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 1.33 by root, Sat Jun 6 19:44:17 2009 UTC vs.
Revision 1.107 by root, Tue Sep 27 00:41:52 2011 UTC

1/* 1/*
2 * libeio implementation 2 * libeio implementation
3 * 3 *
4 * Copyright (c) 2007,2008,2009 Marc Alexander Lehmann <libeio@schmorp.de> 4 * Copyright (c) 2007,2008,2009,2010,2011 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 *
… …
35 * and other provisions required by the GPL. If you do not delete the 35 * and other provisions required by the GPL. If you do not delete the
36 * provisions above, a recipient may use your version of this file under 36 * provisions above, a recipient may use your version of this file under
37 * either the BSD or the GPL. 37 * either the BSD or the GPL.
38 */ 38 */
39 39
40#ifndef _WIN32
41# include "config.h"
42#endif
43
40#include "eio.h" 44#include "eio.h"
45#include "ecb.h"
41 46
42#ifdef EIO_STACKSIZE 47#ifdef EIO_STACKSIZE
43# define XTHREAD_STACKSIZE EIO_STACKSIZE 48# define X_STACKSIZE EIO_STACKSIZE
44#endif 49#endif
45#include "xthread.h" 50#include "xthread.h"
46 51
47#include <errno.h> 52#include <errno.h>
48#include <stddef.h> 53#include <stddef.h>
… …
53#include <sys/stat.h> 58#include <sys/stat.h>
54#include <limits.h> 59#include <limits.h>
55#include <fcntl.h> 60#include <fcntl.h>
56#include <assert.h> 61#include <assert.h>
57 62
63/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
64/* intptr_t only comes from stdint.h, says idiot openbsd coder */
65#if HAVE_STDINT_H
66# include <stdint.h>
67#endif
68
69#ifndef ECANCELED
70# define ECANCELED EDOM
71#endif
72#ifndef ELOOP
73# define ELOOP EDOM
74#endif
75
76#if !defined(ENOTSOCK) && defined(WSAENOTSOCK)
77# define ENOTSOCK WSAENOTSOCK
78#endif
79
80static void eio_destroy (eio_req *req);
81
58#ifndef EIO_FINISH 82#ifndef EIO_FINISH
59# 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
60#endif 84#endif
61 85
62#ifndef EIO_DESTROY 86#ifndef EIO_DESTROY
… …
65 89
66#ifndef EIO_FEED 90#ifndef EIO_FEED
67# 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)
68#endif 92#endif
69 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
70#ifdef _WIN32 105#ifdef _WIN32
71 106
72 /*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
73#else 197#else
74 198
75# include "config.h"
76# include <sys/time.h> 199 #include <sys/time.h>
77# include <sys/select.h> 200 #include <sys/select.h>
201 #include <sys/statvfs.h>
78# include <unistd.h> 202 #include <unistd.h>
203 #include <signal.h>
204 #include <dirent.h>
205
206 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
207 #include <sys/mman.h>
208 #endif
209
210 #define D_NAME(entp) entp->d_name
211
212 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
213 #if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
214 #define _DIRENT_HAVE_D_TYPE /* sigh */
215 #define D_INO(de) (de)->d_fileno
216 #define D_NAMLEN(de) (de)->d_namlen
217 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
218 #define D_INO(de) (de)->d_ino
219 #endif
220
221 #ifdef _D_EXACT_NAMLEN
222 #undef D_NAMLEN
223 #define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
224 #endif
225
226 #ifdef _DIRENT_HAVE_D_TYPE
227 #define D_TYPE(de) (de)->d_type
228 #endif
229
230 #ifndef EIO_STRUCT_DIRENT
231 #define EIO_STRUCT_DIRENT struct dirent
232 #endif
233
234#endif
235
236#if HAVE_UTIMES
79# include <utime.h> 237# include <utime.h>
80# include <signal.h>
81# include <dirent.h>
82
83/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
84# if defined(__FreeBSD__) || defined(__NetBSD__) || defined(__OpenBSD__)
85# define _DIRENT_HAVE_D_TYPE /* sigh */
86# define D_INO(de) (de)->d_fileno
87# define D_NAMLEN(de) (de)->d_namlen
88# elif defined(__linux) || defined(d_ino) || _XOPEN_SOURCE >= 600
89# define D_INO(de) (de)->d_ino
90# endif 238#endif
91 239
92#ifdef _D_EXACT_NAMLEN 240#if HAVE_SYS_SYSCALL_H
93# undef D_NAMLEN 241# include <sys/syscall.h>
94# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
95#endif
96
97# ifdef _DIRENT_HAVE_D_TYPE
98# define D_TYPE(de) (de)->d_type
99# endif 242#endif
100 243
101# ifndef EIO_STRUCT_DIRENT 244#if HAVE_SYS_PRCTL_H
102# define EIO_STRUCT_DIRENT struct dirent 245# include <sys/prctl.h>
103# endif
104
105#endif 246#endif
106 247
107#if HAVE_SENDFILE 248#if HAVE_SENDFILE
108# if __linux 249# if __linux
109# include <sys/sendfile.h> 250# include <sys/sendfile.h>
110# elif __freebsd 251# elif __FreeBSD__ || defined __APPLE__
111# include <sys/socket.h> 252# include <sys/socket.h>
112# include <sys/uio.h> 253# include <sys/uio.h>
113# elif __hpux 254# elif __hpux
114# include <sys/socket.h> 255# include <sys/socket.h>
115# elif __solaris /* not yet */ 256# elif __solaris
116# include <sys/sendfile.h> 257# include <sys/sendfile.h>
117# else 258# else
118# error sendfile support requested but not available 259# error sendfile support requested but not available
119# endif 260# endif
120#endif 261#endif
… …
124#endif 265#endif
125#ifndef D_INO 266#ifndef D_INO
126# define D_INO(de) 0 267# define D_INO(de) 0
127#endif 268#endif
128#ifndef D_NAMLEN 269#ifndef D_NAMLEN
129# define D_NAMLEN(de) strlen ((de)->d_name) 270# define D_NAMLEN(entp) strlen (D_NAME (entp))
130#endif 271#endif
131
132/* number of seconds after which an idle threads exit */
133#define IDLE_TIMEOUT 10
134 272
135/* used for struct dirent, AIX doesn't provide it */ 273/* used for struct dirent, AIX doesn't provide it */
136#ifndef NAME_MAX 274#ifndef NAME_MAX
137# define NAME_MAX 4096 275# define NAME_MAX 4096
138#endif 276#endif
139 277
278/* used for readlink etc. */
279#ifndef PATH_MAX
280# define PATH_MAX 4096
281#endif
282
140/* buffer size for various temporary buffers */ 283/* buffer size for various temporary buffers */
141#define EIO_BUFSIZE 65536 284#define EIO_BUFSIZE 65536
142 285
143#define dBUF \ 286#define dBUF \
144 char *eio_buf; \
145 ETP_WORKER_LOCK (self); \
146 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 287 char *eio_buf = malloc (EIO_BUFSIZE); \
147 ETP_WORKER_UNLOCK (self); \
148 errno = ENOMEM; \ 288 errno = ENOMEM; \
149 if (!eio_buf) \ 289 if (!eio_buf) \
150 return -1; 290 return -1
291
292#define FUBd \
293 free (eio_buf)
151 294
152#define EIO_TICKS ((1000000 + 1023) >> 10) 295#define EIO_TICKS ((1000000 + 1023) >> 10)
153 296
154/*****************************************************************************/ 297/*****************************************************************************/
155 298
156#if __GNUC__ >= 3 299struct tmpbuf
157# 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
158#else 325#else
159# 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);
160#endif 328#endif
161 329
162#define expect_false(expr) expect ((expr) != 0, 0) 330struct eio_pwd
163#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};
164 338
165/*****************************************************************************/ 339/*****************************************************************************/
166 340
167#define ETP_PRI_MIN EIO_PRI_MIN 341#define ETP_PRI_MIN EIO_PRI_MIN
168#define ETP_PRI_MAX EIO_PRI_MAX 342#define ETP_PRI_MAX EIO_PRI_MAX
… …
174static int eio_finish (eio_req *req); 348static int eio_finish (eio_req *req);
175#define ETP_FINISH(req) eio_finish (req) 349#define ETP_FINISH(req) eio_finish (req)
176static void eio_execute (struct etp_worker *self, eio_req *req); 350static void eio_execute (struct etp_worker *self, eio_req *req);
177#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) 351#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
178 352
179#define ETP_WORKER_CLEAR(req) \
180 if (wrk->dbuf) \
181 { \
182 free (wrk->dbuf); \
183 wrk->dbuf = 0; \
184 } \
185 \
186 if (wrk->dirp) \
187 { \
188 closedir (wrk->dirp); \
189 wrk->dirp = 0; \
190 }
191
192#define ETP_WORKER_COMMON \
193 void *dbuf; \
194 DIR *dirp;
195
196/*****************************************************************************/ 353/*****************************************************************************/
197 354
198#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) 355#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
199 356
200/* calculcate time difference in ~1/EIO_TICKS of a second */ 357/* calculate time difference in ~1/EIO_TICKS of a second */
358ecb_inline int
201static int tvdiff (struct timeval *tv1, struct timeval *tv2) 359tvdiff (struct timeval *tv1, struct timeval *tv2)
202{ 360{
203 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 361 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
204 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 362 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
205} 363}
206 364
… …
210static void (*done_poll_cb) (void); 368static void (*done_poll_cb) (void);
211 369
212static unsigned int max_poll_time; /* reslock */ 370static unsigned int max_poll_time; /* reslock */
213static unsigned int max_poll_reqs; /* reslock */ 371static unsigned int max_poll_reqs; /* reslock */
214 372
215static volatile unsigned int nreqs; /* reqlock */ 373static unsigned int nreqs; /* reqlock */
216static volatile unsigned int nready; /* reqlock */ 374static unsigned int nready; /* reqlock */
217static volatile unsigned int npending; /* reqlock */ 375static unsigned int npending; /* reqlock */
218static volatile unsigned int max_idle = 4; 376static unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
377static unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
219 378
220static mutex_t wrklock = X_MUTEX_INIT; 379static xmutex_t wrklock;
221static mutex_t reslock = X_MUTEX_INIT; 380static xmutex_t reslock;
222static mutex_t reqlock = X_MUTEX_INIT; 381static xmutex_t reqlock;
223static cond_t reqwait = X_COND_INIT; 382static xcond_t reqwait;
224 383
225#if !HAVE_PREADWRITE 384#if !HAVE_PREADWRITE
226/* 385/*
227 * make our pread/pwrite emulation safe against themselves, but not against 386 * make our pread/pwrite emulation safe against themselves, but not against
228 * 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,
229 * but that's your problem, not mine. 388 * but that's your problem, not mine.
