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 1.137 by root, Thu Jun 25 17:40:24 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; 382static struct etp_pool eio_pool;
383#define EIO_POOL (&eio_pool)
227 384
228 for (pri = EIO_NUM_PRI; pri--; ) 385/*****************************************************************************/
229 {
230 eio_req *req = q->qs[pri];
231 386
232 if (req) 387static void
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) 388grp_try_feed (eio_req *grp)
245{ 389{
246 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 390 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
247 { 391 {
248 int old_len = grp->size; 392 grp->flags &= ~ETP_FLAG_GROUPADD;
249 393
250 EIO_FEED (grp); 394 EIO_FEED (grp);
251 395
252 /* stop if no progress has been made */ 396 /* stop if no progress has been made */
253 if (old_len == grp->size) 397 if (!(grp->flags & ETP_FLAG_GROUPADD))
254 { 398 {
255 grp->feed = 0; 399 grp->feed = 0;
256 break; 400 break;
257 } 401 }
258 } 402 }
259} 403}
260 404
261static int eio_finish (eio_req *req); 405static int
262
263static int grp_dec (eio_req *grp) 406grp_dec (eio_req *grp)
264{ 407{
265 --grp->size; 408 --grp->size;
266 409
267 /* call feeder, if applicable */ 410 /* call feeder, if applicable */
268 grp_try_feed (grp); 411 grp_try_feed (grp);
269 412
270 /* finish, if done */ 413 /* finish, if done */
271 if (!grp->size && grp->int1) 414 if (!grp->size && grp->flags & ETP_FLAG_DELAYED)
272 return eio_finish (grp); 415 return eio_finish (grp);
273 else 416 else
274 return 0; 417 return 0;
275} 418}
276 419
420static void
277void eio_destroy (eio_req *req) 421eio_destroy (eio_req *req)
278{ 422{
423 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
279 if ((req)->flags & EIO_FLAG_PTR2_FREE) 424 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
280 free (req->ptr2);
281 425
282 EIO_DESTROY (req); 426 EIO_DESTROY (req);
283} 427}
284 428
429static int
285static int eio_finish (eio_req *req) 430eio_finish (eio_req *req)
286{ 431{
287 int res = EIO_FINISH (req); 432 int res = EIO_FINISH (req);
288 433
289 if (req->grp) 434 if (req->grp)
290 { 435 {
… …
298 if (grp->grp_first == req) 443 if (grp->grp_first == req)
299 grp->grp_first = req->grp_next; 444 grp->grp_first = req->grp_next;
300 445
301 res2 = grp_dec (grp); 446 res2 = grp_dec (grp);
302 447
303 if (!res && res2) 448 if (!res)
304 res = res2; 449 res = res2;
305 } 450 }
306 451
307 eio_destroy (req); 452 eio_destroy (req);
308 453
309 return res; 454 return res;
310} 455}
311 456
457void
312void eio_grp_cancel (eio_req *grp) 458eio_grp_cancel (eio_req *grp)
313{ 459{
314 for (grp = grp->grp_first; grp; grp = grp->grp_next) 460 etp_grp_cancel (EIO_POOL, grp);
315 eio_cancel (grp);
316} 461}
317 462
463void
318void eio_cancel (eio_req *req) 464eio_cancel (eio_req *req)
319{ 465{
320 req->flags |= EIO_FLAG_CANCELLED; 466 etp_cancel (EIO_POOL, req);
321
322 eio_grp_cancel (req);
323} 467}
324 468
325X_THREAD_PROC (eio_proc); 469void
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) 470eio_submit (eio_req *req)
363{ 471{
364 ++nreqs; 472 etp_submit (EIO_POOL, 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} 473}
374 474
375static void end_thread (void) 475unsigned int
476eio_nreqs (void)
376{ 477{
377 eio_req *req = calloc (1, sizeof (eio_req)); 478 return etp_nreqs (EIO_POOL);
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} 479}
391 480
481unsigned int
482eio_nready (void)
483{
484 return etp_nready (EIO_POOL);
485}
486
487unsigned int
488eio_npending (void)
489{
490 return etp_npending (EIO_POOL);
491}
492
493unsigned int ecb_cold
494eio_nthreads (void)
495{
496 return etp_nthreads (EIO_POOL);
497}
498
499void ecb_cold
392void eio_set_max_poll_time (double nseconds) 500eio_set_max_poll_time (double nseconds)
393{ 501{
394 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 502 etp_set_max_poll_time (EIO_POOL, nseconds);
395 max_poll_time = nseconds;
396 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
397} 503}
398 504
505void ecb_cold
399void eio_set_max_poll_reqs (unsigned int maxreqs) 506eio_set_max_poll_reqs (unsigned int maxreqs)
400{ 507{
401 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 508 etp_set_max_poll_reqs (EIO_POOL, maxreqs);
402 max_poll_reqs = maxreqs;
403 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
404} 509}
405 510
511void ecb_cold
406void eio_set_max_idle (unsigned int nthreads) 512eio_set_max_idle (unsigned int nthreads)
407{ 513{
408 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 514 etp_set_max_idle (EIO_POOL, nthreads);
409 max_idle = nthreads <= 0 ? 1 : nthreads;
410 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
411} 515}
412 516
517void ecb_cold
518eio_set_idle_timeout (unsigned int seconds)
519{
520 etp_set_idle_timeout (EIO_POOL, seconds);
521}
522
523void ecb_cold
413void eio_set_min_parallel (unsigned int nthreads) 524eio_set_min_parallel (unsigned int nthreads)
414{ 525{
415 if (wanted < nthreads) 526 etp_set_min_parallel (EIO_POOL, nthreads);
416 wanted = nthreads;
417} 527}
418 528
529void ecb_cold
419void eio_set_max_parallel (unsigned int nthreads) 530eio_set_max_parallel (unsigned int nthreads)
420{ 531{
421 if (wanted > nthreads) 532 etp_set_max_parallel (EIO_POOL, nthreads);
422 wanted = nthreads;
423
424 while (started > wanted)
425 end_thread ();
426} 533}
427 534
428int eio_poll (void) 535int eio_poll (void)
429{ 536{
430 int maxreqs = max_poll_reqs; 537 return etp_poll (EIO_POOL);
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} 538}
486 539
487/*****************************************************************************/ 540/*****************************************************************************/
488/* work around various missing functions */ 541/* work around various missing functions */
489 542
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 543#ifndef HAVE_UTIMES
533 544
545# undef utimes
534# define utimes(path,times) aio_utimes (path, times) 546# define utimes(path,times) eio__utimes (path, times)
535# define futimes(fd,times) aio_futimes (fd, times)
536 547
548static int
537static int aio_utimes (const char *filename, const struct timeval times[2]) 549eio__utimes (const char *filename, const struct timeval times[2])
538{ 550{
539 if (times) 551 if (times)
540 { 552 {
541 struct utimbuf buf; 553 struct utimbuf buf;
542 554
… …
547 } 559 }
548 else 560 else
549 return utime (filename, 0); 561 return utime (filename, 0);
550} 562}
551 563
564#endif
565
566#ifndef HAVE_FUTIMES
567
568# undef futimes
569# define futimes(fd,times) eio__futimes (fd, times)
