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 1.105 by root, Mon Sep 26 18:23:18 2011 UTC

1/*
2 * libeio implementation
3 *
4 * Copyright (c) 2007,2008,2009,2010,2011 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 /* rename() uses MoveFile, which fails to overwrite */
138 #define rename(old,neu) eio__rename (old, neu)
139
140 static int
141 eio__rename (const char *old, const char *neu)
142 {
143 if (MoveFileEx (old, neu, MOVEFILE_REPLACE_EXISTING))
144 return 0;
145
146 /* should steal _dosmaperr */
147 switch (GetLastError ())
148 {
149 case ERROR_FILE_NOT_FOUND:
150 case ERROR_PATH_NOT_FOUND:
151 case ERROR_INVALID_DRIVE:
152 case ERROR_NO_MORE_FILES:
153 case ERROR_BAD_NETPATH:
154 case ERROR_BAD_NET_NAME:
155 case ERROR_BAD_PATHNAME:
156 case ERROR_FILENAME_EXCED_RANGE:
157 errno = ENOENT;
158 break;
159
160 default:
161 errno = EACCES;
162 break;
163 }
164
165 return -1;
166 }
167
168 /* we could even stat and see if it exists */
169 static int
170 symlink (const char *old, const char *neu)
171 {
172 #if WINVER >= 0x0600
173 if (CreateSymbolicLink (neu, old, 1))
174 return 0;
175
176 if (CreateSymbolicLink (neu, old, 0))
177 return 0;
178 #endif
179
180 return EIO_ERRNO (ENOENT, -1);
181 }
182
183 /* POSIX API only */
184 #define CreateHardLink(neu,old,flags) 0
185 #define CreateSymbolicLink(neu,old,flags) 0
186
187 struct statvfs
188 {
189 int dummy;
190 };
191
192 #define DT_DIR EIO_DT_DIR
193 #define DT_REG EIO_DT_REG
194 #define D_NAME(entp) entp.cFileName
195 #define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
34 196
35#else 197#else
36 198
37# include "config.h"
38# include <sys/time.h> 199 #include <sys/time.h>
39# include <sys/select.h> 200 #include <sys/select.h>
201 #include <sys/statvfs.h>
40# include <unistd.h> 202 #include <unistd.h>
203 #include <signal.h>
204 #include <dirent.h>
205
206 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
207 #include <sys/mman.h>
208 #endif
209
210 #define D_NAME(entp) entp->d_name
211
212 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
213 #if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
214 #define _DIRENT_HAVE_D_TYPE /* sigh */
215 #define D_INO(de) (de)->d_fileno
216 #define D_NAMLEN(de) (de)->d_namlen
217 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
218 #define D_INO(de) (de)->d_ino
219 #endif
220
221 #ifdef _D_EXACT_NAMLEN
222 #undef D_NAMLEN
223 #define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
224 #endif
225
226 #ifdef _DIRENT_HAVE_D_TYPE
227 #define D_TYPE(de) (de)->d_type
228 #endif
229
230 #ifndef EIO_STRUCT_DIRENT
231 #define EIO_STRUCT_DIRENT struct dirent
232 #endif
233
234#endif
235
236#if HAVE_UTIMES
41# include <utime.h> 237# include <utime.h>
42# include <signal.h>
43
44# ifndef EIO_STRUCT_DIRENT
45# define EIO_STRUCT_DIRENT struct dirent
46# endif 238#endif
47 239
48#endif 240#if HAVE_SYS_SYSCALL_H
49 241# include <sys/syscall.h>
50# ifndef EIO_STRUCT_STAT
51# define EIO_STRUCT_STAT struct stat
52# endif 242#endif
243
244#if HAVE_SYS_PRCTL_H
245# include <sys/prctl.h>
246#endif
53 247
54#if HAVE_SENDFILE 248#if HAVE_SENDFILE
55# if __linux 249# if __linux
56# include <sys/sendfile.h> 250# include <sys/sendfile.h>
57# elif __freebsd 251# elif __FreeBSD__ || defined __APPLE__
58# include <sys/socket.h> 252# include <sys/socket.h>
59# include <sys/uio.h> 253# include <sys/uio.h>
60# elif __hpux 254# elif __hpux
61# include <sys/socket.h> 255# include <sys/socket.h>
62# elif __solaris /* not yet */ 256# elif __solaris
63# include <sys/sendfile.h> 257# include <sys/sendfile.h>
64# else 258# else
65# error sendfile support requested but not available 259# error sendfile support requested but not available
66# endif 260# endif
67#endif 261#endif
68 262
69/* number of seconds after which an idle threads exit */ 263#ifndef D_TYPE
70#define IDLE_TIMEOUT 10 264# define D_TYPE(de) 0
265#endif
266#ifndef D_INO
267# define D_INO(de) 0
268#endif
269#ifndef D_NAMLEN
270# define D_NAMLEN(entp) strlen (D_NAME (entp))
271#endif
71 272
72/* used for struct dirent, AIX doesn't provide it */ 273/* used for struct dirent, AIX doesn't provide it */
73#ifndef NAME_MAX 274#ifndef NAME_MAX
74# define NAME_MAX 4096 275# define NAME_MAX 4096
75#endif 276#endif
76 277
278/* used for readlink etc. */
279#ifndef PATH_MAX
280# define PATH_MAX 4096
281#endif
282
77/* buffer size for various temporary buffers */ 283/* buffer size for various temporary buffers */
78#define EIO_BUFSIZE 65536 284#define EIO_BUFSIZE 65536
79 285
80#define dBUF \ 286#define dBUF \
81 char *eio_buf; \
82 X_LOCK (wrklock); \
83 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 287 char *eio_buf = malloc (EIO_BUFSIZE); \
84 X_UNLOCK (wrklock); \ 288 errno = ENOMEM; \
85 if (!eio_buf) \ 289 if (!eio_buf) \
86 return -1; 290 return -1
291
292#define FUBd \
293 free (eio_buf)
87 294
88#define EIO_TICKS ((1000000 + 1023) >> 10) 295#define EIO_TICKS ((1000000 + 1023) >> 10)
296
297/*****************************************************************************/
298
299struct tmpbuf
300{
301 void *ptr;
302 int len;
303};
304
305static void *
306tmpbuf_get (struct tmpbuf *buf, int len)
307{
308 if (buf->len < len)
309 {
310 free (buf->ptr);
311 buf->ptr = malloc (buf->len = len);
312 }
313
314 return buf->ptr;
315}
316
317/*****************************************************************************/
318
319#define ETP_PRI_MIN EIO_PRI_MIN
320#define ETP_PRI_MAX EIO_PRI_MAX
321
322struct etp_worker;
323
324#define ETP_REQ eio_req
325#define ETP_DESTROY(req) eio_destroy (req)
326static int eio_finish (eio_req *req);
327#define ETP_FINISH(req) eio_finish (req)
328static void eio_execute (struct etp_worker *self, eio_req *req);
329#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
330
331/*****************************************************************************/
332
333#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
334
335/* calculate time difference in ~1/EIO_TICKS of a second */
336ecb_inline int
337tvdiff (struct timeval *tv1, struct timeval *tv2)
338{
339 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
340 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
341}
342
343static unsigned int started, idle, wanted = 4;
89 344
90static void (*want_poll_cb) (void); 345static void (*want_poll_cb) (void);
91static void (*done_poll_cb) (void); 346static void (*done_poll_cb) (void);
92 347
93static unsigned int max_poll_time = 0; 348static unsigned int max_poll_time; /* reslock */
94static unsigned int max_poll_reqs = 0; 349static unsigned int max_poll_reqs; /* reslock */
95 350
96/* calculcate time difference in ~1/EIO_TICKS of a second */ 351static unsigned int nreqs; /* reqlock */
97static int tvdiff (struct timeval *tv1, struct timeval *tv2) 352static unsigned int nready; /* reqlock */
98{ 353static unsigned int npending; /* reqlock */
99 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 354static unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
100 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 355static unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
101}
102 356
103static unsigned int started, idle, wanted; 357static xmutex_t wrklock;
358static xmutex_t reslock;
359static xmutex_t reqlock;
360static xcond_t reqwait;
361
362#if !HAVE_PREADWRITE
363/*
364 * make our pread/pwrite emulation safe against themselves, but not against
365 * normal read/write by using a mutex. slows down execution a lot,
366 * but that's your problem, not mine.
367 */
368static xmutex_t preadwritelock;
369#endif
370
371typedef struct etp_worker
372{
373 struct tmpbuf tmpbuf;
374
375 /* locked by wrklock */
376 struct etp_worker *prev, *next;
377
378 xthread_t tid;
379
380#ifdef ETP_WORKER_COMMON
381 ETP_WORKER_COMMON
382#endif
383} etp_worker;
384
385static etp_worker wrk_first; /* NOT etp */
386
387#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
388#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
104 389
105/* worker threads management */ 390/* worker threads management */
106static mutex_t wrklock = X_MUTEX_INIT;
107 391
108typedef struct worker { 392static void ecb_cold
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) 393etp_worker_clear (etp_worker *wrk)
123{ 394{
124 if (wrk->dirp) 395}
125 {
126 closedir (wrk->dirp);
127 wrk->dirp = 0;
128 }
129 396
130 if (wrk->dbuf) 397static void ecb_cold
131 { 398etp_worker_free (etp_worker *wrk)
399{
132 free (wrk->dbuf); 400 free (wrk->tmpbuf.ptr);
133 wrk->dbuf = 0;
134 }
135}
136 401
137static void worker_free (worker *wrk)
138{
139 wrk->next->prev = wrk->prev; 402 wrk->next->prev = wrk->prev;
140 wrk->prev->next = wrk->next; 403 wrk->prev->next = wrk->next;
141 404
142 free (wrk); 405 free (wrk);
143} 406}
144 407
145static volatile unsigned int nreqs, nready, npending; 408static unsigned int
146static volatile unsigned int max_idle = 4; 409etp_nreqs (void)
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{ 410{
411 int retval;
412 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
413 retval = nreqs;
414 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
154 return nreqs; 415 return retval;
155} 416}
156 417
157unsigned int eio_nready (void) 418static unsigned int
419etp_nready (void)
158{ 420{
159 unsigned int retval; 421 unsigned int retval;
160 422
161 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 423 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
162 retval = nready; 424 retval = nready;
163 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 425 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
164 426
165 return retval; 427 return retval;
166} 428}
167 429
168unsigned int eio_npending (void) 430static unsigned int
431etp_npending (void)
169{ 432{
170 unsigned int retval; 433 unsigned int retval;
171 434
172 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 435 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
173 retval = npending; 436 retval = npending;
174 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 437 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
175 438
176 return retval; 439 return retval;
177} 440}
178 441
179unsigned int eio_nthreads (void) 442static unsigned int
443etp_nthreads (void)
180{ 444{
181 unsigned int retval; 445 unsigned int retval;
182 446
183 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 447 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
184 retval = started; 448 retval = started;
191 * a somewhat faster data structure might be nice, but 455 * a somewhat faster data structure might be nice, but
192 * with 8 priorities this actually needs <20 insns 456 * with 8 priorities this actually needs <20 insns
193 * per shift, the most expensive operation. 457 * per shift, the most expensive operation.