230 */ 389 */
231static mutex_t preadwritelock = X_MUTEX_INIT; 390static xmutex_t preadwritelock;
232#endif 391#endif
233 392
234typedef struct etp_worker 393typedef struct etp_worker
235{ 394{
395 struct tmpbuf tmpbuf;
396
236 /* locked by wrklock */ 397 /* locked by wrklock */
237 struct etp_worker *prev, *next; 398 struct etp_worker *prev, *next;
238 399
239 thread_t tid; 400 xthread_t tid;
240 401
241 /* locked by reslock, reqlock or wrklock */ 402#ifdef ETP_WORKER_COMMON
242 ETP_REQ *req; /* currently processed request */
243
244 ETP_WORKER_COMMON 403 ETP_WORKER_COMMON
404#endif
245} etp_worker; 405} etp_worker;
246 406
247static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */ 407static etp_worker wrk_first; /* NOT etp */
248 408
249#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock) 409#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
250#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock) 410#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
251 411
252/* worker threads management */ 412/* worker threads management */
253 413
414static void ecb_cold
254static void etp_worker_clear (etp_worker *wrk) 415etp_worker_clear (etp_worker *wrk)
255{ 416{
256 ETP_WORKER_CLEAR (wrk);
257} 417}
258 418
419static void ecb_cold
259static void etp_worker_free (etp_worker *wrk) 420etp_worker_free (etp_worker *wrk)
260{ 421{
422 free (wrk->tmpbuf.ptr);
423
261 wrk->next->prev = wrk->prev; 424 wrk->next->prev = wrk->prev;
262 wrk->prev->next = wrk->next; 425 wrk->prev->next = wrk->next;
263 426
264 free (wrk); 427 free (wrk);
265} 428}
266 429
267static unsigned int etp_nreqs (void) 430static unsigned int
431etp_nreqs (void)
268{ 432{
269 int retval; 433 int retval;
270 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 434 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
271 retval = nreqs; 435 retval = nreqs;
272 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 436 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
273 return retval; 437 return retval;
274} 438}
275 439
276static unsigned int etp_nready (void) 440static unsigned int
441etp_nready (void)
277{ 442{
278 unsigned int retval; 443 unsigned int retval;
279 444
280 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 445 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
281 retval = nready; 446 retval = nready;
282 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 447 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
283 448
284 return retval; 449 return retval;
285} 450}
286 451
287static unsigned int etp_npending (void) 452static unsigned int
453etp_npending (void)
288{ 454{
289 unsigned int retval; 455 unsigned int retval;
290 456
291 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 457 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
292 retval = npending; 458 retval = npending;
293 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 459 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
294 460
295 return retval; 461 return retval;
296} 462}
297 463
298static unsigned int etp_nthreads (void) 464static unsigned int
465etp_nthreads (void)
299{ 466{
300 unsigned int retval; 467 unsigned int retval;
301 468
302 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 469 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
303 retval = started; 470 retval = started;
… …
317} etp_reqq; 484} etp_reqq;
318 485
319static etp_reqq req_queue; 486static etp_reqq req_queue;
320static etp_reqq res_queue; 487static etp_reqq res_queue;
321 488
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
322static int reqq_push (etp_reqq *q, ETP_REQ *req) 501reqq_push (etp_reqq *q, ETP_REQ *req)
323{ 502{
324 int pri = req->pri; 503 int pri = req->pri;
325 req->next = 0; 504 req->next = 0;
326 505
327 if (q->qe[pri]) 506 if (q->qe[pri])
… …
333 q->qe[pri] = q->qs[pri] = req; 512 q->qe[pri] = q->qs[pri] = req;
334 513
335 return q->size++; 514 return q->size++;
336} 515}
337 516
517static ETP_REQ * ecb_noinline
338static ETP_REQ *reqq_shift (etp_reqq *q) 518reqq_shift (etp_reqq *q)
339{ 519{
340 int pri; 520 int pri;
341 521
342 if (!q->size) 522 if (!q->size)
343 return 0; 523 return 0;
… …
358 } 538 }
359 539
360 abort (); 540 abort ();
361} 541}
362 542
363static void etp_atfork_prepare (void) 543static int ecb_cold
544etp_init (void (*want_poll)(void), void (*done_poll)(void))
364{ 545{
365 X_LOCK (wrklock); 546 X_MUTEX_CREATE (wrklock);
366 X_LOCK (reqlock); 547 X_MUTEX_CREATE (reslock);
367 X_LOCK (reslock); 548 X_MUTEX_CREATE (reqlock);
368#if !HAVE_PREADWRITE 549 X_COND_CREATE (reqwait);
369 X_LOCK (preadwritelock);
370#endif
371}
372 550
373static void etp_atfork_parent (void) 551 reqq_init (&req_queue);
374{ 552 reqq_init (&res_queue);
375#if !HAVE_PREADWRITE
376 X_UNLOCK (preadwritelock);
377#endif
378 X_UNLOCK (reslock);
379 X_UNLOCK (reqlock);
380 X_UNLOCK (wrklock);
381}
382 553
383static void etp_atfork_child (void) 554 wrk_first.next =
384{ 555 wrk_first.prev = &wrk_first;
385 ETP_REQ *prv;
386
387 while ((prv = reqq_shift (&req_queue)))
388 ETP_DESTROY (prv);
389
390 while ((prv = reqq_shift (&res_queue)))
391 ETP_DESTROY (prv);
392
393 while (wrk_first.next != &wrk_first)
394 {
395 etp_worker *wrk = wrk_first.next;
396
397 if (wrk->req)
398 ETP_DESTROY (wrk->req);
399
400 etp_worker_clear (wrk);
401 etp_worker_free (wrk);
402 }
403 556
404 started = 0; 557 started = 0;
405 idle = 0; 558 idle = 0;
406 nreqs = 0; 559 nreqs = 0;
407 nready = 0; 560 nready = 0;
408 npending = 0; 561 npending = 0;
409 562
410 etp_atfork_parent ();
411}
412
413static void
414etp_once_init (void)
415{
416 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
417}
418
419static int
420etp_init (void (*want_poll)(void), void (*done_poll)(void))
421{
422 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
423
424 pthread_once (&doinit, etp_once_init);
425
426 want_poll_cb = want_poll; 563 want_poll_cb = want_poll;
427 done_poll_cb = done_poll; 564 done_poll_cb = done_poll;
428 565
429 return 0; 566 return 0;
430} 567}
431 568
432X_THREAD_PROC (etp_proc); 569X_THREAD_PROC (etp_proc);
433 570
571static void ecb_cold
434static void etp_start_thread (void) 572etp_start_thread (void)
435{ 573{
436 etp_worker *wrk = calloc (1, sizeof (etp_worker)); 574 etp_worker *wrk = calloc (1, sizeof (etp_worker));
437 575
438 /*TODO*/ 576 /*TODO*/
439 assert (("unable to allocate worker thread data", wrk)); 577 assert (("unable to allocate worker thread data", wrk));
… …
452 free (wrk); 590 free (wrk);
453 591
454 X_UNLOCK (wrklock); 592 X_UNLOCK (wrklock);
455} 593}
456 594
595static void
457static void etp_maybe_start_thread (void) 596etp_maybe_start_thread (void)
458{ 597{
459 if (expect_true (etp_nthreads () >= wanted)) 598 if (ecb_expect_true (etp_nthreads () >= wanted))
460 return; 599 return;
461 600
462 /* todo: maybe use idle here, but might be less exact */ 601 /* todo: maybe use idle here, but might be less exact */
463 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 ()))
464 return; 603 return;
465 604
466 etp_start_thread (); 605 etp_start_thread ();
467} 606}
468 607
608static void ecb_cold
469static void etp_end_thread (void) 609etp_end_thread (void)
470{ 610{
471 eio_req *req = calloc (1, sizeof (eio_req)); 611 eio_req *req = calloc (1, sizeof (eio_req)); /* will be freed by worker */
472 612
473 req->type = -1; 613 req->type = -1;
474 req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 614 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
475 615
476 X_LOCK (reqlock); 616 X_LOCK (reqlock);
… …
481 X_LOCK (wrklock); 621 X_LOCK (wrklock);
482 --started; 622 --started;
483 X_UNLOCK (wrklock); 623 X_UNLOCK (wrklock);
484} 624}
485 625
486static int etp_poll (void) 626static int
627etp_poll (void)
487{ 628{
488 unsigned int maxreqs; 629 unsigned int maxreqs;
489 unsigned int maxtime; 630 unsigned int maxtime;
490 struct timeval tv_start, tv_now; 631 struct timeval tv_start, tv_now;
491 632
… …
521 662
522 X_LOCK (reqlock); 663 X_LOCK (reqlock);
523 --nreqs; 664 --nreqs;
524 X_UNLOCK (reqlock); 665 X_UNLOCK (reqlock);
525 666
526 if (expect_false (req->type == EIO_GROUP && req->size)) 667 if (ecb_expect_false (req->type == EIO_GROUP && req->size))
527 { 668 {
528 req->int1 = 1; /* mark request as delayed */ 669 req->int1 = 1; /* mark request as delayed */
529 continue; 670 continue;
530 } 671 }
531 else 672 else
532 { 673 {
533 int res = ETP_FINISH (req); 674 int res = ETP_FINISH (req);
534 if (expect_false (res)) 675 if (ecb_expect_false (res))
535 return res; 676 return res;
536 } 677 }
537 678
538 if (expect_false (maxreqs && !--maxreqs)) 679 if (ecb_expect_false (maxreqs && !--maxreqs))
539 break; 680 break;
540 681
541 if (maxtime) 682 if (maxtime)
542 { 683 {
543 gettimeofday (&tv_now, 0); 684 gettimeofday (&tv_now, 0);
… …
549 690
550 errno = EAGAIN; 691 errno = EAGAIN;
551 return -1; 692 return -1;
552} 693}
553 694
695static void
554static void etp_cancel (ETP_REQ *req) 696etp_cancel (ETP_REQ *req)
555{ 697{
556 X_LOCK (wrklock); 698 req->cancelled = 1;
557 req->flags |= EIO_FLAG_CANCELLED;
558 X_UNLOCK (wrklock);
559 699
560 eio_grp_cancel (req); 700 eio_grp_cancel (req);
561} 701}
562 702
703static void
563static void etp_submit (ETP_REQ *req) 704etp_submit (ETP_REQ *req)
564{ 705{
565 req->pri -= ETP_PRI_MIN; 706 req->pri -= ETP_PRI_MIN;
566 707
567 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;
568 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;
569 710
570 if (expect_false (req->type == EIO_GROUP)) 711 if (ecb_expect_false (req->type == EIO_GROUP))
571 { 712 {
572 /* I hope this is worth it :/ */ 713 /* I hope this is worth it :/ */
573 X_LOCK (reqlock); 714 X_LOCK (reqlock);
574 ++nreqs; 715 ++nreqs;
575 X_UNLOCK (reqlock); 716 X_UNLOCK (reqlock);
… …
594 735
595 etp_maybe_start_thread (); 736 etp_maybe_start_thread ();
596 } 737 }
597} 738}
598 739
740static void ecb_cold
599static void etp_set_max_poll_time (double nseconds) 741etp_set_max_poll_time (double nseconds)
600{ 742{
601 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 743 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
602 max_poll_time = nseconds; 744 max_poll_time = nseconds * EIO_TICKS;
603 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 745 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
604} 746}
605 747
748static void ecb_cold
606static void etp_set_max_poll_reqs (unsigned int maxreqs) 749etp_set_max_poll_reqs (unsigned int maxreqs)
607{ 750{
608 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 751 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
609 max_poll_reqs = maxreqs; 752 max_poll_reqs = maxreqs;
610 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 753 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
611} 754}