570
571static int
552static int aio_futimes (int fd, const struct timeval tv[2]) 572eio__futimes (int fd, const struct timeval tv[2])
553{ 573{
554 errno = ENOSYS; 574 errno = ENOSYS;
555 return -1; 575 return -1;
556} 576}
557 577
558#endif 578#endif
559 579
560#if !HAVE_FDATASYNC 580#if !HAVE_FDATASYNC
581# undef fdatasync
561# define fdatasync fsync 582# define fdatasync(fd) fsync (fd)
562#endif 583#endif
584
585static int
586eio__syncfs (int fd)
587{
588 int res;
589
590#if HAVE_SYS_SYNCFS
591 res = (int)syscall (__NR_syncfs, (int)(fd));
592#else
593 res = EIO_ENOSYS ();
594#endif
595
596 if (res < 0 && errno == ENOSYS && fd >= 0)
597 sync ();
598
599 return res;
600}
601
602/* sync_file_range always needs emulation */
603static int
604eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
605{
606#if HAVE_SYNC_FILE_RANGE
607 int res;
608
609 if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
610 || EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
611 || EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
612 {
613 flags = 0
614 | (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
615 | (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
616 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
617 }
618
619 res = sync_file_range (fd, offset, nbytes, flags);
620
621 if (!res || errno != ENOSYS)
622 return res;
623#endif
624
625 /* even though we could play tricks with the flags, it's better to always
626 * call fdatasync, as that matches the expectation of its users best */
627 return fdatasync (fd);
628}
629
630static int
631eio__fallocate (int fd, int mode, off_t offset, size_t len)
632{
633#if HAVE_LINUX_FALLOCATE
634 return fallocate (fd, mode, offset, len);
635#else
636 return EIO_ENOSYS ();
637#endif
638}
563 639
564#if !HAVE_READAHEAD 640#if !HAVE_READAHEAD
641# undef readahead
565# define readahead(fd,offset,count) aio_readahead (fd, offset, count, self) 642# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
566 643
644static eio_ssize_t
567static ssize_t aio_readahead (int fd, off_t offset, size_t count, worker *self) 645eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
568{ 646{
569 size_t todo = count; 647 size_t todo = count;
570 dBUF; 648 dBUF;
571 649
572 while (todo > 0) 650 while (todo > 0)
573 { 651 {
574 size_t len = todo < EIO_BUFSIZE ? todo : EIO_BUFSIZE; 652 size_t len = todo < EIO_BUFSIZE ? todo : EIO_BUFSIZE;
575 653
576 pread (fd, aio_buf, len, offset); 654 pread (fd, eio_buf, len, offset);
577 offset += len; 655 offset += len;
578 todo -= len; 656 todo -= len;
579 } 657 }
580 658
581 errno = 0; 659 FUBd;
660
661 /* linux's readahead basically only fails for EBADF or EINVAL (not mmappable) */
662 /* but not for e.g. EIO or eof, so we also never fail */
582 return count; 663 return 0;
583} 664}
584 665
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 666#endif
616 667
617/* sendfile always needs emulation */ 668/* sendfile always needs emulation */
669static eio_ssize_t
618static ssize_t sendfile_ (int ofd, int ifd, off_t offset, size_t count, worker *self) 670eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
619{ 671{
672 eio_ssize_t written = 0;
620 ssize_t res; 673 eio_ssize_t res;
621 674
622 if (!count) 675 if (!count)
623 return 0; 676 return 0;
624 677
678 for (;;)
679 {
680#ifdef __APPLE__
681# undef HAVE_SENDFILE /* broken, as everything on os x */
682#endif
625#if HAVE_SENDFILE 683#if HAVE_SENDFILE
626# if __linux 684# if __linux
685 off_t soffset = offset;
627 res = sendfile (ofd, ifd, &offset, count); 686 res = sendfile (ofd, ifd, &soffset, count);
628 687
629# elif __freebsd 688# elif __FreeBSD__
630 /* 689 /*
631 * Of course, the freebsd sendfile is a dire hack with no thoughts 690 * Of course, the freebsd sendfile is a dire hack with no thoughts
632 * wasted on making it similar to other I/O functions. 691 * wasted on making it similar to other I/O functions.
633 */ 692 */
634 {
635 off_t sbytes; 693 off_t sbytes;
636 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 694 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
637 695
638 if (res < 0 && sbytes) 696 #if 0 /* according to the manpage, this is correct, but broken behaviour */
639 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 697 /* freebsd' sendfile will return 0 on success */
698 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
699 /* not on e.g. EIO or EPIPE - sounds broken */
700 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
640 res = sbytes; 701 res = sbytes;
641 } 702 #endif
703
704 /* according to source inspection, this is correct, and useful behaviour */
705 if (sbytes)
706 res = sbytes;
707
708# elif defined __APPLE__
709 off_t sbytes = count;
710 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
711
712 /* according to the manpage, sbytes is always valid */
713 if (sbytes)
714 res = sbytes;
642 715
643# elif __hpux 716# elif __hpux
644 res = sendfile (ofd, ifd, offset, count, 0, 0); 717 res = sendfile (ofd, ifd, offset, count, 0, 0);
645 718
646# elif __solaris 719# elif __solaris
647 {
648 struct sendfilevec vec; 720 struct sendfilevec vec;
649 size_t sbytes; 721 size_t sbytes;
650 722
651 vec.sfv_fd = ifd; 723 vec.sfv_fd = ifd;
652 vec.sfv_flag = 0; 724 vec.sfv_flag = 0;
653 vec.sfv_off = offset; 725 vec.sfv_off = offset;
654 vec.sfv_len = count; 726 vec.sfv_len = count;
655 727
656 res = sendfilev (ofd, &vec, 1, &sbytes); 728 res = sendfilev (ofd, &vec, 1, &sbytes);
657 729
658 if (res < 0 && sbytes) 730 if (res < 0 && sbytes)
659 res = sbytes; 731 res = sbytes;
660 }
661 732
662# endif 733# endif
734
735#elif defined (_WIN32) && 0
736 /* does not work, just for documentation of what would need to be done */
737 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
738 /* libeio guarantees that the file offset does not change, and windows */
739 /* has no way to get an independent handle to the same file description */
740 HANDLE h = TO_SOCKET (ifd);
741 SetFilePointer (h, offset, 0, FILE_BEGIN);
742 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
743
663#else 744#else
664 res = -1; 745 res = EIO_ENOSYS ();
665 errno = ENOSYS;
666#endif 746#endif
667 747
748 /* we assume sendfile can copy at least 128mb in one go */
749 if (res <= 128 * 1024 * 1024)
750 {
751 if (res > 0)
752 written += res;
753
754 if (written)
755 return written;
756
757 break;
758 }
759 else
760 {
761 /* if we requested more, then probably the kernel was lazy */
762 written += res;
763 offset += res;
764 count -= res;
765
766 if (!count)
767 return written;
768 }
769 }
770
668 if (res < 0 771 if (res < 0