194 */ 458 */
195typedef struct { 459typedef struct {
196 eio_req *qs[EIO_NUM_PRI], *qe[EIO_NUM_PRI]; /* qstart, qend */ 460 ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */
197 int size; 461 int size;
198} reqq; 462} etp_reqq;
199 463
200static reqq req_queue; 464static etp_reqq req_queue;
201static reqq res_queue; 465static etp_reqq res_queue;
202 466
203static int reqq_push (reqq *q, eio_req *req) 467static void ecb_noinline ecb_cold
468reqq_init (etp_reqq *q)
469{
470 int pri;
471
472 for (pri = 0; pri < ETP_NUM_PRI; ++pri)
473 q->qs[pri] = q->qe[pri] = 0;
474
475 q->size = 0;
476}
477
478static int ecb_noinline
479reqq_push (etp_reqq *q, ETP_REQ *req)
204{ 480{
205 int pri = req->pri; 481 int pri = req->pri;
206 req->next = 0; 482 req->next = 0;
207 483
208 if (q->qe[pri]) 484 if (q->qe[pri])
214 q->qe[pri] = q->qs[pri] = req; 490 q->qe[pri] = q->qs[pri] = req;
215 491
216 return q->size++; 492 return q->size++;
217} 493}
218 494
219static eio_req *reqq_shift (reqq *q) 495static ETP_REQ * ecb_noinline
496reqq_shift (etp_reqq *q)
220{ 497{
221 int pri; 498 int pri;
222 499
223 if (!q->size) 500 if (!q->size)
224 return 0; 501 return 0;
225 502
226 --q->size; 503 --q->size;
227 504
228 for (pri = EIO_NUM_PRI; pri--; ) 505 for (pri = ETP_NUM_PRI; pri--; )
229 { 506 {
230 eio_req *req = q->qs[pri]; 507 eio_req *req = q->qs[pri];
231 508
232 if (req) 509 if (req)
233 { 510 {
239 } 516 }
240 517
241 abort (); 518 abort ();
242} 519}
243 520
244static void grp_try_feed (eio_req *grp) 521static int ecb_cold
522etp_init (void (*want_poll)(void), void (*done_poll)(void))
245{ 523{
246 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 524 X_MUTEX_CREATE (wrklock);
247 { 525 X_MUTEX_CREATE (reslock);
248 int old_len = grp->size; 526 X_MUTEX_CREATE (reqlock);
527 X_COND_CREATE (reqwait);
249 528
250 EIO_FEED (grp); 529 reqq_init (&req_queue);
530 reqq_init (&res_queue);
251 531
252 /* stop if no progress has been made */ 532 wrk_first.next =
253 if (old_len == grp->size) 533 wrk_first.prev = &wrk_first;
254 {
255 grp->feed = 0;
256 break;
257 }
258 }
259}
260 534
261static int eio_finish (eio_req *req); 535 started = 0;
536 idle = 0;
537 nreqs = 0;
538 nready = 0;
539 npending = 0;
262 540
263static int grp_dec (eio_req *grp) 541 want_poll_cb = want_poll;
264{ 542 done_poll_cb = done_poll;
265 --grp->size;
266 543
267 /* call feeder, if applicable */
268 grp_try_feed (grp);
269
270 /* finish, if done */
271 if (!grp->size && grp->int1)
272 return eio_finish (grp);
273 else
274 return 0; 544 return 0;
275} 545}
276 546
277void eio_destroy (eio_req *req)
278{
279 if ((req)->flags & EIO_FLAG_PTR2_FREE)
280 free (req->ptr2);
281
282 EIO_DESTROY (req);
283}
284
285static int eio_finish (eio_req *req)
286{
287 int res = EIO_FINISH (req);
288
289 if (req->grp)
290 {
291 int res2;
292 eio_req *grp = req->grp;
293
294 /* unlink request */
295 if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
296 if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
297
298 if (grp->grp_first == req)
299 grp->grp_first = req->grp_next;
300
301 res2 = grp_dec (grp);
302
303 if (!res && res2)
304 res = res2;
305 }
306
307 eio_destroy (req);
308
309 return res;
310}
311
312void eio_grp_cancel (eio_req *grp)
313{
314 for (grp = grp->grp_first; grp; grp = grp->grp_next)
315 eio_cancel (grp);
316}
317
318void eio_cancel (eio_req *req)
319{
320 req->flags |= EIO_FLAG_CANCELLED;
321
322 eio_grp_cancel (req);
323}
324
325X_THREAD_PROC (eio_proc); 547X_THREAD_PROC (etp_proc);
326 548
327static void start_thread (void) 549static void ecb_cold
550etp_start_thread (void)
328{ 551{
329 worker *wrk = calloc (1, sizeof (worker)); 552 etp_worker *wrk = calloc (1, sizeof (etp_worker));
330 553
331 if (!wrk) 554 /*TODO*/
332 croak ("unable to allocate worker thread data"); 555 assert (("unable to allocate worker thread data", wrk));
333 556
334 X_LOCK (wrklock); 557 X_LOCK (wrklock);
335 558
336 if (thread_create (&wrk->tid, eio_proc, (void *)wrk)) 559 if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
337 { 560 {
338 wrk->prev = &wrk_first; 561 wrk->prev = &wrk_first;
339 wrk->next = wrk_first.next; 562 wrk->next = wrk_first.next;
340 wrk_first.next->prev = wrk; 563 wrk_first.next->prev = wrk;
341 wrk_first.next = wrk; 564 wrk_first.next = wrk;
345 free (wrk); 568 free (wrk);
346 569
347 X_UNLOCK (wrklock); 570 X_UNLOCK (wrklock);
348} 571}
349 572
573static void
350static void maybe_start_thread (void) 574etp_maybe_start_thread (void)
351{ 575{
352 if (eio_nthreads () >= wanted) 576 if (ecb_expect_true (etp_nthreads () >= wanted))
353 return; 577 return;
354 578
355 /* todo: maybe use idle here, but might be less exact */ 579 /* todo: maybe use idle here, but might be less exact */
356 if (0 <= (int)eio_nthreads () + (int)eio_npending () - (int)eio_nreqs ()) 580 if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
357 return; 581 return;
358 582
359 start_thread (); 583 etp_start_thread ();
360} 584}
361 585
362void eio_submit (eio_req *req) 586static void ecb_cold
587etp_end_thread (void)
363{ 588{
364 ++nreqs;
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}
374
375static void end_thread (void)
376{
377 eio_req *req = calloc (1, sizeof (eio_req)); 589 eio_req *req = calloc (1, sizeof (eio_req)); /* will be freed by worker */
378 590
379 req->type = EIO_QUIT; 591 req->type = -1;
380 req->pri = EIO_PRI_MAX + EIO_PRI_BIAS; 592 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
381 593
382 X_LOCK (reqlock); 594 X_LOCK (reqlock);
383 reqq_push (&req_queue, req); 595 reqq_push (&req_queue, req);
384 X_COND_SIGNAL (reqwait); 596 X_COND_SIGNAL (reqwait);
385 X_UNLOCK (reqlock); 597 X_UNLOCK (reqlock);
387 X_LOCK (wrklock); 599 X_LOCK (wrklock);
388 --started; 600 --started;
389 X_UNLOCK (wrklock); 601 X_UNLOCK (wrklock);
390} 602}
391 603
392void eio_set_max_poll_time (double nseconds) 604static int
393{
394 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
395 max_poll_time = nseconds;
396 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
397}
398
399void eio_set_max_poll_reqs (unsigned int maxreqs)
400{
401 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
402 max_poll_reqs = maxreqs;
403 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
404}
405
406void eio_set_max_idle (unsigned int nthreads)
407{
408 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
409 max_idle = nthreads <= 0 ? 1 : nthreads;
410 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
411}
412
413void eio_set_min_parallel (unsigned int nthreads)
414{
415 if (wanted < nthreads)
416 wanted = nthreads;
417}
418
419void eio_set_max_parallel (unsigned int nthreads)
420{
421 if (wanted > nthreads)
422 wanted = nthreads;
423
424 while (started > wanted)
425 end_thread ();
426}
427
428int eio_poll (void) 605etp_poll (void)
429{ 606{
430 int maxreqs = max_poll_reqs; 607 unsigned int maxreqs;
608 unsigned int maxtime;
431 struct timeval tv_start, tv_now; 609 struct timeval tv_start, tv_now;
432 eio_req *req;
433 610
611 X_LOCK (reslock);
612 maxreqs = max_poll_reqs;
613 maxtime = max_poll_time;
614 X_UNLOCK (reslock);
615
434 if (max_poll_time) 616 if (maxtime)
435 gettimeofday (&tv_start, 0); 617 gettimeofday (&tv_start, 0);
436 618
437 for (;;) 619 for (;;)
438 { 620 {
621 ETP_REQ *req;
622
439 maybe_start_thread (); 623 etp_maybe_start_thread ();
440 624
441 X_LOCK (reslock); 625 X_LOCK (reslock);
442 req = reqq_shift (&res_queue); 626 req = reqq_shift (&res_queue);
443 627
444 if (req) 628 if (req)
452 X_UNLOCK (reslock); 636 X_UNLOCK (reslock);
453 637
454 if (!req) 638 if (!req)
455 return 0; 639 return 0;
456 640
641 X_LOCK (reqlock);
457 --nreqs; 642 --nreqs;
643 X_UNLOCK (reqlock);
458 644
459 if (req->type == EIO_GROUP && req->size) 645 if (ecb_expect_false (req->type == EIO_GROUP && req->size))
460 { 646 {
461 req->int1 = 1; /* mark request as delayed */ 647 req->int1 = 1; /* mark request as delayed */
462 continue; 648 continue;
463 } 649 }
464 else 650 else
465 { 651 {
466 int res = eio_finish (req); 652 int res = ETP_FINISH (req);
467 if (res) 653 if (ecb_expect_false (res))
468 return res; 654 return res;
469 } 655 }
470 656
471 if (maxreqs && !--maxreqs) 657 if (ecb_expect_false (maxreqs && !--maxreqs))
472 break; 658 break;
473 659
474 if (max_poll_time) 660 if (maxtime)
475 { 661 {
476 gettimeofday (&tv_now, 0); 662 gettimeofday (&tv_now, 0);
477 663
478 if (tvdiff (&tv_start, &tv_now) >= max_poll_time) 664 if (tvdiff (&tv_start, &tv_now) >= maxtime)
479 break; 665 break;
480 } 666 }
481 } 667 }
482 668
483 errno = EAGAIN; 669 errno = EAGAIN;
484 return -1; 670 return -1;
485} 671}
486 672
673static void
674etp_cancel (ETP_REQ *req)
675{
676 req->cancelled = 1;
677
678 eio_grp_cancel (req);
679}
680
681static void
682etp_submit (ETP_REQ *req)
683{
684 req->pri -= ETP_PRI_MIN;
685
686 if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
687 if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
688
