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 1.129 by root, Sun Apr 14 09:43:19 2013 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
126 #define fchmod(fd,mode) EIO_ENOSYS ()
127 #define chown(path,uid,gid) EIO_ENOSYS ()
128 #define fchown(fd,uid,gid) EIO_ENOSYS ()
129 #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
130 #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
131 #define mknod(path,mode,dev) EIO_ENOSYS ()
132 #define sync() EIO_ENOSYS ()
133 #define readlink(path,buf,s) EIO_ENOSYS ()
134 #define statvfs(path,buf) EIO_ENOSYS ()
135 #define fstatvfs(fd,buf) EIO_ENOSYS ()
136
137 #define pread(fd,buf,count,offset) eio__pread (fd, buf, count, offset)
138 #define pwrite(fd,buf,count,offset) eio__pwrite (fd, buf, count, offset)
139
140 #if __GNUC__
141 typedef long long eio_off_t; /* signed for compatibility to msvc */
142 #else
143 typedef __int64 eio_off_t; /* unsigned not supported by msvc */
144 #endif
145
146 static eio_ssize_t
147 eio__pread (int fd, void *buf, eio_ssize_t count, eio_off_t offset)
148 {
149 OVERLAPPED o = { 0 };
150 DWORD got;
151
152 o.Offset = offset;
153 o.OffsetHigh = offset >> 32;
154
155 return ReadFile ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd), buf, count, &got, &o)
156 ? got : -1;
157 }
158
159 static eio_ssize_t
160 eio__pwrite (int fd, void *buf, eio_ssize_t count, eio_off_t offset)
161 {
162 OVERLAPPED o = { 0 };
163 DWORD got;
164
165 o.Offset = offset;
166 o.OffsetHigh = offset >> 32;
167
168 return WriteFile ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd), buf, count, &got, &o)
169 ? got : -1;
170 }
171
172 /* rename() uses MoveFile, which fails to overwrite */
173 #define rename(old,neu) eio__rename (old, neu)
174
175 static int
176 eio__rename (const char *old, const char *neu)
177 {
178 if (MoveFileEx (old, neu, MOVEFILE_REPLACE_EXISTING))
179 return 0;
180
181 /* should steal _dosmaperr */
182 switch (GetLastError ())
183 {
184 case ERROR_FILE_NOT_FOUND:
185 case ERROR_PATH_NOT_FOUND:
186 case ERROR_INVALID_DRIVE:
187 case ERROR_NO_MORE_FILES:
188 case ERROR_BAD_NETPATH:
189 case ERROR_BAD_NET_NAME:
190 case ERROR_BAD_PATHNAME:
191 case ERROR_FILENAME_EXCED_RANGE:
192 errno = ENOENT;
193 break;
194
195 default:
196 errno = EACCES;
197 break;
198 }
199
200 return -1;
201 }
202
203 /* we could even stat and see if it exists */
204 static int
205 symlink (const char *old, const char *neu)
206 {
207 #if WINVER >= 0x0600
208 if (CreateSymbolicLink (neu, old, 1))
209 return 0;
210
211 if (CreateSymbolicLink (neu, old, 0))
212 return 0;
213 #endif
214
215 return EIO_ERRNO (ENOENT, -1);
216 }
217
218 /* POSIX API only, causing trouble for win32 apps */
219 #define CreateHardLink(neu,old,flags) 0 /* not really creating hardlink, still using relative paths? */
220 #define CreateSymbolicLink(neu,old,flags) 0 /* vista+ only */
221
222 struct statvfs
223 {
224 int dummy;
225 };
226
227 #define DT_DIR EIO_DT_DIR
228 #define DT_REG EIO_DT_REG
229 #define D_NAME(entp) entp.cFileName
230 #define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
34 231
35#else 232#else
36 233
37# include "config.h"
38# include <sys/time.h> 234 #include <sys/time.h>
39# include <sys/select.h> 235 #include <sys/select.h>
40# include <unistd.h> 236 #include <unistd.h>
237 #include <signal.h>
238 #include <dirent.h>
239
240 #ifdef ANDROID
241 #include <sys/vfs.h>
242 #define statvfs statfs
243 #define fstatvfs fstatfs
244 #include <asm/page.h> /* supposedly limits.h does #define PAGESIZE PAGESIZE */
245 #else
246 #include <sys/statvfs.h>
247 #endif
248
249 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
250 #include <sys/mman.h>
251 #endif
252
253 #define D_NAME(entp) entp->d_name
254
255 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
256 #if __FreeBSD__ || __NetBSD__ || __OpenBSD__
257 #define _DIRENT_HAVE_D_TYPE /* sigh */
258 #define D_INO(de) (de)->d_fileno
259 #define D_NAMLEN(de) (de)->d_namlen
260 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
261 #define D_INO(de) (de)->d_ino
262 #endif
263
264 #ifdef _D_EXACT_NAMLEN
265 #undef D_NAMLEN
266 #define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
267 #endif
268
269 #ifdef _DIRENT_HAVE_D_TYPE
270 #define D_TYPE(de) (de)->d_type
271 #endif
272
273 #ifndef EIO_STRUCT_DIRENT
274 #define EIO_STRUCT_DIRENT struct dirent
275 #endif
276
277#endif
278
279#if HAVE_UTIMES
41# include <utime.h> 280# include <utime.h>
42# include <signal.h>
43
44# ifndef EIO_STRUCT_DIRENT
45# define EIO_STRUCT_DIRENT struct dirent
46# endif 281#endif
47 282
48#endif 283#if HAVE_SYS_SYSCALL_H
49 284# include <sys/syscall.h>
50# ifndef EIO_STRUCT_STAT
51# define EIO_STRUCT_STAT struct stat
52# endif 285#endif
286
287#if HAVE_SYS_PRCTL_H
288# include <sys/prctl.h>
289#endif
53 290
54#if HAVE_SENDFILE 291#if HAVE_SENDFILE
55# if __linux 292# if __linux
56# include <sys/sendfile.h> 293# include <sys/sendfile.h>
57# elif __freebsd 294# elif __FreeBSD__ || defined __APPLE__
58# include <sys/socket.h> 295# include <sys/socket.h>
59# include <sys/uio.h> 296# include <sys/uio.h>
60# elif __hpux 297# elif __hpux
61# include <sys/socket.h> 298# include <sys/socket.h>
62# elif __solaris /* not yet */ 299# elif __solaris
63# include <sys/sendfile.h> 300# include <sys/sendfile.h>
64# else 301# else
65# error sendfile support requested but not available 302# error sendfile support requested but not available
66# endif 303# endif
67#endif 304#endif
68 305
69/* number of seconds after which an idle threads exit */ 306#ifndef D_TYPE
70#define IDLE_TIMEOUT 10 307# define D_TYPE(de) 0
308#endif
309#ifndef D_INO
310# define D_INO(de) 0
311#endif
312#ifndef D_NAMLEN
313# define D_NAMLEN(entp) strlen (D_NAME (entp))
314#endif
71 315
72/* used for struct dirent, AIX doesn't provide it */ 316/* used for struct dirent, AIX doesn't provide it */
73#ifndef NAME_MAX 317#ifndef NAME_MAX
74# define NAME_MAX 4096 318# define NAME_MAX 4096
75#endif 319#endif
76 320
321/* used for readlink etc. */
322#ifndef PATH_MAX
323# define PATH_MAX 4096
324#endif
325
77/* buffer size for various temporary buffers */ 326/* buffer size for various temporary buffers */
78#define EIO_BUFSIZE 65536 327#define EIO_BUFSIZE 65536
79 328
80#define dBUF \ 329#define dBUF \
81 char *eio_buf; \
82 X_LOCK (wrklock); \
83 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 330 char *eio_buf = malloc (EIO_BUFSIZE); \
84 X_UNLOCK (wrklock); \ 331 errno = ENOMEM; \