669 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 772 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
773 /* BSDs */
774#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
775 || errno == ENOTSUP
776#endif
777#ifdef EOPNOTSUPP /* windows */
778 || errno == EOPNOTSUPP /* BSDs */
779#endif
670#if __solaris 780#if __solaris
671 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 781 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
672#endif 782#endif
673 ) 783 )
674 ) 784 )
… …
678 788
679 res = 0; 789 res = 0;
680 790
681 while (count) 791 while (count)
682 { 792 {
683 ssize_t cnt; 793 eio_ssize_t cnt;
684 794
685 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 795 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
686 796
687 if (cnt <= 0) 797 if (cnt <= 0)
688 { 798 {
… …
700 810
701 offset += cnt; 811 offset += cnt;
702 res += cnt; 812 res += cnt;
703 count -= cnt; 813 count -= cnt;
704 } 814 }
815
816 FUBd;
705 } 817 }
706 818
707 return res; 819 return res;
708} 820}
709 821
710/* read a full directory */ 822#ifdef PAGESIZE
711static void scandir_ (eio_req *req, worker *self) 823# define eio_pagesize() PAGESIZE
824#else
825static intptr_t
826eio_pagesize (void)
712{ 827{
713 DIR *dirp; 828 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;
724 829
725 X_LOCK (wrklock); 830 if (!page)
726 self->dirp = dirp = opendir (req->ptr1); 831 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);
731 832
732 if (dirp && u && names) 833 return page;
733 for (;;) 834}
835#endif
836
837static void
838eio_page_align (void **addr, size_t *length)
839{
840 intptr_t mask = eio_pagesize () - 1;
841
842 /* round down addr */
843 intptr_t adj = mask & (intptr_t)*addr;
844
845 *addr = (void *)((intptr_t)*addr - adj);
846 *length += adj;
847
848 /* round up length */
849 *length = (*length + mask) & ~mask;
850}
851
852#if !_POSIX_MEMLOCK
853# define eio__mlockall(a) EIO_ENOSYS ()
854#else
855
856static int
857eio__mlockall (int flags)
858{
859 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
860 extern int mallopt (int, int);
861 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
862 #endif
863
864 if (EIO_MCL_CURRENT != MCL_CURRENT
865 || EIO_MCL_FUTURE != MCL_FUTURE)
866 {
867 flags = 0
868 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
869 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
870 }
871
872 return mlockall (flags);
873}
874#endif
875
876#if !_POSIX_MEMLOCK_RANGE
877# define eio__mlock(a,b) EIO_ENOSYS ()
878#else
879
880static int
881eio__mlock (void *addr, size_t length)
882{
883 eio_page_align (&addr, &length);
884
885 return mlock (addr, length);
886}
887
888#endif
889
890#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
891# define eio__msync(a,b,c) EIO_ENOSYS ()
892#else
893
894static int
895eio__msync (void *mem, size_t len, int flags)
896{
897 eio_page_align (&mem, &len);
898
899 if (EIO_MS_ASYNC != MS_SYNC
900 || EIO_MS_INVALIDATE != MS_INVALIDATE
901 || EIO_MS_SYNC != MS_SYNC)
902 {
903 flags = 0
904 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
905 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
906 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
907 }
908
909 return msync (mem, len, flags);
910}
911
912#endif
913
914static int
915eio__mtouch (eio_req *req)
916{
917 void *mem = req->ptr2;
918 size_t len = req->size;
919 int flags = req->int1;
920
921 eio_page_align (&mem, &len);
922
923 {
924 intptr_t addr = (intptr_t)mem;
925 intptr_t end = addr + len;
926 intptr_t page = eio_pagesize ();
927
928 if (addr < end)
929 if (flags & EIO_MT_MODIFY) /* modify */
930 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
931 else
932 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
933 }
934
935 return 0;
936}
937
938/*****************************************************************************/
939/* requests implemented outside eio_execute, because they are so large */
940
941static void
942eio__lseek (eio_req *req)
943{
944 /* this usually gets optimised away completely, or your compiler sucks, */
945 /* or the whence constants really are not 0, 1, 2 */
946 int whence = req->int2 == EIO_SEEK_SET ? SEEK_SET
947 : req->int2 == EIO_SEEK_CUR ? SEEK_CUR
948 : req->int2 == EIO_SEEK_END ? SEEK_END
949 : req->int2;
950
951 req->offs = lseek (req->int1, req->offs, whence);
952 req->result = req->offs == (off_t)-1 ? -1 : 0;
953}
954
955/* result will always end up in tmpbuf, there is always space for adding a 0-byte */
956static int
957eio__realpath (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
958{
959 char *res;
960 const char *rel = path;
961 char *tmp1, *tmp2;
962#if SYMLOOP_MAX > 32
963 int symlinks = SYMLOOP_MAX;
964#else
965 int symlinks = 32;
966#endif
967
968 errno = EINVAL;
969 if (!rel)
970 return -1;
971
972 errno = ENOENT;
973 if (!*rel)
974 return -1;
975
976 res = etp_tmpbuf_get (tmpbuf, PATH_MAX * 3);
977#ifdef _WIN32
978 if (_access (rel, 4) != 0)
979 return -1;
980
981 symlinks = GetFullPathName (rel, PATH_MAX * 3, res, 0);
982
983 errno = ENAMETOOLONG;
984 if (symlinks >= PATH_MAX * 3)
985 return -1;
986
987 errno = EIO;
988 if (symlinks <= 0)
989 return -1;
990
991 return symlinks;
992
993#else
994 tmp1 = res + PATH_MAX;
995 tmp2 = tmp1 + PATH_MAX;
996
997#if 0 /* disabled, the musl way to do things is just too racy */
998#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
999 /* on linux we may be able to ask the kernel */
1000 {
1001 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1002
1003 if (fd >= 0)
734 { 1004 {
1005 sprintf (tmp1, "/proc/self/fd/%d", fd);
1006 req->result = readlink (tmp1, res, PATH_MAX);
1007 /* here we should probably stat the open file and the disk file, to make sure they still match */
1008 close (fd);
1009
1010 if (req->result > 0)
1011 goto done;
1012 }
1013 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1014 return -1;
1015 }
1016#endif
1017#endif
1018
1019 if (*rel != '/')
1020 {
1021 int len;
1022
735 errno = 0; 1023 errno = ENOENT;
736 readdir_r (dirp, &u->d, &entp); 1024 if (wd == EIO_INVALID_WD)
1025 return -1;
1026
1027 if (wd == EIO_CWD)
1028 {
1029 if (!getcwd (res, PATH_MAX))
1030 return -1;
737 1031
738 if (!entp) 1032 len = strlen (res);
1033 }
1034 else
1035 memcpy (res, wd->str, len = wd->len);
1036
1037 if (res [1]) /* only use if not / */
1038 res += len;
1039 }
1040
1041 while (*rel)
1042 {
1043 eio_ssize_t len, linklen;
1044 const char *beg = rel;
1045
1046 while (*rel && *rel != '/')
1047 ++rel;
1048
1049 len = rel - beg;
1050
1051 if (!len) /* skip slashes */
1052 {
739 break; 1053 ++rel;
1054 continue;
1055 }
740 1056
741 name = entp->d_name; 1057 if (beg [0] == '.')