689 if (ecb_expect_false (req->type == EIO_GROUP))
690 {
691 /* I hope this is worth it :/ */
692 X_LOCK (reqlock);
693 ++nreqs;
694 X_UNLOCK (reqlock);
695
696 X_LOCK (reslock);
697
698 ++npending;
699
700 if (!reqq_push (&res_queue, req) && want_poll_cb)
701 want_poll_cb ();
702
703 X_UNLOCK (reslock);
704 }
705 else
706 {
707 X_LOCK (reqlock);
708 ++nreqs;
709 ++nready;
710 reqq_push (&req_queue, req);
711 X_COND_SIGNAL (reqwait);
712 X_UNLOCK (reqlock);
713
714 etp_maybe_start_thread ();
715 }
716}
717
718static void ecb_cold
719etp_set_max_poll_time (double nseconds)
720{
721 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
722 max_poll_time = nseconds * EIO_TICKS;
723 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
724}
725
726static void ecb_cold
727etp_set_max_poll_reqs (unsigned int maxreqs)
728{
729 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
730 max_poll_reqs = maxreqs;
731 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
732}
733
734static void ecb_cold
735etp_set_max_idle (unsigned int nthreads)
736{
737 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
738 max_idle = nthreads;
739 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
740}
741
742static void ecb_cold
743etp_set_idle_timeout (unsigned int seconds)
744{
745 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
746 idle_timeout = seconds;
747 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
748}
749
750static void ecb_cold
751etp_set_min_parallel (unsigned int nthreads)
752{
753 if (wanted < nthreads)
754 wanted = nthreads;
755}
756
757static void ecb_cold
758etp_set_max_parallel (unsigned int nthreads)
759{
760 if (wanted > nthreads)
761 wanted = nthreads;
762
763 while (started > wanted)
764 etp_end_thread ();
765}
766
767/*****************************************************************************/
768
769static void
770grp_try_feed (eio_req *grp)
771{
772 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
773 {
774 grp->flags &= ~EIO_FLAG_GROUPADD;
775
776 EIO_FEED (grp);
777
778 /* stop if no progress has been made */
779 if (!(grp->flags & EIO_FLAG_GROUPADD))
780 {
781 grp->feed = 0;
782 break;
783 }
784 }
785}
786
787static int
788grp_dec (eio_req *grp)
789{
790 --grp->size;
791
792 /* call feeder, if applicable */
793 grp_try_feed (grp);
794
795 /* finish, if done */
796 if (!grp->size && grp->int1)
797 return eio_finish (grp);
798 else
799 return 0;
800}
801
802static void
803eio_destroy (eio_req *req)
804{
805 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
806 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
807
808 EIO_DESTROY (req);
809}
810
811static int
812eio_finish (eio_req *req)
813{
814 int res = EIO_FINISH (req);
815
816 if (req->grp)
817 {
818 int res2;
819 eio_req *grp = req->grp;
820
821 /* unlink request */
822 if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
823 if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
824
825 if (grp->grp_first == req)
826 grp->grp_first = req->grp_next;
827
828 res2 = grp_dec (grp);
829
830 if (!res)
831 res = res2;
832 }
833
834 eio_destroy (req);
835
836 return res;
837}
838
839void
840eio_grp_cancel (eio_req *grp)
841{
842 for (grp = grp->grp_first; grp; grp = grp->grp_next)
843 eio_cancel (grp);
844}
845
846void
847eio_cancel (eio_req *req)
848{
849 etp_cancel (req);
850}
851
852void
853eio_submit (eio_req *req)
854{
855 etp_submit (req);
856}
857
858unsigned int
859eio_nreqs (void)
860{
861 return etp_nreqs ();
862}
863
864unsigned int
865eio_nready (void)
866{
867 return etp_nready ();
868}
869
870unsigned int
871eio_npending (void)
872{
873 return etp_npending ();
874}
875
876unsigned int ecb_cold
877eio_nthreads (void)
878{
879 return etp_nthreads ();
880}
881
882void ecb_cold
883eio_set_max_poll_time (double nseconds)
884{
885 etp_set_max_poll_time (nseconds);
886}
887
888void ecb_cold
889eio_set_max_poll_reqs (unsigned int maxreqs)
890{
891 etp_set_max_poll_reqs (maxreqs);
892}
893
894void ecb_cold
895eio_set_max_idle (unsigned int nthreads)
896{
897 etp_set_max_idle (nthreads);
898}
899
900void ecb_cold
901eio_set_idle_timeout (unsigned int seconds)
902{
903 etp_set_idle_timeout (seconds);
904}
905
906void ecb_cold
907eio_set_min_parallel (unsigned int nthreads)
908{
909 etp_set_min_parallel (nthreads);
910}
911
912void ecb_cold
913eio_set_max_parallel (unsigned int nthreads)
914{
915 etp_set_max_parallel (nthreads);
916}
917
918int eio_poll (void)
919{
920 return etp_poll ();
921}
922
487/*****************************************************************************/ 923/*****************************************************************************/
488/* work around various missing functions */ 924/* work around various missing functions */
489 925
490#if !HAVE_PREADWRITE 926#if !HAVE_PREADWRITE
927# undef pread
928# undef pwrite
491# define pread aio_pread 929# define pread eio__pread
492# define pwrite aio_pwrite 930# define pwrite eio__pwrite
493 931
494/* 932static eio_ssize_t
495 * make our pread/pwrite safe against themselves, but not against 933eio__pread (int fd, void *buf, size_t count, off_t offset)
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{ 934{
503 ssize_t res; 935 eio_ssize_t res;
504 off_t ooffset; 936 off_t ooffset;
505 937
506 X_LOCK (preadwritelock); 938 X_LOCK (preadwritelock);
507 ooffset = lseek (fd, 0, SEEK_CUR); 939 ooffset = lseek (fd, 0, SEEK_CUR);
508 lseek (fd, offset, SEEK_SET); 940 lseek (fd, offset, SEEK_SET);
511 X_UNLOCK (preadwritelock); 943 X_UNLOCK (preadwritelock);
512 944
513 return res; 945 return res;
514} 946}
515 947
516static ssize_t pwrite (int fd, void *buf, size_t count, off_t offset) 948static eio_ssize_t
949eio__pwrite (int fd, void *buf, size_t count, off_t offset)
517{ 950{
518 ssize_t res; 951 eio_ssize_t res;
519 off_t ooffset; 952 off_t ooffset;
520 953
521 X_LOCK (preadwritelock); 954 X_LOCK (preadwritelock);
522 ooffset = lseek (fd, 0, SEEK_CUR); 955 ooffset = lseek (fd, 0, SEEK_CUR);
523 lseek (fd, offset, SEEK_SET); 956 lseek (fd, offset, SEEK_SET);
524 res = write (fd, buf, count); 957 res = write (fd, buf, count);
525 lseek (fd, offset, SEEK_SET); 958 lseek (fd, ooffset, SEEK_SET);
526 X_UNLOCK (preadwritelock); 959 X_UNLOCK (preadwritelock);
527 960
528 return res; 961 return res;
529} 962}
530#endif 963#endif
531 964
532#ifndef HAVE_FUTIMES 965#ifndef HAVE_UTIMES
533 966
967# undef utimes
534# define utimes(path,times) aio_utimes (path, times) 968# define utimes(path,times) eio__utimes (path, times)
535# define futimes(fd,times) aio_futimes (fd, times)
536 969
970static int
537static int aio_utimes (const char *filename, const struct timeval times[2]) 971eio__utimes (const char *filename, const struct timeval times[2])
538{ 972{
539 if (times) 973 if (times)
540 { 974 {
541 struct utimbuf buf; 975 struct utimbuf buf;
542 976
547 } 981 }
548 else 982 else
549 return utime (filename, 0); 983 return utime (filename, 0);
550} 984}
551 985
986#endif
987
988#ifndef HAVE_FUTIMES
989
990# undef futimes
991# define futimes(fd,times) eio__futimes (fd, times)
992
993static int
552static int aio_futimes (int fd, const struct timeval tv[2]) 994eio__futimes (int fd, const struct timeval tv[2])
553{ 995{
554 errno = ENOSYS; 996 errno = ENOSYS;
555 return -1; 997 return -1;
556} 998}
557 999
558#endif 1000#endif
559 1001
560#if !HAVE_FDATASYNC 1002#if !HAVE_FDATASYNC
1003# undef fdatasync
561# define fdatasync fsync 1004# define fdatasync(fd) fsync (fd)
562#endif 1005#endif
1006
1007static int
1008eio__syncfs (int fd)
1009{
1010 int res;
1011
1012#if HAVE_SYS_SYNCFS
1013 res = (int)syscall (__NR_syncfs, (int)(fd));
1014#else
1015 res = -1;
1016 errno = ENOSYS;
1017#endif
1018
1019 if (res < 0 && errno == ENOSYS && fd >= 0)
1020 sync ();
1021
1022 return res;
1023}
1024
1025/* sync_file_range always needs emulation */
1026static int
1027eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
1028{
1029#if HAVE_SYNC_FILE_RANGE
1030 int res;
1031
1032 if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
1033 || EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
1034 || EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
1035 {
1036 flags = 0
1037 | (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
1038 | (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
1039 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
1040 }
1041
1042 res = sync_file_range (fd, offset, nbytes, flags);
1043
1044 if (!res || errno != ENOSYS)
1045 return res;
1046#endif
1047
1048 /* even though we could play tricks with the flags, it's better to always
1049 * call fdatasync, as that matches the expectation of its users best */
1050 return fdatasync (fd);
1051}
1052
1053static int
1054eio__fallocate (int fd, int mode, off_t offset, size_t len)
1055{
1056#if HAVE_FALLOCATE
1057 return fallocate (fd, mode, offset, len);
1058#else
1059 errno = ENOSYS;
1060 return -1;
1061#endif
1062}
563 1063
564#if !HAVE_READAHEAD 1064#if !HAVE_READAHEAD