612 755
756static void ecb_cold
613static void etp_set_max_idle (unsigned int nthreads) 757etp_set_max_idle (unsigned int nthreads)
614{ 758{
615 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 759 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
616 max_idle = nthreads <= 0 ? 1 : nthreads; 760 max_idle = nthreads;
617 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 761 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
618} 762}
619 763
764static void ecb_cold
765etp_set_idle_timeout (unsigned int seconds)
766{
767 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
768 idle_timeout = seconds;
769 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
770}
771
772static void ecb_cold
620static void etp_set_min_parallel (unsigned int nthreads) 773etp_set_min_parallel (unsigned int nthreads)
621{ 774{
622 if (wanted < nthreads) 775 if (wanted < nthreads)
623 wanted = nthreads; 776 wanted = nthreads;
624} 777}
625 778
779static void ecb_cold
626static void etp_set_max_parallel (unsigned int nthreads) 780etp_set_max_parallel (unsigned int nthreads)
627{ 781{
628 if (wanted > nthreads) 782 if (wanted > nthreads)
629 wanted = nthreads; 783 wanted = nthreads;
630 784
631 while (started > wanted) 785 while (started > wanted)
632 etp_end_thread (); 786 etp_end_thread ();
633} 787}
634 788
635/*****************************************************************************/ 789/*****************************************************************************/
636 790
791static void
637static void grp_try_feed (eio_req *grp) 792grp_try_feed (eio_req *grp)
638{ 793{
639 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 794 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
640 { 795 {
641 grp->flags &= ~EIO_FLAG_GROUPADD; 796 grp->flags &= ~EIO_FLAG_GROUPADD;
642 797
… …
649 break; 804 break;
650 } 805 }
651 } 806 }
652} 807}
653 808
809static int
654static int grp_dec (eio_req *grp) 810grp_dec (eio_req *grp)
655{ 811{
656 --grp->size; 812 --grp->size;
657 813
658 /* call feeder, if applicable */ 814 /* call feeder, if applicable */
659 grp_try_feed (grp); 815 grp_try_feed (grp);
… …
663 return eio_finish (grp); 819 return eio_finish (grp);
664 else 820 else
665 return 0; 821 return 0;
666} 822}
667 823
824static void
668void eio_destroy (eio_req *req) 825eio_destroy (eio_req *req)
669{ 826{
670 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); 827 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
671 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); 828 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
672 829
673 EIO_DESTROY (req); 830 EIO_DESTROY (req);
674} 831}
675 832
833static int
676static int eio_finish (eio_req *req) 834eio_finish (eio_req *req)
677{ 835{
678 int res = EIO_FINISH (req); 836 int res = EIO_FINISH (req);
679 837
680 if (req->grp) 838 if (req->grp)
681 { 839 {
… …
689 if (grp->grp_first == req) 847 if (grp->grp_first == req)
690 grp->grp_first = req->grp_next; 848 grp->grp_first = req->grp_next;
691 849
692 res2 = grp_dec (grp); 850 res2 = grp_dec (grp);
693 851
694 if (!res && res2) 852 if (!res)
695 res = res2; 853 res = res2;
696 } 854 }
697 855
698 eio_destroy (req); 856 eio_destroy (req);
699 857
700 return res; 858 return res;
701} 859}
702 860
861void
703void eio_grp_cancel (eio_req *grp) 862eio_grp_cancel (eio_req *grp)
704{ 863{
705 for (grp = grp->grp_first; grp; grp = grp->grp_next) 864 for (grp = grp->grp_first; grp; grp = grp->grp_next)
706 eio_cancel (grp); 865 eio_cancel (grp);
707} 866}
708 867
868void
709void eio_cancel (eio_req *req) 869eio_cancel (eio_req *req)
710{ 870{
711 etp_cancel (req); 871 etp_cancel (req);
712} 872}
713 873
874void
714void eio_submit (eio_req *req) 875eio_submit (eio_req *req)
715{ 876{
716 etp_submit (req); 877 etp_submit (req);
717} 878}
718 879
719unsigned int eio_nreqs (void) 880unsigned int
881eio_nreqs (void)
720{ 882{
721 return etp_nreqs (); 883 return etp_nreqs ();
722} 884}
723 885
724unsigned int eio_nready (void) 886unsigned int
887eio_nready (void)
725{ 888{
726 return etp_nready (); 889 return etp_nready ();
727} 890}
728 891
729unsigned int eio_npending (void) 892unsigned int
893eio_npending (void)
730{ 894{
731 return etp_npending (); 895 return etp_npending ();
732} 896}
733 897
734unsigned int eio_nthreads (void) 898unsigned int ecb_cold
899eio_nthreads (void)
735{ 900{
736 return etp_nthreads (); 901 return etp_nthreads ();
737} 902}
738 903
904void ecb_cold
739void eio_set_max_poll_time (double nseconds) 905eio_set_max_poll_time (double nseconds)
740{ 906{
741 etp_set_max_poll_time (nseconds); 907 etp_set_max_poll_time (nseconds);
742} 908}
743 909
910void ecb_cold
744void eio_set_max_poll_reqs (unsigned int maxreqs) 911eio_set_max_poll_reqs (unsigned int maxreqs)
745{ 912{
746 etp_set_max_poll_reqs (maxreqs); 913 etp_set_max_poll_reqs (maxreqs);
747} 914}
748 915
916void ecb_cold
749void eio_set_max_idle (unsigned int nthreads) 917eio_set_max_idle (unsigned int nthreads)
750{ 918{
751 etp_set_max_idle (nthreads); 919 etp_set_max_idle (nthreads);
752} 920}
753 921
922void ecb_cold
923eio_set_idle_timeout (unsigned int seconds)
924{
925 etp_set_idle_timeout (seconds);
926}
927
928void ecb_cold
754void eio_set_min_parallel (unsigned int nthreads) 929eio_set_min_parallel (unsigned int nthreads)
755{ 930{
756 etp_set_min_parallel (nthreads); 931 etp_set_min_parallel (nthreads);
757} 932}
758 933
934void ecb_cold
759void eio_set_max_parallel (unsigned int nthreads) 935eio_set_max_parallel (unsigned int nthreads)
760{ 936{
761 etp_set_max_parallel (nthreads); 937 etp_set_max_parallel (nthreads);
762} 938}
763 939
764int eio_poll (void) 940int eio_poll (void)
… …
773# undef pread 949# undef pread
774# undef pwrite 950# undef pwrite
775# define pread eio__pread 951# define pread eio__pread
776# define pwrite eio__pwrite 952# define pwrite eio__pwrite
777 953
778static ssize_t 954static eio_ssize_t
779eio__pread (int fd, void *buf, size_t count, off_t offset) 955eio__pread (int fd, void *buf, size_t count, off_t offset)
780{ 956{
781 ssize_t res; 957 eio_ssize_t res;
782 off_t ooffset; 958 off_t ooffset;
783 959
784 X_LOCK (preadwritelock); 960 X_LOCK (preadwritelock);
785 ooffset = lseek (fd, 0, SEEK_CUR); 961 ooffset = lseek (fd, 0, SEEK_CUR);
786 lseek (fd, offset, SEEK_SET); 962 lseek (fd, offset, SEEK_SET);
… …
789 X_UNLOCK (preadwritelock); 965 X_UNLOCK (preadwritelock);
790 966
791 return res; 967 return res;
792} 968}
793 969
794static ssize_t 970static eio_ssize_t
795eio__pwrite (int fd, void *buf, size_t count, off_t offset) 971eio__pwrite (int fd, void *buf, size_t count, off_t offset)
796{ 972{
797 ssize_t res; 973 eio_ssize_t res;
798 off_t ooffset; 974 off_t ooffset;
799 975
800 X_LOCK (preadwritelock); 976 X_LOCK (preadwritelock);
801 ooffset = lseek (fd, 0, SEEK_CUR); 977 ooffset = lseek (fd, 0, SEEK_CUR);
802 lseek (fd, offset, SEEK_SET); 978 lseek (fd, offset, SEEK_SET);
… …
806 982
807 return res; 983 return res;
808} 984}
809#endif 985#endif
810 986
811#ifndef HAVE_FUTIMES 987#ifndef HAVE_UTIMES
812 988
813# undef utimes 989# undef utimes
814# undef futimes
815# define utimes(path,times) eio__utimes (path, times) 990# define utimes(path,times) eio__utimes (path, times)
816# define futimes(fd,times) eio__futimes (fd, times)
817 991
818static int 992static int
819eio__utimes (const char *filename, const struct timeval times[2]) 993eio__utimes (const char *filename, const struct timeval times[2])
820{ 994{
821 if (times) 995 if (times)
… …
829 } 1003 }
830 else 1004 else
831 return utime (filename, 0); 1005 return utime (filename, 0);
832} 1006}
833 1007
1008#endif
1009
1010#ifndef HAVE_FUTIMES
1011
1012# undef futimes
1013# define futimes(fd,times) eio__futimes (fd, times)
1014
1015static int
834static int eio__futimes (int fd, const struct timeval tv[2]) 1016eio__futimes (int fd, const struct timeval tv[2])
835{ 1017{
836 errno = ENOSYS; 1018 errno = ENOSYS;
837 return -1; 1019 return -1;
838} 1020}
839 1021
… …
842#if !HAVE_FDATASYNC 1024#if !HAVE_FDATASYNC
843# undef fdatasync 1025# undef fdatasync
844# define fdatasync(fd) fsync (fd) 1026# define fdatasync(fd) fsync (fd)
845#endif 1027#endif
846 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
847/* sync_file_range always needs emulation */ 1047/* sync_file_range always needs emulation */
848int 1048static int
849eio__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)
850{ 1050{
851#if HAVE_SYNC_FILE_RANGE 1051#if HAVE_SYNC_FILE_RANGE
852 int res; 1052 int res;
853 1053
… …
866 if (!res || errno != ENOSYS) 1066 if (!res || errno != ENOSYS)
867 return res; 1067 return res;
868#endif 1068#endif
869 1069
870 /* 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
871 * call fdatasync, as thta matches the expectation of it's users best */ 1071 * call fdatasync, as that matches the expectation of its users best */
872 return fdatasync (fd); 1072 return fdatasync (fd);
1073}
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
873} 1084}
874 1085
875#if !HAVE_READAHEAD 1086#if !HAVE_READAHEAD
876# undef readahead 1087# undef readahead
877# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 1088# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
878 1089
879static ssize_t 1090static eio_ssize_t
880eio__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)
881{ 1092{
882 size_t todo = count; 1093 size_t todo = count;
883 dBUF; 1094 dBUF;
884 1095
… …
889 pread (fd, eio_buf, len, offset); 1100 pread (fd, eio_buf, len, offset);
890 offset += len; 1101 offset += len;
891 todo -= len; 1102 todo -= len;
892 } 1103 }
893 1104
1105 FUBd;
1106
894 errno = 0; 1107 errno = 0;
895 return count; 1108 return count;
896} 1109}
897 1110
898#endif 1111#endif
899 1112
900/* sendfile always needs emulation */ 1113/* sendfile always needs emulation */
901static ssize_t 1114static eio_ssize_t
902eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) 1115eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
903{ 1116{
1117 eio_ssize_t written = 0;
904 ssize_t res; 1118 eio_ssize_t res;
905 1119
906 if (!count) 1120 if (!count)
907 return 0; 1121 return 0;
908 1122
1123 for (;;)
1124 {
1125#ifdef __APPLE__
1126# undef HAVE_SENDFILE /* broken, as everything on os x */
1127#endif
909#if HAVE_SENDFILE 1128#if HAVE_SENDFILE
910# if __linux 1129# if __linux
1130 off_t soffset = offset;
911 res = sendfile (ofd, ifd, &offset, count); 1131 res = sendfile (ofd, ifd, &soffset, count);
912 1132
913# elif __freebsd 1133# elif __FreeBSD__
914 /* 1134 /*
915 * Of course, the freebsd sendfile is a dire hack with no thoughts 1135 * Of course, the freebsd sendfile is a dire hack with no thoughts
916 * wasted on making it similar to other I/O functions. 1136 * wasted on making it similar to other I/O functions.