1058 {
1059 if (len == 1)
1060 continue; /* . - nop */
742 1061
743 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1062 if (beg [1] == '.' && len == 2)
1063 {
1064 /* .. - back up one component, if possible */
1065
1066 while (res != tmpbuf->ptr)
1067 if (*--res == '/')
1068 break;
1069
1070 continue;
1071 }
1072 }
1073
1074 errno = ENAMETOOLONG;
1075 if (res + 1 + len + 1 >= tmp1)
1076 return -1;
1077
1078 /* copy one component */
1079 *res = '/';
1080 memcpy (res + 1, beg, len);
1081
1082 /* zero-terminate, for readlink */
1083 res [len + 1] = 0;
1084
1085 /* now check if it's a symlink */
1086 linklen = readlink (tmpbuf->ptr, tmp1, PATH_MAX);
1087
1088 if (linklen < 0)
744 { 1089 {
745 int len = strlen (name) + 1; 1090 if (errno != EINVAL)
1091 return -1;
746 1092
747 res++; 1093 /* 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; 1094 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 } 1095 }
1120 else 1096 else
1121 { 1097 {
1122 /* read: grow scalar as necessary */ 1098 /* yay, it was a symlink - build new path in tmp2 */
1123 svptr = SvGROW (data, len + dataoffset + 1); 1099 int rellen = strlen (rel);
1100
1101 errno = ENAMETOOLONG;
1102 if (linklen + 1 + rellen >= PATH_MAX)
1103 return -1;
1104
1105 errno = ELOOP;
1106 if (!--symlinks)
1107 return -1;
1108
1109 if (*tmp1 == '/')
1110 res = tmpbuf->ptr; /* symlink resolves to an absolute path */
1111
1112 /* we need to be careful, as rel might point into tmp2 already */
1113 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1114 tmp2 [linklen] = '/';
1115 memcpy (tmp2, tmp1, linklen);
1116
1117 rel = tmp2;
1124 } 1118 }
1119 }
1125 1120
1121 /* special case for the lone root path */
1122 if (res == tmpbuf->ptr)
1123 *res++ = '/';
1124
1125 return res - (char *)tmpbuf->ptr;
1126#endif
1127}
1128
1129static signed char
1130eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1131{
1132 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1133 : a->inode < b->inode ? -1
1134 : a->inode > b->inode ? 1
1135 : 0;
1136}
1137
1138#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1139
1140#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1141#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1142
1143static void
1144eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1145{
1146 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1147 unsigned char *bit = bits;
1148
1149 assert (CHAR_BIT == 8);
1150 assert (sizeof (eio_dirent) * 8 < 256);
1151 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1152 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1153
1154 if (size <= EIO_SORT_FAST)
1155 return;
1156
1157 /* first prepare an array of bits to test in our radix sort */
1158 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1159 /* inode_bits must contain all inodes ORed together */
1160 /* which is used to skip bits that are 0 everywhere, which is very common */
1161 {
1162 eio_ino_t endianness;
1163 int i, j;
1164
1165 /* we store the byte offset of byte n into byte n of "endianness" */
1166 for (i = 0; i < sizeof (eio_ino_t); ++i)
1167 ((unsigned char *)&endianness)[i] = i;
1168
1169 *bit++ = 0;
1170
1171 for (i = 0; i < sizeof (eio_ino_t); ++i)
1172 {
1173 /* shifting off the byte offsets out of "endianness" */
1174 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1175 endianness >>= 8;
1176
1177 for (j = 0; j < 8; ++j)
1178 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1179 *bit++ = offs + j;
1180 }
1181
1182 for (j = 0; j < 8; ++j)
1183 if (score_bits & (1 << j))
1184 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1185 }
1186
1187 /* now actually do the sorting (a variant of MSD radix sort) */
1188 {
1189 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1190 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1191 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1192 int stk_idx = 0;
1193
1194 base_stk [stk_idx] = dents;
1195 end_stk [stk_idx] = dents + size;
1196 bit_stk [stk_idx] = bit - 1;
1197
1198 do
1199 {
1200 base = base_stk [stk_idx];
1201 end = end_stk [stk_idx];
1202 bit = bit_stk [stk_idx];
1203
1204 for (;;)
1205 {
1206 unsigned char O = *bit >> 3;
1207 unsigned char M = 1 << (*bit & 7);
1208
1209 eio_dirent *a = base;
1210 eio_dirent *b = end;
1211
1212 if (b - a < EIO_SORT_CUTOFF)
1213 break;
1214
1215 /* now bit-partition the array on the bit */
1216 /* this ugly asymmetric loop seems to perform much better than typical */
1217 /* partition algos found in the literature */
1218 do
1219 if (!(((unsigned char *)a)[O] & M))
1220 ++a;
1221 else if (!(((unsigned char *)--b)[O] & M))
1222 {
1223 eio_dirent tmp = *a; *a = *b; *b = tmp;
1224 ++a;
1225 }
1226 while (b > a);
1227
1228 /* next bit, or stop, if no bits left in this path */
1229 if (!*--bit)
1230 break;
1231
1232 base_stk [stk_idx] = a;
1233 end_stk [stk_idx] = end;
1234 bit_stk [stk_idx] = bit;
1235 ++stk_idx;
1236
1237 end = a;
1238 }
1239 }
1240 while (stk_idx--);
1241 }
1242}
1243
1244static void
1245eio_dent_insertion_sort (eio_dirent *dents, int size)
1246{
1247 /* first move the smallest element to the front, to act as a sentinel */
1248 {
1249 int i;
1250 eio_dirent *min = dents;
1251
1252 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1253 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1254 if (EIO_DENT_CMP (dents [i], <, *min))
1255 min = &dents [i];
1256
1257 /* swap elements 0 and j (minimum) */
1258 {
1259 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1260 }
1261 }
1262
1263 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1264 {
1265 eio_dirent *i, *j;
1266
1267 for (i = dents + 1; i < dents + size; ++i)
1268 {
1269 eio_dirent value = *i;
1270
1271 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1272 j [1] = j [0];
1273
1274 j [1] = value;
1275 }
1276 }
1277}
1278
1279static void
1280eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1281{
1282 if (size <= 1)
1283 return; /* our insertion sort relies on size > 0 */
1284
1285 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1286 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1287 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1288
1289 /* use an insertion sort at the end, or for small arrays, */
1290 /* as insertion sort is more efficient for small partitions */
1291 eio_dent_insertion_sort (dents, size);
1292}
1293
1294/* read a full directory */
1295static void
1296eio__scandir (eio_req *req, etp_worker *self)
1297{
1298 char *name, *names;
1299 int namesalloc = 4096 - sizeof (void *) * 4;
1300 int namesoffs = 0;
1301 int flags = req->int1;
1302 eio_dirent *dents = 0;
1303 int dentalloc = 128;
1304 int dentoffs = 0;
1305 eio_ino_t inode_bits = 0;
