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

Comparing libeio/eio.c (file contents):
Revision 1.2 by root, Sat May 10 23:33:40 2008 UTC vs.
Revision 1.134 by root, Thu Jun 25 15:59:57 2015 UTC

1/*
2 * libeio implementation
3 *
4 * Copyright (c) 2007,2008,2009,2010,2011,2012,2013 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
6#include "EXTERN.h"
7#include "perl.h"
8#include "XSUB.h"
9
10#include <stddef.h> 53#include <stddef.h>
11#include <stdlib.h> 54#include <stdlib.h>
55#include <string.h>
12#include <errno.h> 56#include <errno.h>
13#include <sys/types.h> 57#include <sys/types.h>
14#include <sys/stat.h> 58#include <sys/stat.h>
15#include <limits.h> 59#include <limits.h>
16#include <fcntl.h> 60#include <fcntl.h>
17#include <sched.h> 61#include <assert.h>
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);
18 81
19#ifndef EIO_FINISH 82#ifndef EIO_FINISH
20# 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
21#endif 84#endif
22 85
… …
26 89
27#ifndef EIO_FEED 90#ifndef EIO_FEED
28# 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)
29#endif 92#endif
30 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
31#ifdef _WIN32 105#ifdef _WIN32
32 106
33 /*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 #define pipe(fds) _pipe (fds, 4096, O_BINARY)
126
127 #define fchmod(fd,mode) EIO_ENOSYS ()
128 #define chown(path,uid,gid) EIO_ENOSYS ()
129 #define fchown(fd,uid,gid) EIO_ENOSYS ()
130 #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
131 #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
132 #define mknod(path,mode,dev) EIO_ENOSYS ()
133 #define sync() EIO_ENOSYS ()
134 #define readlink(path,buf,s) EIO_ENOSYS ()
135 #define statvfs(path,buf) EIO_ENOSYS ()
136 #define fstatvfs(fd,buf) EIO_ENOSYS ()
137
138 #define pread(fd,buf,count,offset) eio__pread (fd, buf, count, offset)
139 #define pwrite(fd,buf,count,offset) eio__pwrite (fd, buf, count, offset)
140
141 #if __GNUC__
142 typedef long long eio_off_t; /* signed for compatibility to msvc */
143 #else
144 typedef __int64 eio_off_t; /* unsigned not supported by msvc */
145 #endif
146
147 static eio_ssize_t
148 eio__pread (int fd, void *buf, eio_ssize_t count, eio_off_t offset)
149 {
150 OVERLAPPED o = { 0 };
151 DWORD got;
152
153 o.Offset = offset;
154 o.OffsetHigh = offset >> 32;
155
156 return ReadFile ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd), buf, count, &got, &o)
157 ? got : -1;
158 }
159
160 static eio_ssize_t
161 eio__pwrite (int fd, void *buf, eio_ssize_t count, eio_off_t offset)
162 {
163 OVERLAPPED o = { 0 };
164 DWORD got;
165
166 o.Offset = offset;
167 o.OffsetHigh = offset >> 32;
168
169 return WriteFile ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd), buf, count, &got, &o)
170 ? got : -1;
171 }
172
173 /* rename() uses MoveFile, which fails to overwrite */
174 #define rename(old,neu) eio__rename (old, neu)
175
176 static int
177 eio__rename (const char *old, const char *neu)
178 {
179 if (MoveFileEx (old, neu, MOVEFILE_REPLACE_EXISTING))
180 return 0;
181
182 /* should steal _dosmaperr */
183 switch (GetLastError ())
184 {
185 case ERROR_FILE_NOT_FOUND:
186 case ERROR_PATH_NOT_FOUND:
187 case ERROR_INVALID_DRIVE:
188 case ERROR_NO_MORE_FILES:
189 case ERROR_BAD_NETPATH:
190 case ERROR_BAD_NET_NAME:
191 case ERROR_BAD_PATHNAME:
192 case ERROR_FILENAME_EXCED_RANGE:
193 errno = ENOENT;
194 break;
195
196 default:
197 errno = EACCES;
198 break;
199 }
200
201 return -1;
202 }
203
204 /* we could even stat and see if it exists */
205 static int
206 symlink (const char *old, const char *neu)
207 {
208 #if WINVER >= 0x0600
209 if (CreateSymbolicLink (neu, old, 1))
210 return 0;
211
212 if (CreateSymbolicLink (neu, old, 0))
213 return 0;
214 #endif
215
216 return EIO_ERRNO (ENOENT, -1);
217 }
218
219 /* POSIX API only, causing trouble for win32 apps */
220 #define CreateHardLink(neu,old,flags) 0 /* not really creating hardlink, still using relative paths? */
221 #define CreateSymbolicLink(neu,old,flags) 0 /* vista+ only */
222
223 struct statvfs
224 {
225 int dummy;
226 };
227
228 #define DT_DIR EIO_DT_DIR
229 #define DT_REG EIO_DT_REG
230 #define D_NAME(entp) entp.cFileName
231 #define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
34 232
35#else 233#else
36 234
37# include "config.h"
38# include <sys/time.h> 235 #include <sys/time.h>
39# include <sys/select.h> 236 #include <sys/select.h>
40# include <unistd.h> 237 #include <unistd.h>
238 #include <signal.h>
239 #include <dirent.h>
240
241 #ifdef ANDROID
242 #include <sys/vfs.h>
243 #define statvfs statfs
244 #define fstatvfs fstatfs
245 #include <asm/page.h> /* supposedly limits.h does #define PAGESIZE PAGESIZE */
246 #else
247 #include <sys/statvfs.h>
248 #endif
249
250 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
251 #include <sys/mman.h>
252 #endif
253
254 #define D_NAME(entp) entp->d_name
255
256 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
257 #if __FreeBSD__ || __NetBSD__ || __OpenBSD__
258 #define _DIRENT_HAVE_D_TYPE /* sigh */
259 #define D_INO(de) (de)->d_fileno
260 #define D_NAMLEN(de) (de)->d_namlen
261 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
262 #define D_INO(de) (de)->d_ino
263 #endif
264
265 #ifdef _D_EXACT_NAMLEN
266 #undef D_NAMLEN
267 #define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
268 #endif
269
270 #ifdef _DIRENT_HAVE_D_TYPE
271 #define D_TYPE(de) (de)->d_type
272 #endif
273
274 #ifndef EIO_STRUCT_DIRENT
275 #define EIO_STRUCT_DIRENT struct dirent
276 #endif
277
278#endif
279
280#if HAVE_UTIMES
41# include <utime.h> 281# include <utime.h>
42# include <signal.h>
43
44# ifndef EIO_STRUCT_DIRENT
45# define EIO_STRUCT_DIRENT struct dirent
46# endif 282#endif
47 283
48#endif 284#if HAVE_SYS_SYSCALL_H
49 285# include <sys/syscall.h>
50# ifndef EIO_STRUCT_STAT
51# define EIO_STRUCT_STAT struct stat
52# endif 286#endif
287
288#if HAVE_SYS_PRCTL_H
289# include <sys/prctl.h>
290#endif
53 291
54#if HAVE_SENDFILE 292#if HAVE_SENDFILE
55# if __linux 293# if __linux
56# include <sys/sendfile.h> 294# include <sys/sendfile.h>
57# elif __freebsd 295# elif __FreeBSD__ || defined __APPLE__
58# include <sys/socket.h> 296# include <sys/socket.h>
59# include <sys/uio.h> 297# include <sys/uio.h>
60# elif __hpux 298# elif __hpux
61# include <sys/socket.h> 299# include <sys/socket.h>
62# elif __solaris /* not yet */ 300# elif __solaris
63# include <sys/sendfile.h> 301# include <sys/sendfile.h>
64# else 302# else
65# error sendfile support requested but not available 303# error sendfile support requested but not available
66# endif 304# endif
67#endif 305#endif
68 306
69/* number of seconds after which an idle threads exit */ 307#ifndef D_TYPE
70#define IDLE_TIMEOUT 10 308# define D_TYPE(de) 0
309#endif
310#ifndef D_INO
311# define D_INO(de) 0
312#endif
313#ifndef D_NAMLEN
314# define D_NAMLEN(entp) strlen (D_NAME (entp))
315#endif
71 316
72/* used for struct dirent, AIX doesn't provide it */ 317/* used for struct dirent, AIX doesn't provide it */
73#ifndef NAME_MAX 318#ifndef NAME_MAX
74# define NAME_MAX 4096 319# define NAME_MAX 4096
75#endif 320#endif
76 321
322/* used for readlink etc. */
323#ifndef PATH_MAX
324# define PATH_MAX 4096
325#endif
326
77/* buffer size for various temporary buffers */ 327/* buffer size for various temporary buffers */
78#define EIO_BUFSIZE 65536 328#define EIO_BUFSIZE 65536
79 329
80#define dBUF \ 330#define dBUF \
81 char *eio_buf; \
82 X_LOCK (wrklock); \
83 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 331 char *eio_buf = malloc (EIO_BUFSIZE); \
84 X_UNLOCK (wrklock); \ 332 errno = ENOMEM; \
85 if (!eio_buf) \ 333 if (!eio_buf) \
86 return -1; 334 return -1
87 335
88#define EIO_TICKS ((1000000 + 1023) >> 10) 336#define FUBd \
337 free (eio_buf)
89 338
90static void (*want_poll_cb) (void); 339/*****************************************************************************/
91static void (*done_poll_cb) (void);
92 340
93static unsigned int max_poll_time = 0; 341struct etp_tmpbuf;
94static unsigned int max_poll_reqs = 0;
95 342
96/* calculcate time difference in ~1/EIO_TICKS of a second */ 343#if _POSIX_VERSION >= 200809L
97static int tvdiff (struct timeval *tv1, struct timeval *tv2) 344 #define HAVE_AT 1
98{ 345 #define WD2FD(wd) ((wd) ? (wd)->fd : AT_FDCWD)
99 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 346 #ifndef O_SEARCH
100 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 347 #define O_SEARCH O_RDONLY
101} 348 #endif
349#else
350 #define HAVE_AT 0
351 static const char *wd_expand (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path);
352#endif
102 353
103static unsigned int started, idle, wanted; 354struct eio_pwd
104
105/* worker threads management */
106static mutex_t wrklock = X_MUTEX_INIT;
107
108typedef struct worker {
109 /* locked by wrklock */
110 struct worker *prev, *next;
111
112 thread_t tid;
113
114 /* locked by reslock, reqlock or wrklock */
115 eio_req *req; /* currently processed request */
116 void *dbuf;
117 DIR *dirp;
118} worker;
119
120static worker wrk_first = { &wrk_first, &wrk_first, 0 };
121
122static void worker_clear (worker *wrk)
123{ 355{
124 if (wrk->dirp) 356#if HAVE_AT
125 { 357 int fd;
126 closedir (wrk->dirp); 358#endif
127 wrk->dirp = 0;
128 }
129
130 if (wrk->dbuf)
131 {
132 free (wrk->dbuf);
133 wrk->dbuf = 0;
134 }
135}
136
137static void worker_free (worker *wrk)
138{
139 wrk->next->prev = wrk->prev;
140 wrk->prev->next = wrk->next;
141
142 free (wrk);
143}
144
145static volatile unsigned int nreqs, nready, npending;
146static volatile unsigned int max_idle = 4;
147
148static mutex_t reslock = X_MUTEX_INIT;
149static mutex_t reqlock = X_MUTEX_INIT;
150static cond_t reqwait = X_COND_INIT;
151
152unsigned int eio_nreqs (void)
153{
154 return nreqs;
155}
156
157unsigned int eio_nready (void)
158{
159 unsigned int retval;
160
161 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
162 retval = nready;
163 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
164
165 return retval;
166}
167
168unsigned int eio_npending (void)
169{
170 unsigned int retval;
171
172 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
173 retval = npending;
174 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
175
176 return retval;
177}
178
179unsigned int eio_nthreads (void)
180{
181 unsigned int retval;
182
183 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
184 retval = started;
185 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
186
187 return retval;
188}
189
190/*
191 * a somewhat faster data structure might be nice, but
192 * with 8 priorities this actually needs <20 insns
193 * per shift, the most expensive operation.
194 */
195typedef struct {
196 eio_req *qs[EIO_NUM_PRI], *qe[EIO_NUM_PRI]; /* qstart, qend */
197 int size; 359 int len;
198} reqq; 360 char str[1]; /* actually, a 0-terminated canonical path */
361};
199 362
200static reqq req_queue; 363/*****************************************************************************/
201static reqq res_queue;
202 364
203static int reqq_push (reqq *q, eio_req *req) 365#define ETP_PRI_MIN EIO_PRI_MIN
204{ 366#define ETP_PRI_MAX EIO_PRI_MAX
205 int pri = req->pri;
206 req->next = 0;
207 367
208 if (q->qe[pri]) 368#define ETP_TYPE_QUIT -1
209 { 369#define ETP_TYPE_GROUP EIO_GROUP
210 q->qe[pri]->next = req;
211 q->qe[pri] = req;
212 }
213 else
214 q->qe[pri] = q->qs[pri] = req;
215 370
216 return q->size++; 371struct etp_worker;
217}
218 372
219static eio_req *reqq_shift (reqq *q) 373#define ETP_REQ eio_req
220{ 374#define ETP_DESTROY(req) eio_destroy (req)
221 int pri; 375static int eio_finish (eio_req *req);
376#define ETP_FINISH(req) eio_finish (req)
377static void eio_execute (struct etp_worker *self, eio_req *req);
378#define ETP_EXECUTE(wrk,req) eio_execute (wrk, req)
222 379
223 if (!q->size) 380#include "etp.c"
224 return 0;
225 381
226 --q->size; 382/*****************************************************************************/
227 383
228 for (pri = EIO_NUM_PRI; pri--; ) 384static void
229 {
230 eio_req *req = q->qs[pri];
231
232 if (req)
233 {
234 if (!(q->qs[pri] = (eio_req *)req->next))
235 q->qe[pri] = 0;
236
237 return req;
238 }
239 }
240
241 abort ();
242}
243
244static void grp_try_feed (eio_req *grp) 385grp_try_feed (eio_req *grp)
245{ 386{
246 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 387 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
247 { 388 {
248 int old_len = grp->size; 389 grp->flags &= ~EIO_FLAG_GROUPADD;
249 390
250 EIO_FEED (grp); 391 EIO_FEED (grp);
251 392
252 /* stop if no progress has been made */ 393 /* stop if no progress has been made */
253 if (old_len == grp->size) 394 if (!(grp->flags & EIO_FLAG_GROUPADD))
254 { 395 {
255 grp->feed = 0; 396 grp->feed = 0;
256 break; 397 break;
257 } 398 }
258 } 399 }
259} 400}
260 401
261static int eio_finish (eio_req *req); 402static int
262
263static int grp_dec (eio_req *grp) 403grp_dec (eio_req *grp)
264{ 404{
265 --grp->size; 405 --grp->size;
266 406
267 /* call feeder, if applicable */ 407 /* call feeder, if applicable */
268 grp_try_feed (grp); 408 grp_try_feed (grp);
… …
272 return eio_finish (grp); 412 return eio_finish (grp);
273 else 413 else
274 return 0; 414 return 0;
275} 415}
276 416
417static void
277void eio_destroy (eio_req *req) 418eio_destroy (eio_req *req)
278{ 419{
420 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
279 if ((req)->flags & EIO_FLAG_PTR2_FREE) 421 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
280 free (req->ptr2);
281 422
282 EIO_DESTROY (req); 423 EIO_DESTROY (req);
283} 424}
284 425
426static int
285static int eio_finish (eio_req *req) 427eio_finish (eio_req *req)
286{ 428{
287 int res = EIO_FINISH (req); 429 int res = EIO_FINISH (req);
288 430
289 if (req->grp) 431 if (req->grp)
290 { 432 {
… …
298 if (grp->grp_first == req) 440 if (grp->grp_first == req)
299 grp->grp_first = req->grp_next; 441 grp->grp_first = req->grp_next;
300 442
301 res2 = grp_dec (grp); 443 res2 = grp_dec (grp);
302 444
303 if (!res && res2) 445 if (!res)
304 res = res2; 446 res = res2;
305 } 447 }
306 448
307 eio_destroy (req); 449 eio_destroy (req);
308 450
309 return res; 451 return res;
310} 452}
311 453
454void
312void eio_grp_cancel (eio_req *grp) 455eio_grp_cancel (eio_req *grp)
313{ 456{
314 for (grp = grp->grp_first; grp; grp = grp->grp_next)
315 eio_cancel (grp); 457 etp_grp_cancel (grp);
316} 458}
317 459
460void
318void eio_cancel (eio_req *req) 461eio_cancel (eio_req *req)
319{ 462{
320 req->flags |= EIO_FLAG_CANCELLED;
321
322 eio_grp_cancel (req); 463 etp_cancel (req);
323} 464}
324 465
325X_THREAD_PROC (eio_proc); 466void
326
327static void start_thread (void)
328{
329 worker *wrk = calloc (1, sizeof (worker));
330
331 if (!wrk)
332 croak ("unable to allocate worker thread data");
333
334 X_LOCK (wrklock);
335
336 if (thread_create (&wrk->tid, eio_proc, (void *)wrk))
337 {
338 wrk->prev = &wrk_first;
339 wrk->next = wrk_first.next;
340 wrk_first.next->prev = wrk;
341 wrk_first.next = wrk;
342 ++started;
343 }
344 else
345 free (wrk);
346
347 X_UNLOCK (wrklock);
348}
349
350static void maybe_start_thread (void)
351{
352 if (eio_nthreads () >= wanted)
353 return;
354
355 /* todo: maybe use idle here, but might be less exact */
356 if (0 <= (int)eio_nthreads () + (int)eio_npending () - (int)eio_nreqs ())
357 return;
358
359 start_thread ();
360}
361
362void eio_submit (eio_req *req) 467eio_submit (eio_req *req)
363{ 468{
364 ++nreqs; 469 etp_submit (req);
365
366 X_LOCK (reqlock);
367 ++nready;
368 reqq_push (&req_queue, req);
369 X_COND_SIGNAL (reqwait);
370 X_UNLOCK (reqlock);
371
372 maybe_start_thread ();
373} 470}
374 471
375static void end_thread (void) 472unsigned int
473eio_nreqs (void)
376{ 474{
377 eio_req *req = calloc (1, sizeof (eio_req)); 475 return etp_nreqs ();
378
379 req->type = EIO_QUIT;
380 req->pri = EIO_PRI_MAX + EIO_PRI_BIAS;
381
382 X_LOCK (reqlock);
383 reqq_push (&req_queue, req);
384 X_COND_SIGNAL (reqwait);
385 X_UNLOCK (reqlock);
386
387 X_LOCK (wrklock);
388 --started;
389 X_UNLOCK (wrklock);
390} 476}
391 477
478unsigned int
479eio_nready (void)
480{
481 return etp_nready ();
482}
483
484unsigned int
485eio_npending (void)
486{
487 return etp_npending ();
488}
489
490unsigned int ecb_cold
491eio_nthreads (void)
492{
493 return etp_nthreads ();
494}
495
496void ecb_cold
392void eio_set_max_poll_time (double nseconds) 497eio_set_max_poll_time (double nseconds)
393{ 498{
394 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
395 max_poll_time = nseconds; 499 etp_set_max_poll_time (nseconds);
396 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
397} 500}
398 501
502void ecb_cold
399void eio_set_max_poll_reqs (unsigned int maxreqs) 503eio_set_max_poll_reqs (unsigned int maxreqs)
400{ 504{
401 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
402 max_poll_reqs = maxreqs; 505 etp_set_max_poll_reqs (maxreqs);
403 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
404} 506}
405 507
508void ecb_cold
406void eio_set_max_idle (unsigned int nthreads) 509eio_set_max_idle (unsigned int nthreads)
407{ 510{
408 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 511 etp_set_max_idle (nthreads);
409 max_idle = nthreads <= 0 ? 1 : nthreads;
410 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
411} 512}
412 513
514void ecb_cold
515eio_set_idle_timeout (unsigned int seconds)
516{
517 etp_set_idle_timeout (seconds);
518}
519
520void ecb_cold
413void eio_set_min_parallel (unsigned int nthreads) 521eio_set_min_parallel (unsigned int nthreads)
414{ 522{
415 if (wanted < nthreads) 523 etp_set_min_parallel (nthreads);
416 wanted = nthreads;
417} 524}
418 525
526void ecb_cold
419void eio_set_max_parallel (unsigned int nthreads) 527eio_set_max_parallel (unsigned int nthreads)
420{ 528{
421 if (wanted > nthreads) 529 etp_set_max_parallel (nthreads);
422 wanted = nthreads;
423
424 while (started > wanted)
425 end_thread ();
426} 530}
427 531
428int eio_poll (void) 532int eio_poll (void)
429{ 533{
430 int maxreqs = max_poll_reqs; 534 return etp_poll ();
431 struct timeval tv_start, tv_now;
432 eio_req *req;
433
434 if (max_poll_time)
435 gettimeofday (&tv_start, 0);
436
437 for (;;)
438 {
439 maybe_start_thread ();
440
441 X_LOCK (reslock);
442 req = reqq_shift (&res_queue);
443
444 if (req)
445 {
446 --npending;
447
448 if (!res_queue.size && done_poll_cb)
449 done_poll_cb ();
450 }
451
452 X_UNLOCK (reslock);
453
454 if (!req)
455 return 0;
456
457 --nreqs;
458
459 if (req->type == EIO_GROUP && req->size)
460 {
461 req->int1 = 1; /* mark request as delayed */
462 continue;
463 }
464 else
465 {
466 int res = eio_finish (req);
467 if (res)
468 return res;
469 }
470
471 if (maxreqs && !--maxreqs)
472 break;
473
474 if (max_poll_time)
475 {
476 gettimeofday (&tv_now, 0);
477
478 if (tvdiff (&tv_start, &tv_now) >= max_poll_time)
479 break;
480 }
481 }
482
483 errno = EAGAIN;
484 return -1;
485} 535}
486 536
487/*****************************************************************************/ 537/*****************************************************************************/
488/* work around various missing functions */ 538/* work around various missing functions */
489 539
490#if !HAVE_PREADWRITE
491# define pread aio_pread
492# define pwrite aio_pwrite
493
494/*
495 * make our pread/pwrite safe against themselves, but not against
496 * normal read/write by using a mutex. slows down execution a lot,
497 * but that's your problem, not mine.
498 */
499static mutex_t preadwritelock = X_MUTEX_INIT;
500
501static ssize_t pread (int fd, void *buf, size_t count, off_t offset)
502{
503 ssize_t res;
504 off_t ooffset;
505
506 X_LOCK (preadwritelock);
507 ooffset = lseek (fd, 0, SEEK_CUR);
508 lseek (fd, offset, SEEK_SET);
509 res = read (fd, buf, count);
510 lseek (fd, ooffset, SEEK_SET);
511 X_UNLOCK (preadwritelock);
512
513 return res;
514}
515
516static ssize_t pwrite (int fd, void *buf, size_t count, off_t offset)
517{
518 ssize_t res;
519 off_t ooffset;
520
521 X_LOCK (preadwritelock);
522 ooffset = lseek (fd, 0, SEEK_CUR);
523 lseek (fd, offset, SEEK_SET);
524 res = write (fd, buf, count);
525 lseek (fd, offset, SEEK_SET);
526 X_UNLOCK (preadwritelock);
527
528 return res;
529}
530#endif
531
532#ifndef HAVE_FUTIMES 540#ifndef HAVE_UTIMES
533 541
542# undef utimes
534# define utimes(path,times) aio_utimes (path, times) 543# define utimes(path,times) eio__utimes (path, times)
535# define futimes(fd,times) aio_futimes (fd, times)
536 544
545static int
537static int aio_utimes (const char *filename, const struct timeval times[2]) 546eio__utimes (const char *filename, const struct timeval times[2])
538{ 547{
539 if (times) 548 if (times)
540 { 549 {
541 struct utimbuf buf; 550 struct utimbuf buf;
542 551
… …
547 } 556 }
548 else 557 else
549 return utime (filename, 0); 558 return utime (filename, 0);
550} 559}
551 560
561#endif
562
563#ifndef HAVE_FUTIMES
564
565# undef futimes
566# define futimes(fd,times) eio__futimes (fd, times)
567
568static int
552static int aio_futimes (int fd, const struct timeval tv[2]) 569eio__futimes (int fd, const struct timeval tv[2])
553{ 570{
554 errno = ENOSYS; 571 errno = ENOSYS;
555 return -1; 572 return -1;
556} 573}
557 574
558#endif 575#endif
559 576
560#if !HAVE_FDATASYNC 577#if !HAVE_FDATASYNC
578# undef fdatasync
561# define fdatasync fsync 579# define fdatasync(fd) fsync (fd)
562#endif 580#endif
581
582static int
583eio__syncfs (int fd)
584{
585 int res;
586
587#if HAVE_SYS_SYNCFS
588 res = (int)syscall (__NR_syncfs, (int)(fd));
589#else
590 res = EIO_ENOSYS ();
591#endif
592
593 if (res < 0 && errno == ENOSYS && fd >= 0)
594 sync ();
595
596 return res;
597}
598
599/* sync_file_range always needs emulation */
600static int
601eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
602{
603#if HAVE_SYNC_FILE_RANGE
604 int res;
605
606 if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
607 || EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
608 || EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
609 {
610 flags = 0
611 | (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
612 | (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
613 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
614 }
615
616 res = sync_file_range (fd, offset, nbytes, flags);
617
618 if (!res || errno != ENOSYS)
619 return res;
620#endif
621
622 /* even though we could play tricks with the flags, it's better to always
623 * call fdatasync, as that matches the expectation of its users best */
624 return fdatasync (fd);
625}
626
627static int
628eio__fallocate (int fd, int mode, off_t offset, size_t len)
629{
630#if HAVE_LINUX_FALLOCATE
631 return fallocate (fd, mode, offset, len);
632#else
633 return EIO_ENOSYS ();
634#endif
635}
563 636
564#if !HAVE_READAHEAD 637#if !HAVE_READAHEAD
638# undef readahead
565# define readahead(fd,offset,count) aio_readahead (fd, offset, count, self) 639# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
566 640
641static eio_ssize_t
567static ssize_t aio_readahead (int fd, off_t offset, size_t count, worker *self) 642eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
568{ 643{
569 size_t todo = count; 644 size_t todo = count;
570 dBUF; 645 dBUF;
571 646
572 while (todo > 0) 647 while (todo > 0)
573 { 648 {
574 size_t len = todo < EIO_BUFSIZE ? todo : EIO_BUFSIZE; 649 size_t len = todo < EIO_BUFSIZE ? todo : EIO_BUFSIZE;
575 650
576 pread (fd, aio_buf, len, offset); 651 pread (fd, eio_buf, len, offset);
577 offset += len; 652 offset += len;
578 todo -= len; 653 todo -= len;
579 } 654 }
580 655
581 errno = 0; 656 FUBd;
657
658 /* linux's readahead basically only fails for EBADF or EINVAL (not mmappable) */
659 /* but not for e.g. EIO or eof, so we also never fail */
582 return count; 660 return 0;
583} 661}
584 662
585#endif
586
587#if !HAVE_READDIR_R
588# define readdir_r aio_readdir_r
589
590static mutex_t readdirlock = X_MUTEX_INIT;
591
592static int readdir_r (DIR *dirp, X_DIRENT *ent, X_DIRENT **res)
593{
594 X_DIRENT *e;
595 int errorno;
596
597 X_LOCK (readdirlock);
598
599 e = readdir (dirp);
600 errorno = errno;
601
602 if (e)
603 {
604 *res = ent;
605 strcpy (ent->d_name, e->d_name);
606 }
607 else
608 *res = 0;
609
610 X_UNLOCK (readdirlock);
611
612 errno = errorno;
613 return e ? 0 : -1;
614}
615#endif 663#endif
616 664
617/* sendfile always needs emulation */ 665/* sendfile always needs emulation */
666static eio_ssize_t
618static ssize_t sendfile_ (int ofd, int ifd, off_t offset, size_t count, worker *self) 667eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
619{ 668{
669 eio_ssize_t written = 0;
620 ssize_t res; 670 eio_ssize_t res;
621 671
622 if (!count) 672 if (!count)
623 return 0; 673 return 0;
624 674
675 for (;;)
676 {
677#ifdef __APPLE__
678# undef HAVE_SENDFILE /* broken, as everything on os x */
679#endif
625#if HAVE_SENDFILE 680#if HAVE_SENDFILE
626# if __linux 681# if __linux
682 off_t soffset = offset;
627 res = sendfile (ofd, ifd, &offset, count); 683 res = sendfile (ofd, ifd, &soffset, count);
628 684
629# elif __freebsd 685# elif __FreeBSD__
630 /* 686 /*
631 * Of course, the freebsd sendfile is a dire hack with no thoughts 687 * Of course, the freebsd sendfile is a dire hack with no thoughts
632 * wasted on making it similar to other I/O functions. 688 * wasted on making it similar to other I/O functions.
633 */ 689 */
634 {
635 off_t sbytes; 690 off_t sbytes;
636 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 691 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
637 692
638 if (res < 0 && sbytes) 693 #if 0 /* according to the manpage, this is correct, but broken behaviour */
639 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 694 /* freebsd' sendfile will return 0 on success */
695 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
696 /* not on e.g. EIO or EPIPE - sounds broken */
697 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
640 res = sbytes; 698 res = sbytes;
641 } 699 #endif
700
701 /* according to source inspection, this is correct, and useful behaviour */
702 if (sbytes)
703 res = sbytes;
704
705# elif defined __APPLE__
706 off_t sbytes = count;
707 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
708
709 /* according to the manpage, sbytes is always valid */
710 if (sbytes)
711 res = sbytes;
642 712
643# elif __hpux 713# elif __hpux
644 res = sendfile (ofd, ifd, offset, count, 0, 0); 714 res = sendfile (ofd, ifd, offset, count, 0, 0);
645 715
646# elif __solaris 716# elif __solaris
647 {
648 struct sendfilevec vec; 717 struct sendfilevec vec;
649 size_t sbytes; 718 size_t sbytes;
650 719
651 vec.sfv_fd = ifd; 720 vec.sfv_fd = ifd;
652 vec.sfv_flag = 0; 721 vec.sfv_flag = 0;
653 vec.sfv_off = offset; 722 vec.sfv_off = offset;
654 vec.sfv_len = count; 723 vec.sfv_len = count;
655 724
656 res = sendfilev (ofd, &vec, 1, &sbytes); 725 res = sendfilev (ofd, &vec, 1, &sbytes);
657 726
658 if (res < 0 && sbytes) 727 if (res < 0 && sbytes)
659 res = sbytes; 728 res = sbytes;
660 }
661 729
662# endif 730# endif
731
732#elif defined (_WIN32) && 0
733 /* does not work, just for documentation of what would need to be done */
734 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
735 /* libeio guarantees that the file offset does not change, and windows */
736 /* has no way to get an independent handle to the same file description */
737 HANDLE h = TO_SOCKET (ifd);
738 SetFilePointer (h, offset, 0, FILE_BEGIN);
739 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
740
663#else 741#else
664 res = -1; 742 res = EIO_ENOSYS ();
665 errno = ENOSYS;
666#endif 743#endif
667 744
745 /* we assume sendfile can copy at least 128mb in one go */
746 if (res <= 128 * 1024 * 1024)
747 {
748 if (res > 0)
749 written += res;
750
751 if (written)
752 return written;
753
754 break;
755 }
756 else
757 {
758 /* if we requested more, then probably the kernel was lazy */
759 written += res;
760 offset += res;
761 count -= res;
762
763 if (!count)
764 return written;
765 }
766 }
767
668 if (res < 0 768 if (res < 0
669 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 769 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
770 /* BSDs */
771#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
772 || errno == ENOTSUP
773#endif
774#ifdef EOPNOTSUPP /* windows */
775 || errno == EOPNOTSUPP /* BSDs */
776#endif
670#if __solaris 777#if __solaris
671 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 778 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
672#endif 779#endif
673 ) 780 )
674 ) 781 )
… …
678 785
679 res = 0; 786 res = 0;
680 787
681 while (count) 788 while (count)
682 { 789 {
683 ssize_t cnt; 790 eio_ssize_t cnt;
684 791
685 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 792 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
686 793
687 if (cnt <= 0) 794 if (cnt <= 0)
688 { 795 {
… …
700 807
701 offset += cnt; 808 offset += cnt;
702 res += cnt; 809 res += cnt;
703 count -= cnt; 810 count -= cnt;
704 } 811 }
812
813 FUBd;
705 } 814 }
706 815
707 return res; 816 return res;
708} 817}
709 818
710/* read a full directory */ 819#ifdef PAGESIZE
711static void scandir_ (eio_req *req, worker *self) 820# define eio_pagesize() PAGESIZE
821#else
822static intptr_t
823eio_pagesize (void)
712{ 824{
713 DIR *dirp; 825 static intptr_t page;
714 union
715 {
716 EIO_STRUCT_DIRENT d;
717 char b [offsetof (EIO_STRUCT_DIRENT, d_name) + NAME_MAX + 1];
718 } *u;
719 EIO_STRUCT_DIRENT *entp;
720 char *name, *names;
721 int memlen = 4096;
722 int memofs = 0;
723 int res = 0;
724 826
725 X_LOCK (wrklock); 827 if (!page)
726 self->dirp = dirp = opendir (req->ptr1); 828 page = sysconf (_SC_PAGESIZE);
727 self->dbuf = u = malloc (sizeof (*u));
728 req->flags |= EIO_FLAG_PTR2_FREE;
729 req->ptr2 = names = malloc (memlen);
730 X_UNLOCK (wrklock);
731 829
732 if (dirp && u && names) 830 return page;
733 for (;;) 831}
832#endif
833
834static void
835eio_page_align (void **addr, size_t *length)
836{
837 intptr_t mask = eio_pagesize () - 1;
838
839 /* round down addr */
840 intptr_t adj = mask & (intptr_t)*addr;
841
842 *addr = (void *)((intptr_t)*addr - adj);
843 *length += adj;
844
845 /* round up length */
846 *length = (*length + mask) & ~mask;
847}
848
849#if !_POSIX_MEMLOCK
850# define eio__mlockall(a) EIO_ENOSYS ()
851#else
852
853static int
854eio__mlockall (int flags)
855{
856 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
857 extern int mallopt (int, int);
858 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
859 #endif
860
861 if (EIO_MCL_CURRENT != MCL_CURRENT
862 || EIO_MCL_FUTURE != MCL_FUTURE)
863 {
864 flags = 0
865 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
866 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
867 }
868
869 return mlockall (flags);
870}
871#endif
872
873#if !_POSIX_MEMLOCK_RANGE
874# define eio__mlock(a,b) EIO_ENOSYS ()
875#else
876
877static int
878eio__mlock (void *addr, size_t length)
879{
880 eio_page_align (&addr, &length);
881
882 return mlock (addr, length);
883}
884
885#endif
886
887#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
888# define eio__msync(a,b,c) EIO_ENOSYS ()
889#else
890
891static int
892eio__msync (void *mem, size_t len, int flags)
893{
894 eio_page_align (&mem, &len);
895
896 if (EIO_MS_ASYNC != MS_SYNC
897 || EIO_MS_INVALIDATE != MS_INVALIDATE
898 || EIO_MS_SYNC != MS_SYNC)
899 {
900 flags = 0
901 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
902 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
903 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
904 }
905
906 return msync (mem, len, flags);
907}
908
909#endif
910
911static int
912eio__mtouch (eio_req *req)
913{
914 void *mem = req->ptr2;
915 size_t len = req->size;
916 int flags = req->int1;
917
918 eio_page_align (&mem, &len);
919
920 {
921 intptr_t addr = (intptr_t)mem;
922 intptr_t end = addr + len;
923 intptr_t page = eio_pagesize ();
924
925 if (addr < end)
926 if (flags & EIO_MT_MODIFY) /* modify */
927 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
928 else
929 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
930 }
931
932 return 0;
933}
934
935/*****************************************************************************/
936/* requests implemented outside eio_execute, because they are so large */
937
938static void
939eio__lseek (eio_req *req)
940{
941 /* this usually gets optimised away completely, or your compiler sucks, */
942 /* or the whence constants really are not 0, 1, 2 */
943 int whence = req->int2 == EIO_SEEK_SET ? SEEK_SET
944 : req->int2 == EIO_SEEK_CUR ? SEEK_CUR
945 : req->int2 == EIO_SEEK_END ? SEEK_END
946 : req->int2;
947
948 req->offs = lseek (req->int1, req->offs, whence);
949 req->result = req->offs == (off_t)-1 ? -1 : 0;
950}
951
952/* result will always end up in tmpbuf, there is always space for adding a 0-byte */
953static int
954eio__realpath (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
955{
956 char *res;
957 const char *rel = path;
958 char *tmp1, *tmp2;
959#if SYMLOOP_MAX > 32
960 int symlinks = SYMLOOP_MAX;
961#else
962 int symlinks = 32;
963#endif
964
965 errno = EINVAL;
966 if (!rel)
967 return -1;
968
969 errno = ENOENT;
970 if (!*rel)
971 return -1;
972
973 res = etp_tmpbuf_get (tmpbuf, PATH_MAX * 3);
974#ifdef _WIN32
975 if (_access (rel, 4) != 0)
976 return -1;
977
978 symlinks = GetFullPathName (rel, PATH_MAX * 3, res, 0);
979
980 errno = ENAMETOOLONG;
981 if (symlinks >= PATH_MAX * 3)
982 return -1;
983
984 errno = EIO;
985 if (symlinks <= 0)
986 return -1;
987
988 return symlinks;
989
990#else
991 tmp1 = res + PATH_MAX;
992 tmp2 = tmp1 + PATH_MAX;
993
994#if 0 /* disabled, the musl way to do things is just too racy */
995#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
996 /* on linux we may be able to ask the kernel */
997 {
998 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
999
1000 if (fd >= 0)
734 { 1001 {
1002 sprintf (tmp1, "/proc/self/fd/%d", fd);
1003 req->result = readlink (tmp1, res, PATH_MAX);
1004 /* here we should probably stat the open file and the disk file, to make sure they still match */
1005 close (fd);
1006
1007 if (req->result > 0)
1008 goto done;
1009 }
1010 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1011 return -1;
1012 }
1013#endif
1014#endif
1015
1016 if (*rel != '/')
1017 {
1018 int len;
1019
735 errno = 0; 1020 errno = ENOENT;
736 readdir_r (dirp, &u->d, &entp); 1021 if (wd == EIO_INVALID_WD)
1022 return -1;
1023
1024 if (wd == EIO_CWD)
1025 {
1026 if (!getcwd (res, PATH_MAX))
1027 return -1;
737 1028
738 if (!entp) 1029 len = strlen (res);
1030 }
1031 else
1032 memcpy (res, wd->str, len = wd->len);
1033
1034 if (res [1]) /* only use if not / */
1035 res += len;
1036 }
1037
1038 while (*rel)
1039 {
1040 eio_ssize_t len, linklen;
1041 const char *beg = rel;
1042
1043 while (*rel && *rel != '/')
1044 ++rel;
1045
1046 len = rel - beg;
1047
1048 if (!len) /* skip slashes */
1049 {
739 break; 1050 ++rel;
1051 continue;
1052 }
740 1053
741 name = entp->d_name; 1054 if (beg [0] == '.')
1055 {
1056 if (len == 1)
1057 continue; /* . - nop */
742 1058
743 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1059 if (beg [1] == '.' && len == 2)
1060 {
1061 /* .. - back up one component, if possible */
1062
1063 while (res != tmpbuf->ptr)
1064 if (*--res == '/')
1065 break;
1066
1067 continue;
1068 }
1069 }
1070
1071 errno = ENAMETOOLONG;
1072 if (res + 1 + len + 1 >= tmp1)
1073 return -1;
1074
1075 /* copy one component */
1076 *res = '/';
1077 memcpy (res + 1, beg, len);
1078
1079 /* zero-terminate, for readlink */
1080 res [len + 1] = 0;
1081
1082 /* now check if it's a symlink */
1083 linklen = readlink (tmpbuf->ptr, tmp1, PATH_MAX);
1084
1085 if (linklen < 0)
744 { 1086 {
745 int len = strlen (name) + 1; 1087 if (errno != EINVAL)
1088 return -1;
746 1089
747 res++; 1090 /* it's a normal directory. hopefully */
748
749 while (memofs + len > memlen)
750 {
751 memlen *= 2;
752 X_LOCK (wrklock);
753 req->ptr2 = names = realloc (names, memlen);
754 X_UNLOCK (wrklock);
755
756 if (!names)
757 break;
758 }
759
760 memcpy (names + memofs, name, len);
761 memofs += len; 1091 res += len + 1;
762 }
763 }
764
765 if (errno)
766 res = -1;
767
768 req->result = res;
769}
770
771/*****************************************************************************/
772
773X_THREAD_PROC (eio_proc)
774{
775 eio_req *req;
776 struct timespec ts;
777 worker *self = (worker *)thr_arg;
778
779 /* try to distribute timeouts somewhat randomly */
780 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
781
782 for (;;)
783 {
784 ts.tv_sec = time (0) + IDLE_TIMEOUT;
785
786 X_LOCK (reqlock);
787
788 for (;;)
789 {
790 self->req = req = reqq_shift (&req_queue);
791
792 if (req)
793 break;
794
795 ++idle;
796
797 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
798 {
799 if (idle > max_idle)
800 {
801 --idle;
802 X_UNLOCK (reqlock);
803 X_LOCK (wrklock);
804 --started;
805 X_UNLOCK (wrklock);
806 goto quit;
807 }
808
809 /* we are allowed to idle, so do so without any timeout */
810 X_COND_WAIT (reqwait, reqlock);
811 ts.tv_sec = time (0) + IDLE_TIMEOUT;
812 }
813
814 --idle;
815 }
816
817 --nready;
818
819 X_UNLOCK (reqlock);
820
821 errno = 0; /* strictly unnecessary */
822
823 if (!EIO_CANCELLED (req))
824 switch (req->type)
825 {
826 case EIO_READ: req->result = req->offs >= 0
827 ? pread (req->int1, req->ptr2, req->size, req->offs)
828 : read (req->int1, req->ptr2, req->size); break;
829 case EIO_WRITE: req->result = req->offs >= 0
830 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
831 : write (req->int1, req->ptr2, req->size); break;
832
833 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
834 case EIO_SENDFILE: req->result = sendfile_ (req->int1, req->int2, req->offs, req->size, self); break;
835
836 case EIO_STAT: req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
837 case EIO_LSTAT: req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
838 case EIO_FSTAT: req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
839
840 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
841 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
842 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
843 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
844 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
845 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
846
847 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
848 case EIO_CLOSE: req->result = close (req->int1); break;
849 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
850 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
851 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
852 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
853 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
854 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
855 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
856 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
857 case EIO_READLINK: req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
858
859 case EIO_SYNC: req->result = 0; sync (); break;
860 case EIO_FSYNC: req->result = fsync (req->int1); break;
861 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
862
863 case EIO_READDIR: scandir_ (req, self); break;
864
865 case EIO_BUSY:
866#ifdef _WIN32
867 Sleep (req->nv1 * 1000.);
868#else
869 {
870 struct timeval tv;
871
872 tv.tv_sec = req->nv1;
873 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;
874
875 req->result = select (0, 0, 0, 0, &tv);
876 }
877#endif
878 break;
879
880 case EIO_UTIME:
881 case EIO_FUTIME:
882 {
883 struct timeval tv[2];
884 struct timeval *times;
885
886 if (req->nv1 != -1. || req->nv2 != -1.)
887 {
888 tv[0].tv_sec = req->nv1;
889 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
890 tv[1].tv_sec = req->nv2;
891 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
892
893 times = tv;
894 }
895 else
896 times = 0;
897
898
899 req->result = req->type == EIO_FUTIME
900 ? futimes (req->int1, times)
901 : utimes (req->ptr1, times);
902 }
903
904 case EIO_GROUP:
905 case EIO_NOP:
906 break;
907
908 case EIO_QUIT:
909 goto quit;
910
911 default:
912 req->result = -1;
913 break;
914 }
915
916 req->errorno = errno;
917
918 X_LOCK (reslock);
919
920 ++npending;
921
922 if (!reqq_push (&res_queue, req) && want_poll_cb)
923 want_poll_cb ();
924
925 self->req = 0;
926 worker_clear (self);
927
928 X_UNLOCK (reslock);
929 }
930
931quit:
932 X_LOCK (wrklock);
933 worker_free (self);
934 X_UNLOCK (wrklock);
935
936 return 0;
937}
938
939/*****************************************************************************/
940
941static void eio_atfork_prepare (void)
942{
943 X_LOCK (wrklock);
944 X_LOCK (reqlock);
945 X_LOCK (reslock);
946#if !HAVE_PREADWRITE
947 X_LOCK (preadwritelock);
948#endif
949#if !HAVE_READDIR_R
950 X_LOCK (readdirlock);
951#endif
952}
953
954static void eio_atfork_parent (void)
955{
956#if !HAVE_READDIR_R
957 X_UNLOCK (readdirlock);
958#endif
959#if !HAVE_PREADWRITE
960 X_UNLOCK (preadwritelock);
961#endif
962 X_UNLOCK (reslock);
963 X_UNLOCK (reqlock);
964 X_UNLOCK (wrklock);
965}
966
967static void eio_atfork_child (void)
968{
969 eio_req *prv;
970
971 while (prv = reqq_shift (&req_queue))
972 eio_destroy (prv);
973
974 while (prv = reqq_shift (&res_queue))
975 eio_destroy (prv);
976
977 while (wrk_first.next != &wrk_first)
978 {
979 worker *wrk = wrk_first.next;
980
981 if (wrk->req)
982 eio_destroy (wrk->req);
983
984 worker_clear (wrk);
985 worker_free (wrk);
986 }
987
988 started = 0;
989 idle = 0;
990 nreqs = 0;
991 nready = 0;
992 npending = 0;
993
994 eio_atfork_parent ();
995}
996
997int eio_init (void (*want_poll)(void), void (*done_poll)(void))
998{
999 want_poll_cb = want_poll;
1000 done_poll_cb = done_poll;
1001
1002#ifdef _WIN32
1003 X_MUTEX_CHECK (wrklock);
1004 X_MUTEX_CHECK (reslock);
1005 X_MUTEX_CHECK (reqlock);
1006 X_MUTEX_CHECK (reqwait);
1007 X_MUTEX_CHECK (preadwritelock);
1008 X_MUTEX_CHECK (readdirlock);
1009
1010 X_COND_CHECK (reqwait);
1011#endif
1012
1013 X_THREAD_ATFORK (eio_atfork_prepare, eio_atfork_parent, eio_atfork_child);
1014}
1015
1016#if 0
1017
1018eio_req *eio_fsync (int fd, eio_cb cb);
1019eio_req *eio_fdatasync (int fd, eio_cb cb);
1020eio_req *eio_dupclose (int fd, eio_cb cb);
1021eio_req *eio_readahead (int fd, off_t offset, size_t length, eio_cb cb);
1022eio_req *eio_read (int fd, off_t offs, size_t length, char *data, eio_cb cb);
1023eio_req *eio_write (int fd, off_t offs, size_t length, char *data, eio_cb cb);
1024eio_req *eio_fstat (int fd, eio_cb cb); /* stat buffer=ptr2 allocates dynamically */
1025eio_req *eio_futime (int fd, double atime, double mtime, eio_cb cb);
1026eio_req *eio_ftruncate (int fd, off_t offset, eio_cb cb);
1027eio_req *eio_fchmod (int fd, mode_t mode, eio_cb cb);
1028eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, eio_cb cb);
1029eio_req *eio_dup2 (int fd, int fd2, eio_cb cb);
1030eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, eio_cb cb);
1031eio_req *eio_open (const char *path, int flags, mode_t mode, eio_cb cb);
1032eio_req *eio_readlink (const char *path, eio_cb cb); /* result=ptr2 allocated dynamically */
1033eio_req *eio_stat (const char *path, eio_cb cb); /* stat buffer=ptr2 allocates dynamically */
1034eio_req *eio_lstat (const char *path, eio_cb cb); /* stat buffer=ptr2 allocates dynamically */
1035eio_req *eio_utime (const char *path, double atime, double mtime, eio_cb cb);
1036eio_req *eio_truncate (const char *path, off_t offset, eio_cb cb);
1037eio_req *eio_chmod (const char *path, mode_t mode, eio_cb cb);
1038eio_req *eio_mkdir (const char *path, mode_t mode, eio_cb cb);
1039eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, eio_cb cb);
1040eio_req *eio_unlink (const char *path, eio_cb cb);
1041eio_req *eio_rmdir (const char *path, eio_cb cb);
1042eio_req *eio_readdir (const char *path, eio_cb cb); /* result=ptr2 allocated dynamically */
1043eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, eio_cb cb);
1044eio_req *eio_busy (double delay, eio_cb cb); /* ties a thread for this long, simulating busyness */
1045eio_req *eio_nop (eio_cb cb); /* does nothing except go through the whole process */
1046void
1047aio_open (SV8 *pathname, int flags, int mode, SV *callback=&PL_sv_undef)
1048 PROTOTYPE: $$$;$
1049 PPCODE:
1050{
1051 dREQ;
1052
1053 req->type = EIO_OPEN;
1054 req->sv1 = newSVsv (pathname);
1055 req->ptr1 = SvPVbyte_nolen (req->sv1);
1056 req->int1 = flags;
1057 req->int2 = mode;
1058
1059 EIO_SEND;
1060}
1061
1062void
1063aio_fsync (SV *fh, SV *callback=&PL_sv_undef)
1064 PROTOTYPE: $;$
1065 ALIAS:
1066 aio_fsync = EIO_FSYNC
1067 aio_fdatasync = EIO_FDATASYNC
1068 PPCODE:
1069{
1070 dREQ;
1071
1072 req->type = ix;
1073 req->sv1 = newSVsv (fh);
1074 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1075
1076 EIO_SEND (req);
1077}
1078
1079void
1080aio_close (SV *fh, SV *callback=&PL_sv_undef)
1081 PROTOTYPE: $;$
1082 PPCODE:
1083{
1084 dREQ;
1085
1086 req->type = EIO_CLOSE;
1087 req->sv1 = newSVsv (fh);
1088 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1089
1090 EIO_SEND (req);
1091}
1092
1093void
1094aio_read (SV *fh, SV *offset, SV *length, SV8 *data, IV dataoffset, SV *callback=&PL_sv_undef)
1095 ALIAS:
1096 aio_read = EIO_READ
1097 aio_write = EIO_WRITE
1098 PROTOTYPE: $$$$$;$
1099 PPCODE:
1100{
1101 STRLEN svlen;
1102 char *svptr = SvPVbyte (data, svlen);
1103 UV len = SvUV (length);
1104
1105 SvUPGRADE (data, SVt_PV);
1106 SvPOK_on (data);
1107
1108 if (dataoffset < 0)
1109 dataoffset += svlen;
1110
1111 if (dataoffset < 0 || dataoffset > svlen)
1112 croak ("dataoffset outside of data scalar");
1113
1114 if (ix == EIO_WRITE)
1115 {
1116 /* write: check length and adjust. */
1117 if (!SvOK (length) || len + dataoffset > svlen)
1118 len = svlen - dataoffset;
1119 } 1092 }
1120 else 1093 else
1121 { 1094 {
1122 /* read: grow scalar as necessary */ 1095 /* yay, it was a symlink - build new path in tmp2 */
1123 svptr = SvGROW (data, len + dataoffset + 1); 1096 int rellen = strlen (rel);
1097
1098 errno = ENAMETOOLONG;
1099 if (linklen + 1 + rellen >= PATH_MAX)
1100 return -1;
1101
1102 errno = ELOOP;
1103 if (!--symlinks)
1104 return -1;
1105
1106 if (*tmp1 == '/')
1107 res = tmpbuf->ptr; /* symlink resolves to an absolute path */
1108
1109 /* we need to be careful, as rel might point into tmp2 already */
1110 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1111 tmp2 [linklen] = '/';
1112 memcpy (tmp2, tmp1, linklen);
1113
1114 rel = tmp2;
1124 } 1115 }
1116 }
1125 1117
1118 /* special case for the lone root path */
1119 if (res == tmpbuf->ptr)
1120 *res++ = '/';
1121
1122 return res - (char *)tmpbuf->ptr;
1123#endif
1124}
1125
1126static signed char
1127eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1128{
1129 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1130 : a->inode < b->inode ? -1
1131 : a->inode > b->inode ? 1
1132 : 0;
1133}
1134
1135#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1136
1137#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1138#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1139
1140static void
1141eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1142{
1143 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1144 unsigned char *bit = bits;
1145
1146 assert (CHAR_BIT == 8);
1147 assert (sizeof (eio_dirent) * 8 < 256);
1148 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1149 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1150
1151 if (size <= EIO_SORT_FAST)
1152 return;
1153
1154 /* first prepare an array of bits to test in our radix sort */
1155 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1156 /* inode_bits must contain all inodes ORed together */
1157 /* which is used to skip bits that are 0 everywhere, which is very common */
1158 {
1159 eio_ino_t endianness;
1160 int i, j;
1161
1162 /* we store the byte offset of byte n into byte n of "endianness" */
1163 for (i = 0; i < sizeof (eio_ino_t); ++i)
1164 ((unsigned char *)&endianness)[i] = i;
1165
1166 *bit++ = 0;
1167
1168 for (i = 0; i < sizeof (eio_ino_t); ++i)
1169 {
1170 /* shifting off the byte offsets out of "endianness" */
1171 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1172 endianness >>= 8;
1173
1174 for (j = 0; j < 8; ++j)
1175 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1176 *bit++ = offs + j;
1177 }
1178
1179 for (j = 0; j < 8; ++j)
1180 if (score_bits & (1 << j))
1181 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1182 }
1183
1184 /* now actually do the sorting (a variant of MSD radix sort) */
1185 {
1186 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1187 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1188 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1189 int stk_idx = 0;
1190
1191 base_stk [stk_idx] = dents;
1192 end_stk [stk_idx] = dents + size;
1193 bit_stk [stk_idx] = bit - 1;
1194
1195 do
1196 {
1197 base = base_stk [stk_idx];
1198 end = end_stk [stk_idx];
1199 bit = bit_stk [stk_idx];
1200
1201 for (;;)
1202 {
1203 unsigned char O = *bit >> 3;
1204 unsigned char M = 1 << (*bit & 7);
1205
1206 eio_dirent *a = base;
1207 eio_dirent *b = end;
1208
1209 if (b - a < EIO_SORT_CUTOFF)
1210 break;
1211
1212 /* now bit-partition the array on the bit */
1213 /* this ugly asymmetric loop seems to perform much better than typical */
1214 /* partition algos found in the literature */
1215 do
1216 if (!(((unsigned char *)a)[O] & M))
1217 ++a;
1218 else if (!(((unsigned char *)--b)[O] & M))
1219 {
1220 eio_dirent tmp = *a; *a = *b; *b = tmp;
1221 ++a;
1222 }
1223 while (b > a);
1224
1225 /* next bit, or stop, if no bits left in this path */
1226 if (!*--bit)
1227 break;
1228
1229 base_stk [stk_idx] = a;
1230 end_stk [stk_idx] = end;
1231 bit_stk [stk_idx] = bit;
1232 ++stk_idx;
1233
1234 end = a;
1235 }
1236 }
1237 while (stk_idx--);
1238 }
1239}
1240
1241static void
1242eio_dent_insertion_sort (eio_dirent *dents, int size)
1243{
1244 /* first move the smallest element to the front, to act as a sentinel */
1245 {
1246 int i;
1247 eio_dirent *min = dents;
1248
1249 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1250 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1251 if (EIO_DENT_CMP (dents [i], <, *min))
1252 min = &dents [i];
1253
1254 /* swap elements 0 and j (minimum) */
1255 {
1256 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1257 }
1258 }
1259
1260 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1261 {
1262 eio_dirent *i, *j;
1263
1264 for (i = dents + 1; i < dents + size; ++i)
1265 {
1266 eio_dirent value = *i;
1267
1268 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1269 j [1] = j [0];
1270
1271 j [1] = value;
1272 }
1273 }
1274}
1275
1276static void
1277eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1278{
1279 if (size <= 1)
1280 return; /* our insertion sort relies on size > 0 */
1281
1282 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1283 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1284 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1285
1286 /* use an insertion sort at the end, or for small arrays, */
1287 /* as insertion sort is more efficient for small partitions */
1288 eio_dent_insertion_sort (dents, size);
1289}
1290
1291/* read a full directory */
1292static void
1293eio__scandir (eio_req *req, etp_worker *self)
1294{
1295 char *name, *names;
1296 int namesalloc = 4096 - sizeof (void *) * 4;
1297 int namesoffs = 0;
1298 int flags = req->int1;
1299 eio_dirent *dents = 0;
1300 int dentalloc = 128;
1301 int dentoffs = 0;
1302 eio_ino_t inode_bits = 0;
1303#ifdef _WIN32
1304 HANDLE dirp;
1305 WIN32_FIND_DATA entp;
1306#else
1307 DIR *dirp;
1308 EIO_STRUCT_DIRENT *entp;
1309#endif
1310
1311 req->result = -1;
1312
1313 if (!(flags & EIO_READDIR_DENTS))
1314 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1315
1316#ifdef _WIN32
1317 {
1318 int len = strlen ((const char *)req->ptr1);
1319 char *path = malloc (MAX_PATH);
1320 const char *fmt;
1321 const char *reqpath = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1322
1323 if (!len)
1324 fmt = "./*";
1325 else if (reqpath[len - 1] == '/' || reqpath[len - 1] == '\\')
1326 fmt = "%s*";
1327 else
1328 fmt = "%s/*";
1329
1330 _snprintf (path, MAX_PATH, fmt, reqpath);
1331 dirp = FindFirstFile (path, &entp);
1332 free (path);
1333
1334 if (dirp == INVALID_HANDLE_VALUE)
1335 {
1336 /* should steal _dosmaperr */
1337 switch (GetLastError ())
1338 {
1339 case ERROR_FILE_NOT_FOUND:
1340 req->result = 0;
1341 break;
1342
1343 case ERROR_INVALID_NAME:
1344 case ERROR_PATH_NOT_FOUND:
1345 case ERROR_NO_MORE_FILES:
1346 errno = ENOENT;
1347 break;
1348
1349 case ERROR_NOT_ENOUGH_MEMORY:
1350 errno = ENOMEM;
1351 break;
1352
1353 default:
1354 errno = EINVAL;
1355 break;
1356 }
1357
1358 return;
1359 }
1360 }
1361#else
1362 #if HAVE_AT
1363 if (req->wd)
1364 {
1365 int fd = openat (WD2FD (req->wd), req->ptr1, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1366
1126 if (len < 0) 1367 if (fd < 0)
1127 croak ("length must not be negative"); 1368 return;
1128 1369
1370 dirp = fdopendir (fd);
1371
1372 if (!dirp)
1373 close (fd);
1374 }
1375 else
1376 dirp = opendir (req->ptr1);
1377 #else
1378 dirp = opendir (wd_expand (&self->tmpbuf, req->wd, req->ptr1));
1379 #endif
1380
1381 if (!dirp)
1382 return;
1383#endif
1384
1385 if (req->flags & EIO_FLAG_PTR1_FREE)
1386 free (req->ptr1);
1387
1388 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1389 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1390 req->ptr2 = names = malloc (namesalloc);
1391
1392 if (!names || (flags && !dents))
1393 return;
1394
1395 for (;;)
1396 {
1397 int done;
1398
1399#ifdef _WIN32
1400 done = !dirp;
1401#else
1402 errno = 0;
1403 entp = readdir (dirp);
1404 done = !entp;
1405#endif
1406
1407 if (done)
1129 { 1408 {
1130 dREQ; 1409#ifndef _WIN32
1410 int old_errno = errno;
1411 closedir (dirp);
1412 errno = old_errno;
1131 1413
1132 req->type = ix; 1414 if (errno)
1133 req->sv1 = newSVsv (fh); 1415 break;
1134 req->int1 = PerlIO_fileno (ix == EIO_READ ? IoIFP (sv_2io (fh)) 1416#endif
1135 : IoOFP (sv_2io (fh)));
1136 req->offs = SvOK (offset) ? SvVAL64 (offset) : -1;
1137 req->size = len;
1138 req->sv2 = SvREFCNT_inc (data);
1139 req->ptr2 = (char *)svptr + dataoffset;
1140 req->stroffset = dataoffset;
1141 1417
1142 if (!SvREADONLY (data)) 1418 /* sort etc. */
1419 req->int1 = flags;
1420 req->result = dentoffs;
1421
1422 if (flags & EIO_READDIR_STAT_ORDER)
1423 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1424 else if (flags & EIO_READDIR_DIRS_FIRST)
1425 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1426 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1427 else
1428 {
1429 /* in this case, all is known, and we just put dirs first and sort them */
1430 eio_dirent *oth = dents + dentoffs;
1431 eio_dirent *dir = dents;
1432
1433 /* now partition dirs to the front, and non-dirs to the back */
1434 /* by walking from both sides and swapping if necessary */
1435 while (oth > dir)
1436 {
1437 if (dir->type == EIO_DT_DIR)
1438 ++dir;
1439 else if ((--oth)->type == EIO_DT_DIR)
1440 {
1441 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1442
1443 ++dir;
1444 }
1445 }
1446
1447 /* now sort the dirs only (dirs all have the same score) */
1448 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1449 }
1450
1451 break;
1452 }
1453
1454 /* now add the entry to our list(s) */
1455 name = D_NAME (entp);
1456
1457 /* skip . and .. entries */
1458 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1459 {
1460 int len = D_NAMLEN (entp) + 1;
1461
1462 while (ecb_expect_false (namesoffs + len > namesalloc))
1143 { 1463 {
1144 SvREADONLY_on (data); 1464 namesalloc *= 2;
1145 req->flags |= FLAG_SV2_RO_OFF; 1465 req->ptr2 = names = realloc (names, namesalloc);
1466
1467 if (!names)
1468 break;
1146 } 1469 }
1147 1470
1148 EIO_SEND; 1471 memcpy (names + namesoffs, name, len);
1472
1473 if (dents)
1474 {
1475 struct eio_dirent *ent;
1476
1477 if (ecb_expect_false (dentoffs == dentalloc))
1478 {
1479 dentalloc *= 2;
1480 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1481
1482 if (!dents)
1483 break;
1484 }
1485
1486 ent = dents + dentoffs;
1487
1488 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1489 ent->namelen = len - 1;
1490 ent->inode = D_INO (entp);
1491
1492 inode_bits |= ent->inode;
1493
1494 switch (D_TYPE (entp))
1495 {
1496 default:
1497 ent->type = EIO_DT_UNKNOWN;
1498 flags |= EIO_READDIR_FOUND_UNKNOWN;
1499 break;
1500
1501 #ifdef DT_FIFO
1502 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1503 #endif
1504 #ifdef DT_CHR
1505 case DT_CHR: ent->type = EIO_DT_CHR; break;
1506 #endif
1507 #ifdef DT_MPC
1508 case DT_MPC: ent->type = EIO_DT_MPC; break;
1509 #endif
1510 #ifdef DT_DIR
1511 case DT_DIR: ent->type = EIO_DT_DIR; break;
1512 #endif
1513 #ifdef DT_NAM
1514 case DT_NAM: ent->type = EIO_DT_NAM; break;
1515 #endif
1516 #ifdef DT_BLK
1517 case DT_BLK: ent->type = EIO_DT_BLK; break;
1518 #endif
1519 #ifdef DT_MPB
1520 case DT_MPB: ent->type = EIO_DT_MPB; break;
1521 #endif
1522 #ifdef DT_REG
1523 case DT_REG: ent->type = EIO_DT_REG; break;
1524 #endif
1525 #ifdef DT_NWK
1526 case DT_NWK: ent->type = EIO_DT_NWK; break;
1527 #endif
1528 #ifdef DT_CMP
1529 case DT_CMP: ent->type = EIO_DT_CMP; break;
1530 #endif
1531 #ifdef DT_LNK
1532 case DT_LNK: ent->type = EIO_DT_LNK; break;
1533 #endif
1534 #ifdef DT_SOCK
1535 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1536 #endif
1537 #ifdef DT_DOOR
1538 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1539 #endif
1540 #ifdef DT_WHT
1541 case DT_WHT: ent->type = EIO_DT_WHT; break;
1542 #endif
1543 }
1544
1545 ent->score = 7;
1546
1547 if (flags & EIO_READDIR_DIRS_FIRST)
1548 {
1549 if (ent->type == EIO_DT_UNKNOWN)
1550 {
1551 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1552 ent->score = 1;
1553 else if (!strchr (name, '.')) /* absence of dots indicate likely dirs */
1554 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1555 }
1556 else if (ent->type == EIO_DT_DIR)
1557 ent->score = 0;
1558 }
1559 }
1560
1561 namesoffs += len;
1562 ++dentoffs;
1149 } 1563 }
1564
1565 if (EIO_CANCELLED (req))
1566 {
1567 errno = ECANCELED;
1568 break;
1569 }
1570
1571#ifdef _WIN32
1572 if (!FindNextFile (dirp, &entp))
1573 {
1574 FindClose (dirp);
1575 dirp = 0;
1576 }
1577#endif
1578 }
1579}
1580
1581/*****************************************************************************/
1582/* working directory stuff */
1583/* various deficiencies in the posix 2008 api force us to */
1584/* keep the absolute path in string form at all times */
1585/* fuck yeah. */
1586
1587#if !HAVE_AT
1588
1589/* a bit like realpath, but usually faster because it doesn'T have to return */
1590/* an absolute or canonical path */
1591static const char *
1592wd_expand (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
1593{
1594 if (!wd || *path == '/')
1595 return path;
1596
1597 if (path [0] == '.' && !path [1])
1598 return wd->str;
1599
1600 {
1601 int l1 = wd->len;
1602 int l2 = strlen (path);
1603
1604 char *res = etp_tmpbuf_get (tmpbuf, l1 + l2 + 2);
1605
1606 memcpy (res, wd->str, l1);
1607 res [l1] = '/';
1608 memcpy (res + l1 + 1, path, l2 + 1);
1609
1610 return res;
1611 }
1612}
1613
1614#endif
1615
1616static eio_wd
1617eio__wd_open_sync (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
1618{
1619 int fd;
1620 eio_wd res;
1621 int len = eio__realpath (tmpbuf, wd, path);
1622
1623 if (len < 0)
1624 return EIO_INVALID_WD;
1625
1626#if HAVE_AT
1627 fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1628
1629 if (fd < 0)
1630 return EIO_INVALID_WD;
1631#endif
1632
1633 res = malloc (sizeof (*res) + len); /* one extra 0-byte */
1634
1635#if HAVE_AT
1636 res->fd = fd;
1637#endif
1638
1639 res->len = len;
1640 memcpy (res->str, tmpbuf->ptr, len);
1641 res->str [len] = 0;
1642
1643 return res;
1644}
1645
1646eio_wd
1647eio_wd_open_sync (eio_wd wd, const char *path)
1648{
1649 struct etp_tmpbuf tmpbuf = { };
1650 wd = eio__wd_open_sync (&tmpbuf, wd, path);
1651 free (tmpbuf.ptr);
1652
1653 return wd;
1150} 1654}
1151 1655
1152void 1656void
1153aio_readlink (SV8 *path, SV *callback=&PL_sv_undef) 1657eio_wd_close_sync (eio_wd wd)
1154 PROTOTYPE: $$;$
1155 PPCODE:
1156{ 1658{
1157 SV *data; 1659 if (wd != EIO_INVALID_WD && wd != EIO_CWD)
1158 dREQ; 1660 {
1159 1661 #if HAVE_AT
1160 data = newSV (NAME_MAX); 1662 close (wd->fd);
1161 SvPOK_on (data); 1663 #endif
1162 1664 free (wd);
1163 req->type = EIO_READLINK; 1665 }
1164 req->sv1 = newSVsv (path);
1165 req->ptr1 = SvPVbyte_nolen (req->sv1);
1166 req->sv2 = data;
1167 req->ptr2 = SvPVbyte_nolen (data);
1168
1169 EIO_SEND;
1170} 1666}
1667
1668#if HAVE_AT
1669
1670/* they forgot these */
1671
1672static int
1673eio__truncateat (int dirfd, const char *path, off_t length)
1674{
1675 int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
1676 int res;
1677
1678 if (fd < 0)
1679 return fd;
1680
1681 res = ftruncate (fd, length);
1682 close (fd);
1683 return res;
1684}
1685
1686static int
1687eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
1688{
1689 int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
1690 int res;
1691
1692 if (fd < 0)
1693 return fd;
1694
1695 res = fstatvfs (fd, buf);
1696 close (fd);
1697 return res;
1698
1699}
1700
1701#endif
1702
1703/*****************************************************************************/
1704
1705#define ALLOC(len) \
1706 if (!req->ptr2) \
1707 { \
1708 X_LOCK (wrklock); \
1709 req->flags |= EIO_FLAG_PTR2_FREE; \
1710 X_UNLOCK (wrklock); \
1711 req->ptr2 = malloc (len); \
1712 if (!req->ptr2) \
1713 { \
1714 errno = ENOMEM; \
1715 req->result = -1; \
1716 break; \
1717 } \
1718 }
1719
1720static void ecb_noinline ecb_cold
1721etp_proc_init (void)
1722{
1723#if HAVE_PRCTL_SET_NAME
1724 /* provide a more sensible "thread name" */
1725 char name[16 + 1];
1726 const int namelen = sizeof (name) - 1;
1727 int len;
1728
1729 prctl (PR_GET_NAME, (unsigned long)name, 0, 0, 0);
1730 name [namelen] = 0;
1731 len = strlen (name);
1732 strcpy (name + (len <= namelen - 4 ? len : namelen - 4), "/eio");
1733 prctl (PR_SET_NAME, (unsigned long)name, 0, 0, 0);
1734#endif
1735}
1736
1737/* TODO: move somehow to etp.c */
1738X_THREAD_PROC (etp_proc)
1739{
1740 ETP_REQ *req;
1741 struct timespec ts;
1742 etp_worker *self = (etp_worker *)thr_arg;
1743
1744 etp_proc_init ();
1745
1746 /* try to distribute timeouts somewhat evenly */
1747 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1748
1749 for (;;)
1750 {
1751 ts.tv_sec = 0;
1752
1753 X_LOCK (reqlock);
1754
1755 for (;;)
1756 {
1757 req = reqq_shift (&req_queue);
1758
1759 if (req)
1760 break;
1761
1762 if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
1763 {
1764 X_UNLOCK (reqlock);
1765 X_LOCK (wrklock);
1766 --started;
1767 X_UNLOCK (wrklock);
1768 goto quit;
1769 }
1770
1771 ++idle;
1772
1773 if (idle <= max_idle)
1774 /* we are allowed to idle, so do so without any timeout */
1775 X_COND_WAIT (reqwait, reqlock);
1776 else
1777 {
1778 /* initialise timeout once */
1779 if (!ts.tv_sec)
1780 ts.tv_sec = time (0) + idle_timeout;
1781
1782 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
1783 ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1784 }
1785
1786 --idle;
1787 }
1788
1789 --nready;
1790
1791 X_UNLOCK (reqlock);
1792
1793 if (req->type == ETP_TYPE_QUIT)
1794 goto quit;
1795
1796 ETP_EXECUTE (self, req);
1797
1798 X_LOCK (reslock);
1799
1800 ++npending;
1801
1802 if (!reqq_push (&res_queue, req) && want_poll_cb)
1803 want_poll_cb ();
1804
1805 etp_worker_clear (self);
1806
1807 X_UNLOCK (reslock);
1808 }
1809
1810quit:
1811 free (req);
1812
1813 X_LOCK (wrklock);
1814 etp_worker_free (self);
1815 X_UNLOCK (wrklock);
1816
1817 return 0;
1818}
1819
1820/*****************************************************************************/
1821
1822int ecb_cold
1823eio_init (void (*want_poll)(void), void (*done_poll)(void))
1824{
1825 return etp_init (want_poll, done_poll);
1826}
1827
1828ecb_inline void
1829eio_api_destroy (eio_req *req)
1830{
1831 free (req);
1832}
1833
1834#define REQ(rtype) \
1835 eio_req *req; \
1836 \
1837 req = (eio_req *)calloc (1, sizeof *req); \
1838 if (!req) \
1839 return 0; \
1840 \
1841 req->type = rtype; \
1842 req->pri = pri; \
1843 req->finish = cb; \
1844 req->data = data; \
1845 req->destroy = eio_api_destroy;
1846
1847#define SEND eio_submit (req); return req
1848
1849#define PATH \
1850 req->flags |= EIO_FLAG_PTR1_FREE; \
1851 req->ptr1 = strdup (path); \
1852 if (!req->ptr1) \
1853 { \
1854 eio_api_destroy (req); \
1855 return 0; \
1856 }
1857
1858#define SINGLEDOT(ptr) (0[(char *)(ptr)] == '.' && !1[(char *)(ptr)])
1859
1860static void
1861eio_execute (etp_worker *self, eio_req *req)
1862{
1863#if HAVE_AT
1864 int dirfd;
1865#else
1866 const char *path;
1867#endif
1868
1869 if (ecb_expect_false (EIO_CANCELLED (req)))
1870 {
1871 req->result = -1;
1872 req->errorno = ECANCELED;
1873 return;
1874 }
1875
1876 if (ecb_expect_false (req->wd == EIO_INVALID_WD))
1877 {
1878 req->result = -1;
1879 req->errorno = ENOENT;
1880 return;
1881 }
1882
1883 if (req->type >= EIO_OPEN)
1884 {
1885 #if HAVE_AT
1886 dirfd = WD2FD (req->wd);
1887 #else
1888 path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1889 #endif
1890 }
1891
1892 switch (req->type)
1893 {
1894 case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
1895 req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
1896 break;
1897 case EIO_WD_CLOSE: req->result = 0;
1898 eio_wd_close_sync (req->wd); break;
1899
1900 case EIO_SEEK: eio__lseek (req); break;
1901 case EIO_READ: ALLOC (req->size);
1902 req->result = req->offs >= 0
1903 ? pread (req->int1, req->ptr2, req->size, req->offs)
1904 : read (req->int1, req->ptr2, req->size); break;
1905 case EIO_WRITE: req->result = req->offs >= 0
1906 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1907 : write (req->int1, req->ptr2, req->size); break;
1908
1909 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1910 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
1911
1912#if HAVE_AT
1913
1914 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1915 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break;
1916 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1917 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, AT_SYMLINK_NOFOLLOW); break;
1918 case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break;
1919 case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break;
1920 case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break;
1921 case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break;
1922
1923 case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
1924 case EIO_RMDIR: /* complications arise because "." cannot be removed, so we might have to expand */
1925 req->result = req->wd && SINGLEDOT (req->ptr1)
1926 ? rmdir (req->wd->str)
1927 : unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break;
1928 case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); break;
1929 case EIO_RENAME: /* complications arise because "." cannot be renamed, so we might have to expand */
1930 req->result = req->wd && SINGLEDOT (req->ptr1)
1931 ? rename (req->wd->str, req->ptr2)
1932 : renameat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2); break;
1933 case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2, 0); break;
1934 case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
1935 case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1936 case EIO_READLINK: ALLOC (PATH_MAX);
1937 req->result = readlinkat (dirfd, req->ptr1, req->ptr2, PATH_MAX); break;
1938 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1939 req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1940 case EIO_UTIME:
1941 case EIO_FUTIME:
1942 {
1943 struct timespec ts[2];
1944 struct timespec *times;
1945
1946 if (req->nv1 != -1. || req->nv2 != -1.)
1947 {
1948 ts[0].tv_sec = req->nv1;
1949 ts[0].tv_nsec = (req->nv1 - ts[0].tv_sec) * 1e9;
1950 ts[1].tv_sec = req->nv2;
1951 ts[1].tv_nsec = (req->nv2 - ts[1].tv_sec) * 1e9;
1952
1953 times = ts;
1954 }
1955 else
1956 times = 0;
1957
1958 req->result = req->type == EIO_FUTIME
1959 ? futimens (req->int1, times)
1960 : utimensat (dirfd, req->ptr1, times, 0);
1961 }
1962 break;
1963
1964#else
1965
1966 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1967 req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
1968 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1969 req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
1970 case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
1971 case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
1972 case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
1973 case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
1974
1975 case EIO_UNLINK: req->result = unlink (path ); break;
1976 case EIO_RMDIR: req->result = rmdir (path ); break;
1977 case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
1978 case EIO_RENAME: req->result = rename (path , req->ptr2); break;
1979 case EIO_LINK: req->result = link (path , req->ptr2); break;
1980 case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
1981 case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
1982 case EIO_READLINK: ALLOC (PATH_MAX);
1983 req->result = readlink (path, req->ptr2, PATH_MAX); break;
1984 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1985 req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
1986
1987 case EIO_UTIME:
1988 case EIO_FUTIME:
1989 {
1990 struct timeval tv[2];
1991 struct timeval *times;
1992
1993 if (req->nv1 != -1. || req->nv2 != -1.)
1994 {
1995 tv[0].tv_sec = req->nv1;
1996 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
1997 tv[1].tv_sec = req->nv2;
1998 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
1999
2000 times = tv;
2001 }
2002 else
2003 times = 0;
2004
2005 req->result = req->type == EIO_FUTIME
2006 ? futimes (req->int1, times)
2007 : utimes (req->ptr1, times);
2008 }
2009 break;
2010
2011#endif
2012
2013 case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
2014 {
2015 ALLOC (req->result);
2016 memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
2017 }
2018 break;
2019
2020 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2021 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
2022
2023 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2024 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2025
2026 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
2027 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
2028 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
2029
2030 case EIO_CLOSE: req->result = close (req->int1); break;
2031 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
2032 case EIO_SYNC: req->result = 0; sync (); break;
2033 case EIO_FSYNC: req->result = fsync (req->int1); break;
2034 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
2035 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
2036 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
2037 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
2038 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
2039 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
2040 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
2041 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
2042
2043 case EIO_READDIR: eio__scandir (req, self); break;
2044
2045 case EIO_BUSY:
2046#ifdef _WIN32
2047 Sleep (req->nv1 * 1e3);
2048#else
2049 {
2050 struct timeval tv;
2051
2052 tv.tv_sec = req->nv1;
2053 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
2054
2055 req->result = select (0, 0, 0, 0, &tv);
2056 }
2057#endif
2058 break;
2059
2060 case EIO_GROUP:
2061 abort (); /* handled in eio_request */
2062
2063 case EIO_NOP:
2064 req->result = 0;
2065 break;
2066
2067 case EIO_CUSTOM:
2068 req->feed (req);
2069 break;
2070
2071 default:
2072 req->result = EIO_ENOSYS ();
2073 break;
2074 }
2075
2076 req->errorno = errno;
2077}
2078
2079#ifndef EIO_NO_WRAPPERS
2080
2081eio_req *eio_wd_open (const char *path, int pri, eio_cb cb, void *data)
2082{
2083 REQ (EIO_WD_OPEN); PATH; SEND;
2084}
2085
2086eio_req *eio_wd_close (eio_wd wd, int pri, eio_cb cb, void *data)
2087{
2088 REQ (EIO_WD_CLOSE); req->wd = wd; SEND;
2089}
2090
2091eio_req *eio_nop (int pri, eio_cb cb, void *data)
2092{
2093 REQ (EIO_NOP); SEND;
2094}
2095
2096eio_req *eio_busy (double delay, int pri, eio_cb cb, void *data)
2097{
2098 REQ (EIO_BUSY); req->nv1 = delay; SEND;
2099}
2100
2101eio_req *eio_sync (int pri, eio_cb cb, void *data)
2102{
2103 REQ (EIO_SYNC); SEND;
2104}
2105
2106eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
2107{
2108 REQ (EIO_FSYNC); req->int1 = fd; SEND;
2109}
2110
2111eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2112{
2113 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2114}
2115
2116eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2117{
2118 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2119}
2120
2121eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
2122{
2123 REQ (EIO_SYNCFS); req->int1 = fd; SEND;
2124}
2125
2126eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2127{
2128 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2129}
2130
2131eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2132{
2133 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2134}
2135
2136eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
2137{
2138 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
2139}
2140
2141eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
2142{
2143 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2144}
2145
2146eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2147{
2148 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
2149}
2150
2151eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
2152{
2153 REQ (EIO_CLOSE); req->int1 = fd; SEND;
2154}
2155
2156eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
2157{
2158 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
2159}
2160
2161eio_req *eio_seek (int fd, off_t offset, int whence, int pri, eio_cb cb, void *data)
2162{
2163 REQ (EIO_SEEK); req->int1 = fd; req->offs = offset; req->int2 = whence; SEND;
2164}
2165
2166eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
2167{
2168 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
2169}
2170
2171eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
2172{
2173 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
2174}
2175
2176eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
2177{
2178 REQ (EIO_FSTAT); req->int1 = fd; SEND;
2179}
2180
2181eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
2182{
2183 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
2184}
2185
2186eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
2187{
2188 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
2189}
2190
2191eio_req *eio_ftruncate (int fd, off_t offset, int pri, eio_cb cb, void *data)
2192{
2193 REQ (EIO_FTRUNCATE); req->int1 = fd; req->offs = offset; SEND;
2194}
2195
2196eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
2197{
2198 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
2199}
2200
2201eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
2202{
2203 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
2204}
2205
2206eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
2207{
2208 REQ (EIO_DUP2); req->int1 = fd; req->int2 = fd2; SEND;
2209}
2210
2211eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, int pri, eio_cb cb, void *data)
2212{
2213 REQ (EIO_SENDFILE); req->int1 = out_fd; req->int2 = in_fd; req->offs = in_offset; req->size = length; SEND;
2214}
2215
2216eio_req *eio_open (const char *path, int flags, mode_t mode, int pri, eio_cb cb, void *data)
2217{
2218 REQ (EIO_OPEN); PATH; req->int1 = flags; req->int2 = (long)mode; SEND;
2219}
2220
2221eio_req *eio_utime (const char *path, double atime, double mtime, int pri, eio_cb cb, void *data)
2222{
2223 REQ (EIO_UTIME); PATH; req->nv1 = atime; req->nv2 = mtime; SEND;
2224}
2225
2226eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
2227{
2228 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
2229}
2230
2231eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
2232{
2233 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
2234}
2235
2236eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
2237{
2238 REQ (EIO_CHMOD); PATH; req->int2 = (long)mode; SEND;
2239}
2240
2241eio_req *eio_mkdir (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
2242{
2243 REQ (EIO_MKDIR); PATH; req->int2 = (long)mode; SEND;
2244}
2245
2246static eio_req *
2247eio__1path (int type, const char *path, int pri, eio_cb cb, void *data)
2248{
2249 REQ (type); PATH; SEND;
2250}
2251
2252eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
2253{
2254 return eio__1path (EIO_READLINK, path, pri, cb, data);
2255}
2256
2257eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2258{
2259 return eio__1path (EIO_REALPATH, path, pri, cb, data);
2260}
2261
2262eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
2263{
2264 return eio__1path (EIO_STAT, path, pri, cb, data);
2265}
2266
2267eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
2268{
2269 return eio__1path (EIO_LSTAT, path, pri, cb, data);
2270}
2271
2272eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
2273{
2274 return eio__1path (EIO_STATVFS, path, pri, cb, data);
2275}
2276
2277eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
2278{
2279 return eio__1path (EIO_UNLINK, path, pri, cb, data);
2280}
2281
2282eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
2283{
2284 return eio__1path (EIO_RMDIR, path, pri, cb, data);
2285}
2286
2287eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
2288{
2289 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
2290}
2291
2292eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
2293{
2294 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
2295}
2296
2297static eio_req *
2298eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2299{
2300 REQ (type); PATH;
2301
2302 req->flags |= EIO_FLAG_PTR2_FREE;
2303 req->ptr2 = strdup (new_path);
2304 if (!req->ptr2)
2305 {
2306 eio_api_destroy (req);
2307 return 0;
2308 }
2309
2310 SEND;
2311}
2312
2313eio_req *eio_link (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2314{
2315 return eio__2path (EIO_LINK, path, new_path, pri, cb, data);
2316}
2317
2318eio_req *eio_symlink (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2319{
2320 return eio__2path (EIO_SYMLINK, path, new_path, pri, cb, data);
2321}
2322
2323eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2324{
2325 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
2326}
2327
2328eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
2329{
2330 REQ (EIO_CUSTOM); req->feed = execute; SEND;
2331}
2332
2333#endif
2334
2335eio_req *eio_grp (eio_cb cb, void *data)
2336{
2337 const int pri = EIO_PRI_MAX;
2338
2339 REQ (EIO_GROUP); SEND;
2340}
2341
2342#undef REQ
2343#undef PATH
2344#undef SEND
2345
2346/*****************************************************************************/
2347/* grp functions */
1171 2348
1172void 2349void
1173aio_sendfile (SV *out_fh, SV *in_fh, SV *in_offset, UV length, SV *callback=&PL_sv_undef)
1174 PROTOTYPE: $$$$;$
1175 PPCODE:
1176{
1177 dREQ;
1178
1179 req->type = EIO_SENDFILE;
1180 req->sv1 = newSVsv (out_fh);
1181 req->int1 = PerlIO_fileno (IoIFP (sv_2io (out_fh)));
1182 req->sv2 = newSVsv (in_fh);
1183 req->int2 = PerlIO_fileno (IoIFP (sv_2io (in_fh)));
1184 req->offs = SvVAL64 (in_offset);
1185 req->size = length;
1186
1187 EIO_SEND;
1188}
1189
1190void
1191aio_readahead (SV *fh, SV *offset, IV length, SV *callback=&PL_sv_undef)
1192 PROTOTYPE: $$$;$
1193 PPCODE:
1194{
1195 dREQ;
1196
1197 req->type = EIO_READAHEAD;
1198 req->sv1 = newSVsv (fh);
1199 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1200 req->offs = SvVAL64 (offset);
1201 req->size = length;
1202
1203 EIO_SEND;
1204}
1205
1206void
1207aio_stat (SV8 *fh_or_path, SV *callback=&PL_sv_undef)
1208 ALIAS:
1209 aio_stat = EIO_STAT
1210 aio_lstat = EIO_LSTAT
1211 PPCODE:
1212{
1213 dREQ;
1214
1215 req->ptr2 = malloc (sizeof (EIO_STRUCT_STAT));
1216 if (!req->ptr2)
1217 {
1218 req_destroy (req);
1219 croak ("out of memory during aio_stat statdata allocation");
1220 }
1221
1222 req->flags |= FLAG_PTR2_FREE;
1223 req->sv1 = newSVsv (fh_or_path);
1224
1225 if (SvPOK (fh_or_path))
1226 {
1227 req->type = ix;
1228 req->ptr1 = SvPVbyte_nolen (req->sv1);
1229 }
1230 else
1231 {
1232 req->type = EIO_FSTAT;
1233 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1234 }
1235
1236 EIO_SEND;
1237}
1238
1239void
1240aio_utime (SV8 *fh_or_path, SV *atime, SV *mtime, SV *callback=&PL_sv_undef)
1241 PPCODE:
1242{
1243 dREQ;
1244
1245 req->nv1 = SvOK (atime) ? SvNV (atime) : -1.;
1246 req->nv2 = SvOK (mtime) ? SvNV (mtime) : -1.;
1247 req->sv1 = newSVsv (fh_or_path);
1248
1249 if (SvPOK (fh_or_path))
1250 {
1251 req->type = EIO_UTIME;
1252 req->ptr1 = SvPVbyte_nolen (req->sv1);
1253 }
1254 else
1255 {
1256 req->type = EIO_FUTIME;
1257 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1258 }
1259
1260 EIO_SEND;
1261}
1262
1263void
1264aio_truncate (SV8 *fh_or_path, SV *offset, SV *callback=&PL_sv_undef)
1265 PPCODE:
1266{
1267 dREQ;
1268
1269 req->sv1 = newSVsv (fh_or_path);
1270 req->offs = SvOK (offset) ? SvVAL64 (offset) : -1;
1271
1272 if (SvPOK (fh_or_path))
1273 {
1274 req->type = EIO_TRUNCATE;
1275 req->ptr1 = SvPVbyte_nolen (req->sv1);
1276 }
1277 else
1278 {
1279 req->type = EIO_FTRUNCATE;
1280 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1281 }
1282
1283 EIO_SEND;
1284}
1285
1286void
1287aio_chmod (SV8 *fh_or_path, int mode, SV *callback=&PL_sv_undef)
1288 ALIAS:
1289 aio_chmod = EIO_CHMOD
1290 aio_fchmod = EIO_FCHMOD
1291 aio_mkdir = EIO_MKDIR
1292 PPCODE:
1293{
1294 dREQ;
1295
1296 req->type = type;
1297 req->int2 = mode;
1298 req->sv1 = newSVsv (fh_or_path);
1299
1300 if (ix == EIO_FCHMOD)
1301 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1302 else
1303 req->ptr1 = SvPVbyte_nolen (req->sv1);
1304
1305 EIO_SEND;
1306}
1307
1308void
1309aio_chown (SV8 *fh_or_path, SV *uid, SV *gid, SV *callback=&PL_sv_undef)
1310 PPCODE:
1311{
1312 dREQ;
1313
1314 req->int2 = SvOK (uid) ? SvIV (uid) : -1;
1315 req->int3 = SvOK (gid) ? SvIV (gid) : -1;
1316 req->sv1 = newSVsv (fh_or_path);
1317
1318 if (SvPOK (fh_or_path))
1319 {
1320 req->type = EIO_CHOWN;
1321 req->ptr1 = SvPVbyte_nolen (req->sv1);
1322 }
1323 else
1324 {
1325 req->type = EIO_FCHOWN;
1326 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1327 }
1328
1329 EIO_SEND;
1330}
1331
1332void
1333aio_unlink (SV8 *pathname, SV *callback=&PL_sv_undef)
1334 ALIAS:
1335 aio_unlink = EIO_UNLINK
1336 aio_rmdir = EIO_RMDIR
1337 aio_readdir = EIO_READDIR
1338 PPCODE:
1339{
1340 dREQ;
1341
1342 req->type = ix;
1343 req->sv1 = newSVsv (pathname);
1344 req->ptr1 = SvPVbyte_nolen (req->sv1);
1345
1346 EIO_SEND;
1347}
1348
1349void
1350aio_link (SV8 *oldpath, SV8 *newpath, SV *callback=&PL_sv_undef)
1351 ALIAS:
1352 aio_link = EIO_LINK
1353 aio_symlink = EIO_SYMLINK
1354 aio_rename = EIO_RENAME
1355 PPCODE:
1356{
1357 dREQ;
1358
1359 req->type = ix;
1360 req->sv1 = newSVsv (oldpath);
1361 req->ptr1 = SvPVbyte_nolen (req->sv1);
1362 req->sv2 = newSVsv (newpath);
1363 req->ptr2 = SvPVbyte_nolen (req->sv2);
1364
1365 EIO_SEND;
1366}
1367
1368void
1369aio_mknod (SV8 *pathname, int mode, UV dev, SV *callback=&PL_sv_undef)
1370 PPCODE:
1371{
1372 dREQ;
1373
1374 req->type = EIO_MKNOD;
1375 req->sv1 = newSVsv (pathname);
1376 req->ptr1 = SvPVbyte_nolen (req->sv1);
1377 req->int2 = (mode_t)mode;
1378 req->offs = dev;
1379
1380 EIO_SEND;
1381}
1382
1383void
1384aio_busy (double delay, SV *callback=&PL_sv_undef)
1385 PPCODE:
1386{
1387 dREQ;
1388
1389 req->type = EIO_BUSY;
1390 req->nv1 = delay < 0. ? 0. : delay;
1391
1392 EIO_SEND;
1393}
1394
1395void
1396aio_group (SV *callback=&PL_sv_undef)
1397 PROTOTYPE: ;$
1398 PPCODE:
1399{
1400 dREQ;
1401
1402 req->type = EIO_GROUP;
1403
1404 req_send (req);
1405 XPUSHs (req_sv (req, AIO_GRP_KLASS));
1406}
1407
1408void
1409aio_nop (SV *callback=&PL_sv_undef)
1410 ALIAS:
1411 aio_nop = EIO_NOP
1412 aio_sync = EIO_SYNC
1413 PPCODE:
1414{
1415 dREQ;
1416
1417#endif
1418
1419void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2350eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1420{ 2351{
1421 grp->int2 = limit; 2352 grp->int2 = limit;
1422 grp->feed = feed; 2353 grp->feed = feed;
1423 2354
1424 grp_try_feed (grp); 2355 grp_try_feed (grp);
1425} 2356}
1426 2357
2358void
1427void eio_grp_limit (eio_req *grp, int limit) 2359eio_grp_limit (eio_req *grp, int limit)
1428{ 2360{
1429 grp->int2 = limit; 2361 grp->int2 = limit;
1430 2362
1431 grp_try_feed (grp); 2363 grp_try_feed (grp);
1432} 2364}
1433 2365
2366void
1434void eio_grp_add (eio_req *grp, eio_req *req) 2367eio_grp_add (eio_req *grp, eio_req *req)
1435{ 2368{
1436 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2369 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
2370
2371 grp->flags |= EIO_FLAG_GROUPADD;
1437 2372
1438 ++grp->size; 2373 ++grp->size;
1439 req->grp = grp; 2374 req->grp = grp;
1440 2375
1441 req->grp_prev = 0; 2376 req->grp_prev = 0;
… …
1445 grp->grp_first->grp_prev = req; 2380 grp->grp_first->grp_prev = req;
1446 2381
1447 grp->grp_first = req; 2382 grp->grp_first = req;
1448} 2383}
1449 2384
2385/*****************************************************************************/
2386/* misc garbage */
1450 2387
2388eio_ssize_t
2389eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2390{
2391 return eio__sendfile (ofd, ifd, offset, count);
2392}
2393

Diff Legend

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