1065# undef readahead
565# define readahead(fd,offset,count) aio_readahead (fd, offset, count, self) 1066# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
566 1067
1068static eio_ssize_t
567static ssize_t aio_readahead (int fd, off_t offset, size_t count, worker *self) 1069eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
568{ 1070{
569 size_t todo = count; 1071 size_t todo = count;
570 dBUF; 1072 dBUF;
571 1073
572 while (todo > 0) 1074 while (todo > 0)
573 { 1075 {
574 size_t len = todo < EIO_BUFSIZE ? todo : EIO_BUFSIZE; 1076 size_t len = todo < EIO_BUFSIZE ? todo : EIO_BUFSIZE;
575 1077
576 pread (fd, aio_buf, len, offset); 1078 pread (fd, eio_buf, len, offset);
577 offset += len; 1079 offset += len;
578 todo -= len; 1080 todo -= len;
579 } 1081 }
580 1082
1083 FUBd;
1084
581 errno = 0; 1085 errno = 0;
582 return count; 1086 return count;
583} 1087}
584 1088
585#endif 1089#endif
586 1090
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
616
617/* sendfile always needs emulation */ 1091/* sendfile always needs emulation */
1092static eio_ssize_t
618static ssize_t sendfile_ (int ofd, int ifd, off_t offset, size_t count, worker *self) 1093eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
619{ 1094{
1095 eio_ssize_t written = 0;
620 ssize_t res; 1096 eio_ssize_t res;
621 1097
622 if (!count) 1098 if (!count)
623 return 0; 1099 return 0;
624 1100
1101 for (;;)
1102 {
1103#ifdef __APPLE__
1104# undef HAVE_SENDFILE /* broken, as everything on os x */
1105#endif
625#if HAVE_SENDFILE 1106#if HAVE_SENDFILE
626# if __linux 1107# if __linux
1108 off_t soffset = offset;
627 res = sendfile (ofd, ifd, &offset, count); 1109 res = sendfile (ofd, ifd, &soffset, count);
628 1110
629# elif __freebsd 1111# elif __FreeBSD__
630 /* 1112 /*
631 * Of course, the freebsd sendfile is a dire hack with no thoughts 1113 * Of course, the freebsd sendfile is a dire hack with no thoughts
632 * wasted on making it similar to other I/O functions. 1114 * wasted on making it similar to other I/O functions.
633 */ 1115 */
634 {
635 off_t sbytes; 1116 off_t sbytes;
636 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 1117 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
637 1118
638 if (res < 0 && sbytes) 1119 #if 0 /* according to the manpage, this is correct, but broken behaviour */
639 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 1120 /* freebsd' sendfile will return 0 on success */
1121 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
1122 /* not on e.g. EIO or EPIPE - sounds broken */
1123 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
640 res = sbytes; 1124 res = sbytes;
641 } 1125 #endif
1126
1127 /* according to source inspection, this is correct, and useful behaviour */
1128 if (sbytes)
1129 res = sbytes;
1130
1131# elif defined (__APPLE__)
1132 off_t sbytes = count;
1133 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
1134
1135 /* according to the manpage, sbytes is always valid */
1136 if (sbytes)
1137 res = sbytes;
642 1138
643# elif __hpux 1139# elif __hpux
644 res = sendfile (ofd, ifd, offset, count, 0, 0); 1140 res = sendfile (ofd, ifd, offset, count, 0, 0);
645 1141
646# elif __solaris 1142# elif __solaris
647 {
648 struct sendfilevec vec; 1143 struct sendfilevec vec;
649 size_t sbytes; 1144 size_t sbytes;
650 1145
651 vec.sfv_fd = ifd; 1146 vec.sfv_fd = ifd;
652 vec.sfv_flag = 0; 1147 vec.sfv_flag = 0;
653 vec.sfv_off = offset; 1148 vec.sfv_off = offset;
654 vec.sfv_len = count; 1149 vec.sfv_len = count;
655 1150
656 res = sendfilev (ofd, &vec, 1, &sbytes); 1151 res = sendfilev (ofd, &vec, 1, &sbytes);
657 1152
658 if (res < 0 && sbytes) 1153 if (res < 0 && sbytes)
659 res = sbytes; 1154 res = sbytes;
660 }
661 1155
662# endif 1156# endif
1157
1158#elif defined (_WIN32) && 0
1159 /* does not work, just for documentation of what would need to be done */
1160 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
1161 /* libeio guarantees that the file offset does not change, and windows */
1162 /* has no way to get an independent handle to the same file description */
1163 HANDLE h = TO_SOCKET (ifd);
1164 SetFilePointer (h, offset, 0, FILE_BEGIN);
1165 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1166
663#else 1167#else
664 res = -1; 1168 res = -1;
665 errno = ENOSYS; 1169 errno = ENOSYS;
666#endif 1170#endif
667 1171
1172 /* we assume sendfile can copy at least 128mb in one go */
1173 if (res <= 128 * 1024 * 1024)
1174 {
1175 if (res > 0)
1176 written += res;
1177
1178 if (written)
1179 return written;
1180
1181 break;
1182 }
1183 else
1184 {
1185 /* if we requested more, then probably the kernel was lazy */
1186 written += res;
1187 offset += res;
1188 count -= res;
1189
1190 if (!count)
1191 return written;
1192 }
1193 }
1194
668 if (res < 0 1195 if (res < 0
669 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1196 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1197 /* BSDs */
1198#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1199 || errno == ENOTSUP
1200#endif
1201#ifdef EOPNOTSUPP /* windows */
1202 || errno == EOPNOTSUPP /* BSDs */
1203#endif
670#if __solaris 1204#if __solaris
671 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1205 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
672#endif 1206#endif
673 ) 1207 )
674 ) 1208 )
678 1212
679 res = 0; 1213 res = 0;
680 1214
681 while (count) 1215 while (count)
682 { 1216 {
683 ssize_t cnt; 1217 eio_ssize_t cnt;
684 1218
685 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 1219 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
686 1220
687 if (cnt <= 0) 1221 if (cnt <= 0)
688 { 1222 {
700 1234
701 offset += cnt; 1235 offset += cnt;
702 res += cnt; 1236 res += cnt;
703 count -= cnt; 1237 count -= cnt;
704 } 1238 }
1239
1240 FUBd;
705 } 1241 }
706 1242
707 return res; 1243 return res;
708} 1244}
709 1245
710/* read a full directory */ 1246#ifdef PAGESIZE
711static void scandir_ (eio_req *req, worker *self) 1247# define eio_pagesize() PAGESIZE
1248#else
1249static intptr_t
1250eio_pagesize (void)
712{ 1251{
713 DIR *dirp; 1252 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 1253
725 X_LOCK (wrklock); 1254 if (!page)
726 self->dirp = dirp = opendir (req->ptr1); 1255 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 1256
732 if (dirp && u && names) 1257 return page;
733 for (;;) 1258}
1259#endif
1260
1261static void
1262eio_page_align (void **addr, size_t *length)
1263{
1264 intptr_t mask = eio_pagesize () - 1;
1265
1266 /* round down addr */
1267 intptr_t adj = mask & (intptr_t)*addr;
1268
1269 *addr = (void *)((intptr_t)*addr - adj);
1270 *length += adj;
1271
1272 /* round up length */
1273 *length = (*length + mask) & ~mask;
1274}
1275
1276#if !_POSIX_MEMLOCK
1277# define eio__mlockall(a) EIO_ENOSYS ()
1278#else
1279
1280static int
1281eio__mlockall (int flags)
1282{
1283 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1284 extern int mallopt (int, int);
1285 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1286 #endif
1287
1288 if (EIO_MCL_CURRENT != MCL_CURRENT
1289 || EIO_MCL_FUTURE != MCL_FUTURE)
1290 {
1291 flags = 0
1292 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1293 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1294 }
1295
1296 return mlockall (flags);
1297}
1298#endif
1299
1300#if !_POSIX_MEMLOCK_RANGE
1301# define eio__mlock(a,b) EIO_ENOSYS ()
1302#else
1303
1304static int
1305eio__mlock (void *addr, size_t length)
1306{
1307 eio_page_align (&addr, &length);
1308
1309 return mlock (addr, length);
1310}
1311
1312#endif
1313
1314#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1315# define eio__msync(a,b,c) EIO_ENOSYS ()
1316#else
1317
1318static int
1319eio__msync (void *mem, size_t len, int flags)
1320{
1321 eio_page_align (&mem, &len);
1322
1323 if (EIO_MS_ASYNC != MS_SYNC
1324 || EIO_MS_INVALIDATE != MS_INVALIDATE
1325 || EIO_MS_SYNC != MS_SYNC)
1326 {
1327 flags = 0
1328 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1329 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1330 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1331 }
1332
1333 return msync (mem, len, flags);
1334}
1335
1336#endif
1337
1338static int
1339eio__mtouch (eio_req *req)
1340{
1341 void *mem = req->ptr2;
1342 size_t len = req->size;
1343 int flags = req->int1;
1344
1345 eio_page_align (&mem, &len);
1346
1347 {
1348 intptr_t addr = (intptr_t)mem;
1349 intptr_t end = addr + len;
1350 intptr_t page = eio_pagesize ();
1351
1352 if (addr < end)
1353 if (flags & EIO_MT_MODIFY) /* modify */
1354 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1355 else
1356 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1357 }
1358
1359 return 0;
1360}
1361
1362/*****************************************************************************/
1363/* requests implemented outside eio_execute, because they are so large */
1364
1365/* copies some absolute path to tmpbuf */
1366static char *
1367eio__getwd (struct tmpbuf *tmpbuf, eio_wd wd)