1306#ifdef _WIN32
1307 HANDLE dirp;
1308 WIN32_FIND_DATA entp;
1309#else
1310 DIR *dirp;
1311 EIO_STRUCT_DIRENT *entp;
1312#endif
1313
1314 req->result = -1;
1315
1316 if (!(flags & EIO_READDIR_DENTS))
1317 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1318
1319#ifdef _WIN32
1320 {
1321 int len = strlen ((const char *)req->ptr1);
1322 char *path = malloc (MAX_PATH);
1323 const char *fmt;
1324 const char *reqpath = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1325
1326 if (!len)
1327 fmt = "./*";
1328 else if (reqpath[len - 1] == '/' || reqpath[len - 1] == '\\')
1329 fmt = "%s*";
1330 else
1331 fmt = "%s/*";
1332
1333 _snprintf (path, MAX_PATH, fmt, reqpath);
1334 dirp = FindFirstFile (path, &entp);
1335 free (path);
1336
1337 if (dirp == INVALID_HANDLE_VALUE)
1338 {
1339 /* should steal _dosmaperr */
1340 switch (GetLastError ())
1341 {
1342 case ERROR_FILE_NOT_FOUND:
1343 req->result = 0;
1344 break;
1345
1346 case ERROR_INVALID_NAME:
1347 case ERROR_PATH_NOT_FOUND:
1348 case ERROR_NO_MORE_FILES:
1349 errno = ENOENT;
1350 break;
1351
1352 case ERROR_NOT_ENOUGH_MEMORY:
1353 errno = ENOMEM;
1354 break;
1355
1356 default:
1357 errno = EINVAL;
1358 break;
1359 }
1360
1361 return;
1362 }
1363 }
1364#else
1365 #if HAVE_AT
1366 if (req->wd)
1367 {
1368 int fd = openat (WD2FD (req->wd), req->ptr1, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1369
1126 if (len < 0) 1370 if (fd < 0)
1127 croak ("length must not be negative"); 1371 return;
1128 1372
1373 dirp = fdopendir (fd);
1374
1375 if (!dirp)
1376 close (fd);
1377 }
1378 else
1379 dirp = opendir (req->ptr1);
1380 #else
1381 dirp = opendir (wd_expand (&self->tmpbuf, req->wd, req->ptr1));
1382 #endif
1383
1384 if (!dirp)
1385 return;
1386#endif
1387
1388 if (req->flags & EIO_FLAG_PTR1_FREE)
1389 free (req->ptr1);
1390
1391 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1392 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1393 req->ptr2 = names = malloc (namesalloc);
1394
1395 if (!names || (flags && !dents))
1396 return;
1397
1398 for (;;)
1399 {
1400 int done;
1401
1402#ifdef _WIN32
1403 done = !dirp;
1404#else
1405 errno = 0;
1406 entp = readdir (dirp);
1407 done = !entp;
1408#endif
1409
1410 if (done)
1129 { 1411 {
1130 dREQ; 1412#ifndef _WIN32
1413 int old_errno = errno;
1414 closedir (dirp);
1415 errno = old_errno;
1131 1416
1132 req->type = ix; 1417 if (errno)
1133 req->sv1 = newSVsv (fh); 1418 break;
1134 req->int1 = PerlIO_fileno (ix == EIO_READ ? IoIFP (sv_2io (fh)) 1419#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 1420
1142 if (!SvREADONLY (data)) 1421 /* sort etc. */
1422 req->int1 = flags;
1423 req->result = dentoffs;
1424
1425 if (flags & EIO_READDIR_STAT_ORDER)
1426 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1427 else if (flags & EIO_READDIR_DIRS_FIRST)
1428 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1429 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1430 else
1431 {
1432 /* in this case, all is known, and we just put dirs first and sort them */
1433 eio_dirent *oth = dents + dentoffs;
1434 eio_dirent *dir = dents;
1435
1436 /* now partition dirs to the front, and non-dirs to the back */
1437 /* by walking from both sides and swapping if necessary */
1438 while (oth > dir)
1439 {
1440 if (dir->type == EIO_DT_DIR)
1441 ++dir;
1442 else if ((--oth)->type == EIO_DT_DIR)
1443 {
1444 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1445
1446 ++dir;
1447 }
1448 }
1449
1450 /* now sort the dirs only (dirs all have the same score) */
1451 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1452 }
1453
1454 break;
1455 }
1456
1457 /* now add the entry to our list(s) */
1458 name = D_NAME (entp);
1459
1460 /* skip . and .. entries */
1461 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1462 {
1463 int len = D_NAMLEN (entp) + 1;
1464
1465 while (ecb_expect_false (namesoffs + len > namesalloc))
1143 { 1466 {
1144 SvREADONLY_on (data); 1467 namesalloc *= 2;
1145 req->flags |= FLAG_SV2_RO_OFF; 1468 req->ptr2 = names = realloc (names, namesalloc);
1469
1470 if (!names)
1471 break;
1146 } 1472 }
1147 1473
1148 EIO_SEND; 1474 memcpy (names + namesoffs, name, len);
1475
1476 if (dents)
1477 {
1478 struct eio_dirent *ent;
1479
1480 if (ecb_expect_false (dentoffs == dentalloc))
1481 {
1482 dentalloc *= 2;
1483 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1484
1485 if (!dents)
1486 break;
1487 }
1488
1489 ent = dents + dentoffs;
1490
1491 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1492 ent->namelen = len - 1;
1493 ent->inode = D_INO (entp);
1494
1495 inode_bits |= ent->inode;
1496
1497 switch (D_TYPE (entp))
1498 {
1499 default:
1500 ent->type = EIO_DT_UNKNOWN;
1501 flags |= EIO_READDIR_FOUND_UNKNOWN;
1502 break;
1503
1504 #ifdef DT_FIFO
1505 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1506 #endif
1507 #ifdef DT_CHR
1508 case DT_CHR: ent->type = EIO_DT_CHR; break;
1509 #endif
1510 #ifdef DT_MPC
1511 case DT_MPC: ent->type = EIO_DT_MPC; break;
1512 #endif
1513 #ifdef DT_DIR
1514 case DT_DIR: ent->type = EIO_DT_DIR; break;
1515 #endif
1516 #ifdef DT_NAM
1517 case DT_NAM: ent->type = EIO_DT_NAM; break;
1518 #endif
1519 #ifdef DT_BLK
1520 case DT_BLK: ent->type = EIO_DT_BLK; break;
1521 #endif
1522 #ifdef DT_MPB
1523 case DT_MPB: ent->type = EIO_DT_MPB; break;
1524 #endif
1525 #ifdef DT_REG
1526 case DT_REG: ent->type = EIO_DT_REG; break;
1527 #endif
1528 #ifdef DT_NWK
1529 case DT_NWK: ent->type = EIO_DT_NWK; break;
1530 #endif
1531 #ifdef DT_CMP
1532 case DT_CMP: ent->type = EIO_DT_CMP; break;
1533 #endif
1534 #ifdef DT_LNK
1535 case DT_LNK: ent->type = EIO_DT_LNK; break;
1536 #endif
1537 #ifdef DT_SOCK
1538 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1539 #endif
1540 #ifdef DT_DOOR
1541 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1542 #endif
1543 #ifdef DT_WHT
1544 case DT_WHT: ent->type = EIO_DT_WHT; break;
1545 #endif
1546 }
1547
1548 ent->score = 7;
1549
1550 if (flags & EIO_READDIR_DIRS_FIRST)
1551 {
1552 if (ent->type == EIO_DT_UNKNOWN)
1553 {
1554 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1555 ent->score = 1;
1556 else if (!strchr (name, '.')) /* absence of dots indicate likely dirs */
1557 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1558 }
1559 else if (ent->type == EIO_DT_DIR)
1560 ent->score = 0;
1561 }
1562 }
1563
1564 namesoffs += len;
1565 ++dentoffs;
1149 } 1566 }
1567
1568 if (EIO_CANCELLED (req))
1569 {
1570 errno = ECANCELED;
1571 break;
1572 }
1573
1574#ifdef _WIN32
1575 if (!FindNextFile (dirp, &entp))
1576 {
1577 FindClose (dirp);
1578 dirp = 0;
1579 }
1580#endif
1581 }
1582}
1583
1584/*****************************************************************************/
1585/* working directory stuff */
1586/* various deficiencies in the posix 2008 api force us to */
1587/* keep the absolute path in string form at all times */
1588/* fuck yeah. */
1589
1590#if !HAVE_AT
1591
1592/* a bit like realpath, but usually faster because it doesn'T have to return */
1593/* an absolute or canonical path */
1594static const char *
1595wd_expand (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
1596{
1597 if (!wd || *path == '/')
1598 return path;
1599
1600 if (path [0] == '.' && !path [1])
1601 return wd->str;
1602
1603 {
1604 int l1 = wd->len;
1605 int l2 = strlen (path);
1606
1607 char *res = etp_tmpbuf_get (tmpbuf, l1 + l2 + 2);
1608
1609 memcpy (res, wd->str, l1);
1610 res [l1] = '/';
1611 memcpy (res + l1 + 1, path, l2 + 1);
1612
1613 return res;
1614 }
1615}
1616
1617#endif
1618
1619static eio_wd
1620eio__wd_open_sync (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
1621{
1622 int fd;
1623 eio_wd res;
1624 int len = eio__realpath (tmpbuf, wd, path);
1625
1626 if (len < 0)
1627 return EIO_INVALID_WD;
1628
1629#if HAVE_AT
1630 fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1631
1632 if (fd < 0)
1633 return EIO_INVALID_WD;
1634#endif
1635
1636 res = malloc (sizeof (*res) + len); /* one extra 0-byte */
1637
1638#if HAVE_AT
1639 res->fd = fd;
1640#endif
1641
1642 res->len = len;
1643 memcpy (res->str, tmpbuf->ptr, len);
1644 res->str [len] = 0;
1645
1646 return res;
1647}
1648
1649eio_wd
1650eio_wd_open_sync (eio_wd wd, const char *path)
1651{
1652 struct etp_tmpbuf tmpbuf = { };
1653 wd = eio__wd_open_sync (&tmpbuf, wd, path);
1654 free (tmpbuf.ptr);
1655
1656 return wd;
1150} 1657}
1151 1658
1152void 1659void
1153aio_readlink (SV8 *path, SV *callback=&PL_sv_undef) 1660eio_wd_close_sync (eio_wd wd)
1154 PROTOTYPE: $$;$
1155 PPCODE:
1156{ 1661{
1157 SV *data; 1662 if (wd != EIO_INVALID_WD && wd != EIO_CWD)
1158 dREQ; 1663 {
1159 1664 #if HAVE_AT
1160 data = newSV (NAME_MAX); 1665 close (wd->fd);
1161 SvPOK_on (data); 1666 #endif
1162 1667 free (wd);
1163 req->type = EIO_READLINK; 1668 }
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} 1669}
1670
1671#if HAVE_AT
1672
1673/* they forgot these */
1674
1675static int
1676eio__truncateat (int dirfd, const char *path, off_t length)
1677{
1678 int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
1679 int res;
1680
1681 if (fd < 0)
1682 return fd;
1683
1684 res = ftruncate (fd, length);
1685 close (fd);
1686 return res;
1687}
1688
1689static int
1690eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
1691{
1692 int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
1693 int res;
1694
1695 if (fd < 0)
1696 return fd;
1697
1698 res = fstatvfs (fd, buf);
1699 close (fd);
1700 return res;
1701
1702}
1703
1704#endif
1705
1706/*****************************************************************************/
1707
1708#define ALLOC(len) \
1709 if (!req->ptr2) \
1710 { \
1711 X_LOCK (wrklock); \
1712 req->flags |= EIO_FLAG_PTR2_FREE; \
1713 X_UNLOCK (wrklock); \
1714 req->ptr2 = malloc (len); \
1715 if (!req->ptr2) \
1716 { \
1717 errno = ENOMEM; \
1718 req->result = -1; \
1719 break; \
1720 } \
1721 }
1722
1723/*****************************************************************************/
1724
1725int ecb_cold
1726eio_init (void (*want_poll)(void), void (*done_poll)(void))
1727{
1728 return etp_init (EIO_POOL, want_poll, done_poll);
1729}
1730
1731ecb_inline void
1732eio_api_destroy (eio_req *req)
1733{
1734 free (req);
1735}
1736
1737#define REQ(rtype) \
1738 eio_req *req; \
1739 \
1740 req = (eio_req *)calloc (1, sizeof *req); \
1741 if (!req) \
1742 return 0; \
1743 \
1744 req->type = rtype; \
1745 req->pri = pri; \
1746 req->finish = cb; \
1747 req->data = data; \
1748 req->destroy = eio_api_destroy;
1749
1750#define SEND eio_submit (req); return req
1751
1752#define PATH \
1753 req->flags |= EIO_FLAG_PTR1_FREE; \
1754 req->ptr1 = strdup (path); \
1755 if (!req->ptr1) \
1756 { \
1757 eio_api_destroy (req); \
1758 return 0; \
1759 }
1760
1761#define SINGLEDOT(ptr) (0[(char *)(ptr)] == '.' && !1[(char *)(ptr)])
1762
1763static void
1764eio_execute (etp_worker *self, eio_req *req)
1765{
1766#if HAVE_AT
1767 int dirfd;
1768#else
1769 const char *path;
1770#endif
1771
1772 if (ecb_expect_false (EIO_CANCELLED (req)))
1773 {
1774 req->result = -1;
1775 req->errorno = ECANCELED;
1776 return;
1777 }
1778
1779 if (ecb_expect_false (req->wd == EIO_INVALID_WD))
1780 {
1781 req->result = -1;
1782 req->errorno = ENOENT;
1783 return;
1784 }
1785
1786 if (req->type >= EIO_OPEN)
1787 {
1788 #if HAVE_AT
1789 dirfd = WD2FD (req->wd);
1790 #else
1791 path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1792 #endif
1793 }
1794
1795 switch (req->type)
1796 {
1797 case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
1798 req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
1799 break;
1800 case EIO_WD_CLOSE: req->result = 0;
1801 eio_wd_close_sync (req->wd); break;
1802
1803 case EIO_SEEK: eio__lseek (req); break;
1804 case EIO_READ: ALLOC (req->size);
1805 req->result = req->offs >= 0
1806 ? pread (req->int1, req->ptr2, req->size, req->offs)
1807 : read (req->int1, req->ptr2, req->size); break;
1808 case EIO_WRITE: req->result = req->offs >= 0
1809 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1810 : write (req->int1, req->ptr2, req->size); break;
1811
1812 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1813 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
1814
1815#if HAVE_AT
1816
1817 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1818 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break;
1819 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1820 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, AT_SYMLINK_NOFOLLOW); break;
1821 case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break;
1822 case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break;
1823 case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break;
1824 case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break;
1825
1826 case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
1827 case EIO_RMDIR: /* complications arise because "." cannot be removed, so we might have to expand */
1828 req->result = req->wd && SINGLEDOT (req->ptr1)
1829 ? rmdir (req->wd->str)
1830 : unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break;
1831 case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); break;
1832 case EIO_RENAME: /* complications arise because "." cannot be renamed, so we might have to expand */
1833 req->result = req->wd && SINGLEDOT (req->ptr1)
1834 ? rename (req->wd->str, req->ptr2)
1835 : renameat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2); break;
1836 case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2, 0); break;
1837 case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
1838 case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1839 case EIO_READLINK: ALLOC (PATH_MAX);
1840 req->result = readlinkat (dirfd, req->ptr1, req->ptr2, PATH_MAX); break;
1841 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1842 req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1843 case EIO_UTIME:
1844 case EIO_FUTIME:
1845 {
1846 struct timespec ts[2];
1847 struct timespec *times;
1848
1849 if (req->nv1 != -1. || req->nv2 != -1.)
1850 {
1851 ts[0].tv_sec = req->nv1;
1852 ts[0].tv_nsec = (req->nv1 - ts[0].tv_sec) * 1e9;
1853 ts[1].tv_sec = req->nv2;
1854 ts[1].tv_nsec = (req->nv2 - ts[1].tv_sec) * 1e9;
1855
1856 times = ts;
1857 }
1858 else
1859 times = 0;
1860
1861 req->result = req->type == EIO_FUTIME
1862 ? futimens (req->int1, times)
1863 : utimensat (dirfd, req->ptr1, times, 0);
1864 }
1865 break;
1866
1867#else
1868
1869 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1870 req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
1871 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1872 req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
1873 case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
1874 case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
1875 case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
1876 case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
1877
1878 case EIO_UNLINK: req->result = unlink (path ); break;
1879 case EIO_RMDIR: req->result = rmdir (path ); break;
1880 case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
1881 case EIO_RENAME: req->result = rename (path , req->ptr2); break;
1882 case EIO_LINK: req->result = link (path , req->ptr2); break;
1883 case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
1884 case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
1885 case EIO_READLINK: ALLOC (PATH_MAX);
1886 req->result = readlink (path, req->ptr2, PATH_MAX); break;
1887 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1888 req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
1889
1890 case EIO_UTIME:
1891 case EIO_FUTIME:
1892 {
1893 struct timeval tv[2];
1894 struct timeval *times;
1895
1896 if (req->nv1 != -1. || req->nv2 != -1.)
1897 {
1898 tv[0].tv_sec = req->nv1;
1899 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
1900 tv[1].tv_sec = req->nv2;
1901 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
1902
1903 times = tv;
1904 }
1905 else
1906 times = 0;
1907
1908 req->result = req->type == EIO_FUTIME
1909 ? futimes (req->int1, times)
1910 : utimes (req->ptr1, times);
1911 }
1912 break;
1913
1914#endif
1915
1916 case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
1917 {
1918 ALLOC (req->result);
1919 memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
1920 }
1921 break;
1922
1923 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1924 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1925
1926 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1927 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1928
1929 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1930 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1931 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1932
1933 case EIO_CLOSE: req->result = close (req->int1); break;
1934 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1935 case EIO_SYNC: req->result = 0; sync (); break;
1936 case EIO_FSYNC: req->result = fsync (req->int1); break;
1937 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1938 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
1939 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1940 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1941 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1942 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1943 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1944 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
1945
1946 case EIO_READDIR: eio__scandir (req, self); break;
1947
1948 case EIO_BUSY:
1949#ifdef _WIN32
1950 Sleep (req->nv1 * 1e3);
1951#else
1952 {
1953 struct timeval tv;
1954
1955 tv.tv_sec = req->nv1;
1956 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1957
1958 req->result = select (0, 0, 0, 0, &tv);
1959 }
1960#endif
1961 break;
1962
1963#if 0
1964 case EIO_GROUP:
1965 abort (); /* handled in eio_request */
1966#endif
1967
1968 case EIO_NOP:
1969 req->result = 0;
1970 break;
1971
1972 case EIO_CUSTOM:
1973 req->feed (req);
1974 break;
1975
1976 default:
1977 req->result = EIO_ENOSYS ();
1978 break;
1979 }
1980
1981 req->errorno = errno;
1982}
1983
1984#ifndef EIO_NO_WRAPPERS
1985
1986eio_req *eio_wd_open (const char *path, int pri, eio_cb cb, void *data)
1987{
1988 REQ (EIO_WD_OPEN); PATH; SEND;
1989}
1990
1991eio_req *eio_wd_close (eio_wd wd, int pri, eio_cb cb, void *data)
1992{
1993 REQ (EIO_WD_CLOSE); req->wd = wd; SEND;
1994}
1995
1996eio_req *eio_nop (int pri, eio_cb cb, void *data)
1997{
1998 REQ (EIO_NOP); SEND;
1999}
2000
2001eio_req *eio_busy (double delay, int pri, eio_cb cb, void *data)
2002{
2003 REQ (EIO_BUSY); req->nv1 = delay; SEND;
2004}
2005
2006eio_req *eio_sync (int pri, eio_cb cb, void *data)
2007{
2008 REQ (EIO_SYNC); SEND;
2009}
2010
2011eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
2012{
2013 REQ (EIO_FSYNC); req->int1 = fd; SEND;
2014}
2015
2016eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2017{
2018 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2019}
2020
2021eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2022{
2023 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2024}
2025
2026eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
2027{
2028 REQ (EIO_SYNCFS); req->int1 = fd; SEND;
2029}
2030
2031eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2032{
2033 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2034}
2035
2036eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2037{
2038 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2039}
2040
2041eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
2042{
2043 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
2044}
2045
2046eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
2047{
2048 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2049}
2050
2051eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2052{
2053 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
2054}
2055
2056eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
2057{
2058 REQ (EIO_CLOSE); req->int1 = fd; SEND;
2059}
2060
2061eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
2062{
2063 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
2064}
2065
2066eio_req *eio_seek (int fd, off_t offset, int whence, int pri, eio_cb cb, void *data)
2067{
2068 REQ (EIO_SEEK); req->int1 = fd; req->offs = offset; req->int2 = whence; SEND;
2069}
2070
2071eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
2072{
2073 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
2074}
2075
2076eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
2077{
2078 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
2079}
2080
2081eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
2082{
2083 REQ (EIO_FSTAT); req->int1 = fd; SEND;
2084}
2085
2086eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
2087{
2088 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
2089}
2090
2091eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
2092{
2093 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
2094}
2095
2096eio_req *eio_ftruncate (int fd, off_t offset, int pri, eio_cb cb, void *data)
2097{
2098 REQ (EIO_FTRUNCATE); req->int1 = fd; req->offs = offset; SEND;
2099}
2100
2101eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
2102{
2103 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
2104}
2105
2106eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
2107{
2108 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
2109}
2110
2111eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
2112{
2113 REQ (EIO_DUP2); req->int1 = fd; req->int2 = fd2; SEND;
2114}
2115
2116eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, int pri, eio_cb cb, void *data)
2117{
2118 REQ (EIO_SENDFILE); req->int1 = out_fd; req->int2 = in_fd; req->offs = in_offset; req->size = length; SEND;
2119}
2120
2121eio_req *eio_open (const char *path, int flags, mode_t mode, int pri, eio_cb cb, void *data)
2122{
2123 REQ (EIO_OPEN); PATH; req->int1 = flags; req->int2 = (long)mode; SEND;
2124}
2125
2126eio_req *eio_utime (const char *path, double atime, double mtime, int pri, eio_cb cb, void *data)
2127{
2128 REQ (EIO_UTIME); PATH; req->nv1 = atime; req->nv2 = mtime; SEND;
2129}
2130
2131eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
2132{
2133 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
2134}
2135
2136eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
2137{
2138 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
2139}
2140
2141eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
2142{
2143 REQ (EIO_CHMOD); PATH; req->int2 = (long)mode; SEND;
2144}
2145
2146eio_req *eio_mkdir (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
2147{
2148 REQ (EIO_MKDIR); PATH; req->int2 = (long)mode; SEND;
2149}
2150
2151static eio_req *
2152eio__1path (int type, const char *path, int pri, eio_cb cb, void *data)
2153{
2154 REQ (type); PATH; SEND;
2155}
2156
2157eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
2158{
2159 return eio__1path (EIO_READLINK, path, pri, cb, data);
2160}
2161
2162eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2163{
2164 return eio__1path (EIO_REALPATH, path, pri, cb, data);
2165}
2166
2167eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
2168{
2169 return eio__1path (EIO_STAT, path, pri, cb, data);
2170}
2171
2172eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
2173{
2174 return eio__1path (EIO_LSTAT, path, pri, cb, data);
2175}
2176
2177eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
2178{
2179 return eio__1path (EIO_STATVFS, path, pri, cb, data);
2180}
2181
2182eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
2183{
2184 return eio__1path (EIO_UNLINK, path, pri, cb, data);
2185}
2186
2187eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
2188{
2189 return eio__1path (EIO_RMDIR, path, pri, cb, data);
2190}
2191
2192eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
2193{
2194 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
2195}
2196
2197eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
2198{
2199 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
2200}
2201
2202static eio_req *
2203eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2204{
2205 REQ (type); PATH;
2206
2207 req->flags |= EIO_FLAG_PTR2_FREE;
2208 req->ptr2 = strdup (new_path);
2209 if (!req->ptr2)
2210 {
2211 eio_api_destroy (req);
2212 return 0;
2213 }
2214
2215 SEND;
2216}
2217
2218eio_req *eio_link (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2219{
2220 return eio__2path (EIO_LINK, path, new_path, pri, cb, data);
2221}
2222
2223eio_req *eio_symlink (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2224{
2225 return eio__2path (EIO_SYMLINK, path, new_path, pri, cb, data);
2226}
2227
2228eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2229{
2230 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
2231}
2232
2233eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
2234{
2235 REQ (EIO_CUSTOM); req->feed = execute; SEND;
2236}
2237
2238#endif
2239
2240eio_req *eio_grp (eio_cb cb, void *data)
2241{
2242 const int pri = EIO_PRI_MAX;
2243
2244 REQ (EIO_GROUP); SEND;
2245}
2246
2247#undef REQ
2248#undef PATH
2249#undef SEND
2250
2251/*****************************************************************************/
2252/* grp functions */
1171 2253
1172void 2254void
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) 2255eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1420{ 2256{
1421 grp->int2 = limit; 2257 grp->int2 = limit;
1422 grp->feed = feed; 2258 grp->feed = feed;
1423 2259
1424 grp_try_feed (grp); 2260 grp_try_feed (grp);
1425} 2261}
1426 2262
2263void
1427void eio_grp_limit (eio_req *grp, int limit) 2264eio_grp_limit (eio_req *grp, int limit)
1428{ 2265{
1429 grp->int2 = limit; 2266 grp->int2 = limit;
1430 2267
1431 grp_try_feed (grp); 2268 grp_try_feed (grp);
1432} 2269}
1433 2270
2271void
1434void eio_grp_add (eio_req *grp, eio_req *req) 2272eio_grp_add (eio_req *grp, eio_req *req)
1435{ 2273{
1436 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2274 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
2275
2276 grp->flags |= ETP_FLAG_GROUPADD;
1437 2277
1438 ++grp->size; 2278 ++grp->size;
1439 req->grp = grp; 2279 req->grp = grp;
1440 2280
1441 req->grp_prev = 0; 2281 req->grp_prev = 0;
… …
1445 grp->grp_first->grp_prev = req; 2285 grp->grp_first->grp_prev = req;
1446 2286
1447 grp->grp_first = req; 2287 grp->grp_first = req;
1448} 2288}
1449 2289
2290/*****************************************************************************/
2291/* misc garbage */
1450 2292
2293eio_ssize_t
2294eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2295{
2296 return eio__sendfile (ofd, ifd, offset, count);
2297}
2298

Diff Legend

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