result = req->type == EIO_FUTIME
1790 ? futimes (req->int1, times)
1791 : utimes (req->ptr1, times);
1792 }
1793 break;
1794
1795 case EIO_GROUP: 2488 case EIO_GROUP:
1796 abort (); /* handled in eio_request */ 2489 abort (); /* handled in eio_request */
1797 2490
1798 case EIO_NOP: 2491 case EIO_NOP:
1799 req->result = 0; 2492 req->result = 0;
… …
1812 req->errorno = errno; 2505 req->errorno = errno;
1813} 2506}
1814 2507
1815#ifndef EIO_NO_WRAPPERS 2508#ifndef EIO_NO_WRAPPERS
1816 2509
2510eio_req *eio_wd_open (const char *path, int pri, eio_cb cb, void *data)
2511{
2512 REQ (EIO_WD_OPEN); PATH; SEND;
2513}
2514
2515eio_req *eio_wd_close (eio_wd wd, int pri, eio_cb cb, void *data)
2516{
2517 REQ (EIO_WD_CLOSE); req->wd = wd; SEND;
2518}
2519
1817eio_req *eio_nop (int pri, eio_cb cb, void *data) 2520eio_req *eio_nop (int pri, eio_cb cb, void *data)
1818{ 2521{
1819 REQ (EIO_NOP); SEND; 2522 REQ (EIO_NOP); SEND;
1820} 2523}
1821 2524
… …
1837eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2540eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1838{ 2541{
1839 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2542 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1840} 2543}
1841 2544
2545eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2546{
2547 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2548}
2549
2550eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
2551{
2552 REQ (EIO_SYNCFS); req->int1 = fd; SEND;
2553}
2554
2555eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2556{
2557 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2558}
2559
1842eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2560eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1843{ 2561{
1844 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2562 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1845} 2563}
1846 2564
… …
1852eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data) 2570eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
1853{ 2571{
1854 REQ (EIO_MLOCKALL); req->int1 = flags; SEND; 2572 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
1855} 2573}
1856 2574
1857eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) 2575eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
1858{ 2576{
1859 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; 2577 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
1860}
1861
1862eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1863{
1864 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1865} 2578}
1866 2579
1867eio_req *eio_close (int fd, int pri, eio_cb cb, void *data) 2580eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
1868{ 2581{
1869 REQ (EIO_CLOSE); req->int1 = fd; SEND; 2582 REQ (EIO_CLOSE); req->int1 = fd; SEND;
… …
1872eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data) 2585eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
1873{ 2586{
1874 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND; 2587 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
1875} 2588}
1876 2589
2590eio_req *eio_seek (int fd, off_t offset, int whence, int pri, eio_cb cb, void *data)
2591{
2592 REQ (EIO_SEEK); req->int1 = fd; req->offs = offset; req->int2 = whence; SEND;
2593}
2594
1877eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data) 2595eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1878{ 2596{
1879 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND; 2597 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1880} 2598}
1881 2599
… …
1907eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) 2625eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1908{ 2626{
1909 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2627 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1910} 2628}
1911 2629
1912eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2630eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1913{ 2631{
1914 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2632 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1915} 2633}
1916 2634
1917eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) 2635eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
… …
1937eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data) 2655eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1938{ 2656{
1939 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2657 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1940} 2658}
1941 2659
1942eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2660eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1943{ 2661{
1944 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2662 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1945} 2663}
1946 2664
1947eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) 2665eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
… …
1961} 2679}
1962 2680
1963eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data) 2681eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1964{ 2682{
1965 return eio__1path (EIO_READLINK, path, pri, cb, data); 2683 return eio__1path (EIO_READLINK, path, pri, cb, data);
2684}
2685
2686eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2687{
2688 return eio__1path (EIO_REALPATH, path, pri, cb, data);
1966} 2689}
1967 2690
1968eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) 2691eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1969{ 2692{
1970 return eio__1path (EIO_STAT, path, pri, cb, data); 2693 return eio__1path (EIO_STAT, path, pri, cb, data);
… …
2029eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2752eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2030{ 2753{
2031 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2754 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
2032} 2755}
2033 2756
2034eio_req *eio_custom (void (*)(eio_req *) execute, int pri, eio_cb cb, void *data); 2757eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
2035{ 2758{
2036 REQ (EIO_CUSTOM); req->feed = execute; SEND; 2759 REQ (EIO_CUSTOM); req->feed = execute; SEND;
2037} 2760}
2038 2761
2039#endif 2762#endif
… …
2050#undef SEND 2773#undef SEND
2051 2774
2052/*****************************************************************************/ 2775/*****************************************************************************/
2053/* grp functions */ 2776/* grp functions */
2054 2777
2778void
2055void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2779eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
2056{ 2780{
2057 grp->int2 = limit; 2781 grp->int2 = limit;
2058 grp->feed = feed; 2782 grp->feed = feed;
2059 2783
2060 grp_try_feed (grp); 2784 grp_try_feed (grp);
2061} 2785}
2062 2786
2787void
2063void eio_grp_limit (eio_req *grp, int limit) 2788eio_grp_limit (eio_req *grp, int limit)
2064{ 2789{
2065 grp->int2 = limit; 2790 grp->int2 = limit;
2066 2791
2067 grp_try_feed (grp); 2792 grp_try_feed (grp);
2068} 2793}
2069 2794
2795void
2070void eio_grp_add (eio_req *grp, eio_req *req) 2796eio_grp_add (eio_req *grp, eio_req *req)
2071{ 2797{
2072 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2798 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
2073 2799
2074 grp->flags |= EIO_FLAG_GROUPADD; 2800 grp->flags |= EIO_FLAG_GROUPADD;
2075 2801
… …
2086} 2812}
2087 2813
2088/*****************************************************************************/ 2814/*****************************************************************************/
2089/* misc garbage */ 2815/* misc garbage */
2090 2816
2817eio_ssize_t
2091ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2818eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2092{ 2819{
2093 etp_worker wrk;
2094 ssize_t ret;
2095
2096 wrk.dbuf = 0;
2097
2098 ret = eio__sendfile (ofd, ifd, offset, count, &wrk); 2820 return eio__sendfile (ofd, ifd, offset, count);
2099
2100 if (wrk.dbuf)
2101 free (wrk.dbuf);
2102
2103 return ret;
2104} 2821}
2105 2822

Diff Legend

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