1368{
1369 if (wd == EIO_CWD)
1370 return getcwd (tmpbuf->ptr, PATH_MAX);
1371
1372#if HAVE_AT
1373 abort (); /*TODO*/
1374#else
1375 strcpy (tmpbuf->ptr, wd);
1376#endif
1377 return tmpbuf->ptr;
1378}
1379
1380/* result will always end up in tmpbuf, there is always space for adding a 0-byte */
1381static int
1382eio__realpath (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
1383{
1384 const char *rel = path;
1385 char *res;
1386 char *tmp1, *tmp2;
1387#if SYMLOOP_MAX > 32
1388 int symlinks = SYMLOOP_MAX;
1389#else
1390 int symlinks = 32;
1391#endif
1392
1393 /*D*/ /*TODO: wd ignored */
1394
1395 errno = EINVAL;
1396 if (!rel)
1397 return -1;
1398
1399 errno = ENOENT;
1400 if (!*rel)
1401 return -1;
1402
1403 res = tmpbuf_get (tmpbuf, PATH_MAX * 3);
1404 tmp1 = res + PATH_MAX;
1405 tmp2 = tmp1 + PATH_MAX;
1406
1407#if 0 /* disabled, the musl way to do things is just too racy */
1408#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1409 /* on linux we may be able to ask the kernel */
1410 {
1411 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1412
1413 if (fd >= 0)
734 { 1414 {
735 errno = 0; 1415 sprintf (tmp1, "/proc/self/fd/%d", fd);
736 readdir_r (dirp, &u->d, &entp); 1416 req->result = readlink (tmp1, res, PATH_MAX);
1417 close (fd);
737 1418
738 if (!entp) 1419 /* here we should probably stat the open file and the disk file, to make sure they still match */
1420
1421 if (req->result > 0)
1422 goto done;
1423 }
1424 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1425 return;
1426 }
1427#endif
1428#endif
1429
1430 if (*rel != '/')
1431 {
1432 if (!eio__getwd (tmpbuf, wd))
1433 return -1;
1434
1435 if (res [1]) /* only use if not / */
1436 res += strlen (res);
1437 }
1438
1439 while (*rel)
1440 {
1441 eio_ssize_t len, linklen;
1442 const char *beg = rel;
1443
1444 while (*rel && *rel != '/')
1445 ++rel;
1446
1447 len = rel - beg;
1448
1449 if (!len) /* skip slashes */
1450 {
739 break; 1451 ++rel;
1452 continue;
1453 }
740 1454
741 name = entp->d_name; 1455 if (beg [0] == '.')
1456 {
1457 if (len == 1)
1458 continue; /* . - nop */
742 1459
743 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1460 if (beg [1] == '.' && len == 2)
1461 {
1462 /* .. - back up one component, if possible */
1463
1464 while (res != tmpbuf->ptr)
1465 if (*--res == '/')
1466 break;
1467
1468 continue;
1469 }
1470 }
1471
1472 errno = ENAMETOOLONG;
1473 if (res + 1 + len + 1 >= tmp1)
1474 return;
1475
1476 /* copy one component */
1477 *res = '/';
1478 memcpy (res + 1, beg, len);
1479
1480 /* zero-terminate, for readlink */
1481 res [len + 1] = 0;
1482
1483 /* now check if it's a symlink */
1484 linklen = readlink (tmpbuf->ptr, tmp1, PATH_MAX);
1485
1486 if (linklen < 0)
744 { 1487 {
745 int len = strlen (name) + 1; 1488 if (errno != EINVAL)
1489 return -1;
746 1490
747 res++; 1491 /* 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; 1492 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 } 1493 }
1120 else 1494 else
1121 { 1495 {
1122 /* read: grow scalar as necessary */ 1496 /* yay, it was a symlink - build new path in tmp2 */
1123 svptr = SvGROW (data, len + dataoffset + 1); 1497 int rellen = strlen (rel);
1498
1499 errno = ENAMETOOLONG;
1500 if (linklen + 1 + rellen >= PATH_MAX)
1501 return -1;
1502
1503 errno = ELOOP;
1504 if (!--symlinks)
1505 return -1;
1506
1507 if (*tmp1 == '/')
1508 res = tmpbuf->ptr; /* symlink resolves to an absolute path */
1509
1510 /* we need to be careful, as rel might point into tmp2 already */
1511 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1512 tmp2 [linklen] = '/';
1513 memcpy (tmp2, tmp1, linklen);
1514
1515 rel = tmp2;
1124 } 1516 }
1517 }
1125 1518
1126 if (len < 0) 1519 /* special case for the lone root path */
1127 croak ("length must not be negative"); 1520 if (res == tmpbuf->ptr)
1521 *res++ = '/';
1128 1522
1523 return res - (char *)tmpbuf->ptr;
1524}
1525
1526static signed char
1527eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1528{
1529 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1530 : a->inode < b->inode ? -1
1531 : a->inode > b->inode ? 1
1532 : 0;
1533}
1534
1535#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1536
1537#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1538#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1539
1540static void
1541eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1542{
1543 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1544 unsigned char *bit = bits;
1545
1546 assert (CHAR_BIT == 8);
1547 assert (sizeof (eio_dirent) * 8 < 256);
1548 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1549 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1550
1551 if (size <= EIO_SORT_FAST)
1552 return;
1553
1554 /* first prepare an array of bits to test in our radix sort */
1555 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1556 /* inode_bits must contain all inodes ORed together */
1557 /* which is used to skip bits that are 0 everywhere, which is very common */
1558 {
1559 eio_ino_t endianness;
1560 int i, j;
1561
1562 /* we store the byte offset of byte n into byte n of "endianness" */
1563 for (i = 0; i < sizeof (eio_ino_t); ++i)
1564 ((unsigned char *)&endianness)[i] = i;
1565
1566 *bit++ = 0;
1567
1568 for (i = 0; i < sizeof (eio_ino_t); ++i)
1569 {
1570 /* shifting off the byte offsets out of "endianness" */
1571 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1572 endianness >>= 8;
1573
1574 for (j = 0; j < 8; ++j)
1575 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1576 *bit++ = offs + j;
1577 }
1578
1579 for (j = 0; j < 8; ++j)
1580 if (score_bits & (1 << j))
1581 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1582 }
1583
1584 /* now actually do the sorting (a variant of MSD radix sort) */
1585 {
1586 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1587 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1588 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1589 int stk_idx = 0;
1590
1591 base_stk [stk_idx] = dents;
1592 end_stk [stk_idx] = dents + size;
1593 bit_stk [stk_idx] = bit - 1;
1594
1595 do
1596 {
1597 base = base_stk [stk_idx];
1598 end = end_stk [stk_idx];
1599 bit = bit_stk [stk_idx];
1600
1601 for (;;)
1602 {
1603 unsigned char O = *bit >> 3;
1604 unsigned char M = 1 << (*bit & 7);
1605
1606 eio_dirent *a = base;
1607 eio_dirent *b = end;
1608
1609 if (b - a < EIO_SORT_CUTOFF)
1610 break;
1611
1612 /* now bit-partition the array on the bit */
1613 /* this ugly asymmetric loop seems to perform much better than typical */
1614 /* partition algos found in the literature */
1615 do
1616 if (!(((unsigned char *)a)[O] & M))
1617 ++a;
1618 else if (!(((unsigned char *)--b)[O] & M))
1619 {
1620 eio_dirent tmp = *a; *a = *b; *b = tmp;
1621 ++a;
1622 }
1623 while (b > a);
1624
1625 /* next bit, or stop, if no bits left in this path */
1626 if (!*--bit)
1627 break;
1628
1629 base_stk [stk_idx] = a;
1630 end_stk [stk_idx] = end;
1631 bit_stk [stk_idx] = bit;
1632 ++stk_idx;
1633
1634 end = a;
1635 }
1636 }
1637 while (stk_idx--);
1638 }
1639}
1640
1641static void
1642eio_dent_insertion_sort (eio_dirent *dents, int size)
1643{
1644 /* first move the smallest element to the front, to act as a sentinel */
1645 {
1646 int i;
1647 eio_dirent *min = dents;
1648
1649 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1650 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1651 if (EIO_DENT_CMP (dents [i], <, *min))
1652 min = &dents [i];
1653
1654 /* swap elements 0 and j (minimum) */
1655 {
1656 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1657 }
1658 }
1659
1660 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1661 {
1662 eio_dirent *i, *j;
1663
1664 for (i = dents + 1; i < dents + size; ++i)
1665 {
1666 eio_dirent value = *i;
1667
1668 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1669 j [1] = j [0];
1670
1671 j [1] = value;
1672 }
1673 }
1674}
1675
1676static void
1677eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1678{
1679 if (size <= 1)
1680 return; /* our insertion sort relies on size > 0 */
1681
1682 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1683 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1684 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1685
1686 /* use an insertion sort at the end, or for small arrays, */
1687 /* as insertion sort is more efficient for small partitions */
1688 eio_dent_insertion_sort (dents, size);
1689}
1690
1691/* read a full directory */
1692static void
1693eio__scandir (eio_req *req)
1694{
1695 char *name, *names;
1696 int namesalloc = 4096 - sizeof (void *) * 4;
1697 int namesoffs = 0;
1698 int flags = req->int1;
1699 eio_dirent *dents = 0;
1700 int dentalloc = 128;
1701 int dentoffs = 0;
1702 eio_ino_t inode_bits = 0;
1703#ifdef _WIN32
1704 HANDLE dirp;
1705 WIN32_FIND_DATA entp;
1706#else
1707 DIR *dirp;
1708 EIO_STRUCT_DIRENT *entp;
1709#endif
1710
1711 req->result = -1;
1712
1713 if (!(flags & EIO_READDIR_DENTS))