85 if (!eio_buf) \ 332 if (!eio_buf) \
86 return -1; 333 return -1
87 334
88#define EIO_TICKS ((1000000 + 1023) >> 10) 335#define FUBd \
336 free (eio_buf)
89 337
90static void (*want_poll_cb) (void); 338/*****************************************************************************/
91static void (*done_poll_cb) (void);
92 339
93static unsigned int max_poll_time = 0; 340struct tmpbuf
94static unsigned int max_poll_reqs = 0;
95
96/* calculcate time difference in ~1/EIO_TICKS of a second */
97static int tvdiff (struct timeval *tv1, struct timeval *tv2)
98{ 341{
99 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 342 void *ptr;
100 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 343 int len;
101} 344};
102 345
103static unsigned int started, idle, wanted; 346static void *
104 347tmpbuf_get (struct tmpbuf *buf, int len)
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{ 348{
124 if (wrk->dirp) 349 if (buf->len < len)
125 { 350 {
126 closedir (wrk->dirp); 351 free (buf->ptr);
127 wrk->dirp = 0; 352 buf->ptr = malloc (buf->len = len);
128 } 353 }
129 354
130 if (wrk->dbuf) 355 return buf->ptr;
131 {
132 free (wrk->dbuf);
133 wrk->dbuf = 0;
134 }
135} 356}
136 357
137static void worker_free (worker *wrk) 358struct tmpbuf;
138{
139 wrk->next->prev = wrk->prev;
140 wrk->prev->next = wrk->next;
141 359
142 free (wrk); 360#if _POSIX_VERSION >= 200809L
143} 361 #define HAVE_AT 1
362 #define WD2FD(wd) ((wd) ? (wd)->fd : AT_FDCWD)
363 #ifndef O_SEARCH
364 #define O_SEARCH O_RDONLY
365 #endif
366#else
367 #define HAVE_AT 0
368 static const char *wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path);
369#endif
144 370
145static volatile unsigned int nreqs, nready, npending; 371struct eio_pwd
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{ 372{
154 return nreqs; 373#if HAVE_AT
155} 374 int fd;
156 375#endif
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; 376 int len;
198} reqq; 377 char str[1]; /* actually, a 0-terminated canonical path */
378};
199 379
200static reqq req_queue; 380/*****************************************************************************/
201static reqq res_queue;
202 381
203static int reqq_push (reqq *q, eio_req *req) 382#define ETP_PRI_MIN EIO_PRI_MIN
204{ 383#define ETP_PRI_MAX EIO_PRI_MAX
205 int pri = req->pri;
206 req->next = 0;
207 384
208 if (q->qe[pri]) 385#define ETP_TYPE_QUIT -1
209 { 386#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;
215 387
216 return q->size++; 388struct etp_worker;
217}
218 389
219static eio_req *reqq_shift (reqq *q) 390#define ETP_REQ eio_req
220{ 391#define ETP_DESTROY(req) eio_destroy (req)
221 int pri; 392static int eio_finish (eio_req *req);
393#define ETP_FINISH(req) eio_finish (req)
394static void eio_execute (struct etp_worker *self, eio_req *req);
395#define ETP_EXECUTE(wrk,req) eio_execute (wrk, req)
222 396
223 if (!q->size) 397#include "etp.c"
224 return 0;
225 398
226 --q->size; 399/*****************************************************************************/
227 400
228 for (pri = EIO_NUM_PRI; pri--; ) 401static 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) 402grp_try_feed (eio_req *grp)
245{ 403{
246 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 404 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
247 { 405 {
248 int old_len = grp->size; 406 grp->flags &= ~EIO_FLAG_GROUPADD;
249 407
250 EIO_FEED (grp); 408 EIO_FEED (grp);
251 409
252 /* stop if no progress has been made */ 410 /* stop if no progress has been made */
253 if (old_len == grp->size) 411 if (!(grp->flags & EIO_FLAG_GROUPADD))
254 { 412 {
255 grp->feed = 0; 413 grp->feed = 0;
256 break; 414 break;
257 } 415 }
258 } 416 }
259} 417}
260 418
261static int eio_finish (eio_req *req); 419static int
262
263static int grp_dec (eio_req *grp) 420grp_dec (eio_req *grp)
264{ 421{
265 --grp->size; 422 --grp->size;
266 423
267 /* call feeder, if applicable */ 424 /* call feeder, if applicable */
268 grp_try_feed (grp); 425 grp_try_feed (grp);
… …
272 return eio_finish (grp); 429 return eio_finish (grp);
273 else 430 else
274 return 0; 431 return 0;
275} 432}
276 433
434static void
277void eio_destroy (eio_req *req) 435eio_destroy (eio_req *req)
278{ 436{
437 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
279 if ((req)->flags & EIO_FLAG_PTR2_FREE) 438 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
280 free (req->ptr2);
281 439
282 EIO_DESTROY (req); 440 EIO_DESTROY (req);
283} 441}
284 442
443static int
285static int eio_finish (eio_req *req) 444eio_finish (eio_req *req)
286{ 445{
287 int res = EIO_FINISH (req); 446 int res = EIO_FINISH (req);
288 447
289 if (req->grp) 448 if (req->grp)
290 { 449 {
… …
298 if (grp->grp_first == req) 457 if (grp->grp_first == req)
299 grp->grp_first = req->grp_next; 458 grp->grp_first = req->grp_next;
300 459
301 res2 = grp_dec (grp); 460 res2 = grp_dec (grp);
302 461
303 if (!res && res2) 462 if (!res)
304 res = res2; 463 res = res2;
305 } 464 }
306 465
307 eio_destroy (req); 466 eio_destroy (req);
308 467
309 return res; 468 return res;
310} 469}
311 470
471void
312void eio_grp_cancel (eio_req *grp) 472eio_grp_cancel (eio_req *grp)
313{ 473{
314 for (grp = grp->grp_first; grp; grp = grp->grp_next)
315 eio_cancel (grp); 474 etp_grp_cancel (grp);
316} 475}
317 476
477void
318void eio_cancel (eio_req *req) 478eio_cancel (eio_req *req)
319{ 479{
320 req->flags |= EIO_FLAG_CANCELLED;
321
322 eio_grp_cancel (req); 480 etp_cancel (req);
323} 481}
324 482
325X_THREAD_PROC (eio_proc); 483void
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) 484eio_submit (eio_req *req)
363{ 485{
364 ++nreqs; 486 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} 487}
374 488
375static void end_thread (void) 489unsigned int
490eio_nreqs (void)
376{ 491{
377 eio_req *req = calloc (1, sizeof (eio_req)); 492 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} 493}
391 494
495unsigned int
496eio_nready (void)
497{
498 return etp_nready ();
499}
500
501unsigned int
502eio_npending (void)
503{
504 return etp_npending ();
505}
506
507unsigned int ecb_cold
508eio_nthreads (void)
509{
510 return etp_nthreads ();
511}
512
513void ecb_cold
392void eio_set_max_poll_time (double nseconds) 514eio_set_max_poll_time (double nseconds)
393{ 515{
394 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
395 max_poll_time = nseconds; 516 etp_set_max_poll_time (nseconds);
396 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