917 */ 1137 */
918 {
919 off_t sbytes; 1138 off_t sbytes;
920 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 1139 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
921 1140
922 if (res < 0 && sbytes) 1141 #if 0 /* according to the manpage, this is correct, but broken behaviour */
923 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 1142 /* freebsd' sendfile will return 0 on success */
1143 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
1144 /* not on e.g. EIO or EPIPE - sounds broken */
1145 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
924 res = sbytes; 1146 res = sbytes;
925 } 1147 #endif
1148
1149 /* according to source inspection, this is correct, and useful behaviour */
1150 if (sbytes)
1151 res = sbytes;
1152
1153# elif defined (__APPLE__)
1154 off_t sbytes = count;
1155 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
1156
1157 /* according to the manpage, sbytes is always valid */
1158 if (sbytes)
1159 res = sbytes;
926 1160
927# elif __hpux 1161# elif __hpux
928 res = sendfile (ofd, ifd, offset, count, 0, 0); 1162 res = sendfile (ofd, ifd, offset, count, 0, 0);
929 1163
930# elif __solaris 1164# elif __solaris
931 {
932 struct sendfilevec vec; 1165 struct sendfilevec vec;
933 size_t sbytes; 1166 size_t sbytes;
934 1167
935 vec.sfv_fd = ifd; 1168 vec.sfv_fd = ifd;
936 vec.sfv_flag = 0; 1169 vec.sfv_flag = 0;
937 vec.sfv_off = offset; 1170 vec.sfv_off = offset;
938 vec.sfv_len = count; 1171 vec.sfv_len = count;
939 1172
940 res = sendfilev (ofd, &vec, 1, &sbytes); 1173 res = sendfilev (ofd, &vec, 1, &sbytes);
941 1174
942 if (res < 0 && sbytes) 1175 if (res < 0 && sbytes)
943 res = sbytes; 1176 res = sbytes;
944 }
945 1177
946# endif 1178# endif
1179
1180#elif defined (_WIN32) && 0
1181 /* does not work, just for documentation of what would need to be done */
1182 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
1183 /* libeio guarantees that the file offset does not change, and windows */
1184 /* has no way to get an independent handle to the same file description */
1185 HANDLE h = TO_SOCKET (ifd);
1186 SetFilePointer (h, offset, 0, FILE_BEGIN);
1187 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1188
947#else 1189#else
948 res = -1; 1190 res = -1;
949 errno = ENOSYS; 1191 errno = ENOSYS;
950#endif 1192#endif
951 1193
1194 /* we assume sendfile can copy at least 128mb in one go */
1195 if (res <= 128 * 1024 * 1024)
1196 {
1197 if (res > 0)
1198 written += res;
1199
1200 if (written)
1201 return written;
1202
1203 break;
1204 }
1205 else
1206 {
1207 /* if we requested more, then probably the kernel was lazy */
1208 written += res;
1209 offset += res;
1210 count -= res;
1211
1212 if (!count)
1213 return written;
1214 }
1215 }
1216
952 if (res < 0 1217 if (res < 0
953 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1218 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1219 /* BSDs */
1220#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1221 || errno == ENOTSUP
1222#endif
1223#ifdef EOPNOTSUPP /* windows */
1224 || errno == EOPNOTSUPP /* BSDs */
1225#endif
954#if __solaris 1226#if __solaris
955 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1227 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
956#endif 1228#endif
957 ) 1229 )
958 ) 1230 )
… …
962 1234
963 res = 0; 1235 res = 0;
964 1236
965 while (count) 1237 while (count)
966 { 1238 {
967 ssize_t cnt; 1239 eio_ssize_t cnt;
968 1240
969 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 1241 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
970 1242
971 if (cnt <= 0) 1243 if (cnt <= 0)
972 { 1244 {
… …
984 1256
985 offset += cnt; 1257 offset += cnt;
986 res += cnt; 1258 res += cnt;
987 count -= cnt; 1259 count -= cnt;
988 } 1260 }
1261
1262 FUBd;
989 } 1263 }
990 1264
991 return res; 1265 return res;
992} 1266}
993 1267
1268#ifdef PAGESIZE
1269# define eio_pagesize() PAGESIZE
1270#else
1271static intptr_t
1272eio_pagesize (void)
1273{
1274 static intptr_t page;
1275
1276 if (!page)
1277 page = sysconf (_SC_PAGESIZE);
1278
1279 return page;
1280}
1281#endif
1282
1283static void
1284eio_page_align (void **addr, size_t *length)
1285{
1286 intptr_t mask = eio_pagesize () - 1;
1287
1288 /* round down addr */
1289 intptr_t adj = mask & (intptr_t)*addr;
1290
1291 *addr = (void *)((intptr_t)*addr - adj);
1292 *length += adj;
1293
1294 /* round up length */
1295 *length = (*length + mask) & ~mask;
1296}
1297
1298#if !_POSIX_MEMLOCK
1299# define eio__mlockall(a) EIO_ENOSYS ()
1300#else
1301
994static int 1302static int
1303eio__mlockall (int flags)
1304{
1305 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1306 extern int mallopt (int, int);
1307 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1308 #endif
1309
1310 if (EIO_MCL_CURRENT != MCL_CURRENT
1311 || EIO_MCL_FUTURE != MCL_FUTURE)
1312 {
1313 flags = 0
1314 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1315 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1316 }
1317
1318 return mlockall (flags);
1319}
1320#endif
1321
1322#if !_POSIX_MEMLOCK_RANGE
1323# define eio__mlock(a,b) EIO_ENOSYS ()
1324#else
1325
1326static int
1327eio__mlock (void *addr, size_t length)
1328{
1329 eio_page_align (&addr, &length);
1330
1331 return mlock (addr, length);
1332}
1333
1334#endif
1335
1336#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1337# define eio__msync(a,b,c) EIO_ENOSYS ()
1338#else
1339
1340static int
1341eio__msync (void *mem, size_t len, int flags)
1342{
1343 eio_page_align (&mem, &len);
1344
1345 if (EIO_MS_ASYNC != MS_SYNC
1346 || EIO_MS_INVALIDATE != MS_INVALIDATE
1347 || EIO_MS_SYNC != MS_SYNC)
1348 {
1349 flags = 0
1350 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1351 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1352 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1353 }
1354
1355 return msync (mem, len, flags);
1356}
1357
1358#endif
1359
1360static int
1361eio__mtouch (eio_req *req)
1362{
1363 void *mem = req->ptr2;
1364 size_t len = req->size;
1365 int flags = req->int1;
1366
1367 eio_page_align (&mem, &len);
1368
1369 {
1370 intptr_t addr = (intptr_t)mem;
1371 intptr_t end = addr + len;
1372 intptr_t page = eio_pagesize ();
1373
1374 if (addr < end)
1375 if (flags & EIO_MT_MODIFY) /* modify */
1376 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1377 else
1378 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1379 }
1380
1381 return 0;
1382}
1383
1384/*****************************************************************************/
1385/* requests implemented outside eio_execute, because they are so large */
1386
1387/* result will always end up in tmpbuf, there is always space for adding a 0-byte */
1388static int
1389eio__realpath (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
1390{
1391 const char *rel = path;
1392 char *res;
1393 char *tmp1, *tmp2;
1394#if SYMLOOP_MAX > 32
1395 int symlinks = SYMLOOP_MAX;
1396#else
1397 int symlinks = 32;
1398#endif
1399
1400 errno = EINVAL;
1401 if (!rel)
1402 return -1;
1403
1404 errno = ENOENT;
1405 if (!*rel)
1406 return -1;
1407
1408 res = tmpbuf_get (tmpbuf, PATH_MAX * 3);
1409 tmp1 = res + PATH_MAX;
1410 tmp2 = tmp1 + PATH_MAX;
1411
1412#if 0 /* disabled, the musl way to do things is just too racy */
1413#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1414 /* on linux we may be able to ask the kernel */
1415 {
1416 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1417
1418 if (fd >= 0)
1419 {
1420 sprintf (tmp1, "/proc/self/fd/%d", fd);
1421 req->result = readlink (tmp1, res, PATH_MAX);
1422 close (fd);
1423
1424 /* here we should probably stat the open file and the disk file, to make sure they still match */
1425
1426 if (req->result > 0)
1427 goto done;
1428 }
1429 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1430 return;
1431 }
1432#endif
1433#endif
1434
1435 if (*rel != '/')
1436 {
1437 int len;
1438
1439 errno = ENOENT;
1440 if (wd == EIO_INVALID_WD)
1441 return -1;
1442
1443 if (wd == EIO_CWD)
1444 {
1445 if (!getcwd (res, PATH_MAX))
1446 return -1;
1447
1448 len = strlen (res);
1449 }
1450 else
1451 memcpy (res, wd->str, len = wd->len);
1452
1453 if (res [1]) /* only use if not / */
1454 res += len;
1455 }
1456
1457 while (*rel)
1458 {
1459 eio_ssize_t len, linklen;
1460 const char *beg = rel;
1461
1462 while (*rel && *rel != '/')
1463 ++rel;
1464
1465 len = rel - beg;
1466
1467 if (!len) /* skip slashes */
1468 {
1469 ++rel;
1470 continue;
1471 }
1472
1473 if (beg [0] == '.')
1474 {
1475 if (len == 1)
1476 continue; /* . - nop */
1477
1478 if (beg [1] == '.' && len == 2)
1479 {
1480 /* .. - back up one component, if possible */
1481
1482 while (res != tmpbuf->ptr)
1483 if (*--res == '/')
1484 break;
1485
1486 continue;
1487 }
1488 }
1489
1490 errno = ENAMETOOLONG;
1491 if (res + 1 + len + 1 >= tmp1)
1492 return;
1493
1494 /* copy one component */
1495 *res = '/';
1496 memcpy (res + 1, beg, len);
1497
1498 /* zero-terminate, for readlink */
1499 res [len + 1] = 0;
1500
1501 /* now check if it's a symlink */
1502 linklen = readlink (tmpbuf->ptr, tmp1, PATH_MAX);
1503
1504 if (linklen < 0)
1505 {
1506 if (errno != EINVAL)
1507 return -1;
1508
1509 /* it's a normal directory. hopefully */
1510 res += len + 1;
1511 }
1512 else
1513 {
1514 /* yay, it was a symlink - build new path in tmp2 */
1515 int rellen = strlen (rel);
1516
1517 errno = ENAMETOOLONG;
1518 if (linklen + 1 + rellen >= PATH_MAX)
1519 return -1;
1520
1521 errno = ELOOP;
1522 if (!--symlinks)
1523 return -1;
1524
1525 if (*tmp1 == '/')
1526 res = tmpbuf->ptr; /* symlink resolves to an absolute path */
1527
1528 /* we need to be careful, as rel might point into tmp2 already */
1529 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1530 tmp2 [linklen] = '/';
1531 memcpy (tmp2, tmp1, linklen);
1532
1533 rel = tmp2;
1534 }
1535 }
1536
1537 /* special case for the lone root path */
1538 if (res == tmpbuf->ptr)
1539 *res++ = '/';
1540
1541 return res - (char *)tmpbuf->ptr;
1542}
1543
1544static signed char
995eio_dent_cmp (const void *a_, const void *b_) 1545eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
996{ 1546{
997 const eio_dirent *a = (const eio_dirent *)a_; 1547 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
998 const eio_dirent *b = (const eio_dirent *)b_; 1548 : a->inode < b->inode ? -1
1549 : a->inode > b->inode ? 1
1550 : 0;
1551}
999 1552
1000 return (int)b->score - (int)a->score ? (int)b->score - (int)a->score 1553#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1001 : a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0; /* int might be < ino_t */ 1554
1555#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1556#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1557
1558static void
1559eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1560{
1561 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1562 unsigned char *bit = bits;
1563
1564 assert (CHAR_BIT == 8);
1565 assert (sizeof (eio_dirent) * 8 < 256);
1566 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1567 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1568
1569 if (size <= EIO_SORT_FAST)
1570 return;
1571
1572 /* first prepare an array of bits to test in our radix sort */
1573 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1574 /* inode_bits must contain all inodes ORed together */
1575 /* which is used to skip bits that are 0 everywhere, which is very common */
1576 {
1577 eio_ino_t endianness;
1578 int i, j;
1579
1580 /* we store the byte offset of byte n into byte n of "endianness" */
1581 for (i = 0; i < sizeof (eio_ino_t); ++i)
1582 ((unsigned char *)&endianness)[i] = i;
1583
1584 *bit++ = 0;
1585
1586 for (i = 0; i < sizeof (eio_ino_t); ++i)
1587 {
1588 /* shifting off the byte offsets out of "endianness" */
1589 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1590 endianness >>= 8;
1591
1592 for (j = 0; j < 8; ++j)
1593 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1594 *bit++ = offs + j;
1595 }
1596
1597 for (j = 0; j < 8; ++j)
1598 if (score_bits & (1 << j))
1599 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1600 }
1601
1602 /* now actually do the sorting (a variant of MSD radix sort) */
1603 {
1604 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1605 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1606 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1607 int stk_idx = 0;
1608
1609 base_stk [stk_idx] = dents;
1610 end_stk [stk_idx] = dents + size;
1611 bit_stk [stk_idx] = bit - 1;
1612
1613 do
1614 {
1615 base = base_stk [stk_idx];
1616 end = end_stk [stk_idx];
1617 bit = bit_stk [stk_idx];
1618
1619 for (;;)
1620 {
1621 unsigned char O = *bit >> 3;
1622 unsigned char M = 1 << (*bit & 7);
1623
1624 eio_dirent *a = base;
1625 eio_dirent *b = end;
1626
1627 if (b - a < EIO_SORT_CUTOFF)
1628 break;
1629
1630 /* now bit-partition the array on the bit */
1631 /* this ugly asymmetric loop seems to perform much better than typical */
1632 /* partition algos found in the literature */
1633 do
1634 if (!(((unsigned char *)a)[O] & M))
1635 ++a;
1636 else if (!(((unsigned char *)--b)[O] & M))
1637 {
1638 eio_dirent tmp = *a; *a = *b; *b = tmp;
1639 ++a;
1640 }
1641 while (b > a);
1642
1643 /* next bit, or stop, if no bits left in this path */
1644 if (!*--bit)
1645 break;
1646
1647 base_stk [stk_idx] = a;
1648 end_stk [stk_idx] = end;
1649 bit_stk [stk_idx] = bit;
1650 ++stk_idx;
1651
1652 end = a;
1653 }
1654 }
1655 while (stk_idx--);
1656 }
1657}
1658
1659static void
1660eio_dent_insertion_sort (eio_dirent *dents, int size)
1661{
1662 /* first move the smallest element to the front, to act as a sentinel */
1663 {
1664 int i;
1665 eio_dirent *min = dents;
1666
1667 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1668 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1669 if (EIO_DENT_CMP (dents [i], <, *min))
1670 min = &dents [i];
1671
1672 /* swap elements 0 and j (minimum) */
1673 {
1674 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1675 }
1676 }
1677
1678 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1679 {
1680 eio_dirent *i, *j;
1681
1682 for (i = dents + 1; i < dents + size; ++i)
1683 {
1684 eio_dirent value = *i;
1685
1686 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1687 j [1] = j [0];
1688
1689 j [1] = value;
1690 }
1691 }
1692}
1693
1694static void
1695eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1696{
1697 if (size <= 1)
1698 return; /* our insertion sort relies on size > 0 */
1699
1700 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1701 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1702 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1703
1704 /* use an insertion sort at the end, or for small arrays, */
1705 /* as insertion sort is more efficient for small partitions */
1706 eio_dent_insertion_sort (dents, size);
1002} 1707}
1003 1708
1004/* read a full directory */ 1709/* read a full directory */
1005static void 1710static void
1006eio__scandir (eio_req *req, etp_worker *self) 1711eio__scandir (eio_req *req, etp_worker *self)
1007{ 1712{
1008 DIR *dirp;
1009 EIO_STRUCT_DIRENT *entp;
1010 unsigned char *name, *names; 1713 char *name, *names;
1011 int namesalloc = 4096; 1714 int namesalloc = 4096 - sizeof (void *) * 4;
1012 int namesoffs = 0; 1715 int namesoffs = 0;
1013 int flags = req->int1; 1716 int flags = req->int1;
1014 eio_dirent *dents = 0; 1717 eio_dirent *dents = 0;
1015 int dentalloc = 128; 1718 int dentalloc = 128;
1016 int dentoffs = 0; 1719 int dentoffs = 0;
1720 eio_ino_t inode_bits = 0;
1721#ifdef _WIN32
1722 HANDLE dirp;
1723 WIN32_FIND_DATA entp;
1724#else
1725 DIR *dirp;
1726 EIO_STRUCT_DIRENT *entp;
1727#endif
1017 1728
1018 req->result = -1; 1729 req->result = -1;
1019 1730
1020 if (!(flags & EIO_READDIR_DENTS)) 1731 if (!(flags & EIO_READDIR_DENTS))
1021 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); 1732 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1022 1733
1023 X_LOCK (wrklock); 1734#ifdef _WIN32
1024 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 1735 {
1736 int len = strlen ((const char *)req->ptr1);
1737 char *path = malloc (MAX_PATH);
1738 const char *fmt;
1739 const char *reqpath = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1740
1741 if (!len)
1742 fmt = "./*";
1743 else if (reqpath[len - 1] == '/' || reqpath[len - 1] == '\\')
1744 fmt = "%s*";
1745 else
1746 fmt = "%s/*";
1747
1748 _snprintf (path, MAX_PATH, fmt, reqpath);
1749 dirp = FindFirstFile (path, &entp);
1750 free (path);
1751
1752 if (dirp == INVALID_HANDLE_VALUE)
1753 {
1754 /* should steal _dosmaperr */
1755 switch (GetLastError ())
1756 {
1757 case ERROR_FILE_NOT_FOUND:
1758 req->result = 0;
1759 break;
1760
1761 case ERROR_INVALID_NAME:
1762 case ERROR_PATH_NOT_FOUND:
1763 case ERROR_NO_MORE_FILES:
1764 errno = ENOENT;
1765 break;
1766
1767 case ERROR_NOT_ENOUGH_MEMORY:
1768 errno = ENOMEM;
1769 break;
1770
1771 default:
1772 errno = EINVAL;
1773 break;
1774 }
1775
1776 return;
1777 }
1778 }
1779#else
1780 #if HAVE_AT
1781 if (req->wd)
1782 {
1783 int fd = openat (WD2FD (req->wd), req->ptr1, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1784
1785 if (fd < 0)
1786 return;
1787
1788 dirp = fdopendir (fd);
1789 }
1790 else
1025 self->dirp = dirp = opendir (req->ptr1); 1791 dirp = opendir (req->ptr1);
1792 #else
1793 dirp = opendir (wd_expand (&self->tmpbuf, req->wd, req->ptr1));
1794 #endif
1795
1796 if (!dirp)
1797 return;
1798#endif
1799
1800 if (req->flags & EIO_FLAG_PTR1_FREE)
1801 free (req->ptr1);
1802
1026 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1803 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1027 req->ptr1 = names = malloc (namesalloc);
1028 req->ptr2 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; 1804 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1029 X_UNLOCK (wrklock); 1805 req->ptr2 = names = malloc (namesalloc);
1030 1806
1031 if (dirp && names && (!flags || dents)) 1807 if (!names || (flags && !dents))
1808 return;
1809
1032 for (;;) 1810 for (;;)
1033 { 1811 {
1812 int done;
1813
1814#ifdef _WIN32
1815 done = !dirp;
1816#else
1034 errno = 0; 1817 errno = 0;
1035 entp = readdir (dirp); 1818 entp = readdir (dirp);
1819 done = !entp;
1820#endif
1036 1821
1037 if (!entp) 1822 if (done)
1038 { 1823 {
1824#ifndef _WIN32
1825 int old_errno = errno;
1826 closedir (dirp);
1827 errno = old_errno;
1828
1039 if (errno) 1829 if (errno)
1040 break; 1830 break;
1831#endif
1041 1832
1042 /* sort etc. */ 1833 /* sort etc. */
1043 req->int1 = flags; 1834 req->int1 = flags;
1044 req->result = dentoffs; 1835 req->result = dentoffs;
1045 1836
1046 if (dents)
1047 {
1048 eio_dirent *ent = dents + dentoffs;
1049
1050 while (ent > dents)
1051 (--ent)->name = names + (size_t)ent->name;
1052 }
1053
1054 if (flags & EIO_READDIR_STAT_ORDER 1837 if (flags & EIO_READDIR_STAT_ORDER)
1055 || !(~flags & (EIO_READDIR_DIRS_FIRST | EIO_READDIR_FOUND_UNKNOWN))) 1838 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1056 {
1057 /* pray your qsort doesn't use quicksort */
1058 qsort (dents, dentoffs, sizeof (*dents), eio_dent_cmp); /* score depends of DIRS_FIRST */
1059 }
1060 else if (flags & EIO_READDIR_DIRS_FIRST) 1839 else if (flags & EIO_READDIR_DIRS_FIRST)
1840 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1841 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1842 else
1061 { 1843 {
1062 /* in this case, all is known, and we just put dirs first and sort them */ 1844 /* in this case, all is known, and we just put dirs first and sort them */
1063 eio_dirent *ent = dents + dentoffs; 1845 eio_dirent *oth = dents + dentoffs;
1064 eio_dirent *dir = dents; 1846 eio_dirent *dir = dents;
1065 1847
1848 /* now partition dirs to the front, and non-dirs to the back */
1849 /* by walking from both sides and swapping if necessary */
1066 while (ent > dir) 1850 while (oth > dir)
1067 { 1851 {
1068 if (dir->type == DT_DIR) 1852 if (dir->type == EIO_DT_DIR)
1069 ++dir; 1853 ++dir;
1070 else 1854 else if ((--oth)->type == EIO_DT_DIR)
1071 { 1855 {
1072 --ent;
1073
1074 if (ent->type == DT_DIR)
1075 {
1076 eio_dirent tmp = *dir; 1856 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1077 *dir = *ent;
1078 *ent = tmp;
1079 1857
1080 ++dir; 1858 ++dir;
1081 }
1082 } 1859 }
1083 } 1860 }
1084 1861
1085 /* now sort the dirs only */ 1862 /* now sort the dirs only (dirs all have the same score) */
1086 qsort (dents, dir - dents, sizeof (*dents), eio_dent_cmp); 1863 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1087 } 1864 }
1088 1865
1089 /* only provide the names array unless DENTS is specified */
1090 if (!(flags & EIO_READDIR_DENTS))
1091 {
1092 X_LOCK (wrklock);
1093 assert (!dents);
1094 req->ptr1 = 0;
1095 req->ptr2 = names;
1096 X_UNLOCK (wrklock);
1097 }
1098
1099 break; 1866 break;
1100 } 1867 }
1101 1868
1102 /* now add the entry to our list(s) */ 1869 /* now add the entry to our list(s) */
1103 name = entp->d_name; 1870 name = D_NAME (entp);
1104 1871
1105 /* skip . and .. entries */ 1872 /* skip . and .. entries */
1106 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1873 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1107 { 1874 {
1108 int len = D_NAMLEN (entp) + 1; 1875 int len = D_NAMLEN (entp) + 1;
1109 1876
1110 while (expect_false (namesoffs + len > namesalloc)) 1877 while (ecb_expect_false (namesoffs + len > namesalloc))
1111 { 1878 {
1112 namesalloc *= 2; 1879 namesalloc *= 2;
1113 X_LOCK (wrklock);
1114 req->ptr1 = names = realloc (names, namesalloc); 1880 req->ptr2 = names = realloc (names, namesalloc);
1115 X_UNLOCK (wrklock);
1116 1881
1117 if (!names) 1882 if (!names)
1118 break; 1883 break;
1119 } 1884 }
1120 1885
1121 memcpy (names + namesoffs, name, len); 1886 memcpy (names + namesoffs, name, len);
1122 1887
1123 if (dents) 1888 if (dents)
1124 { 1889 {
1125 struct eio_dirent *ent; 1890 struct eio_dirent *ent;
1126 1891
1127 if (expect_false (dentoffs == dentalloc)) 1892 if (ecb_expect_false (dentoffs == dentalloc))
1128 { 1893 {
1129 dentalloc *= 2; 1894 dentalloc *= 2;
1130 X_LOCK (wrklock);
1131 req->ptr2 = dents = realloc (dents, dentalloc * sizeof (eio_dirent)); 1895 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1132 X_UNLOCK (wrklock);
1133 1896
1134 if (!dents) 1897 if (!dents)
1135 break; 1898 break;
1136 } 1899 }
1137 1900
1138 ent = dents + dentoffs; 1901 ent = dents + dentoffs;
1139 1902
1140 ent->name = (char *)(size_t)namesoffs; /* rather dirtily we store the offset in the pointer */ 1903 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1141 ent->namelen = len - 1; 1904 ent->namelen = len - 1;
1142 ent->inode = D_INO (entp); 1905 ent->inode = D_INO (entp);
1143 1906
1907 inode_bits |= ent->inode;
1908
1144 switch (D_TYPE (entp)) 1909 switch (D_TYPE (entp))
1145 { 1910 {
1146 default: 1911 default:
1147 ent->type = EIO_DT_UNKNOWN; 1912 ent->type = EIO_DT_UNKNOWN;
1148 flags |= EIO_READDIR_FOUND_UNKNOWN; 1913 flags |= EIO_READDIR_FOUND_UNKNOWN;
1149 break; 1914 break;
1150 1915
1151 #ifdef DT_FIFO 1916 #ifdef DT_FIFO
1152 case DT_FIFO: ent->type = EIO_DT_FIFO; break; 1917 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1153 #endif 1918 #endif
1154 #ifdef DT_CHR 1919 #ifdef DT_CHR
1155 case DT_CHR: ent->type = EIO_DT_CHR; break; 1920 case DT_CHR: ent->type = EIO_DT_CHR; break;
1156 #endif 1921 #endif
1157 #ifdef DT_MPC 1922 #ifdef DT_MPC
1158 case DT_MPC: ent->type = EIO_DT_MPC; break; 1923 case DT_MPC: ent->type = EIO_DT_MPC; break;
1159 #endif 1924 #endif
1160 #ifdef DT_DIR 1925 #ifdef DT_DIR
1161 case DT_DIR: ent->type = EIO_DT_DIR; break; 1926 case DT_DIR: ent->type = EIO_DT_DIR; break;
1162 #endif 1927 #endif
1163 #ifdef DT_NAM 1928 #ifdef DT_NAM
1164 case DT_NAM: ent->type = EIO_DT_NAM; break; 1929 case DT_NAM: ent->type = EIO_DT_NAM; break;
1165 #endif 1930 #endif
1166 #ifdef DT_BLK 1931 #ifdef DT_BLK
1167 case DT_BLK: ent->type = EIO_DT_BLK; break; 1932 case DT_BLK: ent->type = EIO_DT_BLK; break;
1168 #endif 1933 #endif
1169 #ifdef DT_MPB 1934 #ifdef DT_MPB
1170 case DT_MPB: ent->type = EIO_DT_MPB; break; 1935 case DT_MPB: ent->type = EIO_DT_MPB; break;
1171 #endif 1936 #endif
1172 #ifdef DT_REG 1937 #ifdef DT_REG
1173 case DT_REG: ent->type = EIO_DT_REG; break; 1938 case DT_REG: ent->type = EIO_DT_REG; break;
1174 #endif 1939 #endif
1175 #ifdef DT_NWK 1940 #ifdef DT_NWK
1176 case DT_NWK: ent->type = EIO_DT_NWK; break; 1941 case DT_NWK: ent->type = EIO_DT_NWK; break;
1177 #endif 1942 #endif
1178 #ifdef DT_CMP 1943 #ifdef DT_CMP
1179 case DT_CMP: ent->type = EIO_DT_CMP; break; 1944 case DT_CMP: ent->type = EIO_DT_CMP; break;
1180 #endif 1945 #endif
1181 #ifdef DT_LNK 1946 #ifdef DT_LNK
1182 case DT_LNK: ent->type = EIO_DT_LNK; break; 1947 case DT_LNK: ent->type = EIO_DT_LNK; break;
1183 #endif 1948 #endif
1184 #ifdef DT_SOCK 1949 #ifdef DT_SOCK
1185 case DT_SOCK: ent->type = EIO_DT_SOCK; break; 1950 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1186 #endif 1951 #endif
1187 #ifdef DT_DOOR 1952 #ifdef DT_DOOR
1188 case DT_DOOR: ent->type = EIO_DT_DOOR; break; 1953 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1189 #endif 1954 #endif
1190 #ifdef DT_WHT 1955 #ifdef DT_WHT
1191 case DT_WHT: ent->type = EIO_DT_WHT; break; 1956 case DT_WHT: ent->type = EIO_DT_WHT; break;
1192 #endif 1957 #endif
1193 } 1958 }
1194 1959
1195 ent->score = 0; 1960 ent->score = 7;
1196 1961
1197 if (flags & EIO_READDIR_DIRS_FIRST) 1962 if (flags & EIO_READDIR_DIRS_FIRST)
1198 { 1963 {
1199 if (ent->type == EIO_DT_UNKNOWN) 1964 if (ent->type == EIO_DT_UNKNOWN)
1200 { 1965 {
1201 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */ 1966 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1202 ent->score = 98; 1967 ent->score = 1;
1203 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */ 1968 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1204 ent->score = len <= 2 ? len + 60 : len <= 4 ? 50 : len <= 7 ? 40 : 10; /* shorter == more likely dir, but avoid too many classes */ 1969 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1205 } 1970 }
1206 else if (ent->type == EIO_DT_DIR) 1971 else if (ent->type == EIO_DT_DIR)
1207 ent->score = 100; 1972 ent->score = 0;
1208 } 1973 }
1209 } 1974 }
1210 1975
1211 namesoffs += len; 1976 namesoffs += len;
1212 ++dentoffs; 1977 ++dentoffs;
1213 } 1978 }
1979
1980 if (EIO_CANCELLED (req))
1981 {
1982 errno = ECANCELED;
1983 break;
1214 } 1984 }
1215 else
1216 req->result = -1;
1217}
1218 1985
1219#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) 1986#ifdef _WIN32
1220# undef msync 1987 if (!FindNextFile (dirp, &entp))
1221# define msync(a,b,c) ((errno = ENOSYS), -1) 1988 {
1989 FindClose (dirp);
1990 dirp = 0;
1991 }
1222#endif 1992#endif
1993 }
1994}
1223 1995
1224int 1996/*****************************************************************************/
1225eio__mtouch (void *mem, size_t len, int flags) 1997/* working directory stuff */
1226{ 1998/* various deficiencies in the posix 2008 api force us to */
1227 intptr_t addr = (intptr_t)mem; 1999/* keep the absolute path in string form at all times */
1228 intptr_t end = addr + len; 2000/* fuck yeah. */
1229#ifdef PAGESIZE
1230 const intptr_t page = PAGESIZE;
1231#else
1232 static intptr_t page;
1233 2001
1234 if (!page) 2002#if !HAVE_AT
1235 page = sysconf (_SC_PAGESIZE);
1236#endif
1237 2003
1238 addr &= ~(page - 1); /* assume page size is always a power of two */ 2004/* a bit like realpath, but usually faster because it doesn'T have to return */
2005/* an absolute or canonical path */
2006static const char *
2007wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
2008{
2009 if (!wd || *path == '/')
2010 return path;
1239 2011
1240 if (addr < end) 2012 if (path [0] == '.' && !path [1])
1241 if (flags) /* modify */ 2013 return wd->str;
1242 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
1243 else
1244 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
1245 2014
2015 {
2016 int l1 = wd->len;
2017 int l2 = strlen (path);
2018
2019 char *res = tmpbuf_get (tmpbuf, l1 + l2 + 2);
2020
2021 memcpy (res, wd->str, l1);
2022 res [l1] = '/';
2023 memcpy (res + l1 + 1, path, l2 + 1);
2024
2025 return res;
2026 }
2027}
2028
2029#endif
2030
2031static eio_wd
2032eio__wd_open_sync (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
2033{
2034 int fd;
2035 eio_wd res;
2036 int len = eio__realpath (tmpbuf, wd, path);
2037
2038 if (len < 0)
2039 return EIO_INVALID_WD;
2040
2041#if HAVE_AT
2042 fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
2043
2044 if (fd < 0)
2045 return EIO_INVALID_WD;
2046#endif
2047
2048 res = malloc (sizeof (*res) + len); /* one extra 0-byte */
2049
2050#if HAVE_AT
2051 res->fd = fd;
2052#endif
2053
2054 res->len = len;
2055 memcpy (res->str, tmpbuf->ptr, len);
2056 res->str [len] = 0;
2057
2058 return res;
2059}
2060
2061eio_wd
2062eio_wd_open_sync (eio_wd wd, const char *path)
2063{
2064 struct tmpbuf tmpbuf = { 0 };
2065 wd = eio__wd_open_sync (&tmpbuf, wd, path);
2066 free (tmpbuf.ptr);
2067
1246 return 0; 2068 return wd;
1247} 2069}
2070
2071void
2072eio_wd_close_sync (eio_wd wd)
2073{
2074 if (wd != EIO_INVALID_WD && wd != EIO_CWD)
2075 {
2076 #if HAVE_AT
2077 close (wd->fd);
2078 #endif
2079 free (wd);
2080 }
2081}
2082
2083#if HAVE_AT
2084
2085/* they forgot these */
2086
2087static int
2088eio__truncateat (int dirfd, const char *path, off_t length)
2089{
2090 int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
2091 int res;
2092
2093 if (fd < 0)
2094 return fd;
2095
2096 res = ftruncate (fd, length);
2097 close (fd);
2098 return res;
2099}
2100
2101static int
2102eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
2103{
2104 int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
2105 int res;
2106
2107 if (fd < 0)
2108 return fd;
2109
2110 res = fstatvfs (fd, buf);
2111 close (fd);
2112 return res;
2113
2114}
2115
2116#endif
1248 2117
1249/*****************************************************************************/ 2118/*****************************************************************************/
1250 2119
1251#define ALLOC(len) \ 2120#define ALLOC(len) \
1252 if (!req->ptr2) \ 2121 if (!req->ptr2) \
… …
1267{ 2136{
1268 ETP_REQ *req; 2137 ETP_REQ *req;
1269 struct timespec ts; 2138 struct timespec ts;
1270 etp_worker *self = (etp_worker *)thr_arg; 2139 etp_worker *self = (etp_worker *)thr_arg;
1271 2140
2141#if HAVE_PRCTL_SET_NAME
2142 prctl (PR_SET_NAME, (unsigned long)"eio_thread", 0, 0, 0);
2143#endif
2144
1272 /* try to distribute timeouts somewhat randomly */ 2145 /* try to distribute timeouts somewhat evenly */
1273 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 2146 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1274 2147
1275 for (;;) 2148 for (;;)
1276 { 2149 {
2150 ts.tv_sec = 0;
2151
1277 X_LOCK (reqlock); 2152 X_LOCK (reqlock);
1278 2153
1279 for (;;) 2154 for (;;)
1280 { 2155 {
1281 self->req = req = reqq_shift (&req_queue); 2156 req = reqq_shift (&req_queue);
1282 2157
1283 if (req) 2158 if (req)
1284 break; 2159 break;
1285 2160
2161 if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
2162 {
2163 X_UNLOCK (reqlock);
2164 X_LOCK (wrklock);
2165 --started;
2166 X_UNLOCK (wrklock);
2167 goto quit;
2168 }
2169
1286 ++idle; 2170 ++idle;
1287 2171
1288 ts.tv_sec = time (0) + IDLE_TIMEOUT; 2172 if (idle <= max_idle)
1289 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 2173 /* we are allowed to idle, so do so without any timeout */
2174 X_COND_WAIT (reqwait, reqlock);
2175 else
1290 { 2176 {
1291 if (idle > max_idle) 2177 /* initialise timeout once */
1292 { 2178 if (!ts.tv_sec)
1293 --idle; 2179 ts.tv_sec = time (0) + idle_timeout;
1294 X_UNLOCK (reqlock);
1295 X_LOCK (wrklock);
1296 --started;
1297 X_UNLOCK (wrklock);
1298 goto quit;
1299 }
1300 2180
1301 /* we are allowed to idle, so do so without any timeout */
1302 X_COND_WAIT (reqwait, reqlock); 2181 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
2182 ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1303 } 2183 }
1304 2184
1305 --idle; 2185 --idle;
1306 } 2186 }
1307 2187
… …
1310 X_UNLOCK (reqlock); 2190 X_UNLOCK (reqlock);
1311 2191
1312 if (req->type < 0) 2192 if (req->type < 0)
1313 goto quit; 2193 goto quit;
1314 2194
1315 if (!EIO_CANCELLED (req))
1316 ETP_EXECUTE (self, req); 2195 ETP_EXECUTE (self, req);
1317 2196
1318 X_LOCK (reslock); 2197 X_LOCK (reslock);
1319 2198
1320 ++npending; 2199 ++npending;
1321 2200
1322 if (!reqq_push (&res_queue, req) && want_poll_cb) 2201 if (!reqq_push (&res_queue, req) && want_poll_cb)
1323 want_poll_cb (); 2202 want_poll_cb ();
1324 2203
1325 self->req = 0;
1326 etp_worker_clear (self); 2204 etp_worker_clear (self);
1327 2205
1328 X_UNLOCK (reslock); 2206 X_UNLOCK (reslock);
1329 } 2207 }
1330 2208
1331quit: 2209quit:
2210 free (req);
2211
1332 X_LOCK (wrklock); 2212 X_LOCK (wrklock);
1333 etp_worker_free (self); 2213 etp_worker_free (self);
1334 X_UNLOCK (wrklock); 2214 X_UNLOCK (wrklock);
1335 2215
1336 return 0; 2216 return 0;
1337} 2217}
1338 2218
1339/*****************************************************************************/ 2219/*****************************************************************************/
1340 2220
2221int ecb_cold
1341int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 2222eio_init (void (*want_poll)(void), void (*done_poll)(void))
1342{ 2223{
2224#if !HAVE_PREADWRITE
2225 X_MUTEX_CREATE (preadwritelock);
2226#endif
2227
1343 return etp_init (want_poll, done_poll); 2228 return etp_init (want_poll, done_poll);
1344} 2229}
1345 2230
2231ecb_inline void
1346static void eio_api_destroy (eio_req *req) 2232eio_api_destroy (eio_req *req)
1347{ 2233{
1348 free (req); 2234 free (req);
1349} 2235}
1350 2236
1351#define REQ(rtype) \ 2237#define REQ(rtype) \
… …
1370 { \ 2256 { \
1371 eio_api_destroy (req); \ 2257 eio_api_destroy (req); \
1372 return 0; \ 2258 return 0; \
1373 } 2259 }
1374 2260
2261static void
1375static void eio_execute (etp_worker *self, eio_req *req) 2262eio_execute (etp_worker *self, eio_req *req)
1376{ 2263{
1377 errno = 0; 2264#if HAVE_AT
2265 int dirfd;
2266#else
2267 const char *path;
2268#endif
2269
2270 if (ecb_expect_false (EIO_CANCELLED (req)))
2271 {
2272 req->result = -1;
2273 req->errorno = ECANCELED;
2274 return;
2275 }
2276
2277 if (ecb_expect_false (req->wd == EIO_INVALID_WD))
2278 {
2279 req->result = -1;
2280 req->errorno = ENOENT;
2281 return;
2282 }
2283
2284 if (req->type >= EIO_OPEN)
2285 {
2286 #if HAVE_AT
2287 dirfd = WD2FD (req->wd);
2288 #else
2289 path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
2290 #endif
2291 }
1378 2292
1379 switch (req->type) 2293 switch (req->type)
1380 { 2294 {
2295 case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
2296 req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
2297 break;
2298 case EIO_WD_CLOSE: req->result = 0;
2299 eio_wd_close_sync (req->wd); break;
2300
1381 case EIO_READ: ALLOC (req->size); 2301 case EIO_READ: ALLOC (req->size);
1382 req->result = req->offs >= 0 2302 req->result = req->offs >= 0
1383 ? pread (req->int1, req->ptr2, req->size, req->offs) 2303 ? pread (req->int1, req->ptr2, req->size, req->offs)
1384 : read (req->int1, req->ptr2, req->size); break; 2304 : read (req->int1, req->ptr2, req->size); break;
1385 case EIO_WRITE: req->result = req->offs >= 0 2305 case EIO_WRITE: req->result = req->offs >= 0
1386 ? pwrite (req->int1, req->ptr2, req->size, req->offs) 2306 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1387 : write (req->int1, req->ptr2, req->size); break; 2307 : write (req->int1, req->ptr2, req->size); break;
1388 2308
1389 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break; 2309 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1390 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break; 2310 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
2311
2312#if HAVE_AT
1391 2313
1392 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2314 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1393 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2315 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break;
1394 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2316 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1395 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2317 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, AT_SYMLINK_NOFOLLOW); break;
1396 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1397 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1398
1399 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; 2318 case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break;
1400 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1401 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break; 2319 case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break;
1402 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1403 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break; 2320 case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break;
1404 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1405
1406 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break; 2321 case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break;
1407 case EIO_CLOSE: req->result = close (req->int1); break; 2322
1408 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1409 case EIO_UNLINK: req->result = unlink (req->ptr1); break; 2323 case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
1410 case EIO_RMDIR: req->result = rmdir (req->ptr1); break; 2324 case EIO_RMDIR: req->result = unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break;
1411 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; 2325 case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); break;
1412 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; 2326 case EIO_RENAME: req->result = renameat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2); break;
1413 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; 2327 case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2, 0); break;
1414 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; 2328 case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
1415 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break; 2329 case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1416
1417 case EIO_READLINK: ALLOC (NAME_MAX); 2330 case EIO_READLINK: ALLOC (PATH_MAX);
1418 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; 2331 req->result = readlinkat (dirfd, req->ptr1, req->ptr2, PATH_MAX); break;
1419 2332 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1420 case EIO_SYNC: req->result = 0; sync (); break; 2333 req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1421 case EIO_FSYNC: req->result = fsync (req->int1); break;
1422 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1423 case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break;
1424 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;
1425 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1426
1427 case EIO_READDIR: eio__scandir (req, self); break;
1428
1429 case EIO_BUSY: 2334 case EIO_UTIME:
1430#ifdef _WIN32 2335 case EIO_FUTIME:
1431 Sleep (req->nv1 * 1000.); 2336 {
2337 struct timespec ts[2];
2338 struct timespec *times;
2339
2340 if (req->nv1 != -1. || req->nv2 != -1.)
2341 {
2342 ts[0].tv_sec = req->nv1;
2343 ts[0].tv_nsec = (req->nv1 - ts[0].tv_sec) * 1e9;
2344 ts[1].tv_sec = req->nv2;
2345 ts[1].tv_nsec = (req->nv2 - ts[1].tv_sec) * 1e9;
2346
2347 times = ts;
2348 }
2349 else
2350 times = 0;
2351
2352 req->result = req->type == EIO_FUTIME
2353 ? futimens (req->int1, times)
2354 : utimensat (dirfd, req->ptr1, times, 0);
2355 }
2356 break;
2357
1432#else 2358#else
1433 {
1434 struct timeval tv;
1435 2359
1436 tv.tv_sec = req->nv1; 2360 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1437 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; 2361 req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
2362 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2363 req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
2364 case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
2365 case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
2366 case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
2367 case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
1438 2368
1439 req->result = select (0, 0, 0, 0, &tv); 2369 case EIO_UNLINK: req->result = unlink (path ); break;
1440 } 2370 case EIO_RMDIR: req->result = rmdir (path ); break;
1441#endif 2371 case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
1442 break; 2372 case EIO_RENAME: req->result = rename (path , req->ptr2); break;
2373 case EIO_LINK: req->result = link (path , req->ptr2); break;
2374 case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
2375 case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
2376 case EIO_READLINK: ALLOC (PATH_MAX);
2377 req->result = readlink (path, req->ptr2, PATH_MAX); break;
2378 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2379 req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
1443 2380
1444 case EIO_UTIME: 2381 case EIO_UTIME:
1445 case EIO_FUTIME: 2382 case EIO_FUTIME:
1446 { 2383 {
1447 struct timeval tv[2]; 2384 struct timeval tv[2];
1448 struct timeval *times; 2385 struct timeval *times;
1449 2386
1450 if (req->nv1 != -1. || req->nv2 != -1.) 2387 if (req->nv1 != -1. || req->nv2 != -1.)
1451 { 2388 {
1452 tv[0].tv_sec = req->nv1; 2389 tv[0].tv_sec = req->nv1;
1453 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.; 2390 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
1454 tv[1].tv_sec = req->nv2; 2391 tv[1].tv_sec = req->nv2;
1455 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.; 2392 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
1456 2393
1457 times = tv; 2394 times = tv;
1458 } 2395 }
1459 else 2396 else
1460 times = 0; 2397 times = 0;
1461
1462 2398
1463 req->result = req->type == EIO_FUTIME 2399 req->result = req->type == EIO_FUTIME
1464 ? futimes (req->int1, times) 2400 ? futimes (req->int1, times)
1465 : utimes (req->ptr1, times); 2401 : utimes (req->ptr1, times);
1466 } 2402 }
1467 break; 2403 break;
1468 2404
2405#endif
2406
2407 case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
2408 {
2409 ALLOC (req->result);
2410 memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
2411 }
2412 break;
2413
2414 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2415 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
2416
2417 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2418 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2419
2420 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
2421 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
2422 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
2423
2424 case EIO_CLOSE: req->result = close (req->int1); break;
2425 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
2426 case EIO_SYNC: req->result = 0; sync (); break;
2427 case EIO_FSYNC: req->result = fsync (req->int1); break;
2428 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
2429 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
2430 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
2431 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
2432 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
2433 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
2434 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
2435 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
2436
2437 case EIO_READDIR: eio__scandir (req, self); break;
2438
2439 case EIO_BUSY:
2440#ifdef _WIN32
2441 Sleep (req->nv1 * 1e3);
2442#else
2443 {
2444 struct timeval tv;
2445
2446 tv.tv_sec = req->nv1;
2447 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
2448
2449 req->result = select (0, 0, 0, 0, &tv);
2450 }
2451#endif
2452 break;
2453
1469 case EIO_GROUP: 2454 case EIO_GROUP:
1470 abort (); /* handled in eio_request */ 2455 abort (); /* handled in eio_request */
1471 2456
1472 case EIO_NOP: 2457 case EIO_NOP:
1473 req->result = 0; 2458 req->result = 0;
1474 break; 2459 break;
1475 2460
1476 case EIO_CUSTOM: 2461 case EIO_CUSTOM:
1477 ((void (*)(eio_req *))req->feed) (req); 2462 req->feed (req);
1478 break; 2463 break;
1479 2464
1480 default: 2465 default:
2466 errno = ENOSYS;
1481 req->result = -1; 2467 req->result = -1;
1482 break; 2468 break;
1483 } 2469 }
1484 2470
1485 req->errorno = errno; 2471 req->errorno = errno;
… …
1510eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2496eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1511{ 2497{
1512 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2498 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1513} 2499}
1514 2500
2501eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2502{
2503 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2504}
2505
2506eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
2507{
2508 REQ (EIO_SYNCFS); req->int1 = fd; SEND;
2509}
2510
2511eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2512{
2513 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2514}
2515
1515eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2516eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1516{ 2517{
1517 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2518 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1518} 2519}
1519 2520
1520eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) 2521eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
1521{ 2522{
1522 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; 2523 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
1523} 2524}
1524 2525
1525eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) 2526eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
1526{ 2527{
1527 REQ (EIO_FDATASYNC); req->int1 = fd; SEND; 2528 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2529}
2530
2531eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2532{
2533 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
1528} 2534}
1529 2535
1530eio_req *eio_close (int fd, int pri, eio_cb cb, void *data) 2536eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
1531{ 2537{
1532 REQ (EIO_CLOSE); req->int1 = fd; SEND; 2538 REQ (EIO_CLOSE); req->int1 = fd; SEND;
… …
1550eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) 2556eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1551{ 2557{
1552 REQ (EIO_FSTAT); req->int1 = fd; SEND; 2558 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1553} 2559}
1554 2560
2561eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
2562{
2563 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
2564}
2565
1555eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) 2566eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1556{ 2567{
1557 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; 2568 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1558} 2569}
1559 2570
… …
1565eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) 2576eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1566{ 2577{
1567 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2578 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1568} 2579}
1569 2580
1570eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2581eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1571{ 2582{
1572 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2583 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1573} 2584}
1574 2585
1575eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) 2586eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
… …
1595eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data) 2606eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1596{ 2607{
1597 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2608 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1598} 2609}
1599 2610
1600eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2611eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1601{ 2612{
1602 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2613 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1603} 2614}
1604 2615
1605eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) 2616eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
… …
1621eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data) 2632eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1622{ 2633{
1623 return eio__1path (EIO_READLINK, path, pri, cb, data); 2634 return eio__1path (EIO_READLINK, path, pri, cb, data);
1624} 2635}
1625 2636
2637eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2638{
2639 return eio__1path (EIO_REALPATH, path, pri, cb, data);
2640}
2641
1626eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) 2642eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1627{ 2643{
1628 return eio__1path (EIO_STAT, path, pri, cb, data); 2644 return eio__1path (EIO_STAT, path, pri, cb, data);
1629} 2645}
1630 2646
1631eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) 2647eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1632{ 2648{
1633 return eio__1path (EIO_LSTAT, path, pri, cb, data); 2649 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1634} 2650}
1635 2651
2652eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
2653{
2654 return eio__1path (EIO_STATVFS, path, pri, cb, data);
2655}
2656
1636eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) 2657eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1637{ 2658{
1638 return eio__1path (EIO_UNLINK, path, pri, cb, data); 2659 return eio__1path (EIO_UNLINK, path, pri, cb, data);
1639} 2660}
1640 2661
… …
1648 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND; 2669 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1649} 2670}
1650 2671
1651eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) 2672eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1652{ 2673{
1653 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; 2674 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1654} 2675}
1655 2676
1656static eio_req * 2677static eio_req *
1657eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2678eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1658{ 2679{
… …
1682eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2703eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1683{ 2704{
1684 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2705 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1685} 2706}
1686 2707
1687eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) 2708eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
1688{ 2709{
1689 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND; 2710 REQ (EIO_CUSTOM); req->feed = execute; SEND;
1690} 2711}
1691 2712
1692#endif 2713#endif
1693 2714
1694eio_req *eio_grp (eio_cb cb, void *data) 2715eio_req *eio_grp (eio_cb cb, void *data)
… …
1703#undef SEND 2724#undef SEND
1704 2725
1705/*****************************************************************************/ 2726/*****************************************************************************/
1706/* grp functions */ 2727/* grp functions */
1707 2728
2729void
1708void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2730eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1709{ 2731{
1710 grp->int2 = limit; 2732 grp->int2 = limit;
1711 grp->feed = feed; 2733 grp->feed = feed;
1712 2734
1713 grp_try_feed (grp); 2735 grp_try_feed (grp);
1714} 2736}
1715 2737
2738void
1716void eio_grp_limit (eio_req *grp, int limit) 2739eio_grp_limit (eio_req *grp, int limit)
1717{ 2740{
1718 grp->int2 = limit; 2741 grp->int2 = limit;
1719 2742
1720 grp_try_feed (grp); 2743 grp_try_feed (grp);
1721} 2744}
1722 2745
2746void
1723void eio_grp_add (eio_req *grp, eio_req *req) 2747eio_grp_add (eio_req *grp, eio_req *req)
1724{ 2748{
1725 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2749 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1726 2750
1727 grp->flags |= EIO_FLAG_GROUPADD; 2751 grp->flags |= EIO_FLAG_GROUPADD;
1728 2752
… …
1739} 2763}
1740 2764
1741/*****************************************************************************/ 2765/*****************************************************************************/
1742/* misc garbage */ 2766/* misc garbage */
1743 2767
2768eio_ssize_t
1744ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2769eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1745{ 2770{
1746 etp_worker wrk;
1747
1748 wrk.dbuf = 0;
1749
1750 eio__sendfile (ofd, ifd, offset, count, &wrk); 2771 return eio__sendfile (ofd, ifd, offset, count);
1751
1752 if (wrk.dbuf)
1753 free (wrk.dbuf);
1754} 2772}
1755 2773

Diff Legend

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