1714 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1715
1716#ifdef _WIN32
1717 {
1718 int len = strlen ((const char *)req->ptr1);
1719 char *path = malloc (MAX_PATH);
1720 const char *fmt;
1721
1722 if (!len)
1723 fmt = "./*";
1724 else if (((const char *)req->ptr1)[len - 1] == '/' || ((const char *)req->ptr1)[len - 1] == '\\')
1725 fmt = "%s*";
1726 else
1727 fmt = "%s/*";
1728
1729 _snprintf (path, MAX_PATH, fmt, (const char *)req->ptr1);
1730 dirp = FindFirstFile (path, &entp);
1731 free (path);
1732
1733 if (dirp == INVALID_HANDLE_VALUE)
1734 {
1735 /* should steal _dosmaperr */
1736 switch (GetLastError ())
1737 {
1738 case ERROR_FILE_NOT_FOUND:
1739 req->result = 0;
1740 break;
1741
1742 case ERROR_INVALID_NAME:
1743 case ERROR_PATH_NOT_FOUND:
1744 case ERROR_NO_MORE_FILES:
1745 errno = ENOENT;
1746 break;
1747
1748 case ERROR_NOT_ENOUGH_MEMORY:
1749 errno = ENOMEM;
1750 break;
1751
1752 default:
1753 errno = EINVAL;
1754 break;
1755 }
1756
1757 return;
1758 }
1759 }
1760#else
1761 dirp = opendir (req->ptr1);
1762
1763 if (!dirp)
1764 return;
1765#endif
1766
1767 if (req->flags & EIO_FLAG_PTR1_FREE)
1768 free (req->ptr1);
1769
1770 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1771 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1772 req->ptr2 = names = malloc (namesalloc);
1773
1774 if (!names || (flags && !dents))
1775 return;
1776
1777 for (;;)
1778 {
1779 int done;
1780
1781#ifdef _WIN32
1782 done = !dirp;
1783#else
1784 errno = 0;
1785 entp = readdir (dirp);
1786 done = !entp;
1787#endif
1788
1789 if (done)
1129 { 1790 {
1130 dREQ; 1791#ifndef _WIN32
1792 int old_errno = errno;
1793 closedir (dirp);
1794 errno = old_errno;
1131 1795
1132 req->type = ix; 1796 if (errno)
1133 req->sv1 = newSVsv (fh); 1797 break;
1134 req->int1 = PerlIO_fileno (ix == EIO_READ ? IoIFP (sv_2io (fh)) 1798#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 1799
1142 if (!SvREADONLY (data)) 1800 /* sort etc. */
1801 req->int1 = flags;
1802 req->result = dentoffs;
1803
1804 if (flags & EIO_READDIR_STAT_ORDER)
1805 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1806 else if (flags & EIO_READDIR_DIRS_FIRST)
1807 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1808 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1809 else
1810 {
1811 /* in this case, all is known, and we just put dirs first and sort them */
1812 eio_dirent *oth = dents + dentoffs;
1813 eio_dirent *dir = dents;
1814
1815 /* now partition dirs to the front, and non-dirs to the back */
1816 /* by walking from both sides and swapping if necessary */
1817 while (oth > dir)
1818 {
1819 if (dir->type == EIO_DT_DIR)
1820 ++dir;
1821 else if ((--oth)->type == EIO_DT_DIR)
1822 {
1823 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1824
1825 ++dir;
1826 }
1827 }
1828
1829 /* now sort the dirs only (dirs all have the same score) */
1830 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1831 }
1832
1833 break;
1834 }
1835
1836 /* now add the entry to our list(s) */
1837 name = D_NAME (entp);
1838
1839 /* skip . and .. entries */
1840 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1841 {
1842 int len = D_NAMLEN (entp) + 1;
1843
1844 while (ecb_expect_false (namesoffs + len > namesalloc))
1143 { 1845 {
1144 SvREADONLY_on (data); 1846 namesalloc *= 2;
1145 req->flags |= FLAG_SV2_RO_OFF; 1847 req->ptr2 = names = realloc (names, namesalloc);
1848
1849 if (!names)
1850 break;
1146 } 1851 }
1147 1852
1148 EIO_SEND; 1853 memcpy (names + namesoffs, name, len);
1854
1855 if (dents)
1856 {
1857 struct eio_dirent *ent;
1858
1859 if (ecb_expect_false (dentoffs == dentalloc))
1860 {
1861 dentalloc *= 2;
1862 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1863
1864 if (!dents)
1865 break;
1866 }
1867
1868 ent = dents + dentoffs;
1869
1870 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1871 ent->namelen = len - 1;
1872 ent->inode = D_INO (entp);
1873
1874 inode_bits |= ent->inode;
1875
1876 switch (D_TYPE (entp))
1877 {
1878 default:
1879 ent->type = EIO_DT_UNKNOWN;
1880 flags |= EIO_READDIR_FOUND_UNKNOWN;
1881 break;
1882
1883 #ifdef DT_FIFO
1884 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1885 #endif
1886 #ifdef DT_CHR
1887 case DT_CHR: ent->type = EIO_DT_CHR; break;
1888 #endif
1889 #ifdef DT_MPC
1890 case DT_MPC: ent->type = EIO_DT_MPC; break;
1891 #endif
1892 #ifdef DT_DIR
1893 case DT_DIR: ent->type = EIO_DT_DIR; break;
1894 #endif
1895 #ifdef DT_NAM
1896 case DT_NAM: ent->type = EIO_DT_NAM; break;
1897 #endif
1898 #ifdef DT_BLK
1899 case DT_BLK: ent->type = EIO_DT_BLK; break;
1900 #endif
1901 #ifdef DT_MPB
1902 case DT_MPB: ent->type = EIO_DT_MPB; break;
1903 #endif
1904 #ifdef DT_REG
1905 case DT_REG: ent->type = EIO_DT_REG; break;
1906 #endif
1907 #ifdef DT_NWK
1908 case DT_NWK: ent->type = EIO_DT_NWK; break;
1909 #endif
1910 #ifdef DT_CMP
1911 case DT_CMP: ent->type = EIO_DT_CMP; break;
1912 #endif
1913 #ifdef DT_LNK
1914 case DT_LNK: ent->type = EIO_DT_LNK; break;
1915 #endif
1916 #ifdef DT_SOCK
1917 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1918 #endif
1919 #ifdef DT_DOOR
1920 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1921 #endif
1922 #ifdef DT_WHT
1923 case DT_WHT: ent->type = EIO_DT_WHT; break;
1924 #endif
1925 }
1926
1927 ent->score = 7;
1928
1929 if (flags & EIO_READDIR_DIRS_FIRST)
1930 {
1931 if (ent->type == EIO_DT_UNKNOWN)
1932 {
1933 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1934 ent->score = 1;
1935 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1936 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1937 }
1938 else if (ent->type == EIO_DT_DIR)
1939 ent->score = 0;
1940 }
1941 }
1942
1943 namesoffs += len;
1944 ++dentoffs;
1149 } 1945 }
1946
1947 if (EIO_CANCELLED (req))
1948 {
1949 errno = ECANCELED;
1950 break;
1951 }
1952
1953#ifdef _WIN32
1954 if (!FindNextFile (dirp, &entp))
1955 {
1956 FindClose (dirp);
1957 dirp = 0;
1958 }
1959#endif
1960 }
1961}
1962
1963/*****************************************************************************/
1964/* working directory stuff */
1965
1966#if HAVE_AT
1967
1968#define WD2FD(wd) ((wd) ? ((int)(long)(wd)) - 1 : AT_FDCWD)
1969
1970#ifndef O_SEARCH
1971# define O_SEARCH O_RDONLY
1972#endif
1973
1974eio_wd
1975eio_wd_open_sync (eio_wd wd, const char *path)
1976{
1977 int fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1978
1979 return fd >= 0 ? (eio_wd)(long)(fd + 1) : EIO_INVALID_WD;
1980}
1981
1982static eio_wd
1983eio__wd_open_sync (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
1984{
1985 return eio_wd_open_sync (wd, path);
1150} 1986}
1151 1987
1152void 1988void
1153aio_readlink (SV8 *path, SV *callback=&PL_sv_undef) 1989eio_wd_close_sync (eio_wd wd)
1154 PROTOTYPE: $$;$
1155 PPCODE:
1156{ 1990{
1157 SV *data; 1991 int fd = WD2FD (wd);
1158 dREQ;
1159 1992
1160 data = newSV (NAME_MAX); 1993 if (fd >= 0)
1161 SvPOK_on (data); 1994 close (fd);
1995}
1162 1996
1163 req->type = EIO_READLINK; 1997static int
1164 req->sv1 = newSVsv (path); 1998eio__truncateat (int dirfd, const char *path, off_t length)
1165 req->ptr1 = SvPVbyte_nolen (req->sv1); 1999{
1166 req->sv2 = data; 2000 int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
1167 req->ptr2 = SvPVbyte_nolen (data); 2001 int res;
1168 2002
1169 EIO_SEND; 2003 if (fd < 0)
2004 return fd;
2005
2006 res = ftruncate (fd, length);
2007 close (fd);
2008 return res;
2009}
2010
2011static int
2012eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
2013{
2014 int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
2015 int res;
2016
2017 if (fd < 0)
2018 return fd;
2019
2020 res = fstatvfs (fd, buf);
2021 close (fd);
2022 return res;
2023
2024}
2025
2026#else
2027
2028/* on legacy systems, we represent the working directories simply by their path strings */
2029
2030static const char *
2031wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
2032{
2033 if (!wd || *path == '/')
2034 return path;
2035
2036 {
2037 int l1 = strlen ((const char *)wd);
2038 int l2 = strlen (path);
2039
2040 char *res = tmpbuf_get (tmpbuf, l1 + l2 + 2);
2041
2042 memcpy (res, wd, l1);
2043 res [l1] = '/';
2044 memcpy (res + l1 + 1, path, l2 + 1);
2045
2046 return res;
2047 }
2048}
2049
2050static eio_wd
2051eio__wd_open_sync (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
2052{
2053 if (*path == '/') /* absolute paths ignore wd */
2054 path = strdup (path);
2055 else if (path [0] == '.' && !path [1]) /* special case '.', as it is common */
2056 return wd;
2057 else
2058 {
2059 int len = eio__realpath (tmpbuf, wd, path);
2060
2061 path = EIO_INVALID_WD;
2062
2063 if (len >= 0)
2064 {
2065 ((char *)tmpbuf->ptr)[len] = 0;
2066 path = strdup (tmpbuf->ptr);
2067 }
2068 }
2069
2070 if (!path)
2071 path = EIO_INVALID_WD;
2072
2073 return (eio_wd)path;
2074}
2075
2076eio_wd
2077eio_wd_open_sync (eio_wd wd, const char *path)
2078{
2079 struct tmpbuf tmpbuf = { 0 };
2080 wd = eio__wd_open_sync (&tmpbuf, wd, path);
2081 free (tmpbuf.ptr);
2082
2083 return wd;