397} 517}
398 518
519void ecb_cold
399void eio_set_max_poll_reqs (unsigned int maxreqs) 520eio_set_max_poll_reqs (unsigned int maxreqs)
400{ 521{
401 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
402 max_poll_reqs = maxreqs; 522 etp_set_max_poll_reqs (maxreqs);
403 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
404} 523}
405 524
525void ecb_cold
406void eio_set_max_idle (unsigned int nthreads) 526eio_set_max_idle (unsigned int nthreads)
407{ 527{
408 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 528 etp_set_max_idle (nthreads);
409 max_idle = nthreads <= 0 ? 1 : nthreads;
410 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
411} 529}
412 530
531void ecb_cold
532eio_set_idle_timeout (unsigned int seconds)
533{
534 etp_set_idle_timeout (seconds);
535}
536
537void ecb_cold
413void eio_set_min_parallel (unsigned int nthreads) 538eio_set_min_parallel (unsigned int nthreads)
414{ 539{
415 if (wanted < nthreads) 540 etp_set_min_parallel (nthreads);
416 wanted = nthreads;
417} 541}
418 542
543void ecb_cold
419void eio_set_max_parallel (unsigned int nthreads) 544eio_set_max_parallel (unsigned int nthreads)
420{ 545{
421 if (wanted > nthreads) 546 etp_set_max_parallel (nthreads);
422 wanted = nthreads;
423
424 while (started > wanted)
425 end_thread ();
426} 547}
427 548
428int eio_poll (void) 549int eio_poll (void)
429{ 550{
430 int maxreqs = max_poll_reqs; 551 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} 552}
486 553
487/*****************************************************************************/ 554/*****************************************************************************/
488/* work around various missing functions */ 555/* work around various missing functions */
489 556
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 557#ifndef HAVE_UTIMES
533 558
559# undef utimes
534# define utimes(path,times) aio_utimes (path, times) 560# define utimes(path,times) eio__utimes (path, times)
535# define futimes(fd,times) aio_futimes (fd, times)
536 561
562static int
537static int aio_utimes (const char *filename, const struct timeval times[2]) 563eio__utimes (const char *filename, const struct timeval times[2])
538{ 564{
539 if (times) 565 if (times)
540 { 566 {
541 struct utimbuf buf; 567 struct utimbuf buf;
542 568
… …
547 } 573 }
548 else 574 else
549 return utime (filename, 0); 575 return utime (filename, 0);
550} 576}
551 577
578#endif
579
580#ifndef HAVE_FUTIMES
581
582# undef futimes
583# define futimes(fd,times) eio__futimes (fd, times)
584
585static int
552static int aio_futimes (int fd, const struct timeval tv[2]) 586eio__futimes (int fd, const struct timeval tv[2])
553{ 587{
554 errno = ENOSYS; 588 errno = ENOSYS;
555 return -1; 589 return -1;
556} 590}
557 591
558#endif 592#endif
559 593
560#if !HAVE_FDATASYNC 594#if !HAVE_FDATASYNC
595# undef fdatasync
561# define fdatasync fsync 596# define fdatasync(fd) fsync (fd)
562#endif 597#endif
598
599static int
600eio__syncfs (int fd)
601{
602 int res;
603
604#if HAVE_SYS_SYNCFS
605 res = (int)syscall (__NR_syncfs, (int)(fd));
606#else
607 res = EIO_ENOSYS ();
608#endif
609
610 if (res < 0 && errno == ENOSYS && fd >= 0)
611 sync ();
612
613 return res;
614}
615
616/* sync_file_range always needs emulation */
617static int
618eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
619{
620#if HAVE_SYNC_FILE_RANGE
621 int res;
622
623 if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
624 || EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
625 || EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
626 {
627 flags = 0
628 | (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
629 | (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
630 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
631 }
632
633 res = sync_file_range (fd, offset, nbytes, flags);
634
635 if (!res || errno != ENOSYS)
636 return res;
637#endif
638
639 /* even though we could play tricks with the flags, it's better to always
640 * call fdatasync, as that matches the expectation of its users best */
641 return fdatasync (fd);
642}
643
644static int
645eio__fallocate (int fd, int mode, off_t offset, size_t len)
646{
647#if HAVE_LINUX_FALLOCATE
648 return fallocate (fd, mode, offset, len);
649#else
650 return EIO_ENOSYS ();
651#endif
652}
563 653
564#if !HAVE_READAHEAD 654#if !HAVE_READAHEAD
655# undef readahead
565# define readahead(fd,offset,count) aio_readahead (fd, offset, count, self) 656# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
566 657
658static eio_ssize_t
567static ssize_t aio_readahead (int fd, off_t offset, size_t count, worker *self) 659eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
568{ 660{
569 size_t todo = count; 661 size_t todo = count;
570 dBUF; 662 dBUF;
571 663
572 while (todo > 0) 664 while (todo > 0)
573 { 665 {
574 size_t len = todo < EIO_BUFSIZE ? todo : EIO_BUFSIZE; 666 size_t len = todo < EIO_BUFSIZE ? todo : EIO_BUFSIZE;
575 667
576 pread (fd, aio_buf, len, offset); 668 pread (fd, eio_buf, len, offset);
577 offset += len; 669 offset += len;
578 todo -= len; 670 todo -= len;
579 } 671 }
580 672
581 errno = 0; 673 FUBd;
674
675 /* linux's readahead basically only fails for EBADF or EINVAL (not mmappable) */
676 /* but not for e.g. EIO or eof, so we also never fail */
582 return count; 677 return 0;
583} 678}
584 679
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 680#endif
616 681
617/* sendfile always needs emulation */ 682/* sendfile always needs emulation */
683static eio_ssize_t
618static ssize_t sendfile_ (int ofd, int ifd, off_t offset, size_t count, worker *self) 684eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
619{ 685{
686 eio_ssize_t written = 0;
620 ssize_t res; 687 eio_ssize_t res;
621 688
622 if (!count) 689 if (!count)
623 return 0; 690 return 0;
624 691
692 for (;;)
693 {
694#ifdef __APPLE__
695# undef HAVE_SENDFILE /* broken, as everything on os x */
696#endif
625#if HAVE_SENDFILE 697#if HAVE_SENDFILE
626# if __linux 698# if __linux
699 off_t soffset = offset;
627 res = sendfile (ofd, ifd, &offset, count); 700 res = sendfile (ofd, ifd, &soffset, count);
628 701
629# elif __freebsd 702# elif __FreeBSD__
630 /* 703 /*
631 * Of course, the freebsd sendfile is a dire hack with no thoughts 704 * Of course, the freebsd sendfile is a dire hack with no thoughts
632 * wasted on making it similar to other I/O functions. 705 * wasted on making it similar to other I/O functions.
633 */ 706 */
634 {
635 off_t sbytes; 707 off_t sbytes;
636 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 708 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
637 709
638 if (res < 0 && sbytes) 710 #if 0 /* according to the manpage, this is correct, but broken behaviour */
639 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 711 /* freebsd' sendfile will return 0 on success */
712 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
713 /* not on e.g. EIO or EPIPE - sounds broken */
714 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
640 res = sbytes; 715 res = sbytes;
641 } 716 #endif
717
718 /* according to source inspection, this is correct, and useful behaviour */
719 if (sbytes)
720 res = sbytes;
721
722# elif defined __APPLE__
723 off_t sbytes = count;
724 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
725
726 /* according to the manpage, sbytes is always valid */
727 if (sbytes)
728 res = sbytes;
642 729
643# elif __hpux 730# elif __hpux
644 res = sendfile (ofd, ifd, offset, count, 0, 0); 731 res = sendfile (ofd, ifd, offset, count, 0, 0);
645 732
646# elif __solaris 733# elif __solaris
647 {
648 struct sendfilevec vec; 734 struct sendfilevec vec;
649 size_t sbytes; 735 size_t sbytes;
650 736
651 vec.sfv_fd = ifd; 737 vec.sfv_fd = ifd;
652 vec.sfv_flag = 0; 738 vec.sfv_flag = 0;
653 vec.sfv_off = offset; 739 vec.sfv_off = offset;
654 vec.sfv_len = count; 740 vec.sfv_len = count;
655 741
656 res = sendfilev (ofd, &vec, 1, &sbytes); 742 res = sendfilev (ofd, &vec, 1, &sbytes);
657 743
658 if (res < 0 && sbytes) 744 if (res < 0 && sbytes)
659 res = sbytes; 745 res = sbytes;
660 }
661 746
662# endif 747# endif
748
749#elif defined (_WIN32) && 0
750 /* does not work, just for documentation of what would need to be done */
751 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
752 /* libeio guarantees that the file offset does not change, and windows */
753 /* has no way to get an independent handle to the same file description */
754 HANDLE h = TO_SOCKET (ifd);
755 SetFilePointer (h, offset, 0, FILE_BEGIN);
756 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
757
663#else 758#else
664 res = -1; 759 res = EIO_ENOSYS ();
665 errno = ENOSYS;
666#endif 760#endif
667 761
762 /* we assume sendfile can copy at least 128mb in one go */
763 if (res <= 128 * 1024 * 1024)
764 {
765 if (res > 0)
766 written += res;
767
768 if (written)
769 return written;
770
771 break;
772 }
773 else
774 {
775 /* if we requested more, then probably the kernel was lazy */
776 written += res;
777 offset += res;
778 count -= res;
779
780 if (!count)
781 return written;
782 }
783 }
784
668 if (res < 0 785 if (res < 0
669 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 786 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
787 /* BSDs */
788#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
789 || errno == ENOTSUP
790#endif
791#ifdef EOPNOTSUPP /* windows */
792 || errno == EOPNOTSUPP /* BSDs */
793#endif
670#if __solaris 794#if __solaris
671 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 795 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
672#endif 796#endif
673 ) 797 )
674 ) 798 )
… …
678 802
679 res = 0; 803 res = 0;
680 804
681 while (count) 805 while (count)
682 { 806 {
683 ssize_t cnt; 807 eio_ssize_t cnt;
684 808
685 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 809 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
686 810
687 if (cnt <= 0) 811 if (cnt <= 0)
688 { 812 {
… …
700 824
701 offset += cnt; 825 offset += cnt;
702 res += cnt; 826 res += cnt;
703 count -= cnt; 827 count -= cnt;
704 } 828 }
829
830 FUBd;
705 } 831 }
706 832
707 return res; 833 return res;
708} 834}
709 835
710/* read a full directory */ 836#ifdef PAGESIZE
711static void scandir_ (eio_req *req, worker *self) 837# define eio_pagesize() PAGESIZE
838#else
839static intptr_t
840eio_pagesize (void)
712{ 841{
713 DIR *dirp; 842 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;
724 843
725 X_LOCK (wrklock); 844 if (!page)
726 self->dirp = dirp = opendir (req->ptr1); 845 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);
731 846
732 if (dirp && u && names) 847 return page;
733 for (;;) 848}
849#endif
850
851static void
852eio_page_align (void **addr, size_t *length)
853{
854 intptr_t mask = eio_pagesize () - 1;
855
856 /* round down addr */
857 intptr_t adj = mask & (intptr_t)*addr;
858
859 *addr = (void *)((intptr_t)*addr - adj);
860 *length += adj;
861
862 /* round up length */
863 *length = (*length + mask) & ~mask;
864}
865
866#if !_POSIX_MEMLOCK
867# define eio__mlockall(a) EIO_ENOSYS ()
868#else
869
870static int
871eio__mlockall (int flags)
872{
873 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
874 extern int mallopt (int, int);
875 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
876 #endif
877
878 if (EIO_MCL_CURRENT != MCL_CURRENT
879 || EIO_MCL_FUTURE != MCL_FUTURE)
880 {
881 flags = 0
882 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
883 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
884 }
885
886 return mlockall (flags);
887}
888#endif
889
890#if !_POSIX_MEMLOCK_RANGE
891# define eio__mlock(a,b) EIO_ENOSYS ()
892#else
893
894static int
895eio__mlock (void *addr, size_t length)
896{
897 eio_page_align (&addr, &length);
898
899 return mlock (addr, length);
900}
901
902#endif
903
904#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
905# define eio__msync(a,b,c) EIO_ENOSYS ()
906#else
907
908static int
909eio__msync (void *mem, size_t len, int flags)
910{
911 eio_page_align (&mem, &len);
912
913 if (EIO_MS_ASYNC != MS_SYNC
914 || EIO_MS_INVALIDATE != MS_INVALIDATE
915 || EIO_MS_SYNC != MS_SYNC)
916 {
917 flags = 0
918 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
919 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
920 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
921 }
922
923 return msync (mem, len, flags);
924}
925
926#endif
927
928static int
929eio__mtouch (eio_req *req)
930{
931 void *mem = req->ptr2;
932 size_t len = req->size;
933 int flags = req->int1;
934
935 eio_page_align (&mem, &len);
936
937 {
938 intptr_t addr = (intptr_t)mem;
939 intptr_t end = addr + len;
940 intptr_t page = eio_pagesize ();
941
942 if (addr < end)
943 if (flags & EIO_MT_MODIFY) /* modify */
944 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
945 else
946 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
947 }
948
949 return 0;
950}
951
952/*****************************************************************************/
953/* requests implemented outside eio_execute, because they are so large */
954
955static void
956eio__lseek (eio_req *req)
957{
958 /* this usually gets optimised away completely, or your compiler sucks, */
959 /* or the whence constants really are not 0, 1, 2 */
960 int whence = req->int2 == EIO_SEEK_SET ? SEEK_SET
961 : req->int2 == EIO_SEEK_CUR ? SEEK_CUR
962 : req->int2 == EIO_SEEK_END ? SEEK_END
963 : req->int2;
964
965 req->offs = lseek (req->int1, req->offs, whence);
966 req->result = req->offs == (off_t)-1 ? -1 : 0;
967}
968
969/* result will always end up in tmpbuf, there is always space for adding a 0-byte */
970static int
971eio__realpath (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
972{
973 const char *rel = path;
974 char *res;
975 char *tmp1, *tmp2;
976#if SYMLOOP_MAX > 32
977 int symlinks = SYMLOOP_MAX;
978#else
979 int symlinks = 32;
980#endif
981
982 errno = EINVAL;
983 if (!rel)
984 return -1;
985
986 errno = ENOENT;
987 if (!*rel)
988 return -1;
989
990 res = tmpbuf_get (tmpbuf, PATH_MAX * 3);
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}
1124
1125static signed char
1126eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1127{
1128 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1129 : a->inode < b->inode ? -1
1130 : a->inode > b->inode ? 1
1131 : 0;
1132}
1133
1134#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1135
1136#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1137#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1138
1139static void
1140eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1141{
1142 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1143 unsigned char *bit = bits;
1144
1145 assert (CHAR_BIT == 8);
1146 assert (sizeof (eio_dirent) * 8 < 256);
1147 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1148 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1149
1150 if (size <= EIO_SORT_FAST)
1151 return;
1152
1153 /* first prepare an array of bits to test in our radix sort */
1154 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1155 /* inode_bits must contain all inodes ORed together */
1156 /* which is used to skip bits that are 0 everywhere, which is very common */
1157 {
1158 eio_ino_t endianness;
1159 int i, j;
1160
1161 /* we store the byte offset of byte n into byte n of "endianness" */
1162 for (i = 0; i < sizeof (eio_ino_t); ++i)
1163 ((unsigned char *)&endianness)[i] = i;
1164
1165 *bit++ = 0;
1166
1167 for (i = 0; i < sizeof (eio_ino_t); ++i)
1168 {
1169 /* shifting off the byte offsets out of "endianness" */
1170 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1171 endianness >>= 8;
1172
1173 for (j = 0; j < 8; ++j)
1174 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1175 *bit++ = offs + j;
1176 }
1177
1178 for (j = 0; j < 8; ++j)
1179 if (score_bits & (1 << j))
1180 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1181 }
1182
1183 /* now actually do the sorting (a variant of MSD radix sort) */
1184 {
1185 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1186 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1187 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1188 int stk_idx = 0;
1189
1190 base_stk [stk_idx] = dents;
1191 end_stk [stk_idx] = dents + size;
1192 bit_stk [stk_idx] = bit - 1;
1193
1194 do
1195 {
1196 base = base_stk [stk_idx];
1197 end = end_stk [stk_idx];
1198 bit = bit_stk [stk_idx];
1199
1200 for (;;)
1201 {
1202 unsigned char O = *bit >> 3;
1203 unsigned char M = 1 << (*bit & 7);
1204
1205 eio_dirent *a = base;
1206 eio_dirent *b = end;
1207
1208 if (b - a < EIO_SORT_CUTOFF)
1209 break;
1210
1211 /* now bit-partition the array on the bit */
1212 /* this ugly asymmetric loop seems to perform much better than typical */
1213 /* partition algos found in the literature */
1214 do
1215 if (!(((unsigned char *)a)[O] & M))
1216 ++a;
1217 else if (!(((unsigned char *)--b)[O] & M))
1218 {
1219 eio_dirent tmp = *a; *a = *b; *b = tmp;
1220 ++a;
1221 }
1222 while (b > a);
1223
1224 /* next bit, or stop, if no bits left in this path */
1225 if (!*--bit)
1226 break;
1227
1228 base_stk [stk_idx] = a;
1229 end_stk [stk_idx] = end;
1230 bit_stk [stk_idx] = bit;
1231 ++stk_idx;
1232
1233 end = a;
1234 }
1235 }
1236 while (stk_idx--);
1237 }
1238}
1239
1240static void
1241eio_dent_insertion_sort (eio_dirent *dents, int size)
1242{
1243 /* first move the smallest element to the front, to act as a sentinel */
1244 {
1245 int i;
1246 eio_dirent *min = dents;
1247
1248 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1249 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1250 if (EIO_DENT_CMP (dents [i], <, *min))
1251 min = &dents [i];
1252
1253 /* swap elements 0 and j (minimum) */
1254 {
1255 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1256 }
1257 }
1258
1259 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1260 {
1261 eio_dirent *i, *j;
1262
1263 for (i = dents + 1; i < dents + size; ++i)
1264 {
1265 eio_dirent value = *i;
1266
1267 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1268 j [1] = j [0];
1269
1270 j [1] = value;
1271 }
1272 }
1273}
1274
1275static void
1276eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1277{
1278 if (size <= 1)
1279 return; /* our insertion sort relies on size > 0 */
1280
1281 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1282 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1283 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1284
1285 /* use an insertion sort at the end, or for small arrays, */
1286 /* as insertion sort is more efficient for small partitions */
1287 eio_dent_insertion_sort (dents, size);
1288}
1289
1290/* read a full directory */
1291static void
1292eio__scandir (eio_req *req, etp_worker *self)
1293{
1294 char *name, *names;
1295 int namesalloc = 4096 - sizeof (void *) * 4;
1296 int namesoffs = 0;
1297 int flags = req->int1;
1298 eio_dirent *dents = 0;
1299 int dentalloc = 128;
1300 int dentoffs = 0;
1301 eio_ino_t inode_bits = 0;
1302#ifdef _WIN32
1303 HANDLE dirp;
1304 WIN32_FIND_DATA entp;
1305#else
1306 DIR *dirp;
1307 EIO_STRUCT_DIRENT *entp;
1308#endif
1309
1310 req->result = -1;
1311
1312 if (!(flags & EIO_READDIR_DENTS))
1313 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1314
1315#ifdef _WIN32
1316 {
1317 int len = strlen ((const char *)req->ptr1);
1318 char *path = malloc (MAX_PATH);
1319 const char *fmt;
1320 const char *reqpath = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1321
1322 if (!len)
1323 fmt = "./*";
1324 else if (reqpath[len - 1] == '/' || reqpath[len - 1] == '\\')
1325 fmt = "%s*";
1326 else
1327 fmt = "%s/*";
1328
1329 _snprintf (path, MAX_PATH, fmt, reqpath);
1330 dirp = FindFirstFile (path, &entp);
1331 free (path);
1332
1333 if (dirp == INVALID_HANDLE_VALUE)
1334 {
1335 /* should steal _dosmaperr */
1336 switch (GetLastError ())
1337 {
1338 case ERROR_FILE_NOT_FOUND:
1339 req->result = 0;
1340 break;
1341
1342 case ERROR_INVALID_NAME:
1343 case ERROR_PATH_NOT_FOUND:
1344 case ERROR_NO_MORE_FILES:
1345 errno = ENOENT;
1346 break;
1347
1348 case ERROR_NOT_ENOUGH_MEMORY:
1349 errno = ENOMEM;
1350 break;
1351
1352 default:
1353 errno = EINVAL;
1354 break;
1355 }
1356
1357 return;
1358 }
1359 }
1360#else
1361 #if HAVE_AT
1362 if (req->wd)
1363 {
1364 int fd = openat (WD2FD (req->wd), req->ptr1, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1365
1126 if (len < 0) 1366 if (fd < 0)
1127 croak ("length must not be negative"); 1367 return;
1128 1368
1369 dirp = fdopendir (fd);
1370
1371 if (!dirp)
1372 close (fd);
1373 }
1374 else
1375 dirp = opendir (req->ptr1);
1376 #else
1377 dirp = opendir (wd_expand (&self->tmpbuf, req->wd, req->ptr1));
1378 #endif
1379
1380 if (!dirp)
1381 return;
1382#endif
1383
1384 if (req->flags & EIO_FLAG_PTR1_FREE)
1385 free (req->ptr1);
1386
1387 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1388 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1389 req->ptr2 = names = malloc (namesalloc);
1390
1391 if (!names || (flags && !dents))
1392 return;
1393
1394 for (;;)
1395 {
1396 int done;
1397
1398#ifdef _WIN32
1399 done = !dirp;
1400#else
1401 errno = 0;
1402 entp = readdir (dirp);
1403 done = !entp;
1404#endif
1405
1406 if (done)
1129 { 1407 {
1130 dREQ; 1408#ifndef _WIN32
1409 int old_errno = errno;
1410 closedir (dirp);
1411 errno = old_errno;
1131 1412
1132 req->type = ix; 1413 if (errno)
1133 req->sv1 = newSVsv (fh); 1414 break;
1134 req->int1 = PerlIO_fileno (ix == EIO_READ ? IoIFP (sv_2io (fh)) 1415#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 1416
1142 if (!SvREADONLY (data)) 1417 /* sort etc. */
1418 req->int1 = flags;
1419 req->result = dentoffs;
1420
1421 if (flags & EIO_READDIR_STAT_ORDER)
1422 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1423 else if (flags & EIO_READDIR_DIRS_FIRST)
1424 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1425 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1426 else
1427 {
1428 /* in this case, all is known, and we just put dirs first and sort them */
1429 eio_dirent *oth = dents + dentoffs;
1430 eio_dirent *dir = dents;
1431
1432 /* now partition dirs to the front, and non-dirs to the back */
1433 /* by walking from both sides and swapping if necessary */
1434 while (oth > dir)
1435 {
1436 if (dir->type == EIO_DT_DIR)
1437 ++dir;
1438 else if ((--oth)->type == EIO_DT_DIR)
1439 {
1440 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1441
1442 ++dir;
1443 }
1444 }
1445
1446 /* now sort the dirs only (dirs all have the same score) */
1447 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1448 }
1449
1450 break;
1451 }
1452
1453 /* now add the entry to our list(s) */
1454 name = D_NAME (entp);
1455
1456 /* skip . and .. entries */
1457 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1458 {
1459 int len = D_NAMLEN (entp) + 1;
1460
1461 while (ecb_expect_false (namesoffs + len > namesalloc))
1143 { 1462 {
1144 SvREADONLY_on (data); 1463 namesalloc *= 2;
1145 req->flags |= FLAG_SV2_RO_OFF; 1464 req->ptr2 = names = realloc (names, namesalloc);
1465
1466 if (!names)
1467 break;
1146 } 1468 }
1147 1469
1148 EIO_SEND; 1470 memcpy (names + namesoffs, name, len);
1471
1472 if (dents)
1473 {
1474 struct eio_dirent *ent;
1475
1476 if (ecb_expect_false (dentoffs == dentalloc))
1477 {
1478 dentalloc *= 2;
1479 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1480
1481 if (!dents)
1482 break;
1483 }
1484
1485 ent = dents + dentoffs;
1486
1487 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1488 ent->namelen = len - 1;
1489 ent->inode = D_INO (entp);
1490
1491 inode_bits |= ent->inode;
1492
1493 switch (D_TYPE (entp))
1494 {
1495 default:
1496 ent->type = EIO_DT_UNKNOWN;
1497 flags |= EIO_READDIR_FOUND_UNKNOWN;
1498 break;
1499
1500 #ifdef DT_FIFO
1501 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1502 #endif
1503 #ifdef DT_CHR
1504 case DT_CHR: ent->type = EIO_DT_CHR; break;
1505 #endif
1506 #ifdef DT_MPC
1507 case DT_MPC: ent->type = EIO_DT_MPC; break;
1508 #endif
1509 #ifdef DT_DIR
1510 case DT_DIR: ent->type = EIO_DT_DIR; break;
1511 #endif
1512 #ifdef DT_NAM
1513 case DT_NAM: ent->type = EIO_DT_NAM; break;
1514 #endif
1515 #ifdef DT_BLK
1516 case DT_BLK: ent->type = EIO_DT_BLK; break;
1517 #endif
1518 #ifdef DT_MPB
1519 case DT_MPB: ent->type = EIO_DT_MPB; break;
1520 #endif
1521 #ifdef DT_REG
1522 case DT_REG: ent->type = EIO_DT_REG; break;
1523 #endif
1524 #ifdef DT_NWK
1525 case DT_NWK: ent->type = EIO_DT_NWK; break;
1526 #endif
1527 #ifdef DT_CMP
1528 case DT_CMP: ent->type = EIO_DT_CMP; break;
1529 #endif
1530 #ifdef DT_LNK
1531 case DT_LNK: ent->type = EIO_DT_LNK; break;
1532 #endif
1533 #ifdef DT_SOCK
1534 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1535 #endif
1536 #ifdef DT_DOOR
1537 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1538 #endif
1539 #ifdef DT_WHT
1540 case DT_WHT: ent->type = EIO_DT_WHT; break;
1541 #endif
1542 }
1543
1544 ent->score = 7;
1545
1546 if (flags & EIO_READDIR_DIRS_FIRST)
1547 {
1548 if (ent->type == EIO_DT_UNKNOWN)
1549 {
1550 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1551 ent->score = 1;
1552 else if (!strchr (name, '.')) /* absence of dots indicate likely dirs */
1553 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1554 }
1555 else if (ent->type == EIO_DT_DIR)
1556 ent->score = 0;
1557 }
1558 }
1559
1560 namesoffs += len;
1561 ++dentoffs;
1149 } 1562 }
1563
1564 if (EIO_CANCELLED (req))
1565 {
1566 errno = ECANCELED;
1567 break;
1568 }
1569
1570#ifdef _WIN32
1571 if (!FindNextFile (dirp, &entp))
1572 {
1573 FindClose (dirp);
1574 dirp = 0;
1575 }
1576#endif
1577 }
1578}
1579
1580/*****************************************************************************/
1581/* working directory stuff */
1582/* various deficiencies in the posix 2008 api force us to */
1583/* keep the absolute path in string form at all times */
1584/* fuck yeah. */
1585
1586#if !HAVE_AT
1587
1588/* a bit like realpath, but usually faster because it doesn'T have to return */
1589/* an absolute or canonical path */
1590static const char *
1591wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
1592{
1593 if (!wd || *path == '/')
1594 return path;
1595
1596 if (path [0] == '.' && !path [1])
1597 return wd->str;
1598
1599 {
1600 int l1 = wd->len;
1601 int l2 = strlen (path);
1602
1603 char *res = tmpbuf_get (tmpbuf, l1 + l2 + 2);
1604
1605 memcpy (res, wd->str, l1);
1606 res [l1] = '/';
1607 memcpy (res + l1 + 1, path, l2 + 1);
1608
1609 return res;
1610 }
1611}
1612
1613#endif
1614
1615static eio_wd
1616eio__wd_open_sync (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
1617{
1618 int fd;
1619 eio_wd res;
1620 int len = eio__realpath (tmpbuf, wd, path);
1621
1622 if (len < 0)
1623 return EIO_INVALID_WD;
1624
1625#if HAVE_AT
1626 fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1627
1628 if (fd < 0)
1629 return EIO_INVALID_WD;
1630#endif
1631
1632 res = malloc (sizeof (*res) + len); /* one extra 0-byte */
1633
1634#if HAVE_AT
1635 res->fd = fd;
1636#endif
1637
1638 res->len = len;
1639 memcpy (res->str, tmpbuf->ptr, len);
1640 res->str [len] = 0;
1641
1642 return res;
1643}
1644
1645eio_wd
1646eio_wd_open_sync (eio_wd wd, const char *path)
1647{
1648 struct tmpbuf tmpbuf = { 0 };
1649 wd = eio__wd_open_sync (&tmpbuf, wd, path);
1650 free (tmpbuf.ptr);
1651
1652 return wd;
1150} 1653}
1151 1654
1152void 1655void
1153aio_readlink (SV8 *path, SV *callback=&PL_sv_undef) 1656eio_wd_close_sync (eio_wd wd)
1154 PROTOTYPE: $$;$
1155 PPCODE:
1156{ 1657{
1157 SV *data; 1658 if (wd != EIO_INVALID_WD && wd != EIO_CWD)
1158 dREQ; 1659 {
1159 1660 #if HAVE_AT
1160 data = newSV (NAME_MAX); 1661 close (wd->fd);
1161 SvPOK_on (data); 1662 #endif
1162 1663 free (wd);
1163 req->type = EIO_READLINK; 1664 }
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} 1665}
1666
1667#if HAVE_AT
1668
1669/* they forgot these */
1670
1671static int
1672eio__truncateat (int dirfd, const char *path, off_t length)
1673{
1674 int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
1675 int res;
1676
1677 if (fd < 0)
1678 return fd;
1679
1680 res = ftruncate (fd, length);
1681 close (fd);
1682 return res;
1683}
1684
1685static int
1686eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
1687{
1688 int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
1689 int res;
1690
1691 if (fd < 0)
1692 return fd;
1693
1694 res = fstatvfs (fd, buf);
1695 close (fd);
1696 return res;
1697
1698}
1699
1700#endif
1701
1702/*****************************************************************************/
1703
1704#define ALLOC(len) \
1705 if (!req->ptr2) \
1706 { \
1707 X_LOCK (wrklock); \
1708 req->flags |= EIO_FLAG_PTR2_FREE; \
1709 X_UNLOCK (wrklock); \
1710 req->ptr2 = malloc (len); \
1711 if (!req->ptr2) \
1712 { \
1713 errno = ENOMEM; \
1714 req->result = -1; \
1715 break; \
1716 } \
1717 }
1718
1719static void ecb_noinline ecb_cold
1720etp_proc_init (void)
1721{
1722#if HAVE_PRCTL_SET_NAME
1723 /* provide a more sensible "thread name" */
1724 char name[16 + 1];
1725 const int namelen = sizeof (name) - 1;
1726 int len;
1727
1728 prctl (PR_GET_NAME, (unsigned long)name, 0, 0, 0);
1729 name [namelen] = 0;
1730 len = strlen (name);
1731 strcpy (name + (len <= namelen - 4 ? len : namelen - 4), "/eio");
1732 prctl (PR_SET_NAME, (unsigned long)name, 0, 0, 0);
1733#endif
1734}
1735
1736/* TODO: move somehow to etp.c */
1737X_THREAD_PROC (etp_proc)
1738{
1739 ETP_REQ *req;
1740 struct timespec ts;
1741 etp_worker *self = (etp_worker *)thr_arg;
1742
1743 etp_proc_init ();
1744
1745 /* try to distribute timeouts somewhat evenly */
1746 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1747
1748 for (;;)
1749 {
1750 ts.tv_sec = 0;
1751
1752 X_LOCK (reqlock);
1753
1754 for (;;)
1755 {
1756 req = reqq_shift (&req_queue);
1757
1758 if (req)
1759 break;
1760
1761 if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
1762 {
1763 X_UNLOCK (reqlock);
1764 X_LOCK (wrklock);
1765 --started;
1766 X_UNLOCK (wrklock);
1767 goto quit;
1768 }
1769
1770 ++idle;
1771
1772 if (idle <= max_idle)
1773 /* we are allowed to idle, so do so without any timeout */
1774 X_COND_WAIT (reqwait, reqlock);
1775 else
1776 {
1777 /* initialise timeout once */
1778 if (!ts.tv_sec)
1779 ts.tv_sec = time (0) + idle_timeout;
1780
1781 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
1782 ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1783 }
1784
1785 --idle;
1786 }
1787
1788 --nready;
1789
1790 X_UNLOCK (reqlock);
1791
1792 if (req->type == ETP_TYPE_QUIT)
1793 goto quit;
1794
1795 ETP_EXECUTE (self, req);
1796
1797 X_LOCK (reslock);
1798
1799 ++npending;
1800
1801 if (!reqq_push (&res_queue, req) && want_poll_cb)
1802 want_poll_cb ();
1803
1804 etp_worker_clear (self);
1805
1806 X_UNLOCK (reslock);
1807 }
1808
1809quit:
1810 free (req);
1811
1812 X_LOCK (wrklock);
1813 etp_worker_free (self);
1814 X_UNLOCK (wrklock);
1815
1816 return 0;
1817}
1818
1819/*****************************************************************************/
1820
1821int ecb_cold
1822eio_init (void (*want_poll)(void), void (*done_poll)(void))
1823{
1824 return etp_init (want_poll, done_poll);
1825}
1826
1827ecb_inline void
1828eio_api_destroy (eio_req *req)
1829{
1830 free (req);
1831}
1832
1833#define REQ(rtype) \
1834 eio_req *req; \
1835 \
1836 req = (eio_req *)calloc (1, sizeof *req); \
1837 if (!req) \
1838 return 0; \
1839 \
1840 req->type = rtype; \
1841 req->pri = pri; \
1842 req->finish = cb; \
1843 req->data = data; \
1844 req->destroy = eio_api_destroy;
1845
1846#define SEND eio_submit (req); return req
1847
1848#define PATH \
1849 req->flags |= EIO_FLAG_PTR1_FREE; \
1850 req->ptr1 = strdup (path); \
1851 if (!req->ptr1) \
1852 { \
1853 eio_api_destroy (req); \
1854 return 0; \
1855 }
1856
1857#define SINGLEDOT(ptr) (0[(char *)(ptr)] == '.' && !1[(char *)(ptr)])
1858
1859static void
1860eio_execute (etp_worker *self, eio_req *req)
1861{
1862#if HAVE_AT
1863 int dirfd;
1864#else
1865 const char *path;
1866#endif
1867
1868 if (ecb_expect_false (EIO_CANCELLED (req)))
1869 {
1870 req->result = -1;
1871 req->errorno = ECANCELED;
1872 return;
1873 }
1874
1875 if (ecb_expect_false (req->wd == EIO_INVALID_WD))
1876 {
1877 req->result = -1;
1878 req->errorno = ENOENT;
1879 return;
1880 }
1881
1882 if (req->type >= EIO_OPEN)
1883 {
1884 #if HAVE_AT
1885 dirfd = WD2FD (req->wd);
1886 #else
1887 path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1888 #endif
1889 }
1890
1891 switch (req->type)
1892 {
1893 case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
1894 req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
1895 break;
1896 case EIO_WD_CLOSE: req->result = 0;
1897 eio_wd_close_sync (req->wd); break;
1898
1899 case EIO_SEEK: eio__lseek (req); break;
1900 case EIO_READ: ALLOC (req->size);
1901 req->result = req->offs >= 0
1902 ? pread (req->int1, req->ptr2, req->size, req->offs)
1903 : read (req->int1, req->ptr2, req->size); break;
1904 case EIO_WRITE: req->result = req->offs >= 0
1905 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1906 : write (req->int1, req->ptr2, req->size); break;
1907
1908 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1909 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
1910
1911#if HAVE_AT
1912
1913 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1914 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break;
1915 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1916 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, AT_SYMLINK_NOFOLLOW); break;
1917 case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break;
1918 case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break;
1919 case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break;
1920 case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break;
1921
1922 case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
1923 case EIO_RMDIR: /* complications arise because "." cannot be removed, so we might have to expand */
1924 req->result = req->wd && SINGLEDOT (req->ptr1)
1925 ? rmdir (req->wd->str)
1926 : unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break;
1927 case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); break;
1928 case EIO_RENAME: /* complications arise because "." cannot be renamed, so we might have to expand */
1929 req->result = req->wd && SINGLEDOT (req->ptr1)
1930 ? rename (req->wd->str, req->ptr2)
1931 : renameat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2); break;
1932 case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2, 0); break;
1933 case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
1934 case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1935 case EIO_READLINK: ALLOC (PATH_MAX);
1936 req->result = readlinkat (dirfd, req->ptr1, req->ptr2, PATH_MAX); break;
1937 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1938 req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1939 case EIO_UTIME:
1940 case EIO_FUTIME:
1941 {
1942 struct timespec ts[2];
1943 struct timespec *times;
1944
1945 if (req->nv1 != -1. || req->nv2 != -1.)
1946 {
1947 ts[0].tv_sec = req->nv1;
1948 ts[0].tv_nsec = (req->nv1 - ts[0].tv_sec) * 1e9;
1949 ts[1].tv_sec = req->nv2;
1950 ts[1].tv_nsec = (req->nv2 - ts[1].tv_sec) * 1e9;
1951
1952 times = ts;
1953 }
1954 else
1955 times = 0;
1956
1957 req->result = req->type == EIO_FUTIME
1958 ? futimens (req->int1, times)
1959 : utimensat (dirfd, req->ptr1, times, 0);
1960 }
1961 break;
1962
1963#else
1964
1965 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1966 req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
1967 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1968 req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
1969 case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
1970 case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
1971 case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
1972 case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
1973
1974 case EIO_UNLINK: req->result = unlink (path ); break;
1975 case EIO_RMDIR: req->result = rmdir (path ); break;
1976 case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
1977 case EIO_RENAME: req->result = rename (path , req->ptr2); break;
1978 case EIO_LINK: req->result = link (path , req->ptr2); break;
1979 case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
1980 case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
1981 case EIO_READLINK: ALLOC (PATH_MAX);
1982 req->result = readlink (path, req->ptr2, PATH_MAX); break;
1983 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1984 req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
1985
1986 case EIO_UTIME:
1987 case EIO_FUTIME:
1988 {
1989 struct timeval tv[2];
1990 struct timeval *times;
1991
1992 if (req->nv1 != -1. || req->nv2 != -1.)
1993 {
1994 tv[0].tv_sec = req->nv1;
1995 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
1996 tv[1].tv_sec = req->nv2;
1997 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
1998
1999 times = tv;
2000 }
2001 else
2002 times = 0;
2003
2004 req->result = req->type == EIO_FUTIME
2005 ? futimes (req->int1, times)
2006 : utimes (req->ptr1, times);
2007 }
2008 break;
2009
2010#endif
2011
2012 case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
2013 {
2014 ALLOC (req->result);
2015 memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
2016 }
2017 break;
2018
2019 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2020 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
2021
2022 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2023 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2024
2025 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
2026 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
2027 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
2028
2029 case EIO_CLOSE: req->result = close (req->int1); break;
2030 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
2031 case EIO_SYNC: req->result = 0; sync (); break;
2032 case EIO_FSYNC: req->result = fsync (req->int1); break;
2033 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
2034 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
2035 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
2036 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
2037 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
2038 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
2039 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
2040 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
2041
2042 case EIO_READDIR: eio__scandir (req, self); break;
2043
2044 case EIO_BUSY:
2045#ifdef _WIN32
2046 Sleep (req->nv1 * 1e3);
2047#else
2048 {
2049 struct timeval tv;
2050
2051 tv.tv_sec = req->nv1;
2052 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
2053
2054 req->result = select (0, 0, 0, 0, &tv);
2055 }
2056#endif
2057 break;
2058
2059 case EIO_GROUP:
2060 abort (); /* handled in eio_request */
2061
2062 case EIO_NOP:
2063 req->result = 0;
2064 break;
2065
2066 case EIO_CUSTOM:
2067 req->feed (req);
2068 break;
2069
2070 default:
2071 req->result = EIO_ENOSYS ();
2072 break;
2073 }
2074
2075 req->errorno = errno;
2076}
2077
2078#ifndef EIO_NO_WRAPPERS
2079
2080eio_req *eio_wd_open (const char *path, int pri, eio_cb cb, void *data)
2081{
2082 REQ (EIO_WD_OPEN); PATH; SEND;
2083}
2084
2085eio_req *eio_wd_close (eio_wd wd, int pri, eio_cb cb, void *data)
2086{
2087 REQ (EIO_WD_CLOSE); req->wd = wd; SEND;
2088}
2089
2090eio_req *eio_nop (int pri, eio_cb cb, void *data)
2091{
2092 REQ (EIO_NOP); SEND;
2093}
2094
2095eio_req *eio_busy (double delay, int pri, eio_cb cb, void *data)
2096{
2097 REQ (EIO_BUSY); req->nv1 = delay; SEND;
2098}
2099
2100eio_req *eio_sync (int pri, eio_cb cb, void *data)
2101{
2102 REQ (EIO_SYNC); SEND;
2103}
2104
2105eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
2106{
2107 REQ (EIO_FSYNC); req->int1 = fd; SEND;
2108}
2109
2110eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2111{
2112 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2113}
2114
2115eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2116{
2117 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2118}
2119
2120eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
2121{
2122 REQ (EIO_SYNCFS); req->int1 = fd; SEND;
2123}
2124
2125eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2126{
2127 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2128}
2129
2130eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2131{
2132 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2133}
2134
2135eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
2136{
2137 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
2138}
2139
2140eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
2141{
2142 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2143}
2144
2145eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2146{
2147 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
2148}
2149
2150eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
2151{
2152 REQ (EIO_CLOSE); req->int1 = fd; SEND;
2153}
2154
2155eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
2156{
2157 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
2158}
2159
2160eio_req *eio_seek (int fd, off_t offset, int whence, int pri, eio_cb cb, void *data)
2161{
2162 REQ (EIO_SEEK); req->int1 = fd; req->offs = offset; req->int2 = whence; SEND;
2163}
2164
2165eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
2166{
2167 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
2168}
2169
2170eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
2171{
2172 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
2173}
2174
2175eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
2176{
2177 REQ (EIO_FSTAT); req->int1 = fd; SEND;
2178}
2179
2180eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
2181{
2182 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
2183}
2184
2185eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
2186{
2187 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
2188}
2189
2190eio_req *eio_ftruncate (int fd, off_t offset, int pri, eio_cb cb, void *data)
2191{
2192 REQ (EIO_FTRUNCATE); req->int1 = fd; req->offs = offset; SEND;
2193}
2194
2195eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
2196{
2197 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
2198}
2199
2200eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
2201{
2202 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
2203}
2204
2205eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
2206{
2207 REQ (EIO_DUP2); req->int1 = fd; req->int2 = fd2; SEND;
2208}
2209
2210eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, int pri, eio_cb cb, void *data)
2211{
2212 REQ (EIO_SENDFILE); req->int1 = out_fd; req->int2 = in_fd; req->offs = in_offset; req->size = length; SEND;
2213}
2214
2215eio_req *eio_open (const char *path, int flags, mode_t mode, int pri, eio_cb cb, void *data)
2216{
2217 REQ (EIO_OPEN); PATH; req->int1 = flags; req->int2 = (long)mode; SEND;
2218}
2219
2220eio_req *eio_utime (const char *path, double atime, double mtime, int pri, eio_cb cb, void *data)
2221{
2222 REQ (EIO_UTIME); PATH; req->nv1 = atime; req->nv2 = mtime; SEND;
2223}
2224
2225eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
2226{
2227 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
2228}
2229
2230eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
2231{
2232 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
2233}
2234
2235eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
2236{
2237 REQ (EIO_CHMOD); PATH; req->int2 = (long)mode; SEND;
2238}
2239
2240eio_req *eio_mkdir (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
2241{
2242 REQ (EIO_MKDIR); PATH; req->int2 = (long)mode; SEND;
2243}
2244
2245static eio_req *
2246eio__1path (int type, const char *path, int pri, eio_cb cb, void *data)
2247{
2248 REQ (type); PATH; SEND;
2249}
2250
2251eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
2252{
2253 return eio__1path (EIO_READLINK, path, pri, cb, data);
2254}
2255
2256eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2257{
2258 return eio__1path (EIO_REALPATH, path, pri, cb, data);
2259}
2260
2261eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
2262{
2263 return eio__1path (EIO_STAT, path, pri, cb, data);
2264}
2265
2266eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
2267{
2268 return eio__1path (EIO_LSTAT, path, pri, cb, data);
2269}
2270
2271eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
2272{
2273 return eio__1path (EIO_STATVFS, path, pri, cb, data);
2274}
2275
2276eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
2277{
2278 return eio__1path (EIO_UNLINK, path, pri, cb, data);
2279}
2280
2281eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
2282{
2283 return eio__1path (EIO_RMDIR, path, pri, cb, data);
2284}
2285
2286eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
2287{
2288 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
2289}
2290
2291eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
2292{
2293 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
2294}
2295
2296static eio_req *
2297eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2298{
2299 REQ (type); PATH;
2300
2301 req->flags |= EIO_FLAG_PTR2_FREE;
2302 req->ptr2 = strdup (new_path);
2303 if (!req->ptr2)
2304 {
2305 eio_api_destroy (req);
2306 return 0;
2307 }
2308
2309 SEND;
2310}
2311
2312eio_req *eio_link (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2313{
2314 return eio__2path (EIO_LINK, path, new_path, pri, cb, data);
2315}
2316
2317eio_req *eio_symlink (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2318{
2319 return eio__2path (EIO_SYMLINK, path, new_path, pri, cb, data);
2320}
2321
2322eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2323{
2324 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
2325}
2326
2327eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
2328{
2329 REQ (EIO_CUSTOM); req->feed = execute; SEND;
2330}
2331
2332#endif
2333
2334eio_req *eio_grp (eio_cb cb, void *data)
2335{
2336 const int pri = EIO_PRI_MAX;
2337
2338 REQ (EIO_GROUP); SEND;
2339}
2340
2341#undef REQ
2342#undef PATH
2343#undef SEND
2344
2345/*****************************************************************************/
2346/* grp functions */
1171 2347
1172void 2348void
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) 2349eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1420{ 2350{
1421 grp->int2 = limit; 2351 grp->int2 = limit;
1422 grp->feed = feed; 2352 grp->feed = feed;
1423 2353
1424 grp_try_feed (grp); 2354 grp_try_feed (grp);
1425} 2355}
1426 2356
2357void
1427void eio_grp_limit (eio_req *grp, int limit) 2358eio_grp_limit (eio_req *grp, int limit)
1428{ 2359{
1429 grp->int2 = limit; 2360 grp->int2 = limit;
1430 2361
1431 grp_try_feed (grp); 2362 grp_try_feed (grp);
1432} 2363}
1433 2364
2365void
1434void eio_grp_add (eio_req *grp, eio_req *req) 2366eio_grp_add (eio_req *grp, eio_req *req)
1435{ 2367{
1436 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2368 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
2369
2370 grp->flags |= EIO_FLAG_GROUPADD;
1437 2371
1438 ++grp->size; 2372 ++grp->size;
1439 req->grp = grp; 2373 req->grp = grp;
1440 2374
1441 req->grp_prev = 0; 2375 req->grp_prev = 0;
… …
1445 grp->grp_first->grp_prev = req; 2379 grp->grp_first->grp_prev = req;
1446 2380
1447 grp->grp_first = req; 2381 grp->grp_first = req;
1448} 2382}
1449 2383
2384/*****************************************************************************/
2385/* misc garbage */
1450 2386
2387eio_ssize_t
2388eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2389{
2390 return eio__sendfile (ofd, ifd, offset, count);
2391}
2392

Diff Legend

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