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

Comparing libeio/eio.c (file contents):
Revision 1.14 by root, Sat May 17 12:17:25 2008 UTC vs.
Revision 1.126 by root, Fri Dec 28 07:33:41 2012 UTC

1/* 1/*
2 * libeio implementation 2 * libeio implementation
3 * 3 *
4 * Copyright (c) 2007,2008 Marc Alexander Lehmann <libeio@schmorp.de> 4 * Copyright (c) 2007,2008,2009,2010,2011,2012 Marc Alexander Lehmann <libeio@schmorp.de>
5 * All rights reserved. 5 * All rights reserved.
6 * 6 *
7 * Redistribution and use in source and binary forms, with or without modifica- 7 * Redistribution and use in source and binary forms, with or without modifica-
8 * tion, are permitted provided that the following conditions are met: 8 * tion, are permitted provided that the following conditions are met:
9 * 9 *
10 * 1. Redistributions of source code must retain the above copyright notice, 10 * 1. Redistributions of source code must retain the above copyright notice,
11 * this list of conditions and the following disclaimer. 11 * this list of conditions and the following disclaimer.
12 * 12 *
13 * 2. Redistributions in binary form must reproduce the above copyright 13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the 14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution. 15 * documentation and/or other materials provided with the distribution.
16 * 16 *
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR IMPLIED 17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR IMPLIED
18 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MER- 18 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MER-
19 * CHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO 19 * CHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO
20 * EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPE- 20 * EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPE-
21 * CIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, 21 * CIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
… …
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"
46
47#ifdef EIO_STACKSIZE
48# define X_STACKSIZE EIO_STACKSIZE
49#endif
41#include "xthread.h" 50#include "xthread.h"
42 51
43#include <errno.h> 52#include <errno.h>
44#include <stddef.h> 53#include <stddef.h>
45#include <stdlib.h> 54#include <stdlib.h>
… …
49#include <sys/stat.h> 58#include <sys/stat.h>
50#include <limits.h> 59#include <limits.h>
51#include <fcntl.h> 60#include <fcntl.h>
52#include <assert.h> 61#include <assert.h>
53 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
54#ifndef EIO_FINISH 82#ifndef EIO_FINISH
55# 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
56#endif 84#endif
57 85
58#ifndef EIO_DESTROY 86#ifndef EIO_DESTROY
… …
61 89
62#ifndef EIO_FEED 90#ifndef EIO_FEED
63# 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)
64#endif 92#endif
65 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
66#ifdef _WIN32 105#ifdef _WIN32
67 106
68 /*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 #define pread(fd,buf,count,offset) eio__pread (fd, buf, count, offset)
138 #define pwrite(fd,buf,count,offset) eio__pwrite (fd, buf, count, offset)
139
140 #if __GNUC__
141 typedef long long eio_off_t; /* signed for compatibility to msvc */
142 #else
143 typedef __int64 eio_off_t; /* unsigned not supported by msvc */
144 #endif
145
146 static eio_ssize_t
147 eio__pread (int fd, void *buf, eio_ssize_t count, eio_off_t offset)
148 {
149 OVERLAPPED o = { 0 };
150 DWORD got;
151
152 o.Offset = offset;
153 o.OffsetHigh = offset >> 32;
154
155 return ReadFile ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd), buf, count, &got, &o)
156 ? got : -1;
157 }
158
159 static eio_ssize_t
160 eio__pwrite (int fd, void *buf, eio_ssize_t count, eio_off_t offset)
161 {
162 OVERLAPPED o = { 0 };
163 DWORD got;
164
165 o.Offset = offset;
166 o.OffsetHigh = offset >> 32;
167
168 return WriteFile ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd), buf, count, &got, &o)
169 ? got : -1;
170 }
171
172 /* rename() uses MoveFile, which fails to overwrite */
173 #define rename(old,neu) eio__rename (old, neu)
174
175 static int
176 eio__rename (const char *old, const char *neu)
177 {
178 if (MoveFileEx (old, neu, MOVEFILE_REPLACE_EXISTING))
179 return 0;
180
181 /* should steal _dosmaperr */
182 switch (GetLastError ())
183 {
184 case ERROR_FILE_NOT_FOUND:
185 case ERROR_PATH_NOT_FOUND:
186 case ERROR_INVALID_DRIVE:
187 case ERROR_NO_MORE_FILES:
188 case ERROR_BAD_NETPATH:
189 case ERROR_BAD_NET_NAME:
190 case ERROR_BAD_PATHNAME:
191 case ERROR_FILENAME_EXCED_RANGE:
192 errno = ENOENT;
193 break;
194
195 default:
196 errno = EACCES;
197 break;
198 }
199
200 return -1;
201 }
202
203 /* we could even stat and see if it exists */
204 static int
205 symlink (const char *old, const char *neu)
206 {
207 #if WINVER >= 0x0600
208 if (CreateSymbolicLink (neu, old, 1))
209 return 0;
210
211 if (CreateSymbolicLink (neu, old, 0))
212 return 0;
213 #endif
214
215 return EIO_ERRNO (ENOENT, -1);
216 }
217
218 /* POSIX API only */
219 #define CreateHardLink(neu,old,flags) 0
220 #define CreateSymbolicLink(neu,old,flags) 0
221
222 struct statvfs
223 {
224 int dummy;
225 };
226
227 #define DT_DIR EIO_DT_DIR
228 #define DT_REG EIO_DT_REG
229 #define D_NAME(entp) entp.cFileName
230 #define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
69 231
70#else 232#else
71 233
72# include "config.h"
73# include <sys/time.h> 234 #include <sys/time.h>
74# include <sys/select.h> 235 #include <sys/select.h>
236 #include <sys/statvfs.h>
75# include <unistd.h> 237 #include <unistd.h>
238 #include <signal.h>
239 #include <dirent.h>
240
241 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
242 #include <sys/mman.h>
243 #endif
244
245 #define D_NAME(entp) entp->d_name
246
247 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
248 #if __FreeBSD__ || __NetBSD__ || __OpenBSD__
249 #define _DIRENT_HAVE_D_TYPE /* sigh */
250 #define D_INO(de) (de)->d_fileno
251 #define D_NAMLEN(de) (de)->d_namlen
252 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
253 #define D_INO(de) (de)->d_ino
254 #endif
255
256 #ifdef _D_EXACT_NAMLEN
257 #undef D_NAMLEN
258 #define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
259 #endif
260
261 #ifdef _DIRENT_HAVE_D_TYPE
262 #define D_TYPE(de) (de)->d_type
263 #endif
264
265 #ifndef EIO_STRUCT_DIRENT
266 #define EIO_STRUCT_DIRENT struct dirent
267 #endif
268
269#endif
270
271#if HAVE_UTIMES
76# include <utime.h> 272# include <utime.h>
77# include <signal.h>
78# include <dirent.h>
79
80# ifndef EIO_STRUCT_DIRENT
81# define EIO_STRUCT_DIRENT struct dirent
82# endif 273#endif
83 274
275#if HAVE_SYS_SYSCALL_H
276# include <sys/syscall.h>
277#endif
278
279#if HAVE_SYS_PRCTL_H
280# include <sys/prctl.h>
84#endif 281#endif
85 282
86#if HAVE_SENDFILE 283#if HAVE_SENDFILE
87# if __linux 284# if __linux
88# include <sys/sendfile.h> 285# include <sys/sendfile.h>
89# elif __freebsd 286# elif __FreeBSD__ || defined __APPLE__
90# include <sys/socket.h> 287# include <sys/socket.h>
91# include <sys/uio.h> 288# include <sys/uio.h>
92# elif __hpux 289# elif __hpux
93# include <sys/socket.h> 290# include <sys/socket.h>
94# elif __solaris /* not yet */ 291# elif __solaris
95# include <sys/sendfile.h> 292# include <sys/sendfile.h>
96# else 293# else
97# error sendfile support requested but not available 294# error sendfile support requested but not available
98# endif 295# endif
99#endif 296#endif
100 297
101/* number of seconds after which an idle threads exit */ 298#ifndef D_TYPE
102#define IDLE_TIMEOUT 10 299# define D_TYPE(de) 0
300#endif
301#ifndef D_INO
302# define D_INO(de) 0
303#endif
304#ifndef D_NAMLEN
305# define D_NAMLEN(entp) strlen (D_NAME (entp))
306#endif
103 307
104/* used for struct dirent, AIX doesn't provide it */ 308/* used for struct dirent, AIX doesn't provide it */
105#ifndef NAME_MAX 309#ifndef NAME_MAX
106# define NAME_MAX 4096 310# define NAME_MAX 4096
107#endif 311#endif
108 312
313/* used for readlink etc. */
314#ifndef PATH_MAX
315# define PATH_MAX 4096
316#endif
317
109/* buffer size for various temporary buffers */ 318/* buffer size for various temporary buffers */
110#define EIO_BUFSIZE 65536 319#define EIO_BUFSIZE 65536
111 320
112#define dBUF \ 321#define dBUF \
113 char *eio_buf; \
114 ETP_WORKER_LOCK (self); \
115 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 322 char *eio_buf = malloc (EIO_BUFSIZE); \
116 ETP_WORKER_UNLOCK (self); \
117 errno = ENOMEM; \ 323 errno = ENOMEM; \
118 if (!eio_buf) \ 324 if (!eio_buf) \
119 return -1; 325 return -1
326
327#define FUBd \
328 free (eio_buf)
120 329
121#define EIO_TICKS ((1000000 + 1023) >> 10) 330#define EIO_TICKS ((1000000 + 1023) >> 10)
331
332/*****************************************************************************/
333
334struct tmpbuf
335{
336 void *ptr;
337 int len;
338};
339
340static void *
341tmpbuf_get (struct tmpbuf *buf, int len)
342{
343 if (buf->len < len)
344 {
345 free (buf->ptr);
346 buf->ptr = malloc (buf->len = len);
347 }
348
349 return buf->ptr;
350}
351
352struct tmpbuf;
353
354#if _POSIX_VERSION >= 200809L
355 #define HAVE_AT 1
356 #define WD2FD(wd) ((wd) ? (wd)->fd : AT_FDCWD)
357 #ifndef O_SEARCH
358 #define O_SEARCH O_RDONLY
359 #endif
360#else
361 #define HAVE_AT 0
362 static const char *wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path);
363#endif
364
365struct eio_pwd
366{
367#if HAVE_AT
368 int fd;
369#endif
370 int len;
371 char str[1]; /* actually, a 0-terminated canonical path */
372};
122 373
123/*****************************************************************************/ 374/*****************************************************************************/
124 375
125#define ETP_PRI_MIN EIO_PRI_MIN 376#define ETP_PRI_MIN EIO_PRI_MIN
126#define ETP_PRI_MAX EIO_PRI_MAX 377#define ETP_PRI_MAX EIO_PRI_MAX
378
379struct etp_worker;
127 380
128#define ETP_REQ eio_req 381#define ETP_REQ eio_req
129#define ETP_DESTROY(req) eio_destroy (req) 382#define ETP_DESTROY(req) eio_destroy (req)
130static int eio_finish (eio_req *req); 383static int eio_finish (eio_req *req);
131#define ETP_FINISH(req) eio_finish (req) 384#define ETP_FINISH(req) eio_finish (req)
132 385static void eio_execute (struct etp_worker *self, eio_req *req);