1170} 2084}
1171 2085
1172void 2086void
1173aio_sendfile (SV *out_fh, SV *in_fh, SV *in_offset, UV length, SV *callback=&PL_sv_undef) 2087eio_wd_close_sync (eio_wd wd)
1174 PROTOTYPE: $$$$;$
1175 PPCODE:
1176{ 2088{
1177 dREQ; 2089 if (wd != EIO_INVALID_WD)
1178 2090 free (wd);
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} 2091}
2092
2093#endif
2094
2095/*****************************************************************************/
2096
2097#define ALLOC(len) \
2098 if (!req->ptr2) \
2099 { \
2100 X_LOCK (wrklock); \
2101 req->flags |= EIO_FLAG_PTR2_FREE; \
2102 X_UNLOCK (wrklock); \
2103 req->ptr2 = malloc (len); \
2104 if (!req->ptr2) \
2105 { \
2106 errno = ENOMEM; \
2107 req->result = -1; \
2108 break; \
2109 } \
2110 }
2111
2112X_THREAD_PROC (etp_proc)
2113{
2114 ETP_REQ *req;
2115 struct timespec ts;
2116 etp_worker *self = (etp_worker *)thr_arg;
2117
2118#if HAVE_PRCTL_SET_NAME
2119 prctl (PR_SET_NAME, (unsigned long)"eio_thread", 0, 0, 0);
2120#endif
2121
2122 /* try to distribute timeouts somewhat evenly */
2123 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
2124
2125 for (;;)
2126 {
2127 ts.tv_sec = 0;
2128
2129 X_LOCK (reqlock);
2130
2131 for (;;)
2132 {
2133 req = reqq_shift (&req_queue);
2134
2135 if (req)
2136 break;
2137
2138 if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
2139 {
2140 X_UNLOCK (reqlock);
2141 X_LOCK (wrklock);
2142 --started;
2143 X_UNLOCK (wrklock);
2144 goto quit;
2145 }
2146
2147 ++idle;
2148
2149 if (idle <= max_idle)
2150 /* we are allowed to idle, so do so without any timeout */
2151 X_COND_WAIT (reqwait, reqlock);
2152 else
2153 {
2154 /* initialise timeout once */
2155 if (!ts.tv_sec)
2156 ts.tv_sec = time (0) + idle_timeout;
2157
2158 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
2159 ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
2160 }
2161
2162 --idle;
2163 }
2164
2165 --nready;
2166
2167 X_UNLOCK (reqlock);
2168
2169 if (req->type < 0)
2170 goto quit;
2171
2172 ETP_EXECUTE (self, req);
2173
2174 X_LOCK (reslock);
2175
2176 ++npending;
2177
2178 if (!reqq_push (&res_queue, req) && want_poll_cb)
2179 want_poll_cb ();
2180
2181 etp_worker_clear (self);
2182
2183 X_UNLOCK (reslock);
2184 }
2185
2186quit:
2187 free (req);
2188
2189 X_LOCK (wrklock);
2190 etp_worker_free (self);
2191 X_UNLOCK (wrklock);
2192
2193 return 0;
2194}
2195
2196/*****************************************************************************/
2197
2198int ecb_cold
2199eio_init (void (*want_poll)(void), void (*done_poll)(void))
2200{
2201#if !HAVE_PREADWRITE
2202 X_MUTEX_CREATE (preadwritelock);
2203#endif
2204
2205 return etp_init (want_poll, done_poll);
2206}
2207
2208ecb_inline void
2209eio_api_destroy (eio_req *req)
2210{
2211 free (req);
2212}
2213
2214#define REQ(rtype) \
2215 eio_req *req; \
2216 \
2217 req = (eio_req *)calloc (1, sizeof *req); \
2218 if (!req) \
2219 return 0; \
2220 \
2221 req->type = rtype; \
2222 req->pri = pri; \
2223 req->finish = cb; \
2224 req->data = data; \
2225 req->destroy = eio_api_destroy;
2226
2227#define SEND eio_submit (req); return req
2228
2229#define PATH \
2230 req->flags |= EIO_FLAG_PTR1_FREE; \
2231 req->ptr1 = strdup (path); \
2232 if (!req->ptr1) \
2233 { \
2234 eio_api_destroy (req); \
2235 return 0; \
2236 }
2237
2238static void
2239eio_execute (etp_worker *self, eio_req *req)
2240{
2241#if HAVE_AT
2242 int dirfd;
2243#else
2244 const char *path;
2245#endif
2246
2247 if (ecb_expect_false (EIO_CANCELLED (req)))
2248 {
2249 req->result = -1;
2250 req->errorno = ECANCELED;
2251 return;
2252 }
2253
2254 if (ecb_expect_false (req->wd == EIO_INVALID_WD))
2255 {
2256 req->result = -1;
2257 req->errorno = ENOENT;
2258 return;
2259 }
2260
2261 if (req->type >= EIO_OPEN)
2262 {
2263 #if HAVE_AT
2264 dirfd = WD2FD (req->wd);
2265 #else
2266 path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
2267 #endif
2268 }
2269
2270 switch (req->type)
2271 {
2272 case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
2273 req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
2274 break;
2275 case EIO_WD_CLOSE: eio_wd_close_sync (req->wd); break;
2276
2277 case EIO_READ: ALLOC (req->size);
2278 req->result = req->offs >= 0
2279 ? pread (req->int1, req->ptr2, req->size, req->offs)
2280 : read (req->int1, req->ptr2, req->size); break;
2281 case EIO_WRITE: req->result = req->offs >= 0
2282 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
2283 : write (req->int1, req->ptr2, req->size); break;
2284
2285 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
2286 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
2287
2288#if HAVE_AT
2289 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2290 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break;
2291 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2292 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, AT_SYMLINK_NOFOLLOW); break;
2293 case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break;
2294 case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break;
2295 case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break;
2296 case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break;
2297
2298 case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
2299 case EIO_RMDIR: req->result = unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break;
2300 case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); break;
2301 case EIO_RENAME: req->result = renameat (dirfd, req->ptr1, WD2FD (req->int3), req->ptr2); break;
2302 case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD (req->int3), req->ptr2, 0); break;
2303 case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
2304 case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
2305 case EIO_READLINK: ALLOC (PATH_MAX);
2306 req->result = readlinkat (dirfd, req->ptr1, req->ptr2, PATH_MAX); break;
2307 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2308 req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2309#else
2310 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2311 req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
2312 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2313 req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
2314 case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
2315 case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
2316 case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
2317 case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
2318
2319 case EIO_UNLINK: req->result = unlink (path ); break;
2320 case EIO_RMDIR: req->result = rmdir (path ); break;
2321 case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
2322 case EIO_RENAME: req->result = rename (path , req->ptr2); break;
2323 case EIO_LINK: req->result = link (path , req->ptr2); break;
2324 case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
2325 case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
2326 case EIO_READLINK: ALLOC (PATH_MAX);
2327 req->result = readlink (path, req->ptr2, PATH_MAX); break;
2328 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2329 req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
2330#endif
2331
2332 case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
2333 {
2334 ALLOC (req->result);
2335 memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
2336 }
2337 break;
2338
2339 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2340 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
2341
2342 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2343 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2344
2345 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
2346 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
2347 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
2348
2349 case EIO_CLOSE: req->result = close (req->int1); break;
2350 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
2351 case EIO_SYNC: req->result = 0; sync (); break;
2352 case EIO_FSYNC: req->result = fsync (req->int1); break;
2353 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
2354 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
2355 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
2356 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
2357 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
2358 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
2359 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
2360 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
2361
2362 case EIO_READDIR: eio__scandir (req); break;
2363
2364 case EIO_BUSY:
2365#ifdef _WIN32
2366 Sleep (req->nv1 * 1e3);
2367#else
2368 {
2369 struct timeval tv;
2370
2371 tv.tv_sec = req->nv1;
2372 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
2373
2374 req->result = select (0, 0, 0, 0, &tv);
2375 }
2376#endif
2377 break;
2378
2379 case EIO_UTIME:
2380 case EIO_FUTIME:
2381 {
2382 struct timeval tv[2];
2383 struct timeval *times;
2384
2385 if (req->nv1 != -1. || req->nv2 != -1.)
