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

Comparing libeio/eio.c (file contents):
Revision 1.8 by root, Sun May 11 20:08:37 2008 UTC vs.
Revision 1.128 by root, Mon Feb 18 03:16:30 2013 UTC

1/*
2 * libeio implementation
3 *
4 * Copyright (c) 2007,2008,2009,2010,2011,2012 Marc Alexander Lehmann <libeio@schmorp.de>
5 * All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without modifica-
8 * tion, are permitted provided that the following conditions are met:
9 *
10 * 1. Redistributions of source code must retain the above copyright notice,
11 * this list of conditions and the following disclaimer.
12 *
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 *
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-
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-
21 * CIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
22 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
23 * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
24 * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTH-
25 * ERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
26 * OF THE POSSIBILITY OF SUCH DAMAGE.
27 *
28 * Alternatively, the contents of this file may be used under the terms of
29 * the GNU General Public License ("GPL") version 2 or any later version,
30 * in which case the provisions of the GPL are applicable instead of
31 * the above. If you wish to allow the use of your version of this file
32 * only under the terms of the GPL and not to allow others to use your
33 * version of this file under the BSD license, indicate your decision
34 * by deleting the provisions above and replace them with the notice
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
37 * either the BSD or the GPL.
38 */
39
40#ifndef _WIN32
41# include "config.h"
42#endif
43
1#include "eio.h" 44#include "eio.h"
45#include "ecb.h"
46
47#ifdef EIO_STACKSIZE
48# define X_STACKSIZE EIO_STACKSIZE
49#endif
2#include "xthread.h" 50#include "xthread.h"
3 51
4#include <errno.h> 52#include <errno.h>
5#include <stddef.h> 53#include <stddef.h>
6#include <stdlib.h> 54#include <stdlib.h>
… …
10#include <sys/stat.h> 58#include <sys/stat.h>
11#include <limits.h> 59#include <limits.h>
12#include <fcntl.h> 60#include <fcntl.h>
13#include <assert.h> 61#include <assert.h>
14 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
15#ifndef EIO_FINISH 82#ifndef EIO_FINISH
16# 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
17#endif 84#endif
18 85
19#ifndef EIO_DESTROY 86#ifndef EIO_DESTROY
… …
22 89
23#ifndef EIO_FEED 90#ifndef EIO_FEED
24# 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)
25#endif 92#endif
26 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
27#ifdef _WIN32 105#ifdef _WIN32
28 106
29 /*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)
30 231
31#else 232#else
32 233
33# include "config.h"
34# include <sys/time.h> 234 #include <sys/time.h>
35# include <sys/select.h> 235 #include <sys/select.h>
36# include <unistd.h> 236 #include <unistd.h>
237 #include <signal.h>
238 #include <dirent.h>
239
240 #ifdef ANDROID
241 #include <sys/vfs.h>
242 #define statvfs statfs
243 #define fstatvfs fstatfs
244 #include <asm/page.h> /* supposedly limits.h does #define PAGESIZE PAGESIZE */
245 #else
246 #include <sys/statvfs.h>
247 #endif
248
249 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
250 #include <sys/mman.h>
251 #endif
252
253 #define D_NAME(entp) entp->d_name
254
255 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
256 #if __FreeBSD__ || __NetBSD__ || __OpenBSD__
257 #define _DIRENT_HAVE_D_TYPE /* sigh */
258 #define D_INO(de) (de)->d_fileno
259 #define D_NAMLEN(de) (de)->d_namlen
260 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
261 #define D_INO(de) (de)->d_ino
262 #endif
263
264 #ifdef _D_EXACT_NAMLEN
265 #undef D_NAMLEN
266 #define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
267 #endif
268
269 #ifdef _DIRENT_HAVE_D_TYPE
270 #define D_TYPE(de) (de)->d_type
271 #endif
272
273 #ifndef EIO_STRUCT_DIRENT
274 #define EIO_STRUCT_DIRENT struct dirent
275 #endif
276
277#endif
278
279#if HAVE_UTIMES
37# include <utime.h> 280# include <utime.h>
38# include <signal.h>
39# include <dirent.h>
40
41# ifndef EIO_STRUCT_DIRENT
42# define EIO_STRUCT_DIRENT struct dirent
43# endif 281#endif
44 282
45#endif 283#if HAVE_SYS_SYSCALL_H
46 284# include <sys/syscall.h>
47# ifndef EIO_STRUCT_STAT
48# define EIO_STRUCT_STAT struct stat
49# endif 285#endif
286
287#if HAVE_SYS_PRCTL_H
288# include <sys/prctl.h>
289#endif
50 290
51#if HAVE_SENDFILE 291#if HAVE_SENDFILE
52# if __linux 292# if __linux
53# include <sys/sendfile.h> 293# include <sys/sendfile.h>
54# elif __freebsd 294# elif __FreeBSD__ || defined __APPLE__
55# include <sys/socket.h> 295# include <sys/socket.h>
56# include <sys/uio.h> 296# include <sys/uio.h>
57# elif __hpux 297# elif __hpux
58# include <sys/socket.h> 298# include <sys/socket.h>
59# elif __solaris /* not yet */ 299# elif __solaris
60# include <sys/sendfile.h> 300# include <sys/sendfile.h>
61# else 301# else
62# error sendfile support requested but not available 302# error sendfile support requested but not available
63# endif 303# endif
64#endif 304#endif
65 305
66/* number of seconds after which an idle threads exit */ 306#ifndef D_TYPE
67#define IDLE_TIMEOUT 10 307# define D_TYPE(de) 0
308#endif
309#ifndef D_INO
310# define D_INO(de) 0
311#endif
312#ifndef D_NAMLEN
313# define D_NAMLEN(entp) strlen (D_NAME (entp))
314#endif
68 315
69/* used for struct dirent, AIX doesn't provide it */ 316/* used for struct dirent, AIX doesn't provide it */
70#ifndef NAME_MAX 317#ifndef NAME_MAX
71# define NAME_MAX 4096 318# define NAME_MAX 4096
72#endif 319#endif
73 320
321/* used for readlink etc. */
322#ifndef PATH_MAX
323# define PATH_MAX 4096
324#endif
325
74/* buffer size for various temporary buffers */ 326/* buffer size for various temporary buffers */
75#define EIO_BUFSIZE 65536 327#define EIO_BUFSIZE 65536
76 328
77#define dBUF \ 329#define dBUF \
78 char *eio_buf; \
79 X_LOCK (wrklock); \
80 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 330 char *eio_buf = malloc (EIO_BUFSIZE); \
81 X_UNLOCK (wrklock); \ 331 errno = ENOMEM; \
82 if (!eio_buf) \ 332 if (!eio_buf) \
83 return -1; 333 return -1
334
335#define FUBd \
336 free (eio_buf)
84 337
85#define EIO_TICKS ((1000000 + 1023) >> 10) 338#define EIO_TICKS ((1000000 + 1023) >> 10)
339
340/*****************************************************************************/
341
342struct tmpbuf
343{
344 void *ptr;
345 int len;
346};
347
348static void *
349tmpbuf_get (struct tmpbuf *buf, int len)
350{
351 if (buf->len < len)
352 {
353 free (buf->ptr);
354 buf->ptr = malloc (buf->len = len);
355 }
356
357 return buf->ptr;
358}
359
360struct tmpbuf;
361
362#if _POSIX_VERSION >= 200809L
363 #define HAVE_AT 1
364 #define WD2FD(wd) ((wd) ? (wd)->fd : AT_FDCWD)
365 #ifndef O_SEARCH
366 #define O_SEARCH O_RDONLY
367 #endif
368#else
369 #define HAVE_AT 0
370 static const char *wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path);
371#endif
372
373struct eio_pwd
374{
375#if HAVE_AT
376 int fd;
377#endif
378 int len;
379 char str[1]; /* actually, a 0-terminated canonical path */
380};
381
382/*****************************************************************************/
383
384#define ETP_PRI_MIN EIO_PRI_MIN
385#define ETP_PRI_MAX EIO_PRI_MAX
386
387struct etp_worker;
388
389#define ETP_REQ eio_req
390#define ETP_DESTROY(req) eio_destroy (req)
391static int eio_finish (eio_req *req);
392#define ETP_FINISH(req) eio_finish (req)
393static void eio_execute (struct etp_worker *self, eio_req *req);
394#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
395
396/*****************************************************************************/
397
398#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
399
400/* calculate time difference in ~1/EIO_TICKS of a second */
401ecb_inline int
402tvdiff (struct timeval *tv1, struct timeval *tv2)
403{
404 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
405 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
406}
407
408static unsigned int started, idle, wanted = 4;
86 409
87static void (*want_poll_cb) (void); 410static void (*want_poll_cb) (void);
88static void (*done_poll_cb) (void); 411static void (*done_poll_cb) (void);
89 412
90static unsigned int max_poll_time = 0; 413static unsigned int max_poll_time; /* reslock */
91static unsigned int max_poll_reqs = 0; 414static unsigned int max_poll_reqs; /* reslock */
92 415
93/* calculcate time difference in ~1/EIO_TICKS of a second */ 416static unsigned int nreqs; /* reqlock */
94static int tvdiff (struct timeval *tv1, struct timeval *tv2) 417static unsigned int nready; /* reqlock */
95{ 418static unsigned int npending; /* reqlock */
96 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 419static unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
97 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 420static unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
98}
99 421
100static unsigned int started, idle, wanted; 422static xmutex_t wrklock;
423static xmutex_t reslock;
424static xmutex_t reqlock;
425static xcond_t reqwait;
426
427typedef struct etp_worker
428{
429 struct tmpbuf tmpbuf;
430
431 /* locked by wrklock */
432 struct etp_worker *prev, *next;
433
434 xthread_t tid;
435
436#ifdef ETP_WORKER_COMMON
437 ETP_WORKER_COMMON
438#endif
439} etp_worker;
440
441static etp_worker wrk_first; /* NOT etp */
442
443#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
444#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
101 445
102/* worker threads management */ 446/* worker threads management */
103static mutex_t wrklock = X_MUTEX_INIT;
104 447
105typedef struct worker { 448static void
106 /* locked by wrklock */
107 struct worker *prev, *next;
108
109 thread_t tid;
110
111 /* locked by reslock, reqlock or wrklock */
112 eio_req *req; /* currently processed request */
113 void *dbuf;
114 DIR *dirp;
115} worker;
116
117static worker wrk_first = { &wrk_first, &wrk_first, 0 };
118
119static void worker_clear (worker *wrk) 449etp_worker_clear (etp_worker *wrk)
120{ 450{
121 if (wrk->dirp) 451}
122 {
123 closedir (wrk->dirp);
124 wrk->dirp = 0;
125 }
126 452
127 if (wrk->dbuf) 453static void ecb_cold
128 { 454etp_worker_free (etp_worker *wrk)
455{
129 free (wrk->dbuf); 456 free (wrk->tmpbuf.ptr);
130 wrk->dbuf = 0;
131 }
132}
133 457
134static void worker_free (worker *wrk)
135{
136 wrk->next->prev = wrk->prev; 458 wrk->next->prev = wrk->prev;
137 wrk->prev->next = wrk->next; 459 wrk->prev->next = wrk->next;
138 460
139 free (wrk); 461 free (wrk);
140} 462}
141 463
142static volatile unsigned int nreqs, nready, npending; 464static unsigned int
143static volatile unsigned int max_idle = 4; 465etp_nreqs (void)
144
145static mutex_t reslock = X_MUTEX_INIT;
146static mutex_t reqlock = X_MUTEX_INIT;
147static cond_t reqwait = X_COND_INIT;
148
149unsigned int eio_nreqs (void)
150{ 466{
467 int retval;
468 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
469 retval = nreqs;
470 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
151 return nreqs; 471 return retval;
152} 472}
153 473
154unsigned int eio_nready (void) 474static unsigned int
475etp_nready (void)
155{ 476{
156 unsigned int retval; 477 unsigned int retval;
157 478
158 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 479 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
159 retval = nready; 480 retval = nready;
160 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 481 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
161 482
162 return retval; 483 return retval;
163} 484}
164 485
165unsigned int eio_npending (void) 486static unsigned int
487etp_npending (void)
166{ 488{
167 unsigned int retval; 489 unsigned int retval;
168 490
169 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 491 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
170 retval = npending; 492 retval = npending;
171 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 493 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
172 494
173 return retval; 495 return retval;
174} 496}
175 497
176unsigned int eio_nthreads (void) 498static unsigned int
499etp_nthreads (void)
177{ 500{
178 unsigned int retval; 501 unsigned int retval;
179 502
180 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 503 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
181 retval = started; 504 retval = started;
… …
188 * a somewhat faster data structure might be nice, but 511 * a somewhat faster data structure might be nice, but
189 * with 8 priorities this actually needs <20 insns 512 * with 8 priorities this actually needs <20 insns
190 * per shift, the most expensive operation. 513 * per shift, the most expensive operation.
191 */ 514 */
192typedef struct { 515typedef struct {
193 eio_req *qs[EIO_NUM_PRI], *qe[EIO_NUM_PRI]; /* qstart, qend */ 516 ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */
194 int size; 517 int size;
195} reqq; 518} etp_reqq;
196 519
197static reqq req_queue; 520static etp_reqq req_queue;
198static reqq res_queue; 521static etp_reqq res_queue;
199 522
200static int reqq_push (reqq *q, eio_req *req) 523static void ecb_noinline ecb_cold
524reqq_init (etp_reqq *q)
525{
526 int pri;
527
528 for (pri = 0; pri < ETP_NUM_PRI; ++pri)
529 q->qs[pri] = q->qe[pri] = 0;
530
531 q->size = 0;
532}
533
534static int ecb_noinline
535reqq_push (etp_reqq *q, ETP_REQ *req)
201{ 536{
202 int pri = req->pri; 537 int pri = req->pri;
203 req->next = 0; 538 req->next = 0;
204 539
205 if (q->qe[pri]) 540 if (q->qe[pri])
… …
211 q->qe[pri] = q->qs[pri] = req; 546 q->qe[pri] = q->qs[pri] = req;
212 547
213 return q->size++; 548 return q->size++;
214} 549}
215 550
216static eio_req *reqq_shift (reqq *q) 551static ETP_REQ * ecb_noinline
552reqq_shift (etp_reqq *q)
217{ 553{
218 int pri; 554 int pri;
219 555
220 if (!q->size) 556 if (!q->size)
221 return 0; 557 return 0;
222 558
223 --q->size; 559 --q->size;
224 560
225 for (pri = EIO_NUM_PRI; pri--; ) 561 for (pri = ETP_NUM_PRI; pri--; )
226 { 562 {
227 eio_req *req = q->qs[pri]; 563 eio_req *req = q->qs[pri];
228 564
229 if (req) 565 if (req)
230 { 566 {
… …
236 } 572 }
237 573
238 abort (); 574 abort ();
239} 575}
240 576
241static void grp_try_feed (eio_req *grp) 577static int ecb_cold
578etp_init (void (*want_poll)(void), void (*done_poll)(void))
242{ 579{
243 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 580 X_MUTEX_CREATE (wrklock);
244 { 581 X_MUTEX_CREATE (reslock);
245 int old_len = grp->size; 582 X_MUTEX_CREATE (reqlock);
583 X_COND_CREATE (reqwait);
246 584
247 EIO_FEED (grp); 585 reqq_init (&req_queue);
586 reqq_init (&res_queue);
248 587
249 /* stop if no progress has been made */ 588 wrk_first.next =
250 if (old_len == grp->size) 589 wrk_first.prev = &wrk_first;
251 {
252 grp->feed = 0;
253 break;
254 }
255 }
256}
257 590
258static int eio_finish (eio_req *req); 591 started = 0;
592 idle = 0;
593 nreqs = 0;
594 nready = 0;
595 npending = 0;
259 596
260static int grp_dec (eio_req *grp) 597 want_poll_cb = want_poll;
261{ 598 done_poll_cb = done_poll;
262 --grp->size;
263 599
264 /* call feeder, if applicable */
265 grp_try_feed (grp);
266
267 /* finish, if done */
268 if (!grp->size && grp->int1)
269 return eio_finish (grp);
270 else
271 return 0; 600 return 0;
272} 601}
273 602
274void eio_destroy (eio_req *req)
275{
276 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
277 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
278
279 EIO_DESTROY (req);
280}
281
282static int eio_finish (eio_req *req)
283{
284 int res = EIO_FINISH (req);
285
286 if (req->grp)
287 {
288 int res2;
289 eio_req *grp = req->grp;
290
291 /* unlink request */
292 if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
293 if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
294
295 if (grp->grp_first == req)
296 grp->grp_first = req->grp_next;
297
298 res2 = grp_dec (grp);
299
300 if (!res && res2)
301 res = res2;
302 }
303
304 eio_destroy (req);
305
306 return res;
307}
308
309void eio_grp_cancel (eio_req *grp)
310{
311 for (grp = grp->grp_first; grp; grp = grp->grp_next)
312 eio_cancel (grp);
313}
314
315void eio_cancel (eio_req *req)
316{
317 X_LOCK (wrklock);
318 req->flags |= EIO_FLAG_CANCELLED;
319 X_UNLOCK (wrklock);
320
321 eio_grp_cancel (req);
322}
323
324X_THREAD_PROC (eio_proc); 603X_THREAD_PROC (etp_proc);
325 604
326static void start_thread (void) 605static void ecb_cold
606etp_start_thread (void)
327{ 607{
328 worker *wrk = calloc (1, sizeof (worker)); 608 etp_worker *wrk = calloc (1, sizeof (etp_worker));
329 609
330 /*TODO*/ 610 /*TODO*/
331 assert (("unable to allocate worker thread data", !wrk)); 611 assert (("unable to allocate worker thread data", wrk));
332 612
333 X_LOCK (wrklock); 613 X_LOCK (wrklock);
334 614
335 if (thread_create (&wrk->tid, eio_proc, (void *)wrk)) 615 if (xthread_create (&wrk->tid, etp_proc, (void *)wrk))
336 { 616 {
337 wrk->prev = &wrk_first; 617 wrk->prev = &wrk_first;
338 wrk->next = wrk_first.next; 618 wrk->next = wrk_first.next;
339 wrk_first.next->prev = wrk; 619 wrk_first.next->prev = wrk;
340 wrk_first.next = wrk; 620 wrk_first.next = wrk;
… …
344 free (wrk); 624 free (wrk);
345 625
346 X_UNLOCK (wrklock); 626 X_UNLOCK (wrklock);
347} 627}
348 628
629static void
349static void maybe_start_thread (void) 630etp_maybe_start_thread (void)
350{ 631{
351 if (eio_nthreads () >= wanted) 632 if (ecb_expect_true (etp_nthreads () >= wanted))
352 return; 633 return;
353 634
354 /* todo: maybe use idle here, but might be less exact */ 635 /* todo: maybe use idle here, but might be less exact */
355 if (0 <= (int)eio_nthreads () + (int)eio_npending () - (int)eio_nreqs ()) 636 if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
356 return; 637 return;
357 638
358 start_thread (); 639 etp_start_thread ();
359} 640}
360 641
361void eio_submit (eio_req *req) 642static void ecb_cold
643etp_end_thread (void)
362{ 644{
363 ++nreqs;
364
365 X_LOCK (reqlock);
366 ++nready;
367 reqq_push (&req_queue, req);
368 X_COND_SIGNAL (reqwait);
369 X_UNLOCK (reqlock);
370
371 maybe_start_thread ();
372}
373
374static void end_thread (void)
375{
376 eio_req *req = calloc (1, sizeof (eio_req)); 645 eio_req *req = calloc (1, sizeof (eio_req)); /* will be freed by worker */
377 646
378 req->type = EIO_QUIT; 647 req->type = -1;
379 req->pri = EIO_PRI_MAX + EIO_PRI_BIAS; 648 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
380 649
381 X_LOCK (reqlock); 650 X_LOCK (reqlock);
382 reqq_push (&req_queue, req); 651 reqq_push (&req_queue, req);
383 X_COND_SIGNAL (reqwait); 652 X_COND_SIGNAL (reqwait);
384 X_UNLOCK (reqlock); 653 X_UNLOCK (reqlock);
… …
386 X_LOCK (wrklock); 655 X_LOCK (wrklock);
387 --started; 656 --started;
388 X_UNLOCK (wrklock); 657 X_UNLOCK (wrklock);
389} 658}
390 659
391void eio_set_max_poll_time (double nseconds) 660static int
392{
393 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
394 max_poll_time = nseconds;
395 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
396}
397
398void eio_set_max_poll_reqs (unsigned int maxreqs)
399{
400 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
401 max_poll_reqs = maxreqs;
402 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
403}
404
405void eio_set_max_idle (unsigned int nthreads)
406{
407 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
408 max_idle = nthreads <= 0 ? 1 : nthreads;
409 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
410}
411
412void eio_set_min_parallel (unsigned int nthreads)
413{
414 if (wanted < nthreads)
415 wanted = nthreads;
416}
417
418void eio_set_max_parallel (unsigned int nthreads)
419{
420 if (wanted > nthreads)
421 wanted = nthreads;
422
423 while (started > wanted)
424 end_thread ();
425}
426
427int eio_poll (void) 661etp_poll (void)
428{ 662{
429 int maxreqs = max_poll_reqs; 663 unsigned int maxreqs;
664 unsigned int maxtime;
430 struct timeval tv_start, tv_now; 665 struct timeval tv_start, tv_now;
431 eio_req *req;
432 666
667 X_LOCK (reslock);
668 maxreqs = max_poll_reqs;
669 maxtime = max_poll_time;
670 X_UNLOCK (reslock);
671
433 if (max_poll_time) 672 if (maxtime)
434 gettimeofday (&tv_start, 0); 673 gettimeofday (&tv_start, 0);
435 674
436 for (;;) 675 for (;;)
437 { 676 {
677 ETP_REQ *req;
678
438 maybe_start_thread (); 679 etp_maybe_start_thread ();
439 680
440 X_LOCK (reslock); 681 X_LOCK (reslock);
441 req = reqq_shift (&res_queue); 682 req = reqq_shift (&res_queue);
442 683
443 if (req) 684 if (req)
… …
451 X_UNLOCK (reslock); 692 X_UNLOCK (reslock);
452 693
453 if (!req) 694 if (!req)
454 return 0; 695 return 0;
455 696
697 X_LOCK (reqlock);
456 --nreqs; 698 --nreqs;
699 X_UNLOCK (reqlock);
457 700
458 if (req->type == EIO_GROUP && req->size) 701 if (ecb_expect_false (req->type == EIO_GROUP && req->size))
459 { 702 {
460 req->int1 = 1; /* mark request as delayed */ 703 req->int1 = 1; /* mark request as delayed */
461 continue; 704 continue;
462 } 705 }
463 else 706 else
464 { 707 {
465 int res = eio_finish (req); 708 int res = ETP_FINISH (req);
466 if (res) 709 if (ecb_expect_false (res))
467 return res; 710 return res;
468 } 711 }
469 712
470 if (maxreqs && !--maxreqs) 713 if (ecb_expect_false (maxreqs && !--maxreqs))
471 break; 714 break;
472 715
473 if (max_poll_time) 716 if (maxtime)
474 { 717 {
475 gettimeofday (&tv_now, 0); 718 gettimeofday (&tv_now, 0);
476 719
477 if (tvdiff (&tv_start, &tv_now) >= max_poll_time) 720 if (tvdiff (&tv_start, &tv_now) >= maxtime)
478 break; 721 break;
479 } 722 }
480 } 723 }
481 724
482 errno = EAGAIN; 725 errno = EAGAIN;
483 return -1; 726 return -1;
484} 727}
485 728
729static void
730etp_cancel (ETP_REQ *req)
731{
732 req->cancelled = 1;
733
734 eio_grp_cancel (req);
735}
736
737static void
738etp_submit (ETP_REQ *req)
739{
740 req->pri -= ETP_PRI_MIN;
741
742 if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
743 if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
744
745 if (ecb_expect_false (req->type == EIO_GROUP))
746 {
747 /* I hope this is worth it :/ */
748 X_LOCK (reqlock);
749 ++nreqs;
750 X_UNLOCK (reqlock);
751
752 X_LOCK (reslock);
753
754 ++npending;
755
756 if (!reqq_push (&res_queue, req) && want_poll_cb)
757 want_poll_cb ();
758
759 X_UNLOCK (reslock);
760 }
761 else
762 {
763 X_LOCK (reqlock);
764 ++nreqs;
765 ++nready;
766 reqq_push (&req_queue, req);
767 X_COND_SIGNAL (reqwait);
768 X_UNLOCK (reqlock);
769
770 etp_maybe_start_thread ();
771 }
772}
773
774static void ecb_cold
775etp_set_max_poll_time (double nseconds)
776{
777 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
778 max_poll_time = nseconds * EIO_TICKS;
779 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
780}
781
782static void ecb_cold
783etp_set_max_poll_reqs (unsigned int maxreqs)
784{
785 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
786 max_poll_reqs = maxreqs;
787 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
788}
789
790static void ecb_cold
791etp_set_max_idle (unsigned int nthreads)
792{
793 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
794 max_idle = nthreads;
795 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
796}
797
798static void ecb_cold
799etp_set_idle_timeout (unsigned int seconds)
800{
801 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
802 idle_timeout = seconds;
803 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
804}
805
806static void ecb_cold
807etp_set_min_parallel (unsigned int nthreads)
808{
809 if (wanted < nthreads)
810 wanted = nthreads;
811}
812
813static void ecb_cold
814etp_set_max_parallel (unsigned int nthreads)
815{
816 if (wanted > nthreads)
817 wanted = nthreads;
818
819 while (started > wanted)
820 etp_end_thread ();
821}
822
823/*****************************************************************************/
824
825static void
826grp_try_feed (eio_req *grp)
827{
828 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
829 {
830 grp->flags &= ~EIO_FLAG_GROUPADD;
831
832 EIO_FEED (grp);
833
834 /* stop if no progress has been made */
835 if (!(grp->flags & EIO_FLAG_GROUPADD))
836 {
837 grp->feed = 0;
838 break;
839 }
840 }
841}
842
843static int
844grp_dec (eio_req *grp)
845{
846 --grp->size;
847
848 /* call feeder, if applicable */
849 grp_try_feed (grp);
850
851 /* finish, if done */
852 if (!grp->size && grp->int1)
853 return eio_finish (grp);
854 else
855 return 0;
856}
857
858static void
859eio_destroy (eio_req *req)
860{
861 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
862 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
863
864 EIO_DESTROY (req);
865}
866
867static int
868eio_finish (eio_req *req)
869{
870 int res = EIO_FINISH (req);
871
872 if (req->grp)
873 {
874 int res2;
875 eio_req *grp = req->grp;
876
877 /* unlink request */
878 if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
879 if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
880
881 if (grp->grp_first == req)
882 grp->grp_first = req->grp_next;
883
884 res2 = grp_dec (grp);
885
886 if (!res)
887 res = res2;
888 }
889
890 eio_destroy (req);
891
892 return res;
893}
894
895void
896eio_grp_cancel (eio_req *grp)
897{
898 for (grp = grp->grp_first; grp; grp = grp->grp_next)
899 eio_cancel (grp);
900}
901
902void
903eio_cancel (eio_req *req)
904{
905 etp_cancel (req);
906}
907
908void
909eio_submit (eio_req *req)
910{
911 etp_submit (req);
912}
913
914unsigned int
915eio_nreqs (void)
916{
917 return etp_nreqs ();
918}
919
920unsigned int
921eio_nready (void)
922{
923 return etp_nready ();
924}
925
926unsigned int
927eio_npending (void)
928{
929 return etp_npending ();
930}
931
932unsigned int ecb_cold
933eio_nthreads (void)
934{
935 return etp_nthreads ();
936}
937
938void ecb_cold
939eio_set_max_poll_time (double nseconds)
940{
941 etp_set_max_poll_time (nseconds);
942}
943
944void ecb_cold
945eio_set_max_poll_reqs (unsigned int maxreqs)
946{
947 etp_set_max_poll_reqs (maxreqs);
948}
949
950void ecb_cold
951eio_set_max_idle (unsigned int nthreads)
952{
953 etp_set_max_idle (nthreads);
954}
955
956void ecb_cold
957eio_set_idle_timeout (unsigned int seconds)
958{
959 etp_set_idle_timeout (seconds);
960}
961
962void ecb_cold
963eio_set_min_parallel (unsigned int nthreads)
964{
965 etp_set_min_parallel (nthreads);
966}
967
968void ecb_cold
969eio_set_max_parallel (unsigned int nthreads)
970{
971 etp_set_max_parallel (nthreads);
972}
973
974int eio_poll (void)
975{
976 return etp_poll ();
977}
978
486/*****************************************************************************/ 979/*****************************************************************************/
487/* work around various missing functions */ 980/* work around various missing functions */
488 981
489#if !HAVE_PREADWRITE
490# define pread aio_pread
491# define pwrite aio_pwrite
492
493/*
494 * make our pread/pwrite safe against themselves, but not against
495 * normal read/write by using a mutex. slows down execution a lot,
496 * but that's your problem, not mine.
497 */
498static mutex_t preadwritelock = X_MUTEX_INIT;
499
500static ssize_t pread (int fd, void *buf, size_t count, off_t offset)
501{
502 ssize_t res;
503 off_t ooffset;
504
505 X_LOCK (preadwritelock);
506 ooffset = lseek (fd, 0, SEEK_CUR);
507 lseek (fd, offset, SEEK_SET);
508 res = read (fd, buf, count);
509 lseek (fd, ooffset, SEEK_SET);
510 X_UNLOCK (preadwritelock);
511
512 return res;
513}
514
515static ssize_t pwrite (int fd, void *buf, size_t count, off_t offset)
516{
517 ssize_t res;
518 off_t ooffset;
519
520 X_LOCK (preadwritelock);
521 ooffset = lseek (fd, 0, SEEK_CUR);
522 lseek (fd, offset, SEEK_SET);
523 res = write (fd, buf, count);
524 lseek (fd, offset, SEEK_SET);
525 X_UNLOCK (preadwritelock);
526
527 return res;
528}
529#endif
530
531#ifndef HAVE_FUTIMES 982#ifndef HAVE_UTIMES
532 983
984# undef utimes
533# define utimes(path,times) aio_utimes (path, times) 985# define utimes(path,times) eio__utimes (path, times)
534# define futimes(fd,times) aio_futimes (fd, times)
535 986
987static int
536static int aio_utimes (const char *filename, const struct timeval times[2]) 988eio__utimes (const char *filename, const struct timeval times[2])
537{ 989{
538 if (times) 990 if (times)
539 { 991 {
540 struct utimbuf buf; 992 struct utimbuf buf;
541 993
… …
546 } 998 }
547 else 999 else
548 return utime (filename, 0); 1000 return utime (filename, 0);
549} 1001}
550 1002
1003#endif
1004
1005#ifndef HAVE_FUTIMES
1006
1007# undef futimes
1008# define futimes(fd,times) eio__futimes (fd, times)
1009
1010static int
551static int aio_futimes (int fd, const struct timeval tv[2]) 1011eio__futimes (int fd, const struct timeval tv[2])
552{ 1012{
553 errno = ENOSYS; 1013 errno = ENOSYS;
554 return -1; 1014 return -1;
555} 1015}
556 1016
557#endif 1017#endif
558 1018
559#if !HAVE_FDATASYNC 1019#if !HAVE_FDATASYNC
1020# undef fdatasync
560# define fdatasync fsync 1021# define fdatasync(fd) fsync (fd)
561#endif 1022#endif
1023
1024static int
1025eio__syncfs (int fd)
1026{
1027 int res;
1028
1029#if HAVE_SYS_SYNCFS
1030 res = (int)syscall (__NR_syncfs, (int)(fd));
1031#else
1032 res = EIO_ENOSYS ();
1033#endif
1034
1035 if (res < 0 && errno == ENOSYS && fd >= 0)
1036 sync ();
1037
1038 return res;
1039}
1040
1041/* sync_file_range always needs emulation */
1042static int
1043eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
1044{
1045#if HAVE_SYNC_FILE_RANGE
1046 int res;
1047
1048 if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
1049 || EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
1050 || EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
1051 {
1052 flags = 0
1053 | (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
1054 | (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
1055 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
1056 }
1057
1058 res = sync_file_range (fd, offset, nbytes, flags);
1059
1060 if (!res || errno != ENOSYS)
1061 return res;
1062#endif
1063
1064 /* even though we could play tricks with the flags, it's better to always
1065 * call fdatasync, as that matches the expectation of its users best */
1066 return fdatasync (fd);
1067}
1068
1069static int
1070eio__fallocate (int fd, int mode, off_t offset, size_t len)
1071{
1072#if HAVE_LINUX_FALLOCATE
1073 return fallocate (fd, mode, offset, len);
1074#else
1075 return EIO_ENOSYS ();
1076#endif
1077}
562 1078
563#if !HAVE_READAHEAD 1079#if !HAVE_READAHEAD
1080# undef readahead
564# define readahead(fd,offset,count) aio_readahead (fd, offset, count, self) 1081# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
565 1082
1083static eio_ssize_t
566static ssize_t aio_readahead (int fd, off_t offset, size_t count, worker *self) 1084eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
567{ 1085{
568 size_t todo = count; 1086 size_t todo = count;
569 dBUF; 1087 dBUF;
570 1088
571 while (todo > 0) 1089 while (todo > 0)
… …
575 pread (fd, eio_buf, len, offset); 1093 pread (fd, eio_buf, len, offset);
576 offset += len; 1094 offset += len;
577 todo -= len; 1095 todo -= len;
578 } 1096 }
579 1097
580 errno = 0; 1098 FUBd;
1099
1100 /* linux's readahead basically only fails for EBADF or EINVAL (not mmappable) */
1101 /* but not for e.g. EIO or eof, so we also never fail */
581 return count; 1102 return 0;
582} 1103}
583 1104
584#endif
585
586#if !HAVE_READDIR_R
587# define readdir_r aio_readdir_r
588
589static mutex_t readdirlock = X_MUTEX_INIT;
590
591static int readdir_r (DIR *dirp, EIO_STRUCT_DIRENT *ent, EIO_STRUCT_DIRENT **res)
592{
593 EIO_STRUCT_DIRENT *e;
594 int errorno;
595
596 X_LOCK (readdirlock);
597
598 e = readdir (dirp);
599 errorno = errno;
600
601 if (e)
602 {
603 *res = ent;
604 strcpy (ent->d_name, e->d_name);
605 }
606 else
607 *res = 0;
608
609 X_UNLOCK (readdirlock);
610
611 errno = errorno;
612 return e ? 0 : -1;
613}
614#endif 1105#endif
615 1106
616/* sendfile always needs emulation */ 1107/* sendfile always needs emulation */
1108static eio_ssize_t
617static ssize_t sendfile_ (int ofd, int ifd, off_t offset, size_t count, worker *self) 1109eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
618{ 1110{
1111 eio_ssize_t written = 0;
619 ssize_t res; 1112 eio_ssize_t res;
620 1113
621 if (!count) 1114 if (!count)
622 return 0; 1115 return 0;
623 1116
1117 for (;;)
1118 {
1119#ifdef __APPLE__
1120# undef HAVE_SENDFILE /* broken, as everything on os x */
1121#endif
624#if HAVE_SENDFILE 1122#if HAVE_SENDFILE
625# if __linux 1123# if __linux
1124 off_t soffset = offset;
626 res = sendfile (ofd, ifd, &offset, count); 1125 res = sendfile (ofd, ifd, &soffset, count);
627 1126
628# elif __freebsd 1127# elif __FreeBSD__
629 /* 1128 /*
630 * Of course, the freebsd sendfile is a dire hack with no thoughts 1129 * Of course, the freebsd sendfile is a dire hack with no thoughts
631 * wasted on making it similar to other I/O functions. 1130 * wasted on making it similar to other I/O functions.
632 */ 1131 */
633 {
634 off_t sbytes; 1132 off_t sbytes;
635 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 1133 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
636 1134
637 if (res < 0 && sbytes) 1135 #if 0 /* according to the manpage, this is correct, but broken behaviour */
638 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 1136 /* freebsd' sendfile will return 0 on success */
1137 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
1138 /* not on e.g. EIO or EPIPE - sounds broken */
1139 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
639 res = sbytes; 1140 res = sbytes;
640 } 1141 #endif
1142
1143 /* according to source inspection, this is correct, and useful behaviour */
1144 if (sbytes)
1145 res = sbytes;
1146
1147# elif defined __APPLE__
1148 off_t sbytes = count;
1149 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
1150
1151 /* according to the manpage, sbytes is always valid */
1152 if (sbytes)
1153 res = sbytes;
641 1154
642# elif __hpux 1155# elif __hpux
643 res = sendfile (ofd, ifd, offset, count, 0, 0); 1156 res = sendfile (ofd, ifd, offset, count, 0, 0);
644 1157
645# elif __solaris 1158# elif __solaris
646 {
647 struct sendfilevec vec; 1159 struct sendfilevec vec;
648 size_t sbytes; 1160 size_t sbytes;
649 1161
650 vec.sfv_fd = ifd; 1162 vec.sfv_fd = ifd;
651 vec.sfv_flag = 0; 1163 vec.sfv_flag = 0;
652 vec.sfv_off = offset; 1164 vec.sfv_off = offset;
653 vec.sfv_len = count; 1165 vec.sfv_len = count;
654 1166
655 res = sendfilev (ofd, &vec, 1, &sbytes); 1167 res = sendfilev (ofd, &vec, 1, &sbytes);
656 1168
657 if (res < 0 && sbytes) 1169 if (res < 0 && sbytes)
658 res = sbytes; 1170 res = sbytes;
659 }
660 1171
661# endif 1172# endif
1173
1174#elif defined (_WIN32) && 0
1175 /* does not work, just for documentation of what would need to be done */
1176 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
1177 /* libeio guarantees that the file offset does not change, and windows */
1178 /* has no way to get an independent handle to the same file description */
1179 HANDLE h = TO_SOCKET (ifd);
1180 SetFilePointer (h, offset, 0, FILE_BEGIN);
1181 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1182
662#else 1183#else
663 res = -1; 1184 res = EIO_ENOSYS ();
664 errno = ENOSYS;
665#endif 1185#endif
666 1186
1187 /* we assume sendfile can copy at least 128mb in one go */
1188 if (res <= 128 * 1024 * 1024)
1189 {
1190 if (res > 0)
1191 written += res;
1192
1193 if (written)
1194 return written;
1195
1196 break;
1197 }
1198 else
1199 {
1200 /* if we requested more, then probably the kernel was lazy */
1201 written += res;
1202 offset += res;
1203 count -= res;
1204
1205 if (!count)
1206 return written;
1207 }
1208 }
1209
667 if (res < 0 1210 if (res < 0
668 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1211 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1212 /* BSDs */
1213#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1214 || errno == ENOTSUP
1215#endif
1216#ifdef EOPNOTSUPP /* windows */
1217 || errno == EOPNOTSUPP /* BSDs */
1218#endif
669#if __solaris 1219#if __solaris
670 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1220 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
671#endif 1221#endif
672 ) 1222 )
673 ) 1223 )
… …
677 1227
678 res = 0; 1228 res = 0;
679 1229
680 while (count) 1230 while (count)
681 { 1231 {
682 ssize_t cnt; 1232 eio_ssize_t cnt;
683 1233
684 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 1234 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
685 1235
686 if (cnt <= 0) 1236 if (cnt <= 0)
687 { 1237 {
… …
699 1249
700 offset += cnt; 1250 offset += cnt;
701 res += cnt; 1251 res += cnt;
702 count -= cnt; 1252 count -= cnt;
703 } 1253 }
1254
1255 FUBd;
704 } 1256 }
705 1257
706 return res; 1258 return res;
707} 1259}
708 1260
709/* read a full directory */ 1261#ifdef PAGESIZE
710static void scandir_ (eio_req *req, worker *self) 1262# define eio_pagesize() PAGESIZE
1263#else
1264static intptr_t
1265eio_pagesize (void)
711{ 1266{
712 DIR *dirp; 1267 static intptr_t page;
713 union
714 {
715 EIO_STRUCT_DIRENT d;
716 char b [offsetof (EIO_STRUCT_DIRENT, d_name) + NAME_MAX + 1];
717 } *u;
718 EIO_STRUCT_DIRENT *entp;
719 char *name, *names;
720 int memlen = 4096;
721 int memofs = 0;
722 int res = 0;
723 1268
724 X_LOCK (wrklock); 1269 if (!page)
725 self->dirp = dirp = opendir (req->ptr1); 1270 page = sysconf (_SC_PAGESIZE);
726 self->dbuf = u = malloc (sizeof (*u));
727 req->flags |= EIO_FLAG_PTR2_FREE;
728 req->ptr2 = names = malloc (memlen);
729 X_UNLOCK (wrklock);
730 1271
731 if (dirp && u && names) 1272 return page;
732 for (;;) 1273}
1274#endif
1275
1276static void
1277eio_page_align (void **addr, size_t *length)
1278{
1279 intptr_t mask = eio_pagesize () - 1;
1280
1281 /* round down addr */
1282 intptr_t adj = mask & (intptr_t)*addr;
1283
1284 *addr = (void *)((intptr_t)*addr - adj);
1285 *length += adj;
1286
1287 /* round up length */
1288 *length = (*length + mask) & ~mask;
1289}
1290
1291#if !_POSIX_MEMLOCK
1292# define eio__mlockall(a) EIO_ENOSYS ()
1293#else
1294
1295static int
1296eio__mlockall (int flags)
1297{
1298 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1299 extern int mallopt (int, int);
1300 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1301 #endif
1302
1303 if (EIO_MCL_CURRENT != MCL_CURRENT
1304 || EIO_MCL_FUTURE != MCL_FUTURE)
1305 {
1306 flags = 0
1307 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1308 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1309 }
1310
1311 return mlockall (flags);
1312}
1313#endif
1314
1315#if !_POSIX_MEMLOCK_RANGE
1316# define eio__mlock(a,b) EIO_ENOSYS ()
1317#else
1318
1319static int
1320eio__mlock (void *addr, size_t length)
1321{
1322 eio_page_align (&addr, &length);
1323
1324 return mlock (addr, length);
1325}
1326
1327#endif
1328
1329#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1330# define eio__msync(a,b,c) EIO_ENOSYS ()
1331#else
1332
1333static int
1334eio__msync (void *mem, size_t len, int flags)
1335{
1336 eio_page_align (&mem, &len);
1337
1338 if (EIO_MS_ASYNC != MS_SYNC
1339 || EIO_MS_INVALIDATE != MS_INVALIDATE
1340 || EIO_MS_SYNC != MS_SYNC)
1341 {
1342 flags = 0
1343 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1344 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1345 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1346 }
1347
1348 return msync (mem, len, flags);
1349}
1350
1351#endif
1352
1353static int
1354eio__mtouch (eio_req *req)
1355{
1356 void *mem = req->ptr2;
1357 size_t len = req->size;
1358 int flags = req->int1;
1359
1360 eio_page_align (&mem, &len);
1361
1362 {
1363 intptr_t addr = (intptr_t)mem;
1364 intptr_t end = addr + len;
1365 intptr_t page = eio_pagesize ();
1366
1367 if (addr < end)
1368 if (flags & EIO_MT_MODIFY) /* modify */
1369 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1370 else
1371 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1372 }
1373
1374 return 0;
1375}
1376
1377/*****************************************************************************/
1378/* requests implemented outside eio_execute, because they are so large */
1379
1380static void
1381eio__lseek (eio_req *req)
1382{
1383 /* this usually gets optimised away completely, or your compiler sucks, */
1384 /* or the whence constants really are not 0, 1, 2 */
1385 int whence = req->int2 == EIO_SEEK_SET ? SEEK_SET
1386 : req->int2 == EIO_SEEK_CUR ? SEEK_CUR
1387 : req->int2 == EIO_SEEK_END ? SEEK_END
1388 : req->int2;
1389
1390 req->offs = lseek (req->int1, req->offs, whence);
1391 req->result = req->offs == (off_t)-1 ? -1 : 0;
1392}
1393
1394/* result will always end up in tmpbuf, there is always space for adding a 0-byte */
1395static int
1396eio__realpath (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
1397{
1398 const char *rel = path;
1399 char *res;
1400 char *tmp1, *tmp2;
1401#if SYMLOOP_MAX > 32
1402 int symlinks = SYMLOOP_MAX;
1403#else
1404 int symlinks = 32;
1405#endif
1406
1407 errno = EINVAL;
1408 if (!rel)
1409 return -1;
1410
1411 errno = ENOENT;
1412 if (!*rel)
1413 return -1;
1414
1415 res = tmpbuf_get (tmpbuf, PATH_MAX * 3);
1416 tmp1 = res + PATH_MAX;
1417 tmp2 = tmp1 + PATH_MAX;
1418
1419#if 0 /* disabled, the musl way to do things is just too racy */
1420#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1421 /* on linux we may be able to ask the kernel */
1422 {
1423 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1424
1425 if (fd >= 0)
733 { 1426 {
1427 sprintf (tmp1, "/proc/self/fd/%d", fd);
1428 req->result = readlink (tmp1, res, PATH_MAX);
1429 /* here we should probably stat the open file and the disk file, to make sure they still match */
1430 close (fd);
1431
1432 if (req->result > 0)
1433 goto done;
1434 }
1435 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1436 return -1;
1437 }
1438#endif
1439#endif
1440
1441 if (*rel != '/')
1442 {
1443 int len;
1444
734 errno = 0; 1445 errno = ENOENT;
735 readdir_r (dirp, &u->d, &entp); 1446 if (wd == EIO_INVALID_WD)
1447 return -1;
1448
1449 if (wd == EIO_CWD)
1450 {
1451 if (!getcwd (res, PATH_MAX))
1452 return -1;
736 1453
737 if (!entp) 1454 len = strlen (res);
1455 }
1456 else
1457 memcpy (res, wd->str, len = wd->len);
1458
1459 if (res [1]) /* only use if not / */
1460 res += len;
1461 }
1462
1463 while (*rel)
1464 {
1465 eio_ssize_t len, linklen;
1466 const char *beg = rel;
1467
1468 while (*rel && *rel != '/')
1469 ++rel;
1470
1471 len = rel - beg;
1472
1473 if (!len) /* skip slashes */
1474 {
738 break; 1475 ++rel;
1476 continue;
1477 }
739 1478
740 name = entp->d_name; 1479 if (beg [0] == '.')
1480 {
1481 if (len == 1)
1482 continue; /* . - nop */
741 1483
742 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1484 if (beg [1] == '.' && len == 2)
1485 {
1486 /* .. - back up one component, if possible */
1487
1488 while (res != tmpbuf->ptr)
1489 if (*--res == '/')
1490 break;
1491
1492 continue;
1493 }
1494 }
1495
1496 errno = ENAMETOOLONG;
1497 if (res + 1 + len + 1 >= tmp1)
1498 return -1;
1499
1500 /* copy one component */
1501 *res = '/';
1502 memcpy (res + 1, beg, len);
1503
1504 /* zero-terminate, for readlink */
1505 res [len + 1] = 0;
1506
1507 /* now check if it's a symlink */
1508 linklen = readlink (tmpbuf->ptr, tmp1, PATH_MAX);
1509
1510 if (linklen < 0)
743 { 1511 {
1512 if (errno != EINVAL)
1513 return -1;
1514
1515 /* it's a normal directory. hopefully */
1516 res += len + 1;
1517 }
1518 else
1519 {
1520 /* yay, it was a symlink - build new path in tmp2 */
744 int len = strlen (name) + 1; 1521 int rellen = strlen (rel);
745 1522
1523 errno = ENAMETOOLONG;
1524 if (linklen + 1 + rellen >= PATH_MAX)
1525 return -1;
1526
1527 errno = ELOOP;
1528 if (!--symlinks)
1529 return -1;
1530
1531 if (*tmp1 == '/')
1532 res = tmpbuf->ptr; /* symlink resolves to an absolute path */
1533
1534 /* we need to be careful, as rel might point into tmp2 already */
1535 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1536 tmp2 [linklen] = '/';
1537 memcpy (tmp2, tmp1, linklen);
1538
1539 rel = tmp2;
1540 }
1541 }
1542
1543 /* special case for the lone root path */
1544 if (res == tmpbuf->ptr)
1545 *res++ = '/';
1546
1547 return res - (char *)tmpbuf->ptr;
1548}
1549
1550static signed char
1551eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1552{
1553 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1554 : a->inode < b->inode ? -1
1555 : a->inode > b->inode ? 1
1556 : 0;
1557}
1558
1559#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1560
1561#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1562#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1563
1564static void
1565eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1566{
1567 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1568 unsigned char *bit = bits;
1569
1570 assert (CHAR_BIT == 8);
1571 assert (sizeof (eio_dirent) * 8 < 256);
1572 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1573 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1574
1575 if (size <= EIO_SORT_FAST)
1576 return;
1577
1578 /* first prepare an array of bits to test in our radix sort */
1579 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1580 /* inode_bits must contain all inodes ORed together */
1581 /* which is used to skip bits that are 0 everywhere, which is very common */
1582 {
1583 eio_ino_t endianness;
1584 int i, j;
1585
1586 /* we store the byte offset of byte n into byte n of "endianness" */
1587 for (i = 0; i < sizeof (eio_ino_t); ++i)
1588 ((unsigned char *)&endianness)[i] = i;
1589
1590 *bit++ = 0;
1591
1592 for (i = 0; i < sizeof (eio_ino_t); ++i)
1593 {
1594 /* shifting off the byte offsets out of "endianness" */
1595 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1596 endianness >>= 8;
1597
1598 for (j = 0; j < 8; ++j)
1599 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1600 *bit++ = offs + j;
1601 }
1602
1603 for (j = 0; j < 8; ++j)
1604 if (score_bits & (1 << j))
1605 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1606 }
1607
1608 /* now actually do the sorting (a variant of MSD radix sort) */
1609 {
1610 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1611 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1612 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1613 int stk_idx = 0;
1614
1615 base_stk [stk_idx] = dents;
1616 end_stk [stk_idx] = dents + size;
1617 bit_stk [stk_idx] = bit - 1;
1618
1619 do
1620 {
1621 base = base_stk [stk_idx];
1622 end = end_stk [stk_idx];
1623 bit = bit_stk [stk_idx];
1624
1625 for (;;)
1626 {
1627 unsigned char O = *bit >> 3;
1628 unsigned char M = 1 << (*bit & 7);
1629
1630 eio_dirent *a = base;
1631 eio_dirent *b = end;
1632
1633 if (b - a < EIO_SORT_CUTOFF)
746 res++; 1634 break;
747 1635
748 while (memofs + len > memlen) 1636 /* now bit-partition the array on the bit */
1637 /* this ugly asymmetric loop seems to perform much better than typical */
1638 /* partition algos found in the literature */
1639 do
1640 if (!(((unsigned char *)a)[O] & M))
1641 ++a;
1642 else if (!(((unsigned char *)--b)[O] & M))
749 { 1643 {
750 memlen *= 2; 1644 eio_dirent tmp = *a; *a = *b; *b = tmp;
751 X_LOCK (wrklock);
752 req->ptr2 = names = realloc (names, memlen);
753 X_UNLOCK (wrklock);
754
755 if (!names)
756 break; 1645 ++a;
757 } 1646 }
1647 while (b > a);
758 1648
759 memcpy (names + memofs, name, len); 1649 /* next bit, or stop, if no bits left in this path */
760 memofs += len; 1650 if (!*--bit)
1651 break;
1652
1653 base_stk [stk_idx] = a;
1654 end_stk [stk_idx] = end;
1655 bit_stk [stk_idx] = bit;
1656 ++stk_idx;
1657
1658 end = a;
761 } 1659 }
762 } 1660 }
763 1661 while (stk_idx--);
764 if (errno)
765 res = -1;
766 1662 }
1663}
1664
1665static void
1666eio_dent_insertion_sort (eio_dirent *dents, int size)
1667{
1668 /* first move the smallest element to the front, to act as a sentinel */
1669 {
1670 int i;
1671 eio_dirent *min = dents;
1672
1673 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1674 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1675 if (EIO_DENT_CMP (dents [i], <, *min))
1676 min = &dents [i];
1677
1678 /* swap elements 0 and j (minimum) */
1679 {
1680 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1681 }
1682 }
1683
1684 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1685 {
1686 eio_dirent *i, *j;
1687
1688 for (i = dents + 1; i < dents + size; ++i)
1689 {
1690 eio_dirent value = *i;
1691
1692 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1693 j [1] = j [0];
1694
1695 j [1] = value;
1696 }
1697 }
1698}
1699
1700static void
1701eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1702{
1703 if (size <= 1)
1704 return; /* our insertion sort relies on size > 0 */
1705
1706 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1707 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1708 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1709
1710 /* use an insertion sort at the end, or for small arrays, */
1711 /* as insertion sort is more efficient for small partitions */
1712 eio_dent_insertion_sort (dents, size);
1713}
1714
1715/* read a full directory */
1716static void
1717eio__scandir (eio_req *req, etp_worker *self)
1718{
1719 char *name, *names;
1720 int namesalloc = 4096 - sizeof (void *) * 4;
1721 int namesoffs = 0;
1722 int flags = req->int1;
1723 eio_dirent *dents = 0;
1724 int dentalloc = 128;
1725 int dentoffs = 0;
1726 eio_ino_t inode_bits = 0;
1727#ifdef _WIN32
1728 HANDLE dirp;
1729 WIN32_FIND_DATA entp;
1730#else
1731 DIR *dirp;
1732 EIO_STRUCT_DIRENT *entp;
1733#endif
1734
767 req->result = res; 1735 req->result = -1;
1736
1737 if (!(flags & EIO_READDIR_DENTS))
1738 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1739
1740#ifdef _WIN32
1741 {
1742 int len = strlen ((const char *)req->ptr1);
1743 char *path = malloc (MAX_PATH);
1744 const char *fmt;
1745 const char *reqpath = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1746
1747 if (!len)
1748 fmt = "./*";
1749 else if (reqpath[len - 1] == '/' || reqpath[len - 1] == '\\')
1750 fmt = "%s*";
1751 else
1752 fmt = "%s/*";
1753
1754 _snprintf (path, MAX_PATH, fmt, reqpath);
1755 dirp = FindFirstFile (path, &entp);
1756 free (path);
1757
1758 if (dirp == INVALID_HANDLE_VALUE)
1759 {
1760 /* should steal _dosmaperr */
1761 switch (GetLastError ())
1762 {
1763 case ERROR_FILE_NOT_FOUND:
1764 req->result = 0;
1765 break;
1766
1767 case ERROR_INVALID_NAME:
1768 case ERROR_PATH_NOT_FOUND:
1769 case ERROR_NO_MORE_FILES:
1770 errno = ENOENT;
1771 break;
1772
1773 case ERROR_NOT_ENOUGH_MEMORY:
1774 errno = ENOMEM;
1775 break;
1776
1777 default:
1778 errno = EINVAL;
1779 break;
1780 }
1781
1782 return;
1783 }
1784 }
1785#else
1786 #if HAVE_AT
1787 if (req->wd)
1788 {
1789 int fd = openat (WD2FD (req->wd), req->ptr1, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1790
1791 if (fd < 0)
1792 return;
1793
1794 dirp = fdopendir (fd);
1795
1796 if (!dirp)
1797 close (fd);
1798 }
1799 else
1800 dirp = opendir (req->ptr1);
1801 #else
1802 dirp = opendir (wd_expand (&self->tmpbuf, req->wd, req->ptr1));
1803 #endif
1804
1805 if (!dirp)
1806 return;
1807#endif
1808
1809 if (req->flags & EIO_FLAG_PTR1_FREE)
1810 free (req->ptr1);
1811
1812 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1813 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1814 req->ptr2 = names = malloc (namesalloc);
1815
1816 if (!names || (flags && !dents))
1817 return;
1818
1819 for (;;)
1820 {
1821 int done;
1822
1823#ifdef _WIN32
1824 done = !dirp;
1825#else
1826 errno = 0;
1827 entp = readdir (dirp);
1828 done = !entp;
1829#endif
1830
1831 if (done)
1832 {
1833#ifndef _WIN32
1834 int old_errno = errno;
1835 closedir (dirp);
1836 errno = old_errno;
1837
1838 if (errno)
1839 break;
1840#endif
1841
1842 /* sort etc. */
1843 req->int1 = flags;
1844 req->result = dentoffs;
1845
1846 if (flags & EIO_READDIR_STAT_ORDER)
1847 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1848 else if (flags & EIO_READDIR_DIRS_FIRST)
1849 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1850 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1851 else
1852 {
1853 /* in this case, all is known, and we just put dirs first and sort them */
1854 eio_dirent *oth = dents + dentoffs;
1855 eio_dirent *dir = dents;
1856
1857 /* now partition dirs to the front, and non-dirs to the back */
1858 /* by walking from both sides and swapping if necessary */
1859 while (oth > dir)
1860 {
1861 if (dir->type == EIO_DT_DIR)
1862 ++dir;
1863 else if ((--oth)->type == EIO_DT_DIR)
1864 {
1865 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1866
1867 ++dir;
1868 }
1869 }
1870
1871 /* now sort the dirs only (dirs all have the same score) */
1872 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1873 }
1874
1875 break;
1876 }
1877
1878 /* now add the entry to our list(s) */
1879 name = D_NAME (entp);
1880
1881 /* skip . and .. entries */
1882 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1883 {
1884 int len = D_NAMLEN (entp) + 1;
1885
1886 while (ecb_expect_false (namesoffs + len > namesalloc))
1887 {
1888 namesalloc *= 2;
1889 req->ptr2 = names = realloc (names, namesalloc);
1890
1891 if (!names)
1892 break;
1893 }
1894
1895 memcpy (names + namesoffs, name, len);
1896
1897 if (dents)
1898 {
1899 struct eio_dirent *ent;
1900
1901 if (ecb_expect_false (dentoffs == dentalloc))
1902 {
1903 dentalloc *= 2;
1904 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1905
1906 if (!dents)
1907 break;
1908 }
1909
1910 ent = dents + dentoffs;
1911
1912 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1913 ent->namelen = len - 1;
1914 ent->inode = D_INO (entp);
1915
1916 inode_bits |= ent->inode;
1917
1918 switch (D_TYPE (entp))
1919 {
1920 default:
1921 ent->type = EIO_DT_UNKNOWN;
1922 flags |= EIO_READDIR_FOUND_UNKNOWN;
1923 break;
1924
1925 #ifdef DT_FIFO
1926 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1927 #endif
1928 #ifdef DT_CHR
1929 case DT_CHR: ent->type = EIO_DT_CHR; break;
1930 #endif
1931 #ifdef DT_MPC
1932 case DT_MPC: ent->type = EIO_DT_MPC; break;
1933 #endif
1934 #ifdef DT_DIR
1935 case DT_DIR: ent->type = EIO_DT_DIR; break;
1936 #endif
1937 #ifdef DT_NAM
1938 case DT_NAM: ent->type = EIO_DT_NAM; break;
1939 #endif
1940 #ifdef DT_BLK
1941 case DT_BLK: ent->type = EIO_DT_BLK; break;
1942 #endif
1943 #ifdef DT_MPB
1944 case DT_MPB: ent->type = EIO_DT_MPB; break;
1945 #endif
1946 #ifdef DT_REG
1947 case DT_REG: ent->type = EIO_DT_REG; break;
1948 #endif
1949 #ifdef DT_NWK
1950 case DT_NWK: ent->type = EIO_DT_NWK; break;
1951 #endif
1952 #ifdef DT_CMP
1953 case DT_CMP: ent->type = EIO_DT_CMP; break;
1954 #endif
1955 #ifdef DT_LNK
1956 case DT_LNK: ent->type = EIO_DT_LNK; break;
1957 #endif
1958 #ifdef DT_SOCK
1959 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1960 #endif
1961 #ifdef DT_DOOR
1962 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1963 #endif
1964 #ifdef DT_WHT
1965 case DT_WHT: ent->type = EIO_DT_WHT; break;
1966 #endif
1967 }
1968
1969 ent->score = 7;
1970
1971 if (flags & EIO_READDIR_DIRS_FIRST)
1972 {
1973 if (ent->type == EIO_DT_UNKNOWN)
1974 {
1975 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1976 ent->score = 1;
1977 else if (!strchr (name, '.')) /* absence of dots indicate likely dirs */
1978 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1979 }
1980 else if (ent->type == EIO_DT_DIR)
1981 ent->score = 0;
1982 }
1983 }
1984
1985 namesoffs += len;
1986 ++dentoffs;
1987 }
1988
1989 if (EIO_CANCELLED (req))
1990 {
1991 errno = ECANCELED;
1992 break;
1993 }
1994
1995#ifdef _WIN32
1996 if (!FindNextFile (dirp, &entp))
1997 {
1998 FindClose (dirp);
1999 dirp = 0;
2000 }
2001#endif
2002 }
768} 2003}
2004
2005/*****************************************************************************/
2006/* working directory stuff */
2007/* various deficiencies in the posix 2008 api force us to */
2008/* keep the absolute path in string form at all times */
2009/* fuck yeah. */
2010
2011#if !HAVE_AT
2012
2013/* a bit like realpath, but usually faster because it doesn'T have to return */
2014/* an absolute or canonical path */
2015static const char *
2016wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
2017{
2018 if (!wd || *path == '/')
2019 return path;
2020
2021 if (path [0] == '.' && !path [1])
2022 return wd->str;
2023
2024 {
2025 int l1 = wd->len;
2026 int l2 = strlen (path);
2027
2028 char *res = tmpbuf_get (tmpbuf, l1 + l2 + 2);
2029
2030 memcpy (res, wd->str, l1);
2031 res [l1] = '/';
2032 memcpy (res + l1 + 1, path, l2 + 1);
2033
2034 return res;
2035 }
2036}
2037
2038#endif
2039
2040static eio_wd
2041eio__wd_open_sync (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
2042{
2043 int fd;
2044 eio_wd res;
2045 int len = eio__realpath (tmpbuf, wd, path);
2046
2047 if (len < 0)
2048 return EIO_INVALID_WD;
2049
2050#if HAVE_AT
2051 fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
2052
2053 if (fd < 0)
2054 return EIO_INVALID_WD;
2055#endif
2056
2057 res = malloc (sizeof (*res) + len); /* one extra 0-byte */
2058
2059#if HAVE_AT
2060 res->fd = fd;
2061#endif
2062
2063 res->len = len;
2064 memcpy (res->str, tmpbuf->ptr, len);
2065 res->str [len] = 0;
2066
2067 return res;
2068}
2069
2070eio_wd
2071eio_wd_open_sync (eio_wd wd, const char *path)
2072{
2073 struct tmpbuf tmpbuf = { 0 };
2074 wd = eio__wd_open_sync (&tmpbuf, wd, path);
2075 free (tmpbuf.ptr);
2076
2077 return wd;
2078}
2079
2080void
2081eio_wd_close_sync (eio_wd wd)
2082{
2083 if (wd != EIO_INVALID_WD && wd != EIO_CWD)
2084 {
2085 #if HAVE_AT
2086 close (wd->fd);
2087 #endif
2088 free (wd);
2089 }
2090}
2091
2092#if HAVE_AT
2093
2094/* they forgot these */
2095
2096static int
2097eio__truncateat (int dirfd, const char *path, off_t length)
2098{
2099 int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
2100 int res;
2101
2102 if (fd < 0)
2103 return fd;
2104
2105 res = ftruncate (fd, length);
2106 close (fd);
2107 return res;
2108}
2109
2110static int
2111eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
2112{
2113 int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
2114 int res;
2115
2116 if (fd < 0)
2117 return fd;
2118
2119 res = fstatvfs (fd, buf);
2120 close (fd);
2121 return res;
2122
2123}
2124
2125#endif
769 2126
770/*****************************************************************************/ 2127/*****************************************************************************/
771 2128
772#define ALLOC(len) \ 2129#define ALLOC(len) \
773 if (!req->ptr2) \ 2130 if (!req->ptr2) \
… …
782 req->result = -1; \ 2139 req->result = -1; \
783 break; \ 2140 break; \
784 } \ 2141 } \
785 } 2142 }
786 2143
2144static void ecb_noinline ecb_cold
2145etp_proc_init (void)
2146{
2147#if HAVE_PRCTL_SET_NAME
2148 /* provide a more sensible "thread name" */
2149 char name[16 + 1];
2150 const int namelen = sizeof (name) - 1;
2151 int len;
2152
2153 prctl (PR_GET_NAME, (unsigned long)name, 0, 0, 0);
2154 name [namelen] = 0;
2155 len = strlen (name);
2156 strcpy (name + (len <= namelen - 4 ? len : namelen - 4), "/eio");
2157 prctl (PR_SET_NAME, (unsigned long)name, 0, 0, 0);
2158#endif
2159}
2160
787X_THREAD_PROC (eio_proc) 2161X_THREAD_PROC (etp_proc)
788{ 2162{
789 eio_req *req; 2163 ETP_REQ *req;
790 struct timespec ts; 2164 struct timespec ts;
791 worker *self = (worker *)thr_arg; 2165 etp_worker *self = (etp_worker *)thr_arg;
792 2166
2167 etp_proc_init ();
2168
793 /* try to distribute timeouts somewhat randomly */ 2169 /* try to distribute timeouts somewhat evenly */
794 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 2170 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
795 2171
796 for (;;) 2172 for (;;)
797 { 2173 {
798 ts.tv_sec = time (0) + IDLE_TIMEOUT; 2174 ts.tv_sec = 0;
799 2175
800 X_LOCK (reqlock); 2176 X_LOCK (reqlock);
801 2177
802 for (;;) 2178 for (;;)
803 { 2179 {
804 self->req = req = reqq_shift (&req_queue); 2180 req = reqq_shift (&req_queue);
805 2181
806 if (req) 2182 if (req)
807 break; 2183 break;
808 2184
2185 if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
2186 {
2187 X_UNLOCK (reqlock);
2188 X_LOCK (wrklock);
2189 --started;
2190 X_UNLOCK (wrklock);
2191 goto quit;
2192 }
2193
809 ++idle; 2194 ++idle;
810 2195
811 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 2196 if (idle <= max_idle)
2197 /* we are allowed to idle, so do so without any timeout */
2198 X_COND_WAIT (reqwait, reqlock);
2199 else
812 { 2200 {
813 if (idle > max_idle) 2201 /* initialise timeout once */
814 { 2202 if (!ts.tv_sec)
815 --idle; 2203 ts.tv_sec = time (0) + idle_timeout;
816 X_UNLOCK (reqlock);
817 X_LOCK (wrklock);
818 --started;
819 X_UNLOCK (wrklock);
820 goto quit;
821 }
822 2204
823 /* we are allowed to idle, so do so without any timeout */
824 X_COND_WAIT (reqwait, reqlock); 2205 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
825 ts.tv_sec = time (0) + IDLE_TIMEOUT; 2206 ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
826 } 2207 }
827 2208
828 --idle; 2209 --idle;
829 } 2210 }
830 2211
831 --nready; 2212 --nready;
832 2213
833 X_UNLOCK (reqlock); 2214 X_UNLOCK (reqlock);
834 2215
835 errno = 0; /* strictly unnecessary */
836
837 if (!EIO_CANCELLED (req))
838 switch (req->type) 2216 if (req->type < 0)
839 {
840 case EIO_READ: ALLOC (req->size);
841 req->result = req->offs >= 0
842 ? pread (req->int1, req->ptr2, req->size, req->offs)
843 : read (req->int1, req->ptr2, req->size); break;
844 case EIO_WRITE: req->result = req->offs >= 0
845 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
846 : write (req->int1, req->ptr2, req->size); break;
847
848 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
849 case EIO_SENDFILE: req->result = sendfile_ (req->int1, req->int2, req->offs, req->size, self); break;
850
851 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
852 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
853 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
854 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
855 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
856 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
857
858 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
859 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
860 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
861 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
862 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
863 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
864
865 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
866 case EIO_CLOSE: req->result = close (req->int1); break;
867 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
868 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
869 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
870 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
871 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
872 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
873 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
874 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
875
876 case EIO_READLINK: ALLOC (NAME_MAX);
877 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
878
879 case EIO_SYNC: req->result = 0; sync (); break;
880 case EIO_FSYNC: req->result = fsync (req->int1); break;
881 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
882
883 case EIO_READDIR: scandir_ (req, self); break;
884
885 case EIO_BUSY:
886#ifdef _WIN32
887 Sleep (req->nv1 * 1000.);
888#else
889 {
890 struct timeval tv;
891
892 tv.tv_sec = req->nv1;
893 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;
894
895 req->result = select (0, 0, 0, 0, &tv);
896 }
897#endif
898 break;
899
900 case EIO_UTIME:
901 case EIO_FUTIME:
902 {
903 struct timeval tv[2];
904 struct timeval *times;
905
906 if (req->nv1 != -1. || req->nv2 != -1.)
907 {
908 tv[0].tv_sec = req->nv1;
909 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
910 tv[1].tv_sec = req->nv2;
911 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
912
913 times = tv;
914 }
915 else
916 times = 0;
917
918
919 req->result = req->type == EIO_FUTIME
920 ? futimes (req->int1, times)
921 : utimes (req->ptr1, times);
922 }
923
924 case EIO_GROUP:
925 case EIO_NOP:
926 break;
927
928 case EIO_QUIT:
929 goto quit; 2217 goto quit;
930 2218
931 default: 2219 ETP_EXECUTE (self, req);
932 req->result = -1;
933 break;
934 }
935
936 req->errorno = errno;
937 2220
938 X_LOCK (reslock); 2221 X_LOCK (reslock);
939 2222
940 ++npending; 2223 ++npending;
941 2224
942 if (!reqq_push (&res_queue, req) && want_poll_cb) 2225 if (!reqq_push (&res_queue, req) && want_poll_cb)
943 want_poll_cb (); 2226 want_poll_cb ();
944 2227
945 self->req = 0;
946 worker_clear (self); 2228 etp_worker_clear (self);
947 2229
948 X_UNLOCK (reslock); 2230 X_UNLOCK (reslock);
949 } 2231 }
950 2232
951quit: 2233quit:
2234 free (req);
2235
952 X_LOCK (wrklock); 2236 X_LOCK (wrklock);
953 worker_free (self); 2237 etp_worker_free (self);
954 X_UNLOCK (wrklock); 2238 X_UNLOCK (wrklock);
955 2239
956 return 0; 2240 return 0;
957} 2241}
958 2242
959/*****************************************************************************/ 2243/*****************************************************************************/
960 2244
961static void eio_atfork_prepare (void) 2245int ecb_cold
962{
963 X_LOCK (wrklock);
964 X_LOCK (reqlock);
965 X_LOCK (reslock);
966#if !HAVE_PREADWRITE
967 X_LOCK (preadwritelock);
968#endif
969#if !HAVE_READDIR_R
970 X_LOCK (readdirlock);
971#endif
972}
973
974static void eio_atfork_parent (void)
975{
976#if !HAVE_READDIR_R
977 X_UNLOCK (readdirlock);
978#endif
979#if !HAVE_PREADWRITE
980 X_UNLOCK (preadwritelock);
981#endif
982 X_UNLOCK (reslock);
983 X_UNLOCK (reqlock);
984 X_UNLOCK (wrklock);
985}
986
987static void eio_atfork_child (void)
988{
989 eio_req *prv;
990
991 while (prv = reqq_shift (&req_queue))
992 eio_destroy (prv);
993
994 while (prv = reqq_shift (&res_queue))
995 eio_destroy (prv);
996
997 while (wrk_first.next != &wrk_first)
998 {
999 worker *wrk = wrk_first.next;
1000
1001 if (wrk->req)
1002 eio_destroy (wrk->req);
1003
1004 worker_clear (wrk);
1005 worker_free (wrk);
1006 }
1007
1008 started = 0;
1009 idle = 0;
1010 nreqs = 0;
1011 nready = 0;
1012 npending = 0;
1013
1014 eio_atfork_parent ();
1015}
1016
1017int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 2246eio_init (void (*want_poll)(void), void (*done_poll)(void))
1018{ 2247{
1019 want_poll_cb = want_poll; 2248 return etp_init (want_poll, done_poll);
1020 done_poll_cb = done_poll;
1021
1022#ifdef _WIN32
1023 X_MUTEX_CHECK (wrklock);
1024 X_MUTEX_CHECK (reslock);
1025 X_MUTEX_CHECK (reqlock);
1026 X_MUTEX_CHECK (reqwait);
1027 X_MUTEX_CHECK (preadwritelock);
1028 X_MUTEX_CHECK (readdirlock);
1029
1030 X_COND_CHECK (reqwait);
1031#endif
1032
1033 X_THREAD_ATFORK (eio_atfork_prepare, eio_atfork_parent, eio_atfork_child);
1034} 2249}
1035 2250
2251ecb_inline void
1036static void eio_api_destroy (eio_req *req) 2252eio_api_destroy (eio_req *req)
1037{ 2253{
1038 free (req); 2254 free (req);
1039} 2255}
1040 2256
1041#define REQ(rtype) \ 2257#define REQ(rtype) \
1042 eio_req *req; \ 2258 eio_req *req; \
1043 \ 2259 \
1044 req = (eio_req *)calloc (1, sizeof *req); \ 2260 req = (eio_req *)calloc (1, sizeof *req); \
1045 if (!req) \ 2261 if (!req) \
1046 return 0; \ 2262 return 0; \
1047 \ 2263 \
1048 req->type = EIO_ ## rtype; \ 2264 req->type = rtype; \
1049 req->pri = EIO_DEFAULT_PRI + EIO_PRI_BIAS; \ 2265 req->pri = pri; \
1050 req->finish = cb; \ 2266 req->finish = cb; \
2267 req->data = data; \
1051 req->destroy = eio_api_destroy; 2268 req->destroy = eio_api_destroy;
1052 2269
1053#define SEND eio_submit (req); return req 2270#define SEND eio_submit (req); return req
1054 2271
1055#define PATH (void)0 2272#define PATH \
2273 req->flags |= EIO_FLAG_PTR1_FREE; \
2274 req->ptr1 = strdup (path); \
2275 if (!req->ptr1) \
2276 { \
2277 eio_api_destroy (req); \
2278 return 0; \
2279 }
1056 2280
2281#define SINGLEDOT(ptr) (0[(char *)(ptr)] == '.' && !1[(char *)(ptr)])
2282
2283static void
2284eio_execute (etp_worker *self, eio_req *req)
2285{
2286#if HAVE_AT
2287 int dirfd;
2288#else
2289 const char *path;
2290#endif
2291
2292 if (ecb_expect_false (EIO_CANCELLED (req)))
2293 {
2294 req->result = -1;
2295 req->errorno = ECANCELED;
2296 return;
2297 }
2298
2299 if (ecb_expect_false (req->wd == EIO_INVALID_WD))
2300 {
2301 req->result = -1;
2302 req->errorno = ENOENT;
2303 return;
2304 }
2305
2306 if (req->type >= EIO_OPEN)
2307 {
2308 #if HAVE_AT
2309 dirfd = WD2FD (req->wd);
2310 #else
2311 path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
2312 #endif
2313 }
2314
2315 switch (req->type)
2316 {
2317 case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
2318 req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
2319 break;
2320 case EIO_WD_CLOSE: req->result = 0;
2321 eio_wd_close_sync (req->wd); break;
2322
2323 case EIO_SEEK: eio__lseek (req); break;
2324 case EIO_READ: ALLOC (req->size);
2325 req->result = req->offs >= 0
2326 ? pread (req->int1, req->ptr2, req->size, req->offs)
2327 : read (req->int1, req->ptr2, req->size); break;
2328 case EIO_WRITE: req->result = req->offs >= 0
2329 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
2330 : write (req->int1, req->ptr2, req->size); break;
2331
2332 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
2333 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
2334
2335#if HAVE_AT
2336
2337 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2338 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break;
2339 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2340 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, AT_SYMLINK_NOFOLLOW); break;
2341 case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break;
2342 case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break;
2343 case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break;
2344 case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break;
2345
2346 case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
2347 case EIO_RMDIR: /* complications arise because "." cannot be removed, so we might have to expand */
2348 req->result = req->wd && SINGLEDOT (req->ptr1)
2349 ? rmdir (req->wd->str)
2350 : unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break;
2351 case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); break;
2352 case EIO_RENAME: /* complications arise because "." cannot be renamed, so we might have to expand */
2353 req->result = req->wd && SINGLEDOT (req->ptr1)
2354 ? rename (req->wd->str, req->ptr2)
2355 : renameat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2); break;
2356 case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2, 0); break;
2357 case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
2358 case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
2359 case EIO_READLINK: ALLOC (PATH_MAX);
2360 req->result = readlinkat (dirfd, req->ptr1, req->ptr2, PATH_MAX); break;
2361 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2362 req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2363 case EIO_UTIME:
2364 case EIO_FUTIME:
2365 {
2366 struct timespec ts[2];
2367 struct timespec *times;
2368
2369 if (req->nv1 != -1. || req->nv2 != -1.)
2370 {
2371 ts[0].tv_sec = req->nv1;
2372 ts[0].tv_nsec = (req->nv1 - ts[0].tv_sec) * 1e9;
2373 ts[1].tv_sec = req->nv2;
2374 ts[1].tv_nsec = (req->nv2 - ts[1].tv_sec) * 1e9;
2375
2376 times = ts;
2377 }
2378 else
2379 times = 0;
2380
2381 req->result = req->type == EIO_FUTIME
2382 ? futimens (req->int1, times)
2383 : utimensat (dirfd, req->ptr1, times, 0);
2384 }
2385 break;
2386
2387#else
2388
2389 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2390 req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
2391 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2392 req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
2393 case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
2394 case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
2395 case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
2396 case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
2397
2398 case EIO_UNLINK: req->result = unlink (path ); break;
2399 case EIO_RMDIR: req->result = rmdir (path ); break;
2400 case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
2401 case EIO_RENAME: req->result = rename (path , req->ptr2); break;
2402 case EIO_LINK: req->result = link (path , req->ptr2); break;
2403 case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
2404 case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
2405 case EIO_READLINK: ALLOC (PATH_MAX);
2406 req->result = readlink (path, req->ptr2, PATH_MAX); break;
2407 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2408 req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
2409
2410 case EIO_UTIME:
2411 case EIO_FUTIME:
2412 {
2413 struct timeval tv[2];
2414 struct timeval *times;
2415
2416 if (req->nv1 != -1. || req->nv2 != -1.)
2417 {
2418 tv[0].tv_sec = req->nv1;
2419 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
2420 tv[1].tv_sec = req->nv2;
2421 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
2422
2423 times = tv;
2424 }
2425 else
2426 times = 0;
2427
2428 req->result = req->type == EIO_FUTIME
2429 ? futimes (req->int1, times)
2430 : utimes (req->ptr1, times);
2431 }
2432 break;
2433
2434#endif
2435
2436 case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
2437 {
2438 ALLOC (req->result);
2439 memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
2440 }
2441 break;
2442
2443 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2444 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
2445
2446 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2447 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2448
2449 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
2450 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
2451 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
2452
2453 case EIO_CLOSE: req->result = close (req->int1); break;
2454 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
2455 case EIO_SYNC: req->result = 0; sync (); break;
2456 case EIO_FSYNC: req->result = fsync (req->int1); break;
2457 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
2458 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
2459 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
2460 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
2461 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
2462 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
2463 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
2464 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
2465
2466 case EIO_READDIR: eio__scandir (req, self); break;
2467
2468 case EIO_BUSY:
2469#ifdef _WIN32
2470 Sleep (req->nv1 * 1e3);
2471#else
2472 {
2473 struct timeval tv;
2474
2475 tv.tv_sec = req->nv1;
2476 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
2477
2478 req->result = select (0, 0, 0, 0, &tv);
2479 }
2480#endif
2481 break;
2482
2483 case EIO_GROUP:
2484 abort (); /* handled in eio_request */
2485
2486 case EIO_NOP:
2487 req->result = 0;
2488 break;
2489
2490 case EIO_CUSTOM:
2491 req->feed (req);
2492 break;
2493
2494 default:
2495 req->result = EIO_ENOSYS ();
2496 break;
2497 }
2498
2499 req->errorno = errno;
2500}
2501
2502#ifndef EIO_NO_WRAPPERS
2503
2504eio_req *eio_wd_open (const char *path, int pri, eio_cb cb, void *data)
2505{
2506 REQ (EIO_WD_OPEN); PATH; SEND;
2507}
2508
2509eio_req *eio_wd_close (eio_wd wd, int pri, eio_cb cb, void *data)
2510{
2511 REQ (EIO_WD_CLOSE); req->wd = wd; SEND;
2512}
2513
2514eio_req *eio_nop (int pri, eio_cb cb, void *data)
2515{
2516 REQ (EIO_NOP); SEND;
2517}
2518
2519eio_req *eio_busy (double delay, int pri, eio_cb cb, void *data)
2520{
2521 REQ (EIO_BUSY); req->nv1 = delay; SEND;
2522}
2523
1057eio_req *eio_fsync (int fd, eio_cb cb) 2524eio_req *eio_sync (int pri, eio_cb cb, void *data)
1058{ 2525{
2526 REQ (EIO_SYNC); SEND;
2527}
2528
2529eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
2530{
1059 REQ (FSYNC); req->int1 = fd; SEND; 2531 REQ (EIO_FSYNC); req->int1 = fd; SEND;
1060} 2532}
1061 2533
2534eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2535{
2536 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2537}
2538
1062eio_req *eio_fdatasync (int fd, eio_cb cb) 2539eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1063{ 2540{
1064 REQ (FDATASYNC); req->int1 = fd; SEND; 2541 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1065} 2542}
1066 2543
2544eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
2545{
2546 REQ (EIO_SYNCFS); req->int1 = fd; SEND;
2547}
2548
2549eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2550{
2551 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2552}
2553
2554eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2555{
2556 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2557}
2558
2559eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
2560{
2561 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
2562}
2563
2564eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
2565{
2566 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2567}
2568
2569eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2570{
2571 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
2572}
2573
2574eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
2575{
2576 REQ (EIO_CLOSE); req->int1 = fd; SEND;
2577}
2578
1067eio_req *eio_readahead (int fd, off_t offset, size_t length, eio_cb cb) 2579eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
1068{ 2580{
1069 REQ (READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND; 2581 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
1070} 2582}
1071 2583
1072eio_req *eio_read (int fd, off_t offset, size_t length, void *data, eio_cb cb) 2584eio_req *eio_seek (int fd, off_t offset, int whence, int pri, eio_cb cb, void *data)
1073{ 2585{
2586 REQ (EIO_SEEK); req->int1 = fd; req->offs = offset; req->int2 = whence; SEND;
2587}
2588
2589eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
2590{
1074 REQ (READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = data; SEND; 2591 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1075} 2592}
1076 2593
1077eio_req *eio_write (int fd, off_t offset, size_t length, void *data, eio_cb cb) 2594eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1078{ 2595{
1079 REQ (WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = data; SEND; 2596 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1080} 2597}
1081 2598
1082eio_req *eio_fstat (int fd, eio_cb cb) 2599eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1083{ 2600{
1084 REQ (FSTAT); req->int1 = fd; SEND; 2601 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1085} 2602}
1086 2603
2604eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
2605{
2606 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
2607}
2608
1087eio_req *eio_futime (int fd, double atime, double mtime, eio_cb cb) 2609eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1088{ 2610{
1089 REQ (FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; 2611 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1090} 2612}
1091 2613
1092eio_req *eio_ftruncate (int fd, off_t offset, eio_cb cb) 2614eio_req *eio_ftruncate (int fd, off_t offset, int pri, eio_cb cb, void *data)
1093{ 2615{
1094 REQ (FTRUNCATE); req->int1 = fd; req->offs = offset; SEND; 2616 REQ (EIO_FTRUNCATE); req->int1 = fd; req->offs = offset; SEND;
1095} 2617}
1096 2618
1097eio_req *eio_fchmod (int fd, mode_t mode, eio_cb cb) 2619eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1098{ 2620{
1099 REQ (FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2621 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1100} 2622}
1101 2623
1102eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, eio_cb cb) 2624eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1103{ 2625{
1104 REQ (FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2626 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1105} 2627}
1106 2628
1107eio_req *eio_dup2 (int fd, int fd2, eio_cb cb) 2629eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
1108{ 2630{
1109 REQ (DUP2); req->int1 = fd; req->int2 = fd2; SEND; 2631 REQ (EIO_DUP2); req->int1 = fd; req->int2 = fd2; SEND;
1110} 2632}
1111 2633
1112eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, eio_cb cb) 2634eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, int pri, eio_cb cb, void *data)
1113{ 2635{
1114 REQ (SENDFILE); req->int1 = out_fd; req->int2 = in_fd; req->offs = in_offset; req->size = length; SEND; 2636 REQ (EIO_SENDFILE); req->int1 = out_fd; req->int2 = in_fd; req->offs = in_offset; req->size = length; SEND;
1115} 2637}
1116 2638
1117eio_req *eio_open (const char *path, int flags, mode_t mode, eio_cb cb) 2639eio_req *eio_open (const char *path, int flags, mode_t mode, int pri, eio_cb cb, void *data)
1118{ 2640{
1119 REQ (OPEN); PATH; req->int1 = flags; req->int2 = (long)mode; SEND; 2641 REQ (EIO_OPEN); PATH; req->int1 = flags; req->int2 = (long)mode; SEND;
1120} 2642}
1121 2643
1122eio_req *eio_readlink (const char *path, eio_cb cb)
1123{
1124 REQ (READLINK); PATH; SEND;
1125}
1126
1127eio_req *eio_stat (const char *path, eio_cb cb)
1128{
1129 REQ (STAT); PATH; SEND;
1130}
1131
1132eio_req *eio_lstat (const char *path, eio_cb cb)
1133{
1134 REQ (LSTAT); PATH; SEND;
1135}
1136
1137eio_req *eio_utime (const char *path, double atime, double mtime, eio_cb cb) 2644eio_req *eio_utime (const char *path, double atime, double mtime, int pri, eio_cb cb, void *data)
1138{ 2645{
1139 REQ (UTIME); PATH; req->nv1 = atime; req->nv2 = mtime; SEND; 2646 REQ (EIO_UTIME); PATH; req->nv1 = atime; req->nv2 = mtime; SEND;
1140} 2647}
1141 2648
1142eio_req *eio_truncate (const char *path, off_t offset, eio_cb cb) 2649eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1143{ 2650{
1144 REQ (TRUNCATE); PATH; req->offs = offset; SEND; 2651 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1145} 2652}
1146 2653
1147eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, eio_cb cb) 2654eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1148{ 2655{
1149 REQ (CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2656 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1150} 2657}
1151 2658
1152eio_req *eio_chmod (const char *path, mode_t mode, eio_cb cb) 2659eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
1153{ 2660{
1154 REQ (CHMOD); PATH; req->int2 = (long)mode; SEND; 2661 REQ (EIO_CHMOD); PATH; req->int2 = (long)mode; SEND;
1155} 2662}
1156 2663
1157eio_req *eio_mkdir (const char *path, mode_t mode, eio_cb cb) 2664eio_req *eio_mkdir (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
1158{ 2665{
1159 REQ (MKDIR); PATH; req->int2 = (long)mode; SEND; 2666 REQ (EIO_MKDIR); PATH; req->int2 = (long)mode; SEND;
1160} 2667}
1161 2668
1162eio_req *eio_unlink (const char *path, eio_cb cb) 2669static eio_req *
2670eio__1path (int type, const char *path, int pri, eio_cb cb, void *data)
1163{ 2671{
1164 REQ (UNLINK); PATH; SEND;
1165}
1166
1167eio_req *eio_rmdir (const char *path, eio_cb cb)
1168{
1169 REQ (RMDIR); PATH; SEND; 2672 REQ (type); PATH; SEND;
1170} 2673}
1171 2674
2675eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
2676{
2677 return eio__1path (EIO_READLINK, path, pri, cb, data);
2678}
2679
2680eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2681{
2682 return eio__1path (EIO_REALPATH, path, pri, cb, data);
2683}
2684
2685eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
2686{
2687 return eio__1path (EIO_STAT, path, pri, cb, data);
2688}
2689
2690eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
2691{
2692 return eio__1path (EIO_LSTAT, path, pri, cb, data);
2693}
2694
2695eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
2696{
2697 return eio__1path (EIO_STATVFS, path, pri, cb, data);
2698}
2699
2700eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
2701{
2702 return eio__1path (EIO_UNLINK, path, pri, cb, data);
2703}
2704
1172eio_req *eio_readdir (const char *path, eio_cb cb) 2705eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1173{ 2706{
1174 REQ (READDIR); PATH; SEND; 2707 return eio__1path (EIO_RMDIR, path, pri, cb, data);
1175} 2708}
1176 2709
2710eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
2711{
2712 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
2713}
2714
1177eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, eio_cb cb) 2715eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1178{ 2716{
1179 REQ (MKNOD); PATH; req->int2 = (long)mode; req->int2 = (long)dev; SEND; 2717 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1180} 2718}
1181 2719
1182eio_req *eio_busy (double delay, eio_cb cb) 2720static eio_req *
2721eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1183{ 2722{
1184 REQ (BUSY); req->nv1 = delay; SEND; 2723 REQ (type); PATH;
1185}
1186 2724
2725 req->flags |= EIO_FLAG_PTR2_FREE;
2726 req->ptr2 = strdup (new_path);
2727 if (!req->ptr2)
2728 {
2729 eio_api_destroy (req);
2730 return 0;
2731 }
2732
2733 SEND;
2734}
2735
2736eio_req *eio_link (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2737{
2738 return eio__2path (EIO_LINK, path, new_path, pri, cb, data);
2739}
2740
2741eio_req *eio_symlink (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2742{
2743 return eio__2path (EIO_SYMLINK, path, new_path, pri, cb, data);
2744}
2745
2746eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2747{
2748 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
2749}
2750
2751eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
2752{
2753 REQ (EIO_CUSTOM); req->feed = execute; SEND;
2754}
2755
2756#endif
2757
1187eio_req *eio_nop (eio_cb cb) 2758eio_req *eio_grp (eio_cb cb, void *data)
1188{ 2759{
2760 const int pri = EIO_PRI_MAX;
2761
1189 REQ (NOP); SEND; 2762 REQ (EIO_GROUP); SEND;
1190} 2763}
1191 2764
1192#undef REQ 2765#undef REQ
1193#undef PATH 2766#undef PATH
1194#undef SEND 2767#undef SEND
1195 2768
1196#if 0 2769/*****************************************************************************/
2770/* grp functions */
1197 2771
1198void 2772void
1199aio_open (SV8 *pathname, int flags, int mode, SV *callback=&PL_sv_undef)
1200 PROTOTYPE: $$$;$
1201 PPCODE:
1202{
1203 dREQ;
1204
1205 req->type = EIO_OPEN;
1206 req->sv1 = newSVsv (pathname);
1207 req->ptr1 = SvPVbyte_nolen (req->sv1);
1208 req->int1 = flags;
1209 req->int2 = mode;
1210
1211 EIO_SEND;
1212}
1213
1214void
1215aio_fsync (SV *fh, SV *callback=&PL_sv_undef)
1216 PROTOTYPE: $;$
1217 ALIAS:
1218 aio_fsync = EIO_FSYNC
1219 aio_fdatasync = EIO_FDATASYNC
1220 PPCODE:
1221{
1222 dREQ;
1223
1224 req->type = ix;
1225 req->sv1 = newSVsv (fh);
1226 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1227
1228 EIO_SEND (req);
1229}
1230
1231void
1232aio_close (SV *fh, SV *callback=&PL_sv_undef)
1233 PROTOTYPE: $;$
1234 PPCODE:
1235{
1236 dREQ;
1237
1238 req->type = EIO_CLOSE;
1239 req->sv1 = newSVsv (fh);
1240 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1241
1242 EIO_SEND (req);
1243}
1244
1245void
1246aio_read (SV *fh, SV *offset, SV *length, SV8 *data, IV dataoffset, SV *callback=&PL_sv_undef)
1247 ALIAS:
1248 aio_read = EIO_READ
1249 aio_write = EIO_WRITE
1250 PROTOTYPE: $$$$$;$
1251 PPCODE:
1252{
1253 STRLEN svlen;
1254 char *svptr = SvPVbyte (data, svlen);
1255 UV len = SvUV (length);
1256
1257 SvUPGRADE (data, SVt_PV);
1258 SvPOK_on (data);
1259
1260 if (dataoffset < 0)
1261 dataoffset += svlen;
1262
1263 if (dataoffset < 0 || dataoffset > svlen)
1264 croak ("dataoffset outside of data scalar");
1265
1266 if (ix == EIO_WRITE)
1267 {
1268 /* write: check length and adjust. */
1269 if (!SvOK (length) || len + dataoffset > svlen)
1270 len = svlen - dataoffset;
1271 }
1272 else
1273 {
1274 /* read: grow scalar as necessary */
1275 svptr = SvGROW (data, len + dataoffset + 1);
1276 }
1277
1278 if (len < 0)
1279 croak ("length must not be negative");
1280
1281 {
1282 dREQ;
1283
1284 req->type = ix;
1285 req->sv1 = newSVsv (fh);
1286 req->int1 = PerlIO_fileno (ix == EIO_READ ? IoIFP (sv_2io (fh))
1287 : IoOFP (sv_2io (fh)));
1288 req->offs = SvOK (offset) ? SvVAL64 (offset) : -1;
1289 req->size = len;
1290 req->sv2 = SvREFCNT_inc (data);
1291 req->ptr2 = (char *)svptr + dataoffset;
1292 req->stroffset = dataoffset;
1293
1294 if (!SvREADONLY (data))
1295 {
1296 SvREADONLY_on (data);
1297 req->flags |= FLAG_SV2_RO_OFF;
1298 }
1299
1300 EIO_SEND;
1301 }
1302}
1303
1304void
1305aio_readlink (SV8 *path, SV *callback=&PL_sv_undef)
1306 PROTOTYPE: $$;$
1307 PPCODE:
1308{
1309 SV *data;
1310 dREQ;
1311
1312 data = newSV (NAME_MAX);
1313 SvPOK_on (data);
1314
1315 req->type = EIO_READLINK;
1316 req->sv1 = newSVsv (path);
1317 req->ptr1 = SvPVbyte_nolen (req->sv1);
1318 req->sv2 = data;
1319 req->ptr2 = SvPVbyte_nolen (data);
1320
1321 EIO_SEND;
1322}
1323
1324void
1325aio_sendfile (SV *out_fh, SV *in_fh, SV *in_offset, UV length, SV *callback=&PL_sv_undef)
1326 PROTOTYPE: $$$$;$
1327 PPCODE:
1328{
1329 dREQ;
1330
1331 req->type = EIO_SENDFILE;
1332 req->sv1 = newSVsv (out_fh);
1333 req->int1 = PerlIO_fileno (IoIFP (sv_2io (out_fh)));
1334 req->sv2 = newSVsv (in_fh);
1335 req->int2 = PerlIO_fileno (IoIFP (sv_2io (in_fh)));
1336 req->offs = SvVAL64 (in_offset);
1337 req->size = length;
1338
1339 EIO_SEND;
1340}
1341
1342void
1343aio_readahead (SV *fh, SV *offset, IV length, SV *callback=&PL_sv_undef)
1344 PROTOTYPE: $$$;$
1345 PPCODE:
1346{
1347 dREQ;
1348
1349 req->type = EIO_READAHEAD;
1350 req->sv1 = newSVsv (fh);
1351 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1352 req->offs = SvVAL64 (offset);
1353 req->size = length;
1354
1355 EIO_SEND;
1356}
1357
1358void
1359aio_stat (SV8 *fh_or_path, SV *callback=&PL_sv_undef)
1360 ALIAS:
1361 aio_stat = EIO_STAT
1362 aio_lstat = EIO_LSTAT
1363 PPCODE:
1364{
1365 dREQ;
1366
1367 req->ptr2 = malloc (sizeof (EIO_STRUCT_STAT));
1368 if (!req->ptr2)
1369 {
1370 req_destroy (req);
1371 croak ("out of memory during aio_stat statdata allocation");
1372 }
1373
1374 req->flags |= FLAG_PTR2_FREE;
1375 req->sv1 = newSVsv (fh_or_path);
1376
1377 if (SvPOK (fh_or_path))
1378 {
1379 req->type = ix;
1380 req->ptr1 = SvPVbyte_nolen (req->sv1);
1381 }
1382 else
1383 {
1384 req->type = EIO_FSTAT;
1385 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1386 }
1387
1388 EIO_SEND;
1389}
1390
1391void
1392aio_utime (SV8 *fh_or_path, SV *atime, SV *mtime, SV *callback=&PL_sv_undef)
1393 PPCODE:
1394{
1395 dREQ;
1396
1397 req->nv1 = SvOK (atime) ? SvNV (atime) : -1.;
1398 req->nv2 = SvOK (mtime) ? SvNV (mtime) : -1.;
1399 req->sv1 = newSVsv (fh_or_path);
1400
1401 if (SvPOK (fh_or_path))
1402 {
1403 req->type = EIO_UTIME;
1404 req->ptr1 = SvPVbyte_nolen (req->sv1);
1405 }
1406 else
1407 {
1408 req->type = EIO_FUTIME;
1409 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1410 }
1411
1412 EIO_SEND;
1413}
1414
1415void
1416aio_truncate (SV8 *fh_or_path, SV *offset, SV *callback=&PL_sv_undef)
1417 PPCODE:
1418{
1419 dREQ;
1420
1421 req->sv1 = newSVsv (fh_or_path);
1422 req->offs = SvOK (offset) ? SvVAL64 (offset) : -1;
1423
1424 if (SvPOK (fh_or_path))
1425 {
1426 req->type = EIO_TRUNCATE;
1427 req->ptr1 = SvPVbyte_nolen (req->sv1);
1428 }
1429 else
1430 {
1431 req->type = EIO_FTRUNCATE;
1432 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1433 }
1434
1435 EIO_SEND;
1436}
1437
1438void
1439aio_chmod (SV8 *fh_or_path, int mode, SV *callback=&PL_sv_undef)
1440 ALIAS:
1441 aio_chmod = EIO_CHMOD
1442 aio_fchmod = EIO_FCHMOD
1443 aio_mkdir = EIO_MKDIR
1444 PPCODE:
1445{
1446 dREQ;
1447
1448 req->type = type;
1449 req->int2 = mode;
1450 req->sv1 = newSVsv (fh_or_path);
1451
1452 if (ix == EIO_FCHMOD)
1453 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1454 else
1455 req->ptr1 = SvPVbyte_nolen (req->sv1);
1456
1457 EIO_SEND;
1458}
1459
1460void
1461aio_chown (SV8 *fh_or_path, SV *uid, SV *gid, SV *callback=&PL_sv_undef)
1462 PPCODE:
1463{
1464 dREQ;
1465
1466 req->int2 = SvOK (uid) ? SvIV (uid) : -1;
1467 req->int3 = SvOK (gid) ? SvIV (gid) : -1;
1468 req->sv1 = newSVsv (fh_or_path);
1469
1470 if (SvPOK (fh_or_path))
1471 {
1472 req->type = EIO_CHOWN;
1473 req->ptr1 = SvPVbyte_nolen (req->sv1);
1474 }
1475 else
1476 {
1477 req->type = EIO_FCHOWN;
1478 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1479 }
1480
1481 EIO_SEND;
1482}
1483
1484void
1485aio_unlink (SV8 *pathname, SV *callback=&PL_sv_undef)
1486 ALIAS:
1487 aio_unlink = EIO_UNLINK
1488 aio_rmdir = EIO_RMDIR
1489 aio_readdir = EIO_READDIR
1490 PPCODE:
1491{
1492 dREQ;
1493
1494 req->type = ix;
1495 req->sv1 = newSVsv (pathname);
1496 req->ptr1 = SvPVbyte_nolen (req->sv1);
1497
1498 EIO_SEND;
1499}
1500
1501void
1502aio_link (SV8 *oldpath, SV8 *newpath, SV *callback=&PL_sv_undef)
1503 ALIAS:
1504 aio_link = EIO_LINK
1505 aio_symlink = EIO_SYMLINK
1506 aio_rename = EIO_RENAME
1507 PPCODE:
1508{
1509 dREQ;
1510
1511 req->type = ix;
1512 req->sv1 = newSVsv (oldpath);
1513 req->ptr1 = SvPVbyte_nolen (req->sv1);
1514 req->sv2 = newSVsv (newpath);
1515 req->ptr2 = SvPVbyte_nolen (req->sv2);
1516
1517 EIO_SEND;
1518}
1519
1520void
1521aio_mknod (SV8 *pathname, int mode, UV dev, SV *callback=&PL_sv_undef)
1522 PPCODE:
1523{
1524 dREQ;
1525
1526 req->type = EIO_MKNOD;
1527 req->sv1 = newSVsv (pathname);
1528 req->ptr1 = SvPVbyte_nolen (req->sv1);
1529 req->int2 = (mode_t)mode;
1530 req->offs = dev;
1531
1532 EIO_SEND;
1533}
1534
1535void
1536aio_busy (double delay, SV *callback=&PL_sv_undef)
1537 PPCODE:
1538{
1539 dREQ;
1540
1541 req->type = EIO_BUSY;
1542 req->nv1 = delay < 0. ? 0. : delay;
1543
1544 EIO_SEND;
1545}
1546
1547void
1548aio_group (SV *callback=&PL_sv_undef)
1549 PROTOTYPE: ;$
1550 PPCODE:
1551{
1552 dREQ;
1553
1554 req->type = EIO_GROUP;
1555
1556 req_send (req);
1557 XPUSHs (req_sv (req, AIO_GRP_KLASS));
1558}
1559
1560void
1561aio_nop (SV *callback=&PL_sv_undef)
1562 ALIAS:
1563 aio_nop = EIO_NOP
1564 aio_sync = EIO_SYNC
1565 PPCODE:
1566{
1567 dREQ;
1568
1569#endif
1570
1571void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2773eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1572{ 2774{
1573 grp->int2 = limit; 2775 grp->int2 = limit;
1574 grp->feed = feed; 2776 grp->feed = feed;
1575 2777
1576 grp_try_feed (grp); 2778 grp_try_feed (grp);
1577} 2779}
1578 2780
2781void
1579void eio_grp_limit (eio_req *grp, int limit) 2782eio_grp_limit (eio_req *grp, int limit)
1580{ 2783{
1581 grp->int2 = limit; 2784 grp->int2 = limit;
1582 2785
1583 grp_try_feed (grp); 2786 grp_try_feed (grp);
1584} 2787}
1585 2788
2789void
1586void eio_grp_add (eio_req *grp, eio_req *req) 2790eio_grp_add (eio_req *grp, eio_req *req)
1587{ 2791{
1588 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2792 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
2793
2794 grp->flags |= EIO_FLAG_GROUPADD;
1589 2795
1590 ++grp->size; 2796 ++grp->size;
1591 req->grp = grp; 2797 req->grp = grp;
1592 2798
1593 req->grp_prev = 0; 2799 req->grp_prev = 0;
… …
1597 grp->grp_first->grp_prev = req; 2803 grp->grp_first->grp_prev = req;
1598 2804
1599 grp->grp_first = req; 2805 grp->grp_first = req;
1600} 2806}
1601 2807
2808/*****************************************************************************/
2809/* misc garbage */
1602 2810
2811eio_ssize_t
2812eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2813{
2814 return eio__sendfile (ofd, ifd, offset, count);
2815}
2816

Diff Legend

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