133#define ETP_WORKER_CLEAR(req) \ 386#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
134 if (wrk->dbuf) \
135 { \
136 free (wrk->dbuf); \
137 wrk->dbuf = 0; \
138 } \
139 \
140 if (wrk->dirp) \
141 { \
142 closedir (wrk->dirp); \
143 wrk->dirp = 0; \
144 }
145#define ETP_WORKER_COMMON \
146 void *dbuf; \
147 DIR *dirp;
148 387
149/*****************************************************************************/ 388/*****************************************************************************/
150 389
151#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) 390#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
152 391
153/* calculcate time difference in ~1/EIO_TICKS of a second */ 392/* calculate time difference in ~1/EIO_TICKS of a second */
393ecb_inline int
154static int tvdiff (struct timeval *tv1, struct timeval *tv2) 394tvdiff (struct timeval *tv1, struct timeval *tv2)
155{ 395{
156 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 396 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
157 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 397 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
158} 398}
159 399
… …
163static void (*done_poll_cb) (void); 403static void (*done_poll_cb) (void);
164 404
165static unsigned int max_poll_time; /* reslock */ 405static unsigned int max_poll_time; /* reslock */
166static unsigned int max_poll_reqs; /* reslock */ 406static unsigned int max_poll_reqs; /* reslock */
167 407
168static volatile unsigned int nreqs; /* reqlock */ 408static unsigned int nreqs; /* reqlock */
169static volatile unsigned int nready; /* reqlock */ 409static unsigned int nready; /* reqlock */
170static volatile unsigned int npending; /* reqlock */ 410static unsigned int npending; /* reqlock */
171static volatile unsigned int max_idle = 4; 411static unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
412static unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
172 413
173static mutex_t wrklock = X_MUTEX_INIT; 414static xmutex_t wrklock;
174static mutex_t reslock = X_MUTEX_INIT; 415static xmutex_t reslock;
175static mutex_t reqlock = X_MUTEX_INIT; 416static xmutex_t reqlock;
176static cond_t reqwait = X_COND_INIT; 417static xcond_t reqwait;
177 418
178typedef struct etp_worker 419typedef struct etp_worker
179{ 420{
421 struct tmpbuf tmpbuf;
422
180 /* locked by wrklock */ 423 /* locked by wrklock */
181 struct etp_worker *prev, *next; 424 struct etp_worker *prev, *next;
182 425
183 thread_t tid; 426 xthread_t tid;
184 427
185 /* locked by reslock, reqlock or wrklock */ 428#ifdef ETP_WORKER_COMMON
186 ETP_REQ *req; /* currently processed request */
187
188 ETP_WORKER_COMMON 429 ETP_WORKER_COMMON
430#endif
189} etp_worker; 431} etp_worker;
190 432
191static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */ 433static etp_worker wrk_first; /* NOT etp */
192 434
193#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock) 435#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
194#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock) 436#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
195 437
196/* worker threads management */ 438/* worker threads management */
197 439
440static void
198static void etp_worker_clear (etp_worker *wrk) 441etp_worker_clear (etp_worker *wrk)
199{ 442{
200 ETP_WORKER_CLEAR (wrk);
201} 443}
202 444
445static void ecb_cold
203static void etp_worker_free (etp_worker *wrk) 446etp_worker_free (etp_worker *wrk)
204{ 447{
448 free (wrk->tmpbuf.ptr);
449
205 wrk->next->prev = wrk->prev; 450 wrk->next->prev = wrk->prev;
206 wrk->prev->next = wrk->next; 451 wrk->prev->next = wrk->next;
207 452
208 free (wrk); 453 free (wrk);
209} 454}
210 455
211static unsigned int etp_nreqs (void) 456static unsigned int
457etp_nreqs (void)
212{ 458{
213 int retval; 459 int retval;
214 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 460 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
215 retval = nreqs; 461 retval = nreqs;
216 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 462 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
217 return retval; 463 return retval;
218} 464}
219 465
220static unsigned int etp_nready (void) 466static unsigned int
467etp_nready (void)
221{ 468{
222 unsigned int retval; 469 unsigned int retval;
223 470
224 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 471 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
225 retval = nready; 472 retval = nready;
226 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 473 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
227 474
228 return retval; 475 return retval;
229} 476}
230 477
231static unsigned int etp_npending (void) 478static unsigned int
479etp_npending (void)
232{ 480{
233 unsigned int retval; 481 unsigned int retval;
234 482
235 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 483 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
236 retval = npending; 484 retval = npending;
237 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 485 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
238 486
239 return retval; 487 return retval;
240} 488}
241 489
242static unsigned int etp_nthreads (void) 490static unsigned int
491etp_nthreads (void)
243{ 492{
244 unsigned int retval; 493 unsigned int retval;
245 494
246 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 495 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
247 retval = started; 496 retval = started;
… …
261} etp_reqq; 510} etp_reqq;
262 511
263static etp_reqq req_queue; 512static etp_reqq req_queue;
264static etp_reqq res_queue; 513static etp_reqq res_queue;
265 514
515static void ecb_noinline ecb_cold
516reqq_init (etp_reqq *q)
517{
518 int pri;
519
520 for (pri = 0; pri < ETP_NUM_PRI; ++pri)
521 q->qs[pri] = q->qe[pri] = 0;
522
523 q->size = 0;
524}
525
526static int ecb_noinline
266static int reqq_push (etp_reqq *q, ETP_REQ *req) 527reqq_push (etp_reqq *q, ETP_REQ *req)
267{ 528{
268 int pri = req->pri; 529 int pri = req->pri;
269 req->next = 0; 530 req->next = 0;
270 531
271 if (q->qe[pri]) 532 if (q->qe[pri])
… …
277 q->qe[pri] = q->qs[pri] = req; 538 q->qe[pri] = q->qs[pri] = req;
278 539
279 return q->size++; 540 return q->size++;
280} 541}
281 542
543static ETP_REQ * ecb_noinline
282static ETP_REQ *reqq_shift (etp_reqq *q) 544reqq_shift (etp_reqq *q)
283{ 545{
284 int pri; 546 int pri;
285 547
286 if (!q->size) 548 if (!q->size)
287 return 0; 549 return 0;
… …
302 } 564 }
303 565
304 abort (); 566 abort ();
305} 567}
306 568
307static void etp_atfork_prepare (void) 569static int ecb_cold
570etp_init (void (*want_poll)(void), void (*done_poll)(void))
308{ 571{
309 X_LOCK (wrklock); 572 X_MUTEX_CREATE (wrklock);
310 X_LOCK (reqlock); 573 X_MUTEX_CREATE (reslock);
311 X_LOCK (reslock); 574 X_MUTEX_CREATE (reqlock);
312#if !HAVE_PREADWRITE 575 X_COND_CREATE (reqwait);
313 X_LOCK (preadwritelock);
314#endif
315#if !HAVE_READDIR_R
316 X_LOCK (readdirlock);
317#endif
318}
319 576
320static void etp_atfork_parent (void) 577 reqq_init (&req_queue);
321{ 578 reqq_init (&res_queue);
322#if !HAVE_READDIR_R
323 X_UNLOCK (readdirlock);
324#endif
325#if !HAVE_PREADWRITE
326 X_UNLOCK (preadwritelock);
327#endif
328 X_UNLOCK (reslock);
329 X_UNLOCK (reqlock);
330 X_UNLOCK (wrklock);
331}
332 579
333static void etp_atfork_child (void) 580 wrk_first.next =
334{ 581 wrk_first.prev = &wrk_first;
335 ETP_REQ *prv;
336
337 while (prv = reqq_shift (&req_queue))
338 ETP_DESTROY (prv);
339
340 while (prv = reqq_shift (&res_queue))
341 ETP_DESTROY (prv);
342
343 while (wrk_first.next != &wrk_first)
344 {
345 etp_worker *wrk = wrk_first.next;
346
347 if (wrk->req)
348 ETP_DESTROY (wrk->req);
349
350 etp_worker_clear (wrk);
351 etp_worker_free (wrk);
352 }
353 582
354 started = 0; 583 started = 0;
355 idle = 0; 584 idle = 0;
356 nreqs = 0; 585 nreqs = 0;
357 nready = 0; 586 nready = 0;
358 npending = 0; 587 npending = 0;
359 588
360 etp_atfork_parent ();
361}
362
363static void
364etp_once_init (void)
365{
366 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
367}
368
369static int
370etp_init (void (*want_poll)(void), void (*done_poll)(void))
371{
372 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
373
374 pthread_once (&doinit, etp_once_init);
375
376 want_poll_cb = want_poll; 589 want_poll_cb = want_poll;
377 done_poll_cb = done_poll; 590 done_poll_cb = done_poll;
591
592 return 0;
378} 593}
379 594
380X_THREAD_PROC (etp_proc); 595X_THREAD_PROC (etp_proc);
381 596
597static void ecb_cold
382static void etp_start_thread (void) 598etp_start_thread (void)
383{ 599{
384 etp_worker *wrk = calloc (1, sizeof (etp_worker)); 600 etp_worker *wrk = calloc (1, sizeof (etp_worker));
385 601
386 /*TODO*/ 602 /*TODO*/
387 assert (("unable to allocate worker thread data", wrk)); 603 assert (("unable to allocate worker thread data", wrk));
388 604
389 X_LOCK (wrklock); 605 X_LOCK (wrklock);
390 606
391 if (thread_create (&wrk->tid, etp_proc, (void *)wrk)) 607 if (xthread_create (&wrk->tid, etp_proc, (void *)wrk))
392 { 608 {
393 wrk->prev = &wrk_first; 609 wrk->prev = &wrk_first;
394 wrk->next = wrk_first.next; 610 wrk->next = wrk_first.next;
395 wrk_first.next->prev = wrk; 611 wrk_first.next->prev = wrk;
396 wrk_first.next = wrk; 612 wrk_first.next = wrk;
… …
400 free (wrk); 616 free (wrk);
401 617
402 X_UNLOCK (wrklock); 618 X_UNLOCK (wrklock);
403} 619}
404 620
621static void
405static void etp_maybe_start_thread (void) 622etp_maybe_start_thread (void)
406{ 623{
407 if (etp_nthreads () >= wanted) 624 if (ecb_expect_true (etp_nthreads () >= wanted))
408 return; 625 return;
409 626
410 /* todo: maybe use idle here, but might be less exact */ 627 /* todo: maybe use idle here, but might be less exact */
411 if (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()) 628 if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
412 return; 629 return;
413 630
414 etp_start_thread (); 631 etp_start_thread ();
415} 632}
416 633
634static void ecb_cold
417static void etp_end_thread (void) 635etp_end_thread (void)
418{ 636{
419 eio_req *req = calloc (1, sizeof (eio_req)); 637 eio_req *req = calloc (1, sizeof (eio_req)); /* will be freed by worker */
420 638
421 req->type = -1; 639 req->type = -1;
422 req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 640 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
423 641
424 X_LOCK (reqlock); 642 X_LOCK (reqlock);
… …
429 X_LOCK (wrklock); 647 X_LOCK (wrklock);
430 --started; 648 --started;
431 X_UNLOCK (wrklock); 649 X_UNLOCK (wrklock);
432} 650}
433 651
434static int etp_poll (void) 652static int
653etp_poll (void)
435{ 654{
436 unsigned int maxreqs; 655 unsigned int maxreqs;
437 unsigned int maxtime; 656 unsigned int maxtime;
438 struct timeval tv_start, tv_now; 657 struct timeval tv_start, tv_now;
439 658
… …
469 688
470 X_LOCK (reqlock); 689 X_LOCK (reqlock);
471 --nreqs; 690 --nreqs;
472 X_UNLOCK (reqlock); 691 X_UNLOCK (reqlock);
473 692
474 if (req->type == EIO_GROUP && req->size) 693 if (ecb_expect_false (req->type == EIO_GROUP && req->size))
475 { 694 {
476 req->int1 = 1; /* mark request as delayed */ 695 req->int1 = 1; /* mark request as delayed */
477 continue; 696 continue;
478 } 697 }
479 else 698 else
480 { 699 {
481 int res = ETP_FINISH (req); 700 int res = ETP_FINISH (req);
482 if (res) 701 if (ecb_expect_false (res))
483 return res; 702 return res;
484 } 703 }
485 704
486 if (maxreqs && !--maxreqs) 705 if (ecb_expect_false (maxreqs && !--maxreqs))
487 break; 706 break;
488 707
489 if (maxtime) 708 if (maxtime)
490 { 709 {
491 gettimeofday (&tv_now, 0); 710 gettimeofday (&tv_now, 0);
… …
497 716
498 errno = EAGAIN; 717 errno = EAGAIN;
499 return -1; 718 return -1;
500} 719}
501 720
721static void
502static void etp_cancel (ETP_REQ *req) 722etp_cancel (ETP_REQ *req)
503{ 723{
504 X_LOCK (wrklock); 724 req->cancelled = 1;
505 req->flags |= EIO_FLAG_CANCELLED;
506 X_UNLOCK (wrklock);
507 725
508 eio_grp_cancel (req); 726 eio_grp_cancel (req);
509} 727}
510 728
729static void
511static void etp_submit (ETP_REQ *req) 730etp_submit (ETP_REQ *req)
512{ 731{
513 req->pri -= ETP_PRI_MIN; 732 req->pri -= ETP_PRI_MIN;
514 733
515 if (req->pri < ETP_PRI_MIN - ETP_PRI_MIN) req->pri = ETP_PRI_MIN - ETP_PRI_MIN; 734 if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
516 if (req->pri > ETP_PRI_MAX - ETP_PRI_MIN) req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 735 if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
517 736
737 if (ecb_expect_false (req->type == EIO_GROUP))
738 {
739 /* I hope this is worth it :/ */
518 X_LOCK (reqlock); 740 X_LOCK (reqlock);
519 ++nreqs; 741 ++nreqs;
742 X_UNLOCK (reqlock);
743
744 X_LOCK (reslock);
745
746 ++npending;
747
748 if (!reqq_push (&res_queue, req) && want_poll_cb)
749 want_poll_cb ();
750
751 X_UNLOCK (reslock);
752 }
753 else
754 {
755 X_LOCK (reqlock);
756 ++nreqs;
520 ++nready; 757 ++nready;
521 reqq_push (&req_queue, req); 758 reqq_push (&req_queue, req);
522 X_COND_SIGNAL (reqwait); 759 X_COND_SIGNAL (reqwait);
523 X_UNLOCK (reqlock); 760 X_UNLOCK (reqlock);
524 761
525 etp_maybe_start_thread (); 762 etp_maybe_start_thread ();
763 }
526} 764}
527 765
766static void ecb_cold
528static void etp_set_max_poll_time (double nseconds) 767etp_set_max_poll_time (double nseconds)
529{ 768{
530 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 769 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
531 max_poll_time = nseconds; 770 max_poll_time = nseconds * EIO_TICKS;
532 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 771 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
533} 772}
534 773
774static void ecb_cold
535static void etp_set_max_poll_reqs (unsigned int maxreqs) 775etp_set_max_poll_reqs (unsigned int maxreqs)
536{ 776{
537 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 777 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
538 max_poll_reqs = maxreqs; 778 max_poll_reqs = maxreqs;
539 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 779 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
540} 780}
541 781
782static void ecb_cold
542static void etp_set_max_idle (unsigned int nthreads) 783etp_set_max_idle (unsigned int nthreads)
543{ 784{
544 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 785 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
545 max_idle = nthreads <= 0 ? 1 : nthreads; 786 max_idle = nthreads;
546 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 787 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
547} 788}
548 789
790static void ecb_cold
791etp_set_idle_timeout (unsigned int seconds)
792{
793 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
794 idle_timeout = seconds;
795 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
796}
797
798static void ecb_cold
549static void etp_set_min_parallel (unsigned int nthreads) 799etp_set_min_parallel (unsigned int nthreads)
550{ 800{
551 if (wanted < nthreads) 801 if (wanted < nthreads)
552 wanted = nthreads; 802 wanted = nthreads;
553} 803}
554 804
805static void ecb_cold
555static void etp_set_max_parallel (unsigned int nthreads) 806etp_set_max_parallel (unsigned int nthreads)
556{ 807{
557 if (wanted > nthreads) 808 if (wanted > nthreads)
558 wanted = nthreads; 809 wanted = nthreads;
559 810
560 while (started > wanted) 811 while (started > wanted)
561 etp_end_thread (); 812 etp_end_thread ();
562} 813}
563 814
564/*****************************************************************************/ 815/*****************************************************************************/
565 816
817static void
566static void grp_try_feed (eio_req *grp) 818grp_try_feed (eio_req *grp)
567{ 819{
568 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 820 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
569 { 821 {
570 int old_len = grp->size; 822 grp->flags &= ~EIO_FLAG_GROUPADD;
571 823
572 EIO_FEED (grp); 824 EIO_FEED (grp);
573 825
574 /* stop if no progress has been made */ 826 /* stop if no progress has been made */
575 if (old_len == grp->size) 827 if (!(grp->flags & EIO_FLAG_GROUPADD))
576 { 828 {
577 grp->feed = 0; 829 grp->feed = 0;
578 break; 830 break;
579 } 831 }
580 } 832 }
581} 833}
582 834
835static int
583static int grp_dec (eio_req *grp) 836grp_dec (eio_req *grp)
584{ 837{
585 --grp->size; 838 --grp->size;
586 839
587 /* call feeder, if applicable */ 840 /* call feeder, if applicable */
588 grp_try_feed (grp); 841 grp_try_feed (grp);
… …
592 return eio_finish (grp); 845 return eio_finish (grp);
593 else 846 else
594 return 0; 847 return 0;
595} 848}
596 849
850static void
597void eio_destroy (eio_req *req) 851eio_destroy (eio_req *req)
598{ 852{
599 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); 853 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
600 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); 854 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
601 855
602 EIO_DESTROY (req); 856 EIO_DESTROY (req);
603} 857}
604 858
859static int
605static int eio_finish (eio_req *req) 860eio_finish (eio_req *req)
606{ 861{
607 int res = EIO_FINISH (req); 862 int res = EIO_FINISH (req);
608 863
609 if (req->grp) 864 if (req->grp)
610 { 865 {
… …
618 if (grp->grp_first == req) 873 if (grp->grp_first == req)
619 grp->grp_first = req->grp_next; 874 grp->grp_first = req->grp_next;
620 875
621 res2 = grp_dec (grp); 876 res2 = grp_dec (grp);
622 877
623 if (!res && res2) 878 if (!res)
624 res = res2; 879 res = res2;
625 } 880 }
626 881
627 eio_destroy (req); 882 eio_destroy (req);
628 883
629 return res; 884 return res;
630} 885}
631 886
887void
632void eio_grp_cancel (eio_req *grp) 888eio_grp_cancel (eio_req *grp)
633{ 889{
634 for (grp = grp->grp_first; grp; grp = grp->grp_next) 890 for (grp = grp->grp_first; grp; grp = grp->grp_next)
635 eio_cancel (grp); 891 eio_cancel (grp);
636} 892}
637 893
894void
638void eio_cancel (eio_req *req) 895eio_cancel (eio_req *req)
639{ 896{
640 etp_cancel (req); 897 etp_cancel (req);
641} 898}
642 899
900void
643void eio_submit (eio_req *req) 901eio_submit (eio_req *req)
644{ 902{
645 etp_submit (req); 903 etp_submit (req);
646} 904}
647 905
648unsigned int eio_nreqs (void) 906unsigned int
907eio_nreqs (void)
649{ 908{
650 return etp_nreqs (); 909 return etp_nreqs ();
651} 910}
652 911
653unsigned int eio_nready (void) 912unsigned int
913eio_nready (void)
654{ 914{
655 return etp_nready (); 915 return etp_nready ();
656} 916}
657 917
658unsigned int eio_npending (void) 918unsigned int
919eio_npending (void)
659{ 920{
660 return etp_npending (); 921 return etp_npending ();
661} 922}
662 923
663unsigned int eio_nthreads (void) 924unsigned int ecb_cold
925eio_nthreads (void)
664{ 926{
665 return etp_nthreads (); 927 return etp_nthreads ();
666} 928}
667 929
930void ecb_cold
668void eio_set_max_poll_time (double nseconds) 931eio_set_max_poll_time (double nseconds)
669{ 932{
670 etp_set_max_poll_time (nseconds); 933 etp_set_max_poll_time (nseconds);
671} 934}
672 935
936void ecb_cold
673void eio_set_max_poll_reqs (unsigned int maxreqs) 937eio_set_max_poll_reqs (unsigned int maxreqs)
674{ 938{
675 etp_set_max_poll_reqs (maxreqs); 939 etp_set_max_poll_reqs (maxreqs);
676} 940}
677 941
942void ecb_cold
678void eio_set_max_idle (unsigned int nthreads) 943eio_set_max_idle (unsigned int nthreads)
679{ 944{
680 etp_set_max_idle (nthreads); 945 etp_set_max_idle (nthreads);
681} 946}
682 947
948void ecb_cold
949eio_set_idle_timeout (unsigned int seconds)
950{
951 etp_set_idle_timeout (seconds);
952}
953
954void ecb_cold
683void eio_set_min_parallel (unsigned int nthreads) 955eio_set_min_parallel (unsigned int nthreads)
684{ 956{
685 etp_set_min_parallel (nthreads); 957 etp_set_min_parallel (nthreads);
686} 958}
687 959
960void ecb_cold
688void eio_set_max_parallel (unsigned int nthreads) 961eio_set_max_parallel (unsigned int nthreads)
689{ 962{
690 etp_set_max_parallel (nthreads); 963 etp_set_max_parallel (nthreads);
691} 964}
692 965
693int eio_poll (void) 966int eio_poll (void)
… …
696} 969}
697 970
698/*****************************************************************************/ 971/*****************************************************************************/
699/* work around various missing functions */ 972/* work around various missing functions */
700 973
701#if !HAVE_PREADWRITE
702# define pread eio__pread
703# define pwrite eio__pwrite
704
705/*
706 * make our pread/pwrite safe against themselves, but not against
707 * normal read/write by using a mutex. slows down execution a lot,
708 * but that's your problem, not mine.
709 */
710static mutex_t preadwritelock = X_MUTEX_INIT;
711
712static ssize_t
713eio__pread (int fd, void *buf, size_t count, off_t offset)
714{
715 ssize_t res;
716 off_t ooffset;
717
718 X_LOCK (preadwritelock);
719 ooffset = lseek (fd, 0, SEEK_CUR);
720 lseek (fd, offset, SEEK_SET);
721 res = read (fd, buf, count);
722 lseek (fd, ooffset, SEEK_SET);
723 X_UNLOCK (preadwritelock);
724
725 return res;
726}
727
728static ssize_t
729eio__pwrite (int fd, void *buf, size_t count, off_t offset)
730{
731 ssize_t res;
732 off_t ooffset;
733
734 X_LOCK (preadwritelock);
735 ooffset = lseek (fd, 0, SEEK_CUR);
736 lseek (fd, offset, SEEK_SET);
737 res = write (fd, buf, count);
738 lseek (fd, offset, SEEK_SET);
739 X_UNLOCK (preadwritelock);
740
741 return res;
742}
743#endif
744
745#ifndef HAVE_FUTIMES 974#ifndef HAVE_UTIMES
746 975
976# undef utimes
747# define utimes(path,times) eio__utimes (path, times) 977# define utimes(path,times) eio__utimes (path, times)
748# define futimes(fd,times) eio__futimes (fd, times)
749 978
750static int 979static int
751eio__utimes (const char *filename, const struct timeval times[2]) 980eio__utimes (const char *filename, const struct timeval times[2])
752{ 981{
753 if (times) 982 if (times)
… …
761 } 990 }
762 else 991 else
763 return utime (filename, 0); 992 return utime (filename, 0);
764} 993}
765 994
995#endif
996
997#ifndef HAVE_FUTIMES
998
999# undef futimes
1000# define futimes(fd,times) eio__futimes (fd, times)
1001
1002static int
766static int eio__futimes (int fd, const struct timeval tv[2]) 1003eio__futimes (int fd, const struct timeval tv[2])
767{ 1004{
768 errno = ENOSYS; 1005 errno = ENOSYS;
769 return -1; 1006 return -1;
770} 1007}
771 1008
772#endif 1009#endif
773 1010
774#if !HAVE_FDATASYNC 1011#if !HAVE_FDATASYNC
1012# undef fdatasync
775# define fdatasync fsync 1013# define fdatasync(fd) fsync (fd)
776#endif 1014#endif
1015
1016static int
1017eio__syncfs (int fd)
1018{
1019 int res;
1020
1021#if HAVE_SYS_SYNCFS
1022 res = (int)syscall (__NR_syncfs, (int)(fd));
1023#else
1024 res = EIO_ENOSYS ();
1025#endif
1026
1027 if (res < 0 && errno == ENOSYS && fd >= 0)
1028 sync ();
1029
1030 return res;
1031}
1032
1033/* sync_file_range always needs emulation */
1034static int
1035eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
1036{
1037#if HAVE_SYNC_FILE_RANGE
1038 int res;
1039
1040 if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
1041 || EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
1042 || EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
1043 {
1044 flags = 0
1045 | (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
1046 | (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
1047 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
1048 }
1049
1050 res = sync_file_range (fd, offset, nbytes, flags);
1051
1052 if (!res || errno != ENOSYS)
1053 return res;
1054#endif
1055
1056 /* even though we could play tricks with the flags, it's better to always
1057 * call fdatasync, as that matches the expectation of its users best */
1058 return fdatasync (fd);
1059}
1060
1061static int
1062eio__fallocate (int fd, int mode, off_t offset, size_t len)
1063{
1064#if HAVE_LINUX_FALLOCATE
1065 return fallocate (fd, mode, offset, len);
1066#else
1067 return EIO_ENOSYS ();
1068#endif
1069}
777 1070
778#if !HAVE_READAHEAD 1071#if !HAVE_READAHEAD
1072# undef readahead
779# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 1073# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
780 1074
781static ssize_t 1075static eio_ssize_t
782eio__readahead (int fd, off_t offset, size_t count, worker *self) 1076eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
783{ 1077{
784 size_t todo = count; 1078 size_t todo = count;
785 dBUF; 1079 dBUF;
786 1080
787 while (todo > 0) 1081 while (todo > 0)
… …
791 pread (fd, eio_buf, len, offset); 1085 pread (fd, eio_buf, len, offset);
792 offset += len; 1086 offset += len;
793 todo -= len; 1087 todo -= len;
794 } 1088 }
795 1089
796 errno = 0; 1090 FUBd;
1091
1092 /* linux's readahead basically only fails for EBADF or EINVAL (not mmappable) */
1093 /* but not for e.g. EIO or eof, so we also never fail */
797 return count; 1094 return 0;
798} 1095}
799 1096
800#endif
801
802#if !HAVE_READDIR_R
803# define readdir_r eio__readdir_r
804
805static mutex_t readdirlock = X_MUTEX_INIT;
806
807static int
808eio__readdir_r (DIR *dirp, EIO_STRUCT_DIRENT *ent, EIO_STRUCT_DIRENT **res)
809{
810 EIO_STRUCT_DIRENT *e;
811 int errorno;
812
813 X_LOCK (readdirlock);
814
815 e = readdir (dirp);
816 errorno = errno;
817
818 if (e)
819 {
820 *res = ent;
821 strcpy (ent->d_name, e->d_name);
822 }
823 else
824 *res = 0;
825
826 X_UNLOCK (readdirlock);
827
828 errno = errorno;
829 return e ? 0 : -1;
830}
831#endif 1097#endif
832 1098
833/* sendfile always needs emulation */ 1099/* sendfile always needs emulation */
834static ssize_t 1100static eio_ssize_t
835eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) 1101eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
836{ 1102{
1103 eio_ssize_t written = 0;
837 ssize_t res; 1104 eio_ssize_t res;
838 1105
839 if (!count) 1106 if (!count)
840 return 0; 1107 return 0;
841 1108
1109 for (;;)
1110 {
1111#ifdef __APPLE__
1112# undef HAVE_SENDFILE /* broken, as everything on os x */
1113#endif
842#if HAVE_SENDFILE 1114#if HAVE_SENDFILE
843# if __linux 1115# if __linux
1116 off_t soffset = offset;
844 res = sendfile (ofd, ifd, &offset, count); 1117 res = sendfile (ofd, ifd, &soffset, count);
845 1118
846# elif __freebsd 1119# elif __FreeBSD__
847 /* 1120 /*
848 * Of course, the freebsd sendfile is a dire hack with no thoughts 1121 * Of course, the freebsd sendfile is a dire hack with no thoughts
849 * wasted on making it similar to other I/O functions. 1122 * wasted on making it similar to other I/O functions.
850 */ 1123 */
851 {
852 off_t sbytes; 1124 off_t sbytes;
853 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 1125 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
854 1126
855 if (res < 0 && sbytes) 1127 #if 0 /* according to the manpage, this is correct, but broken behaviour */
856 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 1128 /* freebsd' sendfile will return 0 on success */
1129 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
1130 /* not on e.g. EIO or EPIPE - sounds broken */
1131 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
857 res = sbytes; 1132 res = sbytes;
858 } 1133 #endif
1134
1135 /* according to source inspection, this is correct, and useful behaviour */
1136 if (sbytes)
1137 res = sbytes;
1138
1139# elif defined __APPLE__
1140 off_t sbytes = count;
1141 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
1142
1143 /* according to the manpage, sbytes is always valid */
1144 if (sbytes)
1145 res = sbytes;
859 1146
860# elif __hpux 1147# elif __hpux
861 res = sendfile (ofd, ifd, offset, count, 0, 0); 1148 res = sendfile (ofd, ifd, offset, count, 0, 0);
862 1149
863# elif __solaris 1150# elif __solaris
864 {
865 struct sendfilevec vec; 1151 struct sendfilevec vec;
866 size_t sbytes; 1152 size_t sbytes;
867 1153
868 vec.sfv_fd = ifd; 1154 vec.sfv_fd = ifd;
869 vec.sfv_flag = 0; 1155 vec.sfv_flag = 0;
870 vec.sfv_off = offset; 1156 vec.sfv_off = offset;
871 vec.sfv_len = count; 1157 vec.sfv_len = count;
872 1158
873 res = sendfilev (ofd, &vec, 1, &sbytes); 1159 res = sendfilev (ofd, &vec, 1, &sbytes);
874 1160
875 if (res < 0 && sbytes) 1161 if (res < 0 && sbytes)
876 res = sbytes; 1162 res = sbytes;
877 }
878 1163
879# endif 1164# endif
1165
1166#elif defined (_WIN32) && 0
1167 /* does not work, just for documentation of what would need to be done */
1168 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
1169 /* libeio guarantees that the file offset does not change, and windows */
1170 /* has no way to get an independent handle to the same file description */
1171 HANDLE h = TO_SOCKET (ifd);
1172 SetFilePointer (h, offset, 0, FILE_BEGIN);
1173 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1174
880#else 1175#else
881 res = -1; 1176 res = EIO_ENOSYS ();
882 errno = ENOSYS;
883#endif 1177#endif
884 1178
1179 /* we assume sendfile can copy at least 128mb in one go */
1180 if (res <= 128 * 1024 * 1024)
1181 {
1182 if (res > 0)
1183 written += res;
1184
1185 if (written)
1186 return written;
1187
1188 break;
1189 }
1190 else
1191 {
1192 /* if we requested more, then probably the kernel was lazy */
1193 written += res;
1194 offset += res;
1195 count -= res;
1196
1197 if (!count)
1198 return written;
1199 }
1200 }
1201
885 if (res < 0 1202 if (res < 0
886 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1203 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1204 /* BSDs */
1205#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1206 || errno == ENOTSUP
1207#endif
1208#ifdef EOPNOTSUPP /* windows */
1209 || errno == EOPNOTSUPP /* BSDs */
1210#endif
887#if __solaris 1211#if __solaris
888 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1212 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
889#endif 1213#endif
890 ) 1214 )
891 ) 1215 )
… …
895 1219
896 res = 0; 1220 res = 0;
897 1221
898 while (count) 1222 while (count)
899 { 1223 {
900 ssize_t cnt; 1224 eio_ssize_t cnt;
901 1225
902 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 1226 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
903 1227
904 if (cnt <= 0) 1228 if (cnt <= 0)
905 { 1229 {
… …
917 1241
918 offset += cnt; 1242 offset += cnt;
919 res += cnt; 1243 res += cnt;
920 count -= cnt; 1244 count -= cnt;
921 } 1245 }
1246
1247 FUBd;
922 } 1248 }
923 1249
924 return res; 1250 return res;
1251}
1252
1253#ifdef PAGESIZE
1254# define eio_pagesize() PAGESIZE
1255#else
1256static intptr_t
1257eio_pagesize (void)
1258{
1259 static intptr_t page;
1260
1261 if (!page)
1262 page = sysconf (_SC_PAGESIZE);
1263
1264 return page;
1265}
1266#endif
1267
1268static void
1269eio_page_align (void **addr, size_t *length)
1270{
1271 intptr_t mask = eio_pagesize () - 1;
1272
1273 /* round down addr */
1274 intptr_t adj = mask & (intptr_t)*addr;
1275
1276 *addr = (void *)((intptr_t)*addr - adj);
1277 *length += adj;
1278
1279 /* round up length */
1280 *length = (*length + mask) & ~mask;
1281}
1282
1283#if !_POSIX_MEMLOCK
1284# define eio__mlockall(a) EIO_ENOSYS ()
1285#else
1286
1287static int
1288eio__mlockall (int flags)
1289{
1290 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1291 extern int mallopt (int, int);
1292 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1293 #endif
1294
1295 if (EIO_MCL_CURRENT != MCL_CURRENT
1296 || EIO_MCL_FUTURE != MCL_FUTURE)
1297 {
1298 flags = 0
1299 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1300 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1301 }
1302
1303 return mlockall (flags);
1304}
1305#endif
1306
1307#if !_POSIX_MEMLOCK_RANGE
1308# define eio__mlock(a,b) EIO_ENOSYS ()
1309#else
1310
1311static int
1312eio__mlock (void *addr, size_t length)
1313{
1314 eio_page_align (&addr, &length);
1315
1316 return mlock (addr, length);
1317}
1318
1319#endif
1320
1321#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1322# define eio__msync(a,b,c) EIO_ENOSYS ()
1323#else
1324
1325static int
1326eio__msync (void *mem, size_t len, int flags)
1327{
1328 eio_page_align (&mem, &len);
1329
1330 if (EIO_MS_ASYNC != MS_SYNC
1331 || EIO_MS_INVALIDATE != MS_INVALIDATE
1332 || EIO_MS_SYNC != MS_SYNC)
1333 {
1334 flags = 0
1335 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1336 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1337 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1338 }
1339
1340 return msync (mem, len, flags);
1341}
1342
1343#endif
1344
1345static int
1346eio__mtouch (eio_req *req)
1347{
1348 void *mem = req->ptr2;
1349 size_t len = req->size;
1350 int flags = req->int1;
1351
1352 eio_page_align (&mem, &len);
1353
1354 {
1355 intptr_t addr = (intptr_t)mem;
1356 intptr_t end = addr + len;
1357 intptr_t page = eio_pagesize ();
1358
1359 if (addr < end)
1360 if (flags & EIO_MT_MODIFY) /* modify */
1361 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1362 else
1363 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1364 }
1365
1366 return 0;
1367}
1368
1369/*****************************************************************************/
1370/* requests implemented outside eio_execute, because they are so large */
1371
1372static void
1373eio__lseek (eio_req *req)
1374{
1375 /* this usually gets optimised away completely, or your compiler sucks, */
1376 /* or the whence constants really are not 0, 1, 2 */
1377 int whence = req->int2 == EIO_SEEK_SET ? SEEK_SET
1378 : req->int2 == EIO_SEEK_CUR ? SEEK_CUR
1379 : req->int2 == EIO_SEEK_END ? SEEK_END
1380 : req->int2;
1381
1382 req->offs = lseek (req->int1, req->offs, whence);
1383 req->result = req->offs == (off_t)-1 ? -1 : 0;
1384}
1385
1386/* result will always end up in tmpbuf, there is always space for adding a 0-byte */
1387static int
1388eio__realpath (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
1389{
1390 const char *rel = path;
1391 char *res;
1392 char *tmp1, *tmp2;
1393#if SYMLOOP_MAX > 32
1394 int symlinks = SYMLOOP_MAX;
1395#else
1396 int symlinks = 32;
1397#endif
1398
1399 errno = EINVAL;
1400 if (!rel)
1401 return -1;
1402
1403 errno = ENOENT;
1404 if (!*rel)
1405 return -1;
1406
1407 res = tmpbuf_get (tmpbuf, PATH_MAX * 3);
1408 tmp1 = res + PATH_MAX;
1409 tmp2 = tmp1 + PATH_MAX;
1410
1411#if 0 /* disabled, the musl way to do things is just too racy */
1412#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1413 /* on linux we may be able to ask the kernel */
1414 {
1415 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1416
1417 if (fd >= 0)
1418 {
1419 sprintf (tmp1, "/proc/self/fd/%d", fd);
1420 req->result = readlink (tmp1, res, PATH_MAX);
1421 /* here we should probably stat the open file and the disk file, to make sure they still match */
1422 close (fd);
1423
1424 if (req->result > 0)
1425 goto done;
1426 }
1427 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1428 return -1;
1429 }
1430#endif
1431#endif
1432
1433 if (*rel != '/')
1434 {
1435 int len;
1436
1437 errno = ENOENT;
1438 if (wd == EIO_INVALID_WD)
1439 return -1;
1440
1441 if (wd == EIO_CWD)
1442 {
1443 if (!getcwd (res, PATH_MAX))
1444 return -1;
1445
1446 len = strlen (res);
1447 }
1448 else
1449 memcpy (res, wd->str, len = wd->len);
1450
1451 if (res [1]) /* only use if not / */
1452 res += len;
1453 }
1454
1455 while (*rel)
1456 {
1457 eio_ssize_t len, linklen;
1458 const char *beg = rel;
1459
1460 while (*rel && *rel != '/')
1461 ++rel;
1462
1463 len = rel - beg;
1464
1465 if (!len) /* skip slashes */
1466 {
1467 ++rel;
1468 continue;
1469 }
1470
1471 if (beg [0] == '.')
1472 {
1473 if (len == 1)
1474 continue; /* . - nop */
1475
1476 if (beg [1] == '.' && len == 2)
1477 {
1478 /* .. - back up one component, if possible */
1479
1480 while (res != tmpbuf->ptr)
1481 if (*--res == '/')
1482 break;
1483
1484 continue;
1485 }
1486 }
1487
1488 errno = ENAMETOOLONG;
1489 if (res + 1 + len + 1 >= tmp1)
1490 return -1;
1491
1492 /* copy one component */
1493 *res = '/';
1494 memcpy (res + 1, beg, len);
1495
1496 /* zero-terminate, for readlink */
1497 res [len + 1] = 0;
1498
1499 /* now check if it's a symlink */
1500 linklen = readlink (tmpbuf->ptr, tmp1, PATH_MAX);
1501
1502 if (linklen < 0)
1503 {
1504 if (errno != EINVAL)
1505 return -1;
1506
1507 /* it's a normal directory. hopefully */
1508 res += len + 1;
1509 }
1510 else
1511 {
1512 /* yay, it was a symlink - build new path in tmp2 */
1513 int rellen = strlen (rel);
1514
1515 errno = ENAMETOOLONG;
1516 if (linklen + 1 + rellen >= PATH_MAX)
1517 return -1;
1518
1519 errno = ELOOP;
1520 if (!--symlinks)
1521 return -1;
1522
1523 if (*tmp1 == '/')
1524 res = tmpbuf->ptr; /* symlink resolves to an absolute path */
1525
1526 /* we need to be careful, as rel might point into tmp2 already */
1527 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1528 tmp2 [linklen] = '/';
1529 memcpy (tmp2, tmp1, linklen);
1530
1531 rel = tmp2;
1532 }
1533 }
1534
1535 /* special case for the lone root path */
1536 if (res == tmpbuf->ptr)
1537 *res++ = '/';
1538
1539 return res - (char *)tmpbuf->ptr;
1540}
1541
1542static signed char
1543eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1544{
1545 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1546 : a->inode < b->inode ? -1
1547 : a->inode > b->inode ? 1
1548 : 0;
1549}
1550
1551#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1552
1553#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1554#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1555
1556static void
1557eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1558{
1559 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1560 unsigned char *bit = bits;
1561
1562 assert (CHAR_BIT == 8);
1563 assert (sizeof (eio_dirent) * 8 < 256);
1564 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1565 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1566
1567 if (size <= EIO_SORT_FAST)
1568 return;
1569
1570 /* first prepare an array of bits to test in our radix sort */
1571 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1572 /* inode_bits must contain all inodes ORed together */
1573 /* which is used to skip bits that are 0 everywhere, which is very common */
1574 {
1575 eio_ino_t endianness;
1576 int i, j;
1577
1578 /* we store the byte offset of byte n into byte n of "endianness" */
1579 for (i = 0; i < sizeof (eio_ino_t); ++i)
1580 ((unsigned char *)&endianness)[i] = i;
1581
1582 *bit++ = 0;
1583
1584 for (i = 0; i < sizeof (eio_ino_t); ++i)
1585 {
1586 /* shifting off the byte offsets out of "endianness" */
1587 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1588 endianness >>= 8;
1589
1590 for (j = 0; j < 8; ++j)
1591 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1592 *bit++ = offs + j;
1593 }
1594
1595 for (j = 0; j < 8; ++j)
1596 if (score_bits & (1 << j))
1597 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1598 }
1599
1600 /* now actually do the sorting (a variant of MSD radix sort) */
1601 {
1602 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1603 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1604 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1605 int stk_idx = 0;
1606
1607 base_stk [stk_idx] = dents;
1608 end_stk [stk_idx] = dents + size;
1609 bit_stk [stk_idx] = bit - 1;
1610
1611 do
1612 {
1613 base = base_stk [stk_idx];
1614 end = end_stk [stk_idx];
1615 bit = bit_stk [stk_idx];
1616
1617 for (;;)
1618 {
1619 unsigned char O = *bit >> 3;
1620 unsigned char M = 1 << (*bit & 7);
1621
1622 eio_dirent *a = base;
1623 eio_dirent *b = end;
1624
1625 if (b - a < EIO_SORT_CUTOFF)
1626 break;
1627
1628 /* now bit-partition the array on the bit */
1629 /* this ugly asymmetric loop seems to perform much better than typical */
1630 /* partition algos found in the literature */
1631 do
1632 if (!(((unsigned char *)a)[O] & M))
1633 ++a;
1634 else if (!(((unsigned char *)--b)[O] & M))
1635 {
1636 eio_dirent tmp = *a; *a = *b; *b = tmp;
1637 ++a;
1638 }
1639 while (b > a);
1640
1641 /* next bit, or stop, if no bits left in this path */
1642 if (!*--bit)
1643 break;
1644
1645 base_stk [stk_idx] = a;
1646 end_stk [stk_idx] = end;
1647 bit_stk [stk_idx] = bit;
1648 ++stk_idx;
1649
1650 end = a;
1651 }
1652 }
1653 while (stk_idx--);
1654 }
1655}
1656
1657static void
1658eio_dent_insertion_sort (eio_dirent *dents, int size)
1659{
1660 /* first move the smallest element to the front, to act as a sentinel */
1661 {
1662 int i;
1663 eio_dirent *min = dents;
1664
1665 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1666 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1667 if (EIO_DENT_CMP (dents [i], <, *min))
1668 min = &dents [i];
1669
1670 /* swap elements 0 and j (minimum) */
1671 {
1672 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1673 }
1674 }
1675
1676 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1677 {
1678 eio_dirent *i, *j;
1679
1680 for (i = dents + 1; i < dents + size; ++i)
1681 {
1682 eio_dirent value = *i;
1683
1684 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1685 j [1] = j [0];
1686
1687 j [1] = value;
1688 }
1689 }
1690}
1691
1692static void
1693eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1694{
1695 if (size <= 1)
1696 return; /* our insertion sort relies on size > 0 */
1697
1698 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1699 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1700 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1701
1702 /* use an insertion sort at the end, or for small arrays, */
1703 /* as insertion sort is more efficient for small partitions */
1704 eio_dent_insertion_sort (dents, size);
925} 1705}
926 1706
927/* read a full directory */ 1707/* read a full directory */
928static void 1708static void
929eio__scandir (eio_req *req, etp_worker *self) 1709eio__scandir (eio_req *req, etp_worker *self)
930{ 1710{
1711 char *name, *names;
1712 int namesalloc = 4096 - sizeof (void *) * 4;
1713 int namesoffs = 0;
1714 int flags = req->int1;
1715 eio_dirent *dents = 0;
1716 int dentalloc = 128;
1717 int dentoffs = 0;
1718 eio_ino_t inode_bits = 0;
1719#ifdef _WIN32
1720 HANDLE dirp;
1721 WIN32_FIND_DATA entp;
1722#else
931 DIR *dirp; 1723 DIR *dirp;
932 union
933 {
934 EIO_STRUCT_DIRENT d;
935 char b [offsetof (EIO_STRUCT_DIRENT, d_name) + NAME_MAX + 1];
936 } *u;
937 EIO_STRUCT_DIRENT *entp; 1724 EIO_STRUCT_DIRENT *entp;
938 char *name, *names; 1725#endif
939 int memlen = 4096;
940 int memofs = 0;
941 int res = 0;
942 1726
943 X_LOCK (wrklock); 1727 req->result = -1;
944 self->dirp = dirp = opendir (req->ptr1);
945 self->dbuf = u = malloc (sizeof (*u));
946 req->flags |= EIO_FLAG_PTR2_FREE;
947 req->ptr2 = names = malloc (memlen);
948 X_UNLOCK (wrklock);
949 1728
950 if (dirp && u && names) 1729 if (!(flags & EIO_READDIR_DENTS))
951 for (;;) 1730 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1731
1732#ifdef _WIN32
1733 {
1734 int len = strlen ((const char *)req->ptr1);
1735 char *path = malloc (MAX_PATH);
1736 const char *fmt;
1737 const char *reqpath = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1738
1739 if (!len)
1740 fmt = "./*";
1741 else if (reqpath[len - 1] == '/' || reqpath[len - 1] == '\\')
1742 fmt = "%s*";
1743 else
1744 fmt = "%s/*";
1745
1746 _snprintf (path, MAX_PATH, fmt, reqpath);
1747 dirp = FindFirstFile (path, &entp);
1748 free (path);
1749
1750 if (dirp == INVALID_HANDLE_VALUE)
1751 {
1752 /* should steal _dosmaperr */
1753 switch (GetLastError ())
1754 {
1755 case ERROR_FILE_NOT_FOUND:
1756 req->result = 0;
1757 break;
1758
1759 case ERROR_INVALID_NAME:
1760 case ERROR_PATH_NOT_FOUND:
1761 case ERROR_NO_MORE_FILES:
1762 errno = ENOENT;
1763 break;
1764
1765 case ERROR_NOT_ENOUGH_MEMORY:
1766 errno = ENOMEM;
1767 break;
1768
1769 default:
1770 errno = EINVAL;
1771 break;
1772 }
1773
1774 return;
1775 }
1776 }
1777#else
1778 #if HAVE_AT
1779 if (req->wd)
952 { 1780 {
1781 int fd = openat (WD2FD (req->wd), req->ptr1, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1782
1783 if (fd < 0)
1784 return;
1785
1786 dirp = fdopendir (fd);
1787
1788 if (!dirp)
1789 close (fd);
1790 }
1791 else
1792 dirp = opendir (req->ptr1);
1793 #else
1794 dirp = opendir (wd_expand (&self->tmpbuf, req->wd, req->ptr1));
1795 #endif
1796
1797 if (!dirp)
1798 return;
1799#endif
1800
1801 if (req->flags & EIO_FLAG_PTR1_FREE)
1802 free (req->ptr1);
1803
1804 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1805 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1806 req->ptr2 = names = malloc (namesalloc);
1807
1808 if (!names || (flags && !dents))
1809 return;
1810
1811 for (;;)
1812 {
1813 int done;
1814
1815#ifdef _WIN32
1816 done = !dirp;
1817#else
953 errno = 0; 1818 errno = 0;
954 readdir_r (dirp, &u->d, &entp); 1819 entp = readdir (dirp);
1820 done = !entp;
1821#endif
955 1822
1823 if (done)
1824 {
1825#ifndef _WIN32
1826 int old_errno = errno;
1827 closedir (dirp);
1828 errno = old_errno;
1829
956 if (!entp) 1830 if (errno)
1831 break;
1832#endif
1833
1834 /* sort etc. */
1835 req->int1 = flags;
1836 req->result = dentoffs;
1837
1838 if (flags & EIO_READDIR_STAT_ORDER)
1839 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1840 else if (flags & EIO_READDIR_DIRS_FIRST)
1841 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1842 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1843 else
1844 {
1845 /* in this case, all is known, and we just put dirs first and sort them */
1846 eio_dirent *oth = dents + dentoffs;
1847 eio_dirent *dir = dents;
1848
1849 /* now partition dirs to the front, and non-dirs to the back */
1850 /* by walking from both sides and swapping if necessary */
1851 while (oth > dir)
1852 {
1853 if (dir->type == EIO_DT_DIR)
1854 ++dir;
1855 else if ((--oth)->type == EIO_DT_DIR)
1856 {
1857 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1858
1859 ++dir;
1860 }
1861 }
1862
1863 /* now sort the dirs only (dirs all have the same score) */
1864 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1865 }
1866
957 break; 1867 break;
1868 }
958 1869
959 name = entp->d_name; 1870 /* now add the entry to our list(s) */
1871 name = D_NAME (entp);
960 1872
1873 /* skip . and .. entries */
961 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1874 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
962 { 1875 {
963 int len = strlen (name) + 1; 1876 int len = D_NAMLEN (entp) + 1;
964 1877
965 res++; 1878 while (ecb_expect_false (namesoffs + len > namesalloc))
966
967 while (memofs + len > memlen)
968 { 1879 {
969 memlen *= 2; 1880 namesalloc *= 2;
970 X_LOCK (wrklock);
971 req->ptr2 = names = realloc (names, memlen); 1881 req->ptr2 = names = realloc (names, namesalloc);
972 X_UNLOCK (wrklock);
973 1882
974 if (!names) 1883 if (!names)
1884 break;
1885 }
1886
1887 memcpy (names + namesoffs, name, len);
1888
1889 if (dents)
1890 {
1891 struct eio_dirent *ent;
1892
1893 if (ecb_expect_false (dentoffs == dentalloc))
1894 {
1895 dentalloc *= 2;
1896 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1897
1898 if (!dents)
975 break; 1899 break;
976 } 1900 }
977 1901
978 memcpy (names + memofs, name, len); 1902 ent = dents + dentoffs;
979 memofs += len; 1903
1904 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1905 ent->namelen = len - 1;
1906 ent->inode = D_INO (entp);
1907
1908 inode_bits |= ent->inode;
1909
1910 switch (D_TYPE (entp))
1911 {
1912 default:
1913 ent->type = EIO_DT_UNKNOWN;
1914 flags |= EIO_READDIR_FOUND_UNKNOWN;
1915 break;
1916
1917 #ifdef DT_FIFO
1918 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1919 #endif
1920 #ifdef DT_CHR
1921 case DT_CHR: ent->type = EIO_DT_CHR; break;
1922 #endif
1923 #ifdef DT_MPC
1924 case DT_MPC: ent->type = EIO_DT_MPC; break;
1925 #endif
1926 #ifdef DT_DIR
1927 case DT_DIR: ent->type = EIO_DT_DIR; break;
1928 #endif
1929 #ifdef DT_NAM
1930 case DT_NAM: ent->type = EIO_DT_NAM; break;
1931 #endif
1932 #ifdef DT_BLK
1933 case DT_BLK: ent->type = EIO_DT_BLK; break;
1934 #endif
1935 #ifdef DT_MPB
1936 case DT_MPB: ent->type = EIO_DT_MPB; break;
1937 #endif
1938 #ifdef DT_REG
1939 case DT_REG: ent->type = EIO_DT_REG; break;
1940 #endif
1941 #ifdef DT_NWK
1942 case DT_NWK: ent->type = EIO_DT_NWK; break;
1943 #endif
1944 #ifdef DT_CMP
1945 case DT_CMP: ent->type = EIO_DT_CMP; break;
1946 #endif
1947 #ifdef DT_LNK
1948 case DT_LNK: ent->type = EIO_DT_LNK; break;
1949 #endif
1950 #ifdef DT_SOCK
1951 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1952 #endif
1953 #ifdef DT_DOOR
1954 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1955 #endif
1956 #ifdef DT_WHT
1957 case DT_WHT: ent->type = EIO_DT_WHT; break;
1958 #endif
1959 }
1960
1961 ent->score = 7;
1962
1963 if (flags & EIO_READDIR_DIRS_FIRST)
1964 {
1965 if (ent->type == EIO_DT_UNKNOWN)
1966 {
1967 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1968 ent->score = 1;
1969 else if (!strchr (name, '.')) /* absence of dots indicate likely dirs */
1970 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1971 }
1972 else if (ent->type == EIO_DT_DIR)
1973 ent->score = 0;
1974 }
980 } 1975 }
1976
1977 namesoffs += len;
1978 ++dentoffs;
981 } 1979 }
982 1980
983 if (errno) 1981 if (EIO_CANCELLED (req))
984 res = -1; 1982 {
1983 errno = ECANCELED;
1984 break;
1985 }
1986
1987#ifdef _WIN32
1988 if (!FindNextFile (dirp, &entp))
1989 {
1990 FindClose (dirp);
1991 dirp = 0;
1992 }
1993#endif
1994 }
1995}
1996
1997/*****************************************************************************/
1998/* working directory stuff */
1999/* various deficiencies in the posix 2008 api force us to */
2000/* keep the absolute path in string form at all times */
2001/* fuck yeah. */
2002
2003#if !HAVE_AT
2004
2005/* a bit like realpath, but usually faster because it doesn'T have to return */
2006/* an absolute or canonical path */
2007static const char *
2008wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
2009{
2010 if (!wd || *path == '/')
2011 return path;
2012
2013 if (path [0] == '.' && !path [1])
2014 return wd->str;
2015
985 2016 {
986 req->result = res; 2017 int l1 = wd->len;
2018 int l2 = strlen (path);
2019
2020 char *res = tmpbuf_get (tmpbuf, l1 + l2 + 2);
2021
2022 memcpy (res, wd->str, l1);
2023 res [l1] = '/';
2024 memcpy (res + l1 + 1, path, l2 + 1);
2025
2026 return res;
2027 }
987} 2028}
2029
2030#endif
2031
2032static eio_wd
2033eio__wd_open_sync (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
2034{
2035 int fd;
2036 eio_wd res;
2037 int len = eio__realpath (tmpbuf, wd, path);
2038
2039 if (len < 0)
2040 return EIO_INVALID_WD;
2041
2042#if HAVE_AT
2043 fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
2044
2045 if (fd < 0)
2046 return EIO_INVALID_WD;
2047#endif
2048
2049 res = malloc (sizeof (*res) + len); /* one extra 0-byte */
2050
2051#if HAVE_AT
2052 res->fd = fd;
2053#endif
2054
2055 res->len = len;
2056 memcpy (res->str, tmpbuf->ptr, len);
2057 res->str [len] = 0;
2058
2059 return res;
2060}
2061
2062eio_wd
2063eio_wd_open_sync (eio_wd wd, const char *path)
2064{
2065 struct tmpbuf tmpbuf = { 0 };
2066 wd = eio__wd_open_sync (&tmpbuf, wd, path);
2067 free (tmpbuf.ptr);
2068
2069 return wd;
2070}
2071
2072void
2073eio_wd_close_sync (eio_wd wd)
2074{
2075 if (wd != EIO_INVALID_WD && wd != EIO_CWD)
2076 {
2077 #if HAVE_AT
2078 close (wd->fd);
2079 #endif
2080 free (wd);
2081 }
2082}
2083
2084#if HAVE_AT
2085
2086/* they forgot these */
2087
2088static int
2089eio__truncateat (int dirfd, const char *path, off_t length)
2090{
2091 int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
2092 int res;
2093
2094 if (fd < 0)
2095 return fd;
2096
2097 res = ftruncate (fd, length);
2098 close (fd);
2099 return res;
2100}
2101
2102static int
2103eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
2104{
2105 int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
2106 int res;
2107
2108 if (fd < 0)
2109 return fd;
2110
2111 res = fstatvfs (fd, buf);
2112 close (fd);
2113 return res;
2114
2115}
2116
2117#endif
988 2118
989/*****************************************************************************/ 2119/*****************************************************************************/
990 2120
991#define ALLOC(len) \ 2121#define ALLOC(len) \
992 if (!req->ptr2) \ 2122 if (!req->ptr2) \
… …
1001 req->result = -1; \ 2131 req->result = -1; \
1002 break; \ 2132 break; \
1003 } \ 2133 } \
1004 } 2134 }
1005 2135
2136static void ecb_noinline ecb_cold
2137etp_proc_init (void)
2138{
2139#if HAVE_PRCTL_SET_NAME
2140 /* provide a more sensible "thread name" */
2141 char name[16 + 1];
2142 const int namelen = sizeof (name) - 1;
2143 int len;
2144
2145 prctl (PR_GET_NAME, (unsigned long)name, 0, 0, 0);
2146 name [namelen] = 0;
2147 len = strlen (name);
2148 strcpy (name + (len <= namelen - 4 ? len : namelen - 4), "/eio");
2149 prctl (PR_SET_NAME, (unsigned long)name, 0, 0, 0);
2150#endif
2151}
2152
1006X_THREAD_PROC (etp_proc) 2153X_THREAD_PROC (etp_proc)
1007{ 2154{
1008 ETP_REQ *req; 2155 ETP_REQ *req;
1009 struct timespec ts; 2156 struct timespec ts;
1010 etp_worker *self = (etp_worker *)thr_arg; 2157 etp_worker *self = (etp_worker *)thr_arg;
1011 2158
2159 etp_proc_init ();
2160
1012 /* try to distribute timeouts somewhat randomly */ 2161 /* try to distribute timeouts somewhat evenly */
1013 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 2162 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1014 2163
1015 for (;;) 2164 for (;;)
1016 { 2165 {
2166 ts.tv_sec = 0;
2167
1017 X_LOCK (reqlock); 2168 X_LOCK (reqlock);
1018 2169
1019 for (;;) 2170 for (;;)
1020 { 2171 {
1021 self->req = req = reqq_shift (&req_queue); 2172 req = reqq_shift (&req_queue);
1022 2173
1023 if (req) 2174 if (req)
1024 break; 2175 break;
1025 2176
2177 if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
2178 {
2179 X_UNLOCK (reqlock);
2180 X_LOCK (wrklock);
2181 --started;
2182 X_UNLOCK (wrklock);
2183 goto quit;
2184 }
2185
1026 ++idle; 2186 ++idle;
1027 2187
1028 ts.tv_sec = time (0) + IDLE_TIMEOUT; 2188 if (idle <= max_idle)
1029 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 2189 /* we are allowed to idle, so do so without any timeout */
2190 X_COND_WAIT (reqwait, reqlock);
2191 else
1030 { 2192 {
1031 if (idle > max_idle) 2193 /* initialise timeout once */
1032 { 2194 if (!ts.tv_sec)
1033 --idle; 2195 ts.tv_sec = time (0) + idle_timeout;
1034 X_UNLOCK (reqlock);
1035 X_LOCK (wrklock);
1036 --started;
1037 X_UNLOCK (wrklock);
1038 goto quit;
1039 }
1040 2196
1041 /* we are allowed to idle, so do so without any timeout */
1042 X_COND_WAIT (reqwait, reqlock); 2197 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
2198 ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1043 } 2199 }
1044 2200
1045 --idle; 2201 --idle;
1046 } 2202 }
1047 2203
1048 --nready; 2204 --nready;
1049 2205
1050 X_UNLOCK (reqlock); 2206 X_UNLOCK (reqlock);
1051 2207
1052 errno = 0; /* strictly unnecessary */
1053
1054 if (!EIO_CANCELLED (req))
1055 switch (req->type) 2208 if (req->type < 0)
1056 {
1057 case EIO_READ: ALLOC (req->size);
1058 req->result = req->offs >= 0
1059 ? pread (req->int1, req->ptr2, req->size, req->offs)
1060 : read (req->int1, req->ptr2, req->size); break;
1061 case EIO_WRITE: req->result = req->offs >= 0
1062 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1063 : write (req->int1, req->ptr2, req->size); break;
1064
1065 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1066 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;
1067
1068 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1069 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1070 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1071 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1072 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1073 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1074
1075 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1076 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1077 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1078 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1079 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1080 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1081
1082 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
1083 case EIO_CLOSE: req->result = close (req->int1); break;
1084 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1085 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
1086 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1087 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1088 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1089 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1090 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1091 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1092
1093 case EIO_READLINK: ALLOC (NAME_MAX);
1094 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
1095
1096 case EIO_SYNC: req->result = 0; sync (); break;
1097 case EIO_FSYNC: req->result = fsync (req->int1); break;
1098 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1099
1100 case EIO_READDIR: eio__scandir (req, self); break;
1101
1102 case EIO_BUSY:
1103#ifdef _WIN32
1104 Sleep (req->nv1 * 1000.);
1105#else
1106 {
1107 struct timeval tv;
1108
1109 tv.tv_sec = req->nv1;
1110 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;
1111
1112 req->result = select (0, 0, 0, 0, &tv);
1113 }
1114#endif
1115 break;
1116
1117 case EIO_UTIME:
1118 case EIO_FUTIME:
1119 {
1120 struct timeval tv[2];
1121 struct timeval *times;
1122
1123 if (req->nv1 != -1. || req->nv2 != -1.)
1124 {
1125 tv[0].tv_sec = req->nv1;
1126 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
1127 tv[1].tv_sec = req->nv2;
1128 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
1129
1130 times = tv;
1131 }
1132 else
1133 times = 0;
1134
1135
1136 req->result = req->type == EIO_FUTIME
1137 ? futimes (req->int1, times)
1138 : utimes (req->ptr1, times);
1139 }
1140
1141 case EIO_GROUP:
1142 case EIO_NOP:
1143 req->result = 0;
1144 break;
1145
1146 case -1:
1147 goto quit; 2209 goto quit;
1148 2210
1149 default: 2211 ETP_EXECUTE (self, req);
1150 req->result = -1;
1151 break;
1152 }
1153
1154 req->errorno = errno;
1155 2212
1156 X_LOCK (reslock); 2213 X_LOCK (reslock);
1157 2214
1158 ++npending; 2215 ++npending;
1159 2216
1160 if (!reqq_push (&res_queue, req) && want_poll_cb) 2217 if (!reqq_push (&res_queue, req) && want_poll_cb)
1161 want_poll_cb (); 2218 want_poll_cb ();
1162 2219
1163 self->req = 0;
1164 etp_worker_clear (self); 2220 etp_worker_clear (self);
1165 2221
1166 X_UNLOCK (reslock); 2222 X_UNLOCK (reslock);
1167 } 2223 }
1168 2224
1169quit: 2225quit:
2226 free (req);
2227
1170 X_LOCK (wrklock); 2228 X_LOCK (wrklock);
1171 etp_worker_free (self); 2229 etp_worker_free (self);
1172 X_UNLOCK (wrklock); 2230 X_UNLOCK (wrklock);
1173 2231
1174 return 0; 2232 return 0;
1175} 2233}
1176 2234
1177/*****************************************************************************/ 2235/*****************************************************************************/
1178 2236
2237int ecb_cold
1179int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 2238eio_init (void (*want_poll)(void), void (*done_poll)(void))
1180{ 2239{
1181 etp_init (want_poll, done_poll); 2240 return etp_init (want_poll, done_poll);
1182} 2241}
1183 2242
2243ecb_inline void
1184static void eio_api_destroy (eio_req *req) 2244eio_api_destroy (eio_req *req)
1185{ 2245{
1186 free (req); 2246 free (req);
1187} 2247}
1188 2248
1189#define REQ(rtype) \ 2249#define REQ(rtype) \
1190 eio_req *req; \ 2250 eio_req *req; \
1191 \ 2251 \
1192 req = (eio_req *)calloc (1, sizeof *req); \ 2252 req = (eio_req *)calloc (1, sizeof *req); \
1193 if (!req) \ 2253 if (!req) \
1194 return 0; \ 2254 return 0; \
… …
1208 { \ 2268 { \
1209 eio_api_destroy (req); \ 2269 eio_api_destroy (req); \
1210 return 0; \ 2270 return 0; \
1211 } 2271 }
1212 2272
2273#define SINGLEDOT(ptr) (0[(char *)(ptr)] == '.' && !1[(char *)(ptr)])
2274
2275static void
2276eio_execute (etp_worker *self, eio_req *req)
2277{
2278#if HAVE_AT
2279 int dirfd;
2280#else
2281 const char *path;
2282#endif
2283
2284 if (ecb_expect_false (EIO_CANCELLED (req)))
2285 {
2286 req->result = -1;
2287 req->errorno = ECANCELED;
2288 return;
2289 }
2290
2291 if (ecb_expect_false (req->wd == EIO_INVALID_WD))
2292 {
2293 req->result = -1;
2294 req->errorno = ENOENT;
2295 return;
2296 }
2297
2298 if (req->type >= EIO_OPEN)
2299 {
2300 #if HAVE_AT
2301 dirfd = WD2FD (req->wd);
2302 #else
2303 path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
2304 #endif
2305 }
2306
2307 switch (req->type)
2308 {
2309 case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
2310 req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
2311 break;
2312 case EIO_WD_CLOSE: req->result = 0;
2313 eio_wd_close_sync (req->wd); break;
2314
2315 case EIO_SEEK: eio__lseek (req); break;
2316 case EIO_READ: ALLOC (req->size);
2317 req->result = req->offs >= 0
2318 ? pread (req->int1, req->ptr2, req->size, req->offs)
2319 : read (req->int1, req->ptr2, req->size); break;
2320 case EIO_WRITE: req->result = req->offs >= 0
2321 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
2322 : write (req->int1, req->ptr2, req->size); break;
2323
2324 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
2325 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
2326
2327#if HAVE_AT
2328
2329 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2330 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break;
2331 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2332 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, AT_SYMLINK_NOFOLLOW); break;
2333 case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break;
2334 case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break;
2335 case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break;
2336 case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break;
2337
2338 case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
2339 case EIO_RMDIR: /* complications arise because "." cannot be removed, so we might have to expand */
2340 req->result = req->wd && SINGLEDOT (req->ptr1)
2341 ? rmdir (req->wd->str)
2342 : unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break;
2343 case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); break;
2344 case EIO_RENAME: /* complications arise because "." cannot be renamed, so we might have to expand */
2345 req->result = req->wd && SINGLEDOT (req->ptr1)
2346 ? rename (req->wd->str, req->ptr2)
2347 : renameat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2); break;
2348 case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2, 0); break;
2349 case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
2350 case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
2351 case EIO_READLINK: ALLOC (PATH_MAX);
2352 req->result = readlinkat (dirfd, req->ptr1, req->ptr2, PATH_MAX); break;
2353 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2354 req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2355 case EIO_UTIME:
2356 case EIO_FUTIME:
2357 {
2358 struct timespec ts[2];
2359 struct timespec *times;
2360
2361 if (req->nv1 != -1. || req->nv2 != -1.)
2362 {
2363 ts[0].tv_sec = req->nv1;
2364 ts[0].tv_nsec = (req->nv1 - ts[0].tv_sec) * 1e9;
2365 ts[1].tv_sec = req->nv2;
2366 ts[1].tv_nsec = (req->nv2 - ts[1].tv_sec) * 1e9;
2367
2368 times = ts;
2369 }
2370 else
2371 times = 0;
2372
2373 req->result = req->type == EIO_FUTIME
2374 ? futimens (req->int1, times)
2375 : utimensat (dirfd, req->ptr1, times, 0);
2376 }
2377 break;
2378
2379#else
2380
2381 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2382 req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
2383 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2384 req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
2385 case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
2386 case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
2387 case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
2388 case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
2389
2390 case EIO_UNLINK: req->result = unlink (path ); break;
2391 case EIO_RMDIR: req->result = rmdir (path ); break;
2392 case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
2393 case EIO_RENAME: req->result = rename (path , req->ptr2); break;
2394 case EIO_LINK: req->result = link (path , req->ptr2); break;
2395 case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
2396 case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
2397 case EIO_READLINK: ALLOC (PATH_MAX);
2398 req->result = readlink (path, req->ptr2, PATH_MAX); break;
2399 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2400 req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
2401
2402 case EIO_UTIME:
2403 case EIO_FUTIME:
2404 {
2405 struct timeval tv[2];
2406 struct timeval *times;
2407
2408 if (req->nv1 != -1. || req->nv2 != -1.)
2409 {
2410 tv[0].tv_sec = req->nv1;
2411 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
2412 tv[1].tv_sec = req->nv2;
2413 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
2414
2415 times = tv;
2416 }
2417 else
2418 times = 0;
2419
2420 req->result = req->type == EIO_FUTIME
2421 ? futimes (req->int1, times)
2422 : utimes (req->ptr1, times);
2423 }
2424 break;
2425
2426#endif
2427
2428 case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
2429 {
2430 ALLOC (req->result);
2431 memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
2432 }
2433 break;
2434
2435 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2436 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
2437
2438 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2439 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2440
2441 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
2442 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
2443 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
2444
2445 case EIO_CLOSE: req->result = close (req->int1); break;
2446 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
2447 case EIO_SYNC: req->result = 0; sync (); break;
2448 case EIO_FSYNC: req->result = fsync (req->int1); break;
2449 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
2450 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
2451 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
2452 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
2453 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
2454 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
2455 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
2456 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
2457
2458 case EIO_READDIR: eio__scandir (req, self); break;
2459
2460 case EIO_BUSY:
2461#ifdef _WIN32
2462 Sleep (req->nv1 * 1e3);
2463#else
2464 {
2465 struct timeval tv;
2466
2467 tv.tv_sec = req->nv1;
2468 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
2469
2470 req->result = select (0, 0, 0, 0, &tv);
2471 }
2472#endif
2473 break;
2474
2475 case EIO_GROUP:
2476 abort (); /* handled in eio_request */
2477
2478 case EIO_NOP:
2479 req->result = 0;
2480 break;
2481
2482 case EIO_CUSTOM:
2483 req->feed (req);
2484 break;
2485
2486 default:
2487 req->result = EIO_ENOSYS ();
2488 break;
2489 }
2490
2491 req->errorno = errno;
2492}
2493
1213#ifndef EIO_NO_WRAPPERS 2494#ifndef EIO_NO_WRAPPERS
1214 2495
2496eio_req *eio_wd_open (const char *path, int pri, eio_cb cb, void *data)
2497{
2498 REQ (EIO_WD_OPEN); PATH; SEND;
2499}
2500
2501eio_req *eio_wd_close (eio_wd wd, int pri, eio_cb cb, void *data)
2502{
2503 REQ (EIO_WD_CLOSE); req->wd = wd; SEND;
2504}
2505
1215eio_req *eio_nop (int pri, eio_cb cb, void *data) 2506eio_req *eio_nop (int pri, eio_cb cb, void *data)
1216{ 2507{
1217 REQ (EIO_NOP); SEND; 2508 REQ (EIO_NOP); SEND;
1218} 2509}
1219 2510
… …
1230eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data) 2521eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
1231{ 2522{
1232 REQ (EIO_FSYNC); req->int1 = fd; SEND; 2523 REQ (EIO_FSYNC); req->int1 = fd; SEND;
1233} 2524}
1234 2525
2526eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2527{
2528 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2529}
2530
1235eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) 2531eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1236{ 2532{
1237 REQ (EIO_FDATASYNC); req->int1 = fd; SEND; 2533 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1238} 2534}
1239 2535
2536eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
2537{
2538 REQ (EIO_SYNCFS); req->int1 = fd; SEND;
2539}
2540
2541eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2542{
2543 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2544}
2545
2546eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2547{
2548 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2549}
2550
2551eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
2552{
2553 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
2554}
2555
2556eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
2557{
2558 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2559}
2560
2561eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2562{
2563 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
2564}
2565
1240eio_req *eio_close (int fd, int pri, eio_cb cb, void *data) 2566eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
1241{ 2567{
1242 REQ (EIO_CLOSE); req->int1 = fd; SEND; 2568 REQ (EIO_CLOSE); req->int1 = fd; SEND;
1243} 2569}
1244 2570
1245eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data) 2571eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
1246{ 2572{
1247 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND; 2573 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
1248} 2574}
1249 2575
2576eio_req *eio_seek (int fd, off_t offset, int whence, int pri, eio_cb cb, void *data)
2577{
2578 REQ (EIO_SEEK); req->int1 = fd; req->offs = offset; req->int2 = whence; SEND;
2579}
2580
1250eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data) 2581eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1251{ 2582{
1252 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND; 2583 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1253} 2584}
1254 2585
… …
1260eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) 2591eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1261{ 2592{
1262 REQ (EIO_FSTAT); req->int1 = fd; SEND; 2593 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1263} 2594}
1264 2595
2596eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
2597{
2598 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
2599}
2600
1265eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) 2601eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1266{ 2602{
1267 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; 2603 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1268} 2604}
1269 2605
… …
1275eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) 2611eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1276{ 2612{
1277 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2613 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1278} 2614}
1279 2615
1280eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2616eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1281{ 2617{
1282 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2618 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1283} 2619}
1284 2620
1285eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) 2621eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
… …
1305eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data) 2641eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1306{ 2642{
1307 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2643 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1308} 2644}
1309 2645
1310eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2646eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1311{ 2647{
1312 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2648 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1313} 2649}
1314 2650
1315eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) 2651eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
… …
1331eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data) 2667eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1332{ 2668{
1333 return eio__1path (EIO_READLINK, path, pri, cb, data); 2669 return eio__1path (EIO_READLINK, path, pri, cb, data);
1334} 2670}
1335 2671
2672eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2673{
2674 return eio__1path (EIO_REALPATH, path, pri, cb, data);
2675}
2676
1336eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) 2677eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1337{ 2678{
1338 return eio__1path (EIO_STAT, path, pri, cb, data); 2679 return eio__1path (EIO_STAT, path, pri, cb, data);
1339} 2680}
1340 2681
1341eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) 2682eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1342{ 2683{
1343 return eio__1path (EIO_LSTAT, path, pri, cb, data); 2684 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1344} 2685}
1345 2686
2687eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
2688{
2689 return eio__1path (EIO_STATVFS, path, pri, cb, data);
2690}
2691
1346eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) 2692eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1347{ 2693{
1348 return eio__1path (EIO_UNLINK, path, pri, cb, data); 2694 return eio__1path (EIO_UNLINK, path, pri, cb, data);
1349} 2695}
1350 2696
1351eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data) 2697eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1352{ 2698{
1353 return eio__1path (EIO_RMDIR, path, pri, cb, data); 2699 return eio__1path (EIO_RMDIR, path, pri, cb, data);
1354} 2700}
1355 2701
1356eio_req *eio_readdir (const char *path, int pri, eio_cb cb, void *data) 2702eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1357{ 2703{
1358 return eio__1path (EIO_READDIR, path, pri, cb, data); 2704 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1359} 2705}
1360 2706
1361eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) 2707eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1362{ 2708{
1363 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int2 = (long)dev; SEND; 2709 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1364} 2710}
1365 2711
1366static eio_req * 2712static eio_req *
1367eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2713eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1368{ 2714{
… …
1392eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2738eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1393{ 2739{
1394 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2740 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1395} 2741}
1396 2742
2743eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
2744{
2745 REQ (EIO_CUSTOM); req->feed = execute; SEND;
2746}
2747
1397#endif 2748#endif
1398 2749
1399eio_req *eio_grp (eio_cb cb, void *data) 2750eio_req *eio_grp (eio_cb cb, void *data)
1400{ 2751{
1401 const int pri = EIO_PRI_MAX; 2752 const int pri = EIO_PRI_MAX;
… …
1408#undef SEND 2759#undef SEND
1409 2760
1410/*****************************************************************************/ 2761/*****************************************************************************/
1411/* grp functions */ 2762/* grp functions */
1412 2763
2764void
1413void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2765eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1414{ 2766{
1415 grp->int2 = limit; 2767 grp->int2 = limit;
1416 grp->feed = feed; 2768 grp->feed = feed;
1417 2769
1418 grp_try_feed (grp); 2770 grp_try_feed (grp);
1419} 2771}
1420 2772
2773void
1421void eio_grp_limit (eio_req *grp, int limit) 2774eio_grp_limit (eio_req *grp, int limit)
1422{ 2775{
1423 grp->int2 = limit; 2776 grp->int2 = limit;
1424 2777
1425 grp_try_feed (grp); 2778 grp_try_feed (grp);
1426} 2779}
1427 2780
2781void
1428void eio_grp_add (eio_req *grp, eio_req *req) 2782eio_grp_add (eio_req *grp, eio_req *req)
1429{ 2783{
1430 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2784 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
2785
2786 grp->flags |= EIO_FLAG_GROUPADD;
1431 2787
1432 ++grp->size; 2788 ++grp->size;
1433 req->grp = grp; 2789 req->grp = grp;
1434 2790
1435 req->grp_prev = 0; 2791 req->grp_prev = 0;
… …
1442} 2798}
1443 2799
1444/*****************************************************************************/ 2800/*****************************************************************************/
1445/* misc garbage */ 2801/* misc garbage */
1446 2802
2803eio_ssize_t
1447ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2804eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1448{ 2805{
1449 etp_worker wrk;
1450
1451 wrk.dbuf = 0;
1452
1453 eio__sendfile (ofd, ifd, offset, count, &wrk); 2806 return eio__sendfile (ofd, ifd, offset, count);
1454
1455 if (wrk.dbuf)
1456 free (wrk.dbuf);
1457} 2807}
1458 2808

Diff Legend

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