2386 {
2387 tv[0].tv_sec = req->nv1;
2388 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
2389 tv[1].tv_sec = req->nv2;
2390 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
2391
2392 times = tv;
2393 }
2394 else
2395 times = 0;
2396
2397 req->result = req->type == EIO_FUTIME
2398 ? futimes (req->int1, times)
2399 : utimes (req->ptr1, times);
2400 }
2401 break;
2402
2403 case EIO_GROUP:
2404 abort (); /* handled in eio_request */
2405
2406 case EIO_NOP:
2407 req->result = 0;
2408 break;
2409
2410 case EIO_CUSTOM:
2411 req->feed (req);
2412 break;
2413
2414 default:
2415 errno = ENOSYS;
2416 req->result = -1;
2417 break;
2418 }
2419
2420 req->errorno = errno;
2421}
2422
2423#ifndef EIO_NO_WRAPPERS
2424
2425eio_req *eio_nop (int pri, eio_cb cb, void *data)
2426{
2427 REQ (EIO_NOP); SEND;
2428}
2429
2430eio_req *eio_busy (double delay, int pri, eio_cb cb, void *data)
2431{
2432 REQ (EIO_BUSY); req->nv1 = delay; SEND;
2433}
2434
2435eio_req *eio_sync (int pri, eio_cb cb, void *data)
2436{
2437 REQ (EIO_SYNC); SEND;
2438}
2439
2440eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
2441{
2442 REQ (EIO_FSYNC); req->int1 = fd; SEND;
2443}
2444
2445eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2446{
2447 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2448}
2449
2450eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2451{
2452 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2453}
2454
2455eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
2456{
2457 REQ (EIO_SYNCFS); req->int1 = fd; SEND;
2458}
2459
2460eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2461{
2462 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2463}
2464
2465eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2466{
2467 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2468}
2469
2470eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
2471{
2472 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
2473}
2474
2475eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
2476{
2477 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2478}
2479
2480eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2481{
2482 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
2483}
2484
2485eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
2486{
2487 REQ (EIO_CLOSE); req->int1 = fd; SEND;
2488}
2489
2490eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
2491{
2492 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
2493}
2494
2495eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
2496{
2497 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
2498}
2499
2500eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
2501{
2502 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
2503}
2504
2505eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
2506{
2507 REQ (EIO_FSTAT); req->int1 = fd; SEND;
2508}
2509
2510eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
2511{
2512 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
2513}
2514
2515eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
2516{
2517 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
2518}
2519
2520eio_req *eio_ftruncate (int fd, off_t offset, int pri, eio_cb cb, void *data)
2521{
2522 REQ (EIO_FTRUNCATE); req->int1 = fd; req->offs = offset; SEND;
2523}
2524
2525eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
2526{
2527 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
2528}
2529
2530eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
2531{
2532 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
2533}
2534
2535eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
2536{
2537 REQ (EIO_DUP2); req->int1 = fd; req->int2 = fd2; SEND;
2538}
2539
2540eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, int pri, eio_cb cb, void *data)
2541{
2542 REQ (EIO_SENDFILE); req->int1 = out_fd; req->int2 = in_fd; req->offs = in_offset; req->size = length; SEND;
2543}
2544
2545eio_req *eio_open (const char *path, int flags, mode_t mode, int pri, eio_cb cb, void *data)
2546{
2547 REQ (EIO_OPEN); PATH; req->int1 = flags; req->int2 = (long)mode; SEND;
2548}
2549
2550eio_req *eio_utime (const char *path, double atime, double mtime, int pri, eio_cb cb, void *data)
2551{
2552 REQ (EIO_UTIME); PATH; req->nv1 = atime; req->nv2 = mtime; SEND;
2553}
2554
2555eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
2556{
2557 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
2558}
2559
2560eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
2561{
2562 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
2563}
2564
2565eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
2566{
2567 REQ (EIO_CHMOD); PATH; req->int2 = (long)mode; SEND;
2568}
2569
2570eio_req *eio_mkdir (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
2571{
2572 REQ (EIO_MKDIR); PATH; req->int2 = (long)mode; SEND;
2573}
2574
2575static eio_req *
2576eio__1path (int type, const char *path, int pri, eio_cb cb, void *data)
2577{
2578 REQ (type); PATH; SEND;
2579}
2580
2581eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
2582{
2583 return eio__1path (EIO_READLINK, path, pri, cb, data);
2584}
2585
2586eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2587{
2588 return eio__1path (EIO_REALPATH, path, pri, cb, data);
2589}
2590
2591eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
2592{
2593 return eio__1path (EIO_STAT, path, pri, cb, data);
2594}
2595
2596eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
2597{
2598 return eio__1path (EIO_LSTAT, path, pri, cb, data);
2599}
2600
2601eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
2602{
2603 return eio__1path (EIO_STATVFS, path, pri, cb, data);
2604}
2605
2606eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
2607{
2608 return eio__1path (EIO_UNLINK, path, pri, cb, data);
2609}
2610
2611eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
2612{
2613 return eio__1path (EIO_RMDIR, path, pri, cb, data);
2614}
2615
2616eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
2617{
2618 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
2619}
2620
2621eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
2622{
2623 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
2624}
2625
2626static eio_req *
2627eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2628{
2629 REQ (type); PATH;
2630
2631 req->flags |= EIO_FLAG_PTR2_FREE;
2632 req->ptr2 = strdup (new_path);
2633 if (!req->ptr2)
2634 {
2635 eio_api_destroy (req);
2636 return 0;
2637 }
2638
2639 SEND;
2640}
2641
2642eio_req *eio_link (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2643{
2644 return eio__2path (EIO_LINK, path, new_path, pri, cb, data);
2645}
2646
2647eio_req *eio_symlink (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2648{
2649 return eio__2path (EIO_SYMLINK, path, new_path, pri, cb, data);
2650}
2651
2652eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2653{
2654 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
2655}
2656
2657eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
2658{
2659 REQ (EIO_CUSTOM); req->feed = execute; SEND;
2660}
2661
2662#endif
2663
2664eio_req *eio_grp (eio_cb cb, void *data)
2665{
2666 const int pri = EIO_PRI_MAX;
2667
2668 REQ (EIO_GROUP); SEND;
2669}
2670
2671#undef REQ
2672#undef PATH
2673#undef SEND
2674
2675/*****************************************************************************/
2676/* grp functions */
1189 2677
1190void 2678void
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) 2679eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1420{ 2680{
1421 grp->int2 = limit; 2681 grp->int2 = limit;
1422 grp->feed = feed; 2682 grp->feed = feed;
1423 2683
1424 grp_try_feed (grp); 2684 grp_try_feed (grp);
1425} 2685}
1426 2686
2687void
1427void eio_grp_limit (eio_req *grp, int limit) 2688eio_grp_limit (eio_req *grp, int limit)
1428{ 2689{
1429 grp->int2 = limit; 2690 grp->int2 = limit;
1430 2691
1431 grp_try_feed (grp); 2692 grp_try_feed (grp);
1432} 2693}
1433 2694
2695void
1434void eio_grp_add (eio_req *grp, eio_req *req) 2696eio_grp_add (eio_req *grp, eio_req *req)
1435{ 2697{
1436 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2698 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
2699
2700 grp->flags |= EIO_FLAG_GROUPADD;
1437 2701
1438 ++grp->size; 2702 ++grp->size;
1439 req->grp = grp; 2703 req->grp = grp;
1440 2704
1441 req->grp_prev = 0; 2705 req->grp_prev = 0;
1445 grp->grp_first->grp_prev = req; 2709 grp->grp_first->grp_prev = req;
1446 2710
1447 grp->grp_first = req; 2711 grp->grp_first = req;
1448} 2712}
1449 2713
2714/*****************************************************************************/
2715/* misc garbage */
1450 2716
2717eio_ssize_t
2718eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2719{
2720 return eio__sendfile (ofd, ifd, offset, count);
2721}
2722

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines