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

Comparing libeio/eio.c (file contents):
Revision 1.8 by root, Sun May 11 20:08:37 2008 UTC vs.
Revision 1.104 by root, Mon Sep 26 16:54:25 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#include <stddef.h> 53#include <stddef.h>
6#include <stdlib.h> 54#include <stdlib.h>
10#include <sys/stat.h> 58#include <sys/stat.h>
11#include <limits.h> 59#include <limits.h>
12#include <fcntl.h> 60#include <fcntl.h>
13#include <assert.h> 61#include <assert.h>
14 62
63/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
64/* intptr_t only comes from stdint.h, says idiot openbsd coder */
65#if HAVE_STDINT_H
66# include <stdint.h>
67#endif
68
69#ifndef ECANCELED
70# define ECANCELED EDOM
71#endif
72#ifndef ELOOP
73# define ELOOP EDOM
74#endif
75
76#if !defined(ENOTSOCK) && defined(WSAENOTSOCK)
77# define ENOTSOCK WSAENOTSOCK
78#endif
79
80static void eio_destroy (eio_req *req);
81
15#ifndef EIO_FINISH 82#ifndef EIO_FINISH
16# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0 83# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
17#endif 84#endif
18 85
19#ifndef EIO_DESTROY 86#ifndef EIO_DESTROY
22 89
23#ifndef EIO_FEED 90#ifndef EIO_FEED
24# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0) 91# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)
25#endif 92#endif
26 93
94#ifndef EIO_FD_TO_WIN32_HANDLE
95# define EIO_FD_TO_WIN32_HANDLE(fd) _get_osfhandle (fd)
96#endif
97#ifndef EIO_WIN32_HANDLE_TO_FD
98# define EIO_WIN32_HANDLE_TO_FD(handle) _open_osfhandle (handle, 0)
99#endif
100
101#define EIO_ERRNO(errval,retval) ((errno = errval), retval)
102
103#define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1)
104
27#ifdef _WIN32 105#ifdef _WIN32
28 106
29 /*doh*/ 107 #undef PAGESIZE
108 #define PAGESIZE 4096 /* GetSystemInfo? */
109
110 /* TODO: look at how perl does stat (non-sloppy), unlink (ro-files), utime, link */
111
112 #ifdef EIO_STRUCT_STATI64
113 /* look at perl's non-sloppy stat */
114 #define stat(path,buf) _stati64 (path,buf)
115 #define fstat(fd,buf) _fstati64 (fd,buf)
116 #endif
117 #define lstat(path,buf) stat (path,buf)
118 #define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
119 #define mkdir(path,mode) _mkdir (path)
120 #define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
121
122 #define chmod(path,mode) _chmod (path, mode)
123 #define dup(fd) _dup (fd)
124 #define dup2(fd1,fd2) _dup2 (fd1, fd2)
125
126 #define fchmod(fd,mode) EIO_ENOSYS ()
127 #define chown(path,uid,gid) EIO_ENOSYS ()
128 #define fchown(fd,uid,gid) EIO_ENOSYS ()
129 #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
130 #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
131 #define mknod(path,mode,dev) EIO_ENOSYS ()
132 #define sync() EIO_ENOSYS ()
133 #define readlink(path,buf,s) EIO_ENOSYS ()
134 #define statvfs(path,buf) EIO_ENOSYS ()
135 #define fstatvfs(fd,buf) EIO_ENOSYS ()
136
137 /* 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)
30 196
31#else 197#else
32 198
33# include "config.h"
34# include <sys/time.h> 199 #include <sys/time.h>
35# include <sys/select.h> 200 #include <sys/select.h>
201 #include <sys/statvfs.h>
36# 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
37# include <utime.h> 237# include <utime.h>
38# include <signal.h>
39# include <dirent.h>
40
41# ifndef EIO_STRUCT_DIRENT
42# define EIO_STRUCT_DIRENT struct dirent
43# endif 238#endif
44 239
45#endif 240#if HAVE_SYS_SYSCALL_H
46 241# include <sys/syscall.h>
47# ifndef EIO_STRUCT_STAT
48# define EIO_STRUCT_STAT struct stat
49# endif 242#endif
243
244#if HAVE_SYS_PRCTL_H
245# include <sys/prctl.h>
246#endif
50 247
51#if HAVE_SENDFILE 248#if HAVE_SENDFILE
52# if __linux 249# if __linux
53# include <sys/sendfile.h> 250# include <sys/sendfile.h>
54# elif __freebsd 251# elif __FreeBSD__ || defined __APPLE__
55# include <sys/socket.h> 252# include <sys/socket.h>
56# include <sys/uio.h> 253# include <sys/uio.h>
57# elif __hpux 254# elif __hpux
58# include <sys/socket.h> 255# include <sys/socket.h>
59# elif __solaris /* not yet */ 256# elif __solaris
60# include <sys/sendfile.h> 257# include <sys/sendfile.h>
61# else 258# else
62# error sendfile support requested but not available 259# error sendfile support requested but not available
63# endif 260# endif
64#endif 261#endif
65 262
66/* number of seconds after which an idle threads exit */ 263#ifndef D_TYPE
67#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
68 272
69/* used for struct dirent, AIX doesn't provide it */ 273/* used for struct dirent, AIX doesn't provide it */
70#ifndef NAME_MAX 274#ifndef NAME_MAX
71# define NAME_MAX 4096 275# define NAME_MAX 4096
72#endif 276#endif
73 277
278/* used for readlink etc. */
279#ifndef PATH_MAX
280# define PATH_MAX 4096
281#endif
282
74/* buffer size for various temporary buffers */ 283/* buffer size for various temporary buffers */
75#define EIO_BUFSIZE 65536 284#define EIO_BUFSIZE 65536
76 285
77#define dBUF \ 286#define dBUF \
78 char *eio_buf; \
79 X_LOCK (wrklock); \
80 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 287 char *eio_buf = malloc (EIO_BUFSIZE); \
81 X_UNLOCK (wrklock); \ 288 errno = ENOMEM; \
82 if (!eio_buf) \ 289 if (!eio_buf) \
83 return -1; 290 return -1
291
292#define FUBd \
293 free (eio_buf)
84 294
85#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;
86 344
87static void (*want_poll_cb) (void); 345static void (*want_poll_cb) (void);
88static void (*done_poll_cb) (void); 346static void (*done_poll_cb) (void);
89 347
90static unsigned int max_poll_time = 0; 348static unsigned int max_poll_time; /* reslock */
91static unsigned int max_poll_reqs = 0; 349static unsigned int max_poll_reqs; /* reslock */
92 350
93/* calculcate time difference in ~1/EIO_TICKS of a second */ 351static unsigned int nreqs; /* reqlock */
94static int tvdiff (struct timeval *tv1, struct timeval *tv2) 352static unsigned int nready; /* reqlock */
95{ 353static unsigned int npending; /* reqlock */
96 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 354static unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
97 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 355static unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
98}
99 356
100static 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)
101 389
102/* worker threads management */ 390/* worker threads management */
103static mutex_t wrklock = X_MUTEX_INIT;
104 391
105typedef struct worker { 392static void ecb_cold
106 /* locked by wrklock */
107 struct worker *prev, *next;
108
109 thread_t tid;
110
111 /* locked by reslock, reqlock or wrklock */
112 eio_req *req; /* currently processed request */
113 void *dbuf;
114 DIR *dirp;
115} worker;
116
117static worker wrk_first = { &wrk_first, &wrk_first, 0 };
118
119static void worker_clear (worker *wrk) 393etp_worker_clear (etp_worker *wrk)
120{ 394{
121 if (wrk->dirp) 395}
122 {
123 closedir (wrk->dirp);
124 wrk->dirp = 0;
125 }
126 396
127 if (wrk->dbuf) 397static void ecb_cold
128 { 398etp_worker_free (etp_worker *wrk)
399{
129 free (wrk->dbuf); 400 free (wrk->tmpbuf.ptr);
130 wrk->dbuf = 0;
131 }
132}
133 401
134static void worker_free (worker *wrk)
135{
136 wrk->next->prev = wrk->prev; 402 wrk->next->prev = wrk->prev;
137 wrk->prev->next = wrk->next; 403 wrk->prev->next = wrk->next;
138 404
139 free (wrk); 405 free (wrk);
140} 406}
141 407
142static volatile unsigned int nreqs, nready, npending; 408static unsigned int
143static volatile unsigned int max_idle = 4; 409etp_nreqs (void)
144
145static mutex_t reslock = X_MUTEX_INIT;
146static mutex_t reqlock = X_MUTEX_INIT;
147static cond_t reqwait = X_COND_INIT;
148
149unsigned int eio_nreqs (void)
150{ 410{
411 int retval;
412 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
413 retval = nreqs;
414 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
151 return nreqs; 415 return retval;
152} 416}
153 417
154unsigned int eio_nready (void) 418static unsigned int
419etp_nready (void)
155{ 420{
156 unsigned int retval; 421 unsigned int retval;
157 422
158 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 423 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
159 retval = nready; 424 retval = nready;
160 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 425 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
161 426
162 return retval; 427 return retval;
163} 428}
164 429
165unsigned int eio_npending (void) 430static unsigned int
431etp_npending (void)
166{ 432{
167 unsigned int retval; 433 unsigned int retval;
168 434
169 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 435 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
170 retval = npending; 436 retval = npending;
171 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 437 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
172 438
173 return retval; 439 return retval;
174} 440}
175 441
176unsigned int eio_nthreads (void) 442static unsigned int
443etp_nthreads (void)
177{ 444{
178 unsigned int retval; 445 unsigned int retval;
179 446
180 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 447 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
181 retval = started; 448 retval = started;
188 * a somewhat faster data structure might be nice, but 455 * a somewhat faster data structure might be nice, but
189 * with 8 priorities this actually needs <20 insns 456 * with 8 priorities this actually needs <20 insns
190 * per shift, the most expensive operation. 457 * per shift, the most expensive operation.
191 */ 458 */
192typedef struct { 459typedef struct {
193 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 */
194 int size; 461 int size;
195} reqq; 462} etp_reqq;
196 463
197static reqq req_queue; 464static etp_reqq req_queue;
198static reqq res_queue; 465static etp_reqq res_queue;
199 466
200static 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)
201{ 480{
202 int pri = req->pri; 481 int pri = req->pri;
203 req->next = 0; 482 req->next = 0;
204 483
205 if (q->qe[pri]) 484 if (q->qe[pri])
211 q->qe[pri] = q->qs[pri] = req; 490 q->qe[pri] = q->qs[pri] = req;
212 491
213 return q->size++; 492 return q->size++;
214} 493}
215 494
216static eio_req *reqq_shift (reqq *q) 495static ETP_REQ * ecb_noinline
496reqq_shift (etp_reqq *q)
217{ 497{
218 int pri; 498 int pri;
219 499
220 if (!q->size) 500 if (!q->size)
221 return 0; 501 return 0;
222 502
223 --q->size; 503 --q->size;
224 504
225 for (pri = EIO_NUM_PRI; pri--; ) 505 for (pri = ETP_NUM_PRI; pri--; )
226 { 506 {
227 eio_req *req = q->qs[pri]; 507 eio_req *req = q->qs[pri];
228 508
229 if (req) 509 if (req)
230 { 510 {
236 } 516 }
237 517
238 abort (); 518 abort ();
239} 519}
240 520
241static void grp_try_feed (eio_req *grp) 521static int ecb_cold
522etp_init (void (*want_poll)(void), void (*done_poll)(void))
242{ 523{
243 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 524 X_MUTEX_CREATE (wrklock);
244 { 525 X_MUTEX_CREATE (reslock);
245 int old_len = grp->size; 526 X_MUTEX_CREATE (reqlock);
527 X_COND_CREATE (reqwait);
246 528
247 EIO_FEED (grp); 529 reqq_init (&req_queue);
530 reqq_init (&res_queue);
248 531
249 /* stop if no progress has been made */ 532 wrk_first.next =
250 if (old_len == grp->size) 533 wrk_first.prev = &wrk_first;
251 {
252 grp->feed = 0;
253 break;
254 }
255 }
256}
257 534
258static int eio_finish (eio_req *req); 535 started = 0;
536 idle = 0;
537 nreqs = 0;
538 nready = 0;
539 npending = 0;
259 540
260static int grp_dec (eio_req *grp) 541 want_poll_cb = want_poll;
261{ 542 done_poll_cb = done_poll;
262 --grp->size;
263 543
264 /* call feeder, if applicable */
265 grp_try_feed (grp);
266
267 /* finish, if done */
268 if (!grp->size && grp->int1)
269 return eio_finish (grp);
270 else
271 return 0; 544 return 0;
272} 545}
273 546
274void eio_destroy (eio_req *req)
275{
276 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
277 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
278
279 EIO_DESTROY (req);
280}
281
282static int eio_finish (eio_req *req)
283{
284 int res = EIO_FINISH (req);
285
286 if (req->grp)
287 {
288 int res2;
289 eio_req *grp = req->grp;
290
291 /* unlink request */
292 if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
293 if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
294
295 if (grp->grp_first == req)
296 grp->grp_first = req->grp_next;
297
298 res2 = grp_dec (grp);
299
300 if (!res && res2)
301 res = res2;
302 }
303
304 eio_destroy (req);
305
306 return res;
307}
308
309void eio_grp_cancel (eio_req *grp)
310{
311 for (grp = grp->grp_first; grp; grp = grp->grp_next)
312 eio_cancel (grp);
313}
314
315void eio_cancel (eio_req *req)
316{
317 X_LOCK (wrklock);
318 req->flags |= EIO_FLAG_CANCELLED;
319 X_UNLOCK (wrklock);
320
321 eio_grp_cancel (req);
322}
323
324X_THREAD_PROC (eio_proc); 547X_THREAD_PROC (etp_proc);
325 548
326static void start_thread (void) 549static void ecb_cold
550etp_start_thread (void)
327{ 551{
328 worker *wrk = calloc (1, sizeof (worker)); 552 etp_worker *wrk = calloc (1, sizeof (etp_worker));
329 553
330 /*TODO*/ 554 /*TODO*/
331 assert (("unable to allocate worker thread data", !wrk)); 555 assert (("unable to allocate worker thread data", wrk));
332 556
333 X_LOCK (wrklock); 557 X_LOCK (wrklock);
334 558
335 if (thread_create (&wrk->tid, eio_proc, (void *)wrk)) 559 if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
336 { 560 {
337 wrk->prev = &wrk_first; 561 wrk->prev = &wrk_first;
338 wrk->next = wrk_first.next; 562 wrk->next = wrk_first.next;
339 wrk_first.next->prev = wrk; 563 wrk_first.next->prev = wrk;
340 wrk_first.next = wrk; 564 wrk_first.next = wrk;
344 free (wrk); 568 free (wrk);
345 569
346 X_UNLOCK (wrklock); 570 X_UNLOCK (wrklock);
347} 571}
348 572
573static void
349static void maybe_start_thread (void) 574etp_maybe_start_thread (void)
350{ 575{
351 if (eio_nthreads () >= wanted) 576 if (ecb_expect_true (etp_nthreads () >= wanted))
352 return; 577 return;
353 578
354 /* todo: maybe use idle here, but might be less exact */ 579 /* todo: maybe use idle here, but might be less exact */
355 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 ()))
356 return; 581 return;
357 582
358 start_thread (); 583 etp_start_thread ();
359} 584}
360 585
361void eio_submit (eio_req *req) 586static void ecb_cold
362{ 587etp_end_thread (void)
363 ++nreqs;
364
365 X_LOCK (reqlock);
366 ++nready;
367 reqq_push (&req_queue, req);
368 X_COND_SIGNAL (reqwait);
369 X_UNLOCK (reqlock);
370
371 maybe_start_thread ();
372}
373
374static void end_thread (void)
375{ 588{
376 eio_req *req = calloc (1, sizeof (eio_req)); 589 eio_req *req = calloc (1, sizeof (eio_req));
377 590
378 req->type = EIO_QUIT; 591 req->type = -1;
379 req->pri = EIO_PRI_MAX + EIO_PRI_BIAS; 592 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
380 593
381 X_LOCK (reqlock); 594 X_LOCK (reqlock);
382 reqq_push (&req_queue, req); 595 reqq_push (&req_queue, req);
383 X_COND_SIGNAL (reqwait); 596 X_COND_SIGNAL (reqwait);
384 X_UNLOCK (reqlock); 597 X_UNLOCK (reqlock);
386 X_LOCK (wrklock); 599 X_LOCK (wrklock);
387 --started; 600 --started;
388 X_UNLOCK (wrklock); 601 X_UNLOCK (wrklock);
389} 602}
390 603
391void eio_set_max_poll_time (double nseconds) 604static int
392{
393 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
394 max_poll_time = nseconds;
395 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
396}
397
398void eio_set_max_poll_reqs (unsigned int maxreqs)
399{
400 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
401 max_poll_reqs = maxreqs;
402 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
403}
404
405void eio_set_max_idle (unsigned int nthreads)
406{
407 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
408 max_idle = nthreads <= 0 ? 1 : nthreads;
409 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
410}
411
412void eio_set_min_parallel (unsigned int nthreads)
413{
414 if (wanted < nthreads)
415 wanted = nthreads;
416}
417
418void eio_set_max_parallel (unsigned int nthreads)
419{
420 if (wanted > nthreads)
421 wanted = nthreads;
422
423 while (started > wanted)
424 end_thread ();
425}
426
427int eio_poll (void) 605etp_poll (void)
428{ 606{
429 int maxreqs = max_poll_reqs; 607 unsigned int maxreqs;
608 unsigned int maxtime;
430 struct timeval tv_start, tv_now; 609 struct timeval tv_start, tv_now;
431 eio_req *req;
432 610
611 X_LOCK (reslock);
612 maxreqs = max_poll_reqs;
613 maxtime = max_poll_time;
614 X_UNLOCK (reslock);
615
433 if (max_poll_time) 616 if (maxtime)
434 gettimeofday (&tv_start, 0); 617 gettimeofday (&tv_start, 0);
435 618
436 for (;;) 619 for (;;)
437 { 620 {
621 ETP_REQ *req;
622
438 maybe_start_thread (); 623 etp_maybe_start_thread ();
439 624
440 X_LOCK (reslock); 625 X_LOCK (reslock);
441 req = reqq_shift (&res_queue); 626 req = reqq_shift (&res_queue);
442 627
443 if (req) 628 if (req)
451 X_UNLOCK (reslock); 636 X_UNLOCK (reslock);
452 637
453 if (!req) 638 if (!req)
454 return 0; 639 return 0;
455 640
641 X_LOCK (reqlock);
456 --nreqs; 642 --nreqs;
643 X_UNLOCK (reqlock);
457 644
458 if (req->type == EIO_GROUP && req->size) 645 if (ecb_expect_false (req->type == EIO_GROUP && req->size))
459 { 646 {
460 req->int1 = 1; /* mark request as delayed */ 647 req->int1 = 1; /* mark request as delayed */
461 continue; 648 continue;
462 } 649 }
463 else 650 else
464 { 651 {
465 int res = eio_finish (req); 652 int res = ETP_FINISH (req);
466 if (res) 653 if (ecb_expect_false (res))
467 return res; 654 return res;
468 } 655 }
469 656
470 if (maxreqs && !--maxreqs) 657 if (ecb_expect_false (maxreqs && !--maxreqs))
471 break; 658 break;
472 659
473 if (max_poll_time) 660 if (maxtime)
474 { 661 {
475 gettimeofday (&tv_now, 0); 662 gettimeofday (&tv_now, 0);
476 663
477 if (tvdiff (&tv_start, &tv_now) >= max_poll_time) 664 if (tvdiff (&tv_start, &tv_now) >= maxtime)
478 break; 665 break;
479 } 666 }
480 } 667 }
481 668
482 errno = EAGAIN; 669 errno = EAGAIN;
483 return -1; 670 return -1;
484} 671}
485 672
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
486/*****************************************************************************/ 923/*****************************************************************************/
487/* work around various missing functions */ 924/* work around various missing functions */
488 925
489#if !HAVE_PREADWRITE 926#if !HAVE_PREADWRITE
927# undef pread
928# undef pwrite
490# define pread aio_pread 929# define pread eio__pread
491# define pwrite aio_pwrite 930# define pwrite eio__pwrite
492 931
493/* 932static eio_ssize_t
494 * make our pread/pwrite safe against themselves, but not against 933eio__pread (int fd, void *buf, size_t count, off_t offset)
495 * normal read/write by using a mutex. slows down execution a lot,
496 * but that's your problem, not mine.
497 */
498static mutex_t preadwritelock = X_MUTEX_INIT;
499
500static ssize_t pread (int fd, void *buf, size_t count, off_t offset)
501{ 934{
502 ssize_t res; 935 eio_ssize_t res;
503 off_t ooffset; 936 off_t ooffset;
504 937
505 X_LOCK (preadwritelock); 938 X_LOCK (preadwritelock);
506 ooffset = lseek (fd, 0, SEEK_CUR); 939 ooffset = lseek (fd, 0, SEEK_CUR);
507 lseek (fd, offset, SEEK_SET); 940 lseek (fd, offset, SEEK_SET);
510 X_UNLOCK (preadwritelock); 943 X_UNLOCK (preadwritelock);
511 944
512 return res; 945 return res;
513} 946}
514 947
515static 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)
516{ 950{
517 ssize_t res; 951 eio_ssize_t res;
518 off_t ooffset; 952 off_t ooffset;
519 953
520 X_LOCK (preadwritelock); 954 X_LOCK (preadwritelock);
521 ooffset = lseek (fd, 0, SEEK_CUR); 955 ooffset = lseek (fd, 0, SEEK_CUR);
522 lseek (fd, offset, SEEK_SET); 956 lseek (fd, offset, SEEK_SET);
523 res = write (fd, buf, count); 957 res = write (fd, buf, count);
524 lseek (fd, offset, SEEK_SET); 958 lseek (fd, ooffset, SEEK_SET);
525 X_UNLOCK (preadwritelock); 959 X_UNLOCK (preadwritelock);
526 960
527 return res; 961 return res;
528} 962}
529#endif 963#endif
530 964
531#ifndef HAVE_FUTIMES 965#ifndef HAVE_UTIMES
532 966
967# undef utimes
533# define utimes(path,times) aio_utimes (path, times) 968# define utimes(path,times) eio__utimes (path, times)
534# define futimes(fd,times) aio_futimes (fd, times)
535 969
970static int
536static int aio_utimes (const char *filename, const struct timeval times[2]) 971eio__utimes (const char *filename, const struct timeval times[2])
537{ 972{
538 if (times) 973 if (times)
539 { 974 {
540 struct utimbuf buf; 975 struct utimbuf buf;
541 976
546 } 981 }
547 else 982 else
548 return utime (filename, 0); 983 return utime (filename, 0);
549} 984}
550 985
986#endif
987
988#ifndef HAVE_FUTIMES
989
990# undef futimes
991# define futimes(fd,times) eio__futimes (fd, times)
992
993static int
551static int aio_futimes (int fd, const struct timeval tv[2]) 994eio__futimes (int fd, const struct timeval tv[2])
552{ 995{
553 errno = ENOSYS; 996 errno = ENOSYS;
554 return -1; 997 return -1;
555} 998}
556 999
557#endif 1000#endif
558 1001
559#if !HAVE_FDATASYNC 1002#if !HAVE_FDATASYNC
1003# undef fdatasync
560# define fdatasync fsync 1004# define fdatasync(fd) fsync (fd)
561#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}
562 1063
563#if !HAVE_READAHEAD 1064#if !HAVE_READAHEAD
1065# undef readahead
564# define readahead(fd,offset,count) aio_readahead (fd, offset, count, self) 1066# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
565 1067
1068static eio_ssize_t
566static 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)
567{ 1070{
568 size_t todo = count; 1071 size_t todo = count;
569 dBUF; 1072 dBUF;
570 1073
571 while (todo > 0) 1074 while (todo > 0)
575 pread (fd, eio_buf, len, offset); 1078 pread (fd, eio_buf, len, offset);
576 offset += len; 1079 offset += len;
577 todo -= len; 1080 todo -= len;
578 } 1081 }
579 1082
1083 FUBd;
1084
580 errno = 0; 1085 errno = 0;
581 return count; 1086 return count;
582} 1087}
583 1088
584#endif 1089#endif
585 1090
586#if !HAVE_READDIR_R
587# define readdir_r aio_readdir_r
588
589static mutex_t readdirlock = X_MUTEX_INIT;
590
591static int readdir_r (DIR *dirp, EIO_STRUCT_DIRENT *ent, EIO_STRUCT_DIRENT **res)
592{
593 EIO_STRUCT_DIRENT *e;
594 int errorno;
595
596 X_LOCK (readdirlock);
597
598 e = readdir (dirp);
599 errorno = errno;
600
601 if (e)
602 {
603 *res = ent;
604 strcpy (ent->d_name, e->d_name);
605 }
606 else
607 *res = 0;
608
609 X_UNLOCK (readdirlock);
610
611 errno = errorno;
612 return e ? 0 : -1;
613}
614#endif
615
616/* sendfile always needs emulation */ 1091/* sendfile always needs emulation */
1092static eio_ssize_t
617static 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)
618{ 1094{
1095 eio_ssize_t written = 0;
619 ssize_t res; 1096 eio_ssize_t res;
620 1097
621 if (!count) 1098 if (!count)
622 return 0; 1099 return 0;
623 1100
1101 for (;;)
1102 {
1103#ifdef __APPLE__
1104# undef HAVE_SENDFILE /* broken, as everything on os x */
1105#endif
624#if HAVE_SENDFILE 1106#if HAVE_SENDFILE
625# if __linux 1107# if __linux
1108 off_t soffset = offset;
626 res = sendfile (ofd, ifd, &offset, count); 1109 res = sendfile (ofd, ifd, &soffset, count);
627 1110
628# elif __freebsd 1111# elif __FreeBSD__
629 /* 1112 /*
630 * 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
631 * wasted on making it similar to other I/O functions. 1114 * wasted on making it similar to other I/O functions.
632 */ 1115 */
633 {
634 off_t sbytes; 1116 off_t sbytes;
635 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 1117 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
636 1118
637 if (res < 0 && sbytes) 1119 #if 0 /* according to the manpage, this is correct, but broken behaviour */
638 /* 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)
639 res = sbytes; 1124 res = sbytes;
640 } 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;
641 1138
642# elif __hpux 1139# elif __hpux
643 res = sendfile (ofd, ifd, offset, count, 0, 0); 1140 res = sendfile (ofd, ifd, offset, count, 0, 0);
644 1141
645# elif __solaris 1142# elif __solaris
646 {
647 struct sendfilevec vec; 1143 struct sendfilevec vec;
648 size_t sbytes; 1144 size_t sbytes;
649 1145
650 vec.sfv_fd = ifd; 1146 vec.sfv_fd = ifd;
651 vec.sfv_flag = 0; 1147 vec.sfv_flag = 0;
652 vec.sfv_off = offset; 1148 vec.sfv_off = offset;
653 vec.sfv_len = count; 1149 vec.sfv_len = count;
654 1150
655 res = sendfilev (ofd, &vec, 1, &sbytes); 1151 res = sendfilev (ofd, &vec, 1, &sbytes);
656 1152
657 if (res < 0 && sbytes) 1153 if (res < 0 && sbytes)
658 res = sbytes; 1154 res = sbytes;
659 }
660 1155
661# 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
662#else 1167#else
663 res = -1; 1168 res = -1;
664 errno = ENOSYS; 1169 errno = ENOSYS;
665#endif 1170#endif
666 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
667 if (res < 0 1195 if (res < 0
668 && (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
669#if __solaris 1204#if __solaris
670 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1205 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
671#endif 1206#endif
672 ) 1207 )
673 ) 1208 )
677 1212
678 res = 0; 1213 res = 0;
679 1214
680 while (count) 1215 while (count)
681 { 1216 {
682 ssize_t cnt; 1217 eio_ssize_t cnt;
683 1218
684 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);
685 1220
686 if (cnt <= 0) 1221 if (cnt <= 0)
687 { 1222 {
699 1234
700 offset += cnt; 1235 offset += cnt;
701 res += cnt; 1236 res += cnt;
702 count -= cnt; 1237 count -= cnt;
703 } 1238 }
1239
1240 FUBd;
704 } 1241 }
705 1242
706 return res; 1243 return res;
707} 1244}
708 1245
709/* read a full directory */ 1246#ifdef PAGESIZE
710static void scandir_ (eio_req *req, worker *self) 1247# define eio_pagesize() PAGESIZE
1248#else
1249static intptr_t
1250eio_pagesize (void)
711{ 1251{
712 DIR *dirp; 1252 static intptr_t page;
713 union
714 {
715 EIO_STRUCT_DIRENT d;
716 char b [offsetof (EIO_STRUCT_DIRENT, d_name) + NAME_MAX + 1];
717 } *u;
718 EIO_STRUCT_DIRENT *entp;
719 char *name, *names;
720 int memlen = 4096;
721 int memofs = 0;
722 int res = 0;
723 1253
724 X_LOCK (wrklock); 1254 if (!page)
725 self->dirp = dirp = opendir (req->ptr1); 1255 page = sysconf (_SC_PAGESIZE);
726 self->dbuf = u = malloc (sizeof (*u));
727 req->flags |= EIO_FLAG_PTR2_FREE;
728 req->ptr2 = names = malloc (memlen);
729 X_UNLOCK (wrklock);
730 1256
731 if (dirp && u && names) 1257 return page;
732 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)
733 { 1414 {
734 errno = 0; 1415 sprintf (tmp1, "/proc/self/fd/%d", fd);
735 readdir_r (dirp, &u->d, &entp); 1416 req->result = readlink (tmp1, res, PATH_MAX);
1417 close (fd);
736 1418
737 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 {
738 break; 1451 ++rel;
1452 continue;
1453 }
739 1454
740 name = entp->d_name; 1455 if (beg [0] == '.')
1456 {
1457 if (len == 1)
1458 continue; /* . - nop */
741 1459
742 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)
743 { 1487 {
1488 if (errno != EINVAL)
1489 return -1;
1490
1491 /* it's a normal directory. hopefully */
1492 res += len + 1;
1493 }
1494 else
1495 {
1496 /* yay, it was a symlink - build new path in tmp2 */
744 int len = strlen (name) + 1; 1497 int rellen = strlen (rel);
745 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;
1516 }
1517 }
1518
1519 /* special case for the lone root path */
1520 if (res == tmpbuf->ptr)
1521 *res++ = '/';
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)
746 res++; 1610 break;
747 1611
748 while (memofs + len > memlen) 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))
749 { 1619 {
750 memlen *= 2; 1620 eio_dirent tmp = *a; *a = *b; *b = tmp;
751 X_LOCK (wrklock);
752 req->ptr2 = names = realloc (names, memlen);
753 X_UNLOCK (wrklock);
754
755 if (!names)
756 break; 1621 ++a;
757 } 1622 }
1623 while (b > a);
758 1624
759 memcpy (names + memofs, name, len); 1625 /* next bit, or stop, if no bits left in this path */
760 memofs += len; 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;
761 } 1635 }
762 } 1636 }
763 1637 while (stk_idx--);
764 if (errno)
765 res = -1;
766 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
767 req->result = res; 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)
1790 {
1791#ifndef _WIN32
1792 int old_errno = errno;
1793 closedir (dirp);
1794 errno = old_errno;
1795
1796 if (errno)
1797 break;
1798#endif
1799
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))
1845 {
1846 namesalloc *= 2;
1847 req->ptr2 = names = realloc (names, namesalloc);
1848
1849 if (!names)
1850 break;
1851 }
1852
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;
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 }
768} 1961}
1962
1963/*****************************************************************************/
1964/* working directory stuff */
1965
1966#if HAVE_AT
1967
1968#define WD2FD(wd) (wd ? ((int)wd) - 1 : AT_FDCWD)
1969
1970eio_wd
1971eio_wd_open_sync (eio_wd wd, const char *path)
1972{
1973 int fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1974
1975 return fd >= 0 ? (eio_wd)(fd + 1) : EIO_INVALID_WD;
1976}
1977
1978void
1979eio_wd_close_sync (eio_wd wd)
1980{
1981 int fd = WD2FD (wd);
1982
1983 if (fd >= 0)
1984 close (fd);
1985}
1986
1987#else
1988
1989/* on legacy systems, we represent the working directories simply by their path strings */
1990
1991static const char *
1992wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
1993{
1994 if (!wd || *path == '/')
1995 return path;
1996
1997 {
1998 int l1 = strlen ((const char *)wd);
1999 int l2 = strlen (path);
2000
2001 char *res = tmpbuf_get (tmpbuf, l1 + l2 + 2);
2002
2003 memcpy (res, wd, l1);
2004 res [l1] = '/';
2005 memcpy (res + l1 + 1, path, l2 + 1);
2006
2007 return res;
2008 }
2009}
2010
2011eio_wd
2012eio__wd_open_sync (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
2013{
2014 if (*path == '/') /* absolute paths ignore wd */
2015 path = strdup (path);
2016 else if (path [0] == '.' && !path [1]) /* special case '.', as it is common */
2017 return wd;
2018 else
2019 {
2020 int len = eio__realpath (tmpbuf, wd, path);
2021
2022 path = EIO_INVALID_WD;
2023
2024 if (len >= 0)
2025 {
2026 ((char *)tmpbuf->ptr)[len] = 0;
2027 path = strdup (tmpbuf->ptr);
2028 }
2029 }
2030
2031 if (!path)
2032 path = EIO_INVALID_WD;
2033
2034 return (eio_wd)path;
2035}
2036
2037eio_wd
2038eio_wd_open_sync (eio_wd wd, const char *path)
2039{
2040 struct tmpbuf tmpbuf = { 0 };
2041 wd = eio__wd_open_sync (&tmpbuf, wd, path);
2042 free (tmpbuf.ptr);
2043
2044 return wd;
2045}
2046
2047void
2048eio_wd_close_sync (eio_wd wd)
2049{
2050 if (wd != EIO_INVALID_WD)
2051 free (wd);
2052}
2053
2054#endif
769 2055
770/*****************************************************************************/ 2056/*****************************************************************************/
771 2057
772#define ALLOC(len) \ 2058#define ALLOC(len) \
773 if (!req->ptr2) \ 2059 if (!req->ptr2) \
782 req->result = -1; \ 2068 req->result = -1; \
783 break; \ 2069 break; \
784 } \ 2070 } \
785 } 2071 }
786 2072
787X_THREAD_PROC (eio_proc) 2073X_THREAD_PROC (etp_proc)
788{ 2074{
789 eio_req *req; 2075 ETP_REQ *req;
790 struct timespec ts; 2076 struct timespec ts;
791 worker *self = (worker *)thr_arg; 2077 etp_worker *self = (etp_worker *)thr_arg;
792 2078
2079#if HAVE_PRCTL_SET_NAME
2080 prctl (PR_SET_NAME, (unsigned long)"eio_thread", 0, 0, 0);
2081#endif
2082
793 /* try to distribute timeouts somewhat randomly */ 2083 /* try to distribute timeouts somewhat evenly */
794 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 2084 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
795 2085
796 for (;;) 2086 for (;;)
797 { 2087 {
798 ts.tv_sec = time (0) + IDLE_TIMEOUT; 2088 ts.tv_sec = 0;
799 2089
800 X_LOCK (reqlock); 2090 X_LOCK (reqlock);
801 2091
802 for (;;) 2092 for (;;)
803 { 2093 {
804 self->req = req = reqq_shift (&req_queue); 2094 req = reqq_shift (&req_queue);
805 2095
806 if (req) 2096 if (req)
807 break; 2097 break;
808 2098
2099 if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
2100 {
2101 X_UNLOCK (reqlock);
2102 X_LOCK (wrklock);
2103 --started;
2104 X_UNLOCK (wrklock);
2105 goto quit;
2106 }
2107
809 ++idle; 2108 ++idle;
810 2109
811 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 2110 if (idle <= max_idle)
2111 /* we are allowed to idle, so do so without any timeout */
2112 X_COND_WAIT (reqwait, reqlock);
2113 else
812 { 2114 {
813 if (idle > max_idle) 2115 /* initialise timeout once */
814 { 2116 if (!ts.tv_sec)
815 --idle; 2117 ts.tv_sec = time (0) + idle_timeout;
816 X_UNLOCK (reqlock);
817 X_LOCK (wrklock);
818 --started;
819 X_UNLOCK (wrklock);
820 goto quit;
821 }
822 2118
823 /* we are allowed to idle, so do so without any timeout */
824 X_COND_WAIT (reqwait, reqlock); 2119 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
825 ts.tv_sec = time (0) + IDLE_TIMEOUT; 2120 ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
826 } 2121 }
827 2122
828 --idle; 2123 --idle;
829 } 2124 }
830 2125
831 --nready; 2126 --nready;
832 2127
833 X_UNLOCK (reqlock); 2128 X_UNLOCK (reqlock);
834 2129
835 errno = 0; /* strictly unnecessary */
836
837 if (!EIO_CANCELLED (req))
838 switch (req->type) 2130 if (req->type < 0)
839 {
840 case EIO_READ: ALLOC (req->size);
841 req->result = req->offs >= 0
842 ? pread (req->int1, req->ptr2, req->size, req->offs)
843 : read (req->int1, req->ptr2, req->size); break;
844 case EIO_WRITE: req->result = req->offs >= 0
845 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
846 : write (req->int1, req->ptr2, req->size); break;
847
848 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
849 case EIO_SENDFILE: req->result = sendfile_ (req->int1, req->int2, req->offs, req->size, self); break;
850
851 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
852 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
853 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
854 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
855 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
856 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
857
858 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
859 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
860 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
861 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
862 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
863 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
864
865 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
866 case EIO_CLOSE: req->result = close (req->int1); break;
867 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
868 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
869 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
870 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
871 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
872 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
873 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
874 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
875
876 case EIO_READLINK: ALLOC (NAME_MAX);
877 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
878
879 case EIO_SYNC: req->result = 0; sync (); break;
880 case EIO_FSYNC: req->result = fsync (req->int1); break;
881 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
882
883 case EIO_READDIR: scandir_ (req, self); break;
884
885 case EIO_BUSY:
886#ifdef _WIN32
887 Sleep (req->nv1 * 1000.);
888#else
889 {
890 struct timeval tv;
891
892 tv.tv_sec = req->nv1;
893 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;
894
895 req->result = select (0, 0, 0, 0, &tv);
896 }
897#endif
898 break;
899
900 case EIO_UTIME:
901 case EIO_FUTIME:
902 {
903 struct timeval tv[2];
904 struct timeval *times;
905
906 if (req->nv1 != -1. || req->nv2 != -1.)
907 {
908 tv[0].tv_sec = req->nv1;
909 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
910 tv[1].tv_sec = req->nv2;
911 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
912
913 times = tv;
914 }
915 else
916 times = 0;
917
918
919 req->result = req->type == EIO_FUTIME
920 ? futimes (req->int1, times)
921 : utimes (req->ptr1, times);
922 }
923
924 case EIO_GROUP:
925 case EIO_NOP:
926 break;
927
928 case EIO_QUIT:
929 goto quit; 2131 goto quit;
930 2132
931 default: 2133 ETP_EXECUTE (self, req);
932 req->result = -1;
933 break;
934 }
935
936 req->errorno = errno;
937 2134
938 X_LOCK (reslock); 2135 X_LOCK (reslock);
939 2136
940 ++npending; 2137 ++npending;
941 2138
942 if (!reqq_push (&res_queue, req) && want_poll_cb) 2139 if (!reqq_push (&res_queue, req) && want_poll_cb)
943 want_poll_cb (); 2140 want_poll_cb ();
944 2141
945 self->req = 0;
946 worker_clear (self); 2142 etp_worker_clear (self);
947 2143
948 X_UNLOCK (reslock); 2144 X_UNLOCK (reslock);
949 } 2145 }
950 2146
951quit: 2147quit:
2148 free (req);
2149
952 X_LOCK (wrklock); 2150 X_LOCK (wrklock);
953 worker_free (self); 2151 etp_worker_free (self);
954 X_UNLOCK (wrklock); 2152 X_UNLOCK (wrklock);
955 2153
956 return 0; 2154 return 0;
957} 2155}
958 2156
959/*****************************************************************************/ 2157/*****************************************************************************/
960 2158
961static void eio_atfork_prepare (void) 2159int ecb_cold
2160eio_init (void (*want_poll)(void), void (*done_poll)(void))
962{ 2161{
963 X_LOCK (wrklock);
964 X_LOCK (reqlock);
965 X_LOCK (reslock);
966#if !HAVE_PREADWRITE 2162#if !HAVE_PREADWRITE
967 X_LOCK (preadwritelock);
968#endif
969#if !HAVE_READDIR_R
970 X_LOCK (readdirlock);
971#endif
972}
973
974static void eio_atfork_parent (void)
975{
976#if !HAVE_READDIR_R
977 X_UNLOCK (readdirlock);
978#endif
979#if !HAVE_PREADWRITE
980 X_UNLOCK (preadwritelock);
981#endif
982 X_UNLOCK (reslock);
983 X_UNLOCK (reqlock);
984 X_UNLOCK (wrklock);
985}
986
987static void eio_atfork_child (void)
988{
989 eio_req *prv;
990
991 while (prv = reqq_shift (&req_queue))
992 eio_destroy (prv);
993
994 while (prv = reqq_shift (&res_queue))
995 eio_destroy (prv);
996
997 while (wrk_first.next != &wrk_first)
998 {
999 worker *wrk = wrk_first.next;
1000
1001 if (wrk->req)
1002 eio_destroy (wrk->req);
1003
1004 worker_clear (wrk);
1005 worker_free (wrk);
1006 }
1007
1008 started = 0;
1009 idle = 0;
1010 nreqs = 0;
1011 nready = 0;
1012 npending = 0;
1013
1014 eio_atfork_parent ();
1015}
1016
1017int eio_init (void (*want_poll)(void), void (*done_poll)(void))
1018{
1019 want_poll_cb = want_poll;
1020 done_poll_cb = done_poll;
1021
1022#ifdef _WIN32
1023 X_MUTEX_CHECK (wrklock);
1024 X_MUTEX_CHECK (reslock);
1025 X_MUTEX_CHECK (reqlock);
1026 X_MUTEX_CHECK (reqwait);
1027 X_MUTEX_CHECK (preadwritelock); 2163 X_MUTEX_CREATE (preadwritelock);
1028 X_MUTEX_CHECK (readdirlock);
1029
1030 X_COND_CHECK (reqwait);
1031#endif 2164#endif
1032 2165
1033 X_THREAD_ATFORK (eio_atfork_prepare, eio_atfork_parent, eio_atfork_child); 2166 return etp_init (want_poll, done_poll);
1034} 2167}
1035 2168
2169ecb_inline void
1036static void eio_api_destroy (eio_req *req) 2170eio_api_destroy (eio_req *req)
1037{ 2171{
1038 free (req); 2172 free (req);
1039} 2173}
1040 2174
1041#define REQ(rtype) \ 2175#define REQ(rtype) \
1043 \ 2177 \
1044 req = (eio_req *)calloc (1, sizeof *req); \ 2178 req = (eio_req *)calloc (1, sizeof *req); \
1045 if (!req) \ 2179 if (!req) \
1046 return 0; \ 2180 return 0; \
1047 \ 2181 \
1048 req->type = EIO_ ## rtype; \ 2182 req->type = rtype; \
1049 req->pri = EIO_DEFAULT_PRI + EIO_PRI_BIAS; \ 2183 req->pri = pri; \
1050 req->finish = cb; \ 2184 req->finish = cb; \
2185 req->data = data; \
1051 req->destroy = eio_api_destroy; 2186 req->destroy = eio_api_destroy;
1052 2187
1053#define SEND eio_submit (req); return req 2188#define SEND eio_submit (req); return req
1054 2189
1055#define PATH (void)0 2190#define PATH \
2191 req->flags |= EIO_FLAG_PTR1_FREE; \
2192 req->ptr1 = strdup (path); \
2193 if (!req->ptr1) \
2194 { \
2195 eio_api_destroy (req); \
2196 return 0; \
2197 }
1056 2198
2199static void
2200eio_execute (etp_worker *self, eio_req *req)
2201{
2202#if HAVE_AT
2203 int dirfd;
2204#else
2205 const char *path;
2206#endif
2207
2208 if (ecb_expect_false (EIO_CANCELLED (req)))
2209 {
2210 req->result = -1;
2211 req->errorno = ECANCELED;
2212 return;
2213 }
2214
2215 if (ecb_expect_false (req->wd == EIO_INVALID_WD))
2216 {
2217 req->result = -1;
2218 req->errorno = ENOENT;
2219 return;
2220 }
2221
2222 if (req->type >= EIO_OPEN)
2223 {
2224 #if HAVE_AT
2225 dirfd = WD2FD (req->wd);
2226 #else
2227 path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
2228 #endif
2229 }
2230
2231 switch (req->type)
2232 {
2233 case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
2234 req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
2235 break;
2236 case EIO_WD_CLOSE: eio_wd_close_sync (req->wd); break;
2237
2238 case EIO_READ: ALLOC (req->size);
2239 req->result = req->offs >= 0
2240 ? pread (req->int1, req->ptr2, req->size, req->offs)
2241 : read (req->int1, req->ptr2, req->size); break;
2242 case EIO_WRITE: req->result = req->offs >= 0
2243 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
2244 : write (req->int1, req->ptr2, req->size); break;
2245
2246 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
2247 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
2248
2249#if HAVE_AT
2250 case EIO_GETPATH: abort ();
2251 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2252 req->result = fstatat (dirfd, path, (EIO_STRUCT_STAT *)req->ptr2); break;
2253 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2254 req->result = lstat (dirfd, path, (EIO_STRUCT_STAT *)req->ptr2); break;
2255#if 0/*D*/
2256 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2257 req->result = statvfs (dirfd, path, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2258#endif
2259 case EIO_CHOWN: req->result = chown (dirfd, path, req->int2, req->int3); break;
2260 case EIO_CHMOD: req->result = chmod (dirfd, path, (mode_t)req->int2); break;
2261 case EIO_TRUNCATE: req->result = truncate (dirfd, path, req->offs); break;
2262 case EIO_OPEN: req->result = open (dirfd, path, req->int1, (mode_t)req->int2); break;
2263
2264 case EIO_UNLINK: req->result = unlink (dirfd, path); break;
2265 case EIO_RMDIR: req->result = rmdir (dirfd, path); break;
2266 case EIO_MKDIR: req->result = mkdir (dirfd, path, (mode_t)req->int2); break;
2267 case EIO_RENAME: req->result = rename (dirfd, path, req->ptr2); break;
2268 case EIO_LINK: req->result = link (dirfd, path, req->ptr2); break;
2269 case EIO_SYMLINK: req->result = symlink (dirfd, path, req->ptr2); break;
2270 case EIO_MKNOD: req->result = mknod (dirfd, path, (mode_t)req->int2, (dev_t)req->offs); break;
2271#else
2272 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2273 req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
2274 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2275 req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
2276 case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
2277 case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
2278 case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
2279 case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
2280
2281 case EIO_UNLINK: req->result = unlink (path ); break;
2282 case EIO_RMDIR: req->result = rmdir (path ); break;
2283 case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
2284 case EIO_RENAME: req->result = rename (path , req->ptr2); break;
2285 case EIO_LINK: req->result = link (path , req->ptr2); break;
2286 case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
2287 case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
2288
2289 case EIO_READLINK: ALLOC (PATH_MAX);
2290 req->result = readlink (path, req->ptr2, PATH_MAX); break;
2291#endif
2292
2293 case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
2294 {
2295 ALLOC (req->result);
2296 memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
2297 }
2298 break;
2299
2300 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS)); /*D*/
2301 req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
2302
2303 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2304 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
2305
2306 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2307 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2308
2309 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
2310 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
2311 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
2312
2313 case EIO_CLOSE: req->result = close (req->int1); break;
2314 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
2315 case EIO_SYNC: req->result = 0; sync (); break;
2316 case EIO_FSYNC: req->result = fsync (req->int1); break;
2317 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
2318 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
2319 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
2320 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
2321 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
2322 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
2323 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
2324 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
2325
2326 case EIO_READDIR: eio__scandir (req); break;
2327
2328 case EIO_BUSY:
2329#ifdef _WIN32
2330 Sleep (req->nv1 * 1e3);
2331#else
2332 {
2333 struct timeval tv;
2334
2335 tv.tv_sec = req->nv1;
2336 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
2337
2338 req->result = select (0, 0, 0, 0, &tv);
2339 }
2340#endif
2341 break;
2342
2343 case EIO_UTIME:
2344 case EIO_FUTIME:
2345 {
2346 struct timeval tv[2];
2347 struct timeval *times;
2348
2349 if (req->nv1 != -1. || req->nv2 != -1.)
2350 {
2351 tv[0].tv_sec = req->nv1;
2352 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
2353 tv[1].tv_sec = req->nv2;
2354 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
2355
2356 times = tv;
2357 }
2358 else
2359 times = 0;
2360
2361 req->result = req->type == EIO_FUTIME
2362 ? futimes (req->int1, times)
2363 : utimes (req->ptr1, times);
2364 }
2365 break;
2366
2367 case EIO_GROUP:
2368 abort (); /* handled in eio_request */
2369
2370 case EIO_NOP:
2371 req->result = 0;
2372 break;
2373
2374 case EIO_CUSTOM:
2375 req->feed (req);
2376 break;
2377
2378 default:
2379 errno = ENOSYS;
2380 req->result = -1;
2381 break;
2382 }
2383
2384 req->errorno = errno;
2385}
2386
2387#ifndef EIO_NO_WRAPPERS
2388
2389eio_req *eio_nop (int pri, eio_cb cb, void *data)
2390{
2391 REQ (EIO_NOP); SEND;
2392}
2393
2394eio_req *eio_busy (double delay, int pri, eio_cb cb, void *data)
2395{
2396 REQ (EIO_BUSY); req->nv1 = delay; SEND;
2397}
2398
1057eio_req *eio_fsync (int fd, eio_cb cb) 2399eio_req *eio_sync (int pri, eio_cb cb, void *data)
1058{ 2400{
2401 REQ (EIO_SYNC); SEND;
2402}
2403
2404eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
2405{
1059 REQ (FSYNC); req->int1 = fd; SEND; 2406 REQ (EIO_FSYNC); req->int1 = fd; SEND;
1060} 2407}
1061 2408
2409eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2410{
2411 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2412}
2413
1062eio_req *eio_fdatasync (int fd, eio_cb cb) 2414eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1063{ 2415{
1064 REQ (FDATASYNC); req->int1 = fd; SEND; 2416 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1065} 2417}
1066 2418
2419eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
2420{
2421 REQ (EIO_SYNCFS); req->int1 = fd; SEND;
2422}
2423
2424eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2425{
2426 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2427}
2428
2429eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2430{
2431 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2432}
2433
2434eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
2435{
2436 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
2437}
2438
2439eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
2440{
2441 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2442}
2443
2444eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2445{
2446 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
2447}
2448
2449eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
2450{
2451 REQ (EIO_CLOSE); req->int1 = fd; SEND;
2452}
2453
1067eio_req *eio_readahead (int fd, off_t offset, size_t length, eio_cb cb) 2454eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
1068{ 2455{
1069 REQ (READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND; 2456 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
1070} 2457}
1071 2458
1072eio_req *eio_read (int fd, off_t offset, size_t length, void *data, eio_cb cb) 2459eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1073{ 2460{
1074 REQ (READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = data; SEND; 2461 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1075} 2462}
1076 2463
1077eio_req *eio_write (int fd, off_t offset, size_t length, void *data, eio_cb cb) 2464eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1078{ 2465{
1079 REQ (WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = data; SEND; 2466 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1080} 2467}
1081 2468
1082eio_req *eio_fstat (int fd, eio_cb cb) 2469eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1083{ 2470{
1084 REQ (FSTAT); req->int1 = fd; SEND; 2471 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1085} 2472}
1086 2473
2474eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
2475{
2476 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
2477}
2478
1087eio_req *eio_futime (int fd, double atime, double mtime, eio_cb cb) 2479eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1088{ 2480{
1089 REQ (FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; 2481 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1090} 2482}
1091 2483
1092eio_req *eio_ftruncate (int fd, off_t offset, eio_cb cb) 2484eio_req *eio_ftruncate (int fd, off_t offset, int pri, eio_cb cb, void *data)
1093{ 2485{
1094 REQ (FTRUNCATE); req->int1 = fd; req->offs = offset; SEND; 2486 REQ (EIO_FTRUNCATE); req->int1 = fd; req->offs = offset; SEND;
1095} 2487}
1096 2488
1097eio_req *eio_fchmod (int fd, mode_t mode, eio_cb cb) 2489eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1098{ 2490{
1099 REQ (FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2491 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1100} 2492}
1101 2493
1102eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, eio_cb cb) 2494eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1103{ 2495{
1104 REQ (FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2496 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1105} 2497}
1106 2498
1107eio_req *eio_dup2 (int fd, int fd2, eio_cb cb) 2499eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
1108{ 2500{
1109 REQ (DUP2); req->int1 = fd; req->int2 = fd2; SEND; 2501 REQ (EIO_DUP2); req->int1 = fd; req->int2 = fd2; SEND;
1110} 2502}
1111 2503
1112eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, eio_cb cb) 2504eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, int pri, eio_cb cb, void *data)
1113{ 2505{
1114 REQ (SENDFILE); req->int1 = out_fd; req->int2 = in_fd; req->offs = in_offset; req->size = length; SEND; 2506 REQ (EIO_SENDFILE); req->int1 = out_fd; req->int2 = in_fd; req->offs = in_offset; req->size = length; SEND;
1115} 2507}
1116 2508
1117eio_req *eio_open (const char *path, int flags, mode_t mode, eio_cb cb) 2509eio_req *eio_open (const char *path, int flags, mode_t mode, int pri, eio_cb cb, void *data)
1118{ 2510{
1119 REQ (OPEN); PATH; req->int1 = flags; req->int2 = (long)mode; SEND; 2511 REQ (EIO_OPEN); PATH; req->int1 = flags; req->int2 = (long)mode; SEND;
1120} 2512}
1121 2513
1122eio_req *eio_readlink (const char *path, eio_cb cb)
1123{
1124 REQ (READLINK); PATH; SEND;
1125}
1126
1127eio_req *eio_stat (const char *path, eio_cb cb)
1128{
1129 REQ (STAT); PATH; SEND;
1130}
1131
1132eio_req *eio_lstat (const char *path, eio_cb cb)
1133{
1134 REQ (LSTAT); PATH; SEND;
1135}
1136
1137eio_req *eio_utime (const char *path, double atime, double mtime, eio_cb cb) 2514eio_req *eio_utime (const char *path, double atime, double mtime, int pri, eio_cb cb, void *data)
1138{ 2515{
1139 REQ (UTIME); PATH; req->nv1 = atime; req->nv2 = mtime; SEND; 2516 REQ (EIO_UTIME); PATH; req->nv1 = atime; req->nv2 = mtime; SEND;
1140} 2517}
1141 2518
1142eio_req *eio_truncate (const char *path, off_t offset, eio_cb cb) 2519eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1143{ 2520{
1144 REQ (TRUNCATE); PATH; req->offs = offset; SEND; 2521 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1145} 2522}
1146 2523
1147eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, eio_cb cb) 2524eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1148{ 2525{
1149 REQ (CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2526 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1150} 2527}
1151 2528
1152eio_req *eio_chmod (const char *path, mode_t mode, eio_cb cb) 2529eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
1153{ 2530{
1154 REQ (CHMOD); PATH; req->int2 = (long)mode; SEND; 2531 REQ (EIO_CHMOD); PATH; req->int2 = (long)mode; SEND;
1155} 2532}
1156 2533
1157eio_req *eio_mkdir (const char *path, mode_t mode, eio_cb cb) 2534eio_req *eio_mkdir (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
1158{ 2535{
1159 REQ (MKDIR); PATH; req->int2 = (long)mode; SEND; 2536 REQ (EIO_MKDIR); PATH; req->int2 = (long)mode; SEND;
1160} 2537}
1161 2538
1162eio_req *eio_unlink (const char *path, eio_cb cb) 2539static eio_req *
2540eio__1path (int type, const char *path, int pri, eio_cb cb, void *data)
1163{ 2541{
1164 REQ (UNLINK); PATH; SEND;
1165}
1166
1167eio_req *eio_rmdir (const char *path, eio_cb cb)
1168{
1169 REQ (RMDIR); PATH; SEND; 2542 REQ (type); PATH; SEND;
1170} 2543}
1171 2544
2545eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
2546{
2547 return eio__1path (EIO_READLINK, path, pri, cb, data);
2548}
2549
2550eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2551{
2552 return eio__1path (EIO_REALPATH, path, pri, cb, data);
2553}
2554
2555eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
2556{
2557 return eio__1path (EIO_STAT, path, pri, cb, data);
2558}
2559
2560eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
2561{
2562 return eio__1path (EIO_LSTAT, path, pri, cb, data);
2563}
2564
2565eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
2566{
2567 return eio__1path (EIO_STATVFS, path, pri, cb, data);
2568}
2569
2570eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
2571{
2572 return eio__1path (EIO_UNLINK, path, pri, cb, data);
2573}
2574
1172eio_req *eio_readdir (const char *path, eio_cb cb) 2575eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1173{ 2576{
1174 REQ (READDIR); PATH; SEND; 2577 return eio__1path (EIO_RMDIR, path, pri, cb, data);
1175} 2578}
1176 2579
2580eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
2581{
2582 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
2583}
2584
1177eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, eio_cb cb) 2585eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1178{ 2586{
1179 REQ (MKNOD); PATH; req->int2 = (long)mode; req->int2 = (long)dev; SEND; 2587 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1180} 2588}
1181 2589
1182eio_req *eio_busy (double delay, eio_cb cb) 2590static eio_req *
2591eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1183{ 2592{
1184 REQ (BUSY); req->nv1 = delay; SEND; 2593 REQ (type); PATH;
1185}
1186 2594
2595 req->flags |= EIO_FLAG_PTR2_FREE;
2596 req->ptr2 = strdup (new_path);
2597 if (!req->ptr2)
2598 {
2599 eio_api_destroy (req);
2600 return 0;
2601 }
2602
2603 SEND;
2604}
2605
2606eio_req *eio_link (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2607{
2608 return eio__2path (EIO_LINK, path, new_path, pri, cb, data);
2609}
2610
2611eio_req *eio_symlink (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2612{
2613 return eio__2path (EIO_SYMLINK, path, new_path, pri, cb, data);
2614}
2615
2616eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2617{
2618 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
2619}
2620
2621eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
2622{
2623 REQ (EIO_CUSTOM); req->feed = execute; SEND;
2624}
2625
2626#endif
2627
1187eio_req *eio_nop (eio_cb cb) 2628eio_req *eio_grp (eio_cb cb, void *data)
1188{ 2629{
2630 const int pri = EIO_PRI_MAX;
2631
1189 REQ (NOP); SEND; 2632 REQ (EIO_GROUP); SEND;
1190} 2633}
1191 2634
1192#undef REQ 2635#undef REQ
1193#undef PATH 2636#undef PATH
1194#undef SEND 2637#undef SEND
1195 2638
1196#if 0 2639/*****************************************************************************/
2640/* grp functions */
1197 2641
1198void 2642void
1199aio_open (SV8 *pathname, int flags, int mode, SV *callback=&PL_sv_undef)
1200 PROTOTYPE: $$$;$
1201 PPCODE:
1202{
1203 dREQ;
1204
1205 req->type = EIO_OPEN;
1206 req->sv1 = newSVsv (pathname);
1207 req->ptr1 = SvPVbyte_nolen (req->sv1);
1208 req->int1 = flags;
1209 req->int2 = mode;
1210
1211 EIO_SEND;
1212}
1213
1214void
1215aio_fsync (SV *fh, SV *callback=&PL_sv_undef)
1216 PROTOTYPE: $;$
1217 ALIAS:
1218 aio_fsync = EIO_FSYNC
1219 aio_fdatasync = EIO_FDATASYNC
1220 PPCODE:
1221{
1222 dREQ;
1223
1224 req->type = ix;
1225 req->sv1 = newSVsv (fh);
1226 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1227
1228 EIO_SEND (req);
1229}
1230
1231void
1232aio_close (SV *fh, SV *callback=&PL_sv_undef)
1233 PROTOTYPE: $;$
1234 PPCODE:
1235{
1236 dREQ;
1237
1238 req->type = EIO_CLOSE;
1239 req->sv1 = newSVsv (fh);
1240 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1241
1242 EIO_SEND (req);
1243}
1244
1245void
1246aio_read (SV *fh, SV *offset, SV *length, SV8 *data, IV dataoffset, SV *callback=&PL_sv_undef)
1247 ALIAS:
1248 aio_read = EIO_READ
1249 aio_write = EIO_WRITE
1250 PROTOTYPE: $$$$$;$
1251 PPCODE:
1252{
1253 STRLEN svlen;
1254 char *svptr = SvPVbyte (data, svlen);
1255 UV len = SvUV (length);
1256
1257 SvUPGRADE (data, SVt_PV);
1258 SvPOK_on (data);
1259
1260 if (dataoffset < 0)
1261 dataoffset += svlen;
1262
1263 if (dataoffset < 0 || dataoffset > svlen)
1264 croak ("dataoffset outside of data scalar");
1265
1266 if (ix == EIO_WRITE)
1267 {
1268 /* write: check length and adjust. */
1269 if (!SvOK (length) || len + dataoffset > svlen)
1270 len = svlen - dataoffset;
1271 }
1272 else
1273 {
1274 /* read: grow scalar as necessary */
1275 svptr = SvGROW (data, len + dataoffset + 1);
1276 }
1277
1278 if (len < 0)
1279 croak ("length must not be negative");
1280
1281 {
1282 dREQ;
1283
1284 req->type = ix;
1285 req->sv1 = newSVsv (fh);
1286 req->int1 = PerlIO_fileno (ix == EIO_READ ? IoIFP (sv_2io (fh))
1287 : IoOFP (sv_2io (fh)));
1288 req->offs = SvOK (offset) ? SvVAL64 (offset) : -1;
1289 req->size = len;
1290 req->sv2 = SvREFCNT_inc (data);
1291 req->ptr2 = (char *)svptr + dataoffset;
1292 req->stroffset = dataoffset;
1293
1294 if (!SvREADONLY (data))
1295 {
1296 SvREADONLY_on (data);
1297 req->flags |= FLAG_SV2_RO_OFF;
1298 }
1299
1300 EIO_SEND;
1301 }
1302}
1303
1304void
1305aio_readlink (SV8 *path, SV *callback=&PL_sv_undef)
1306 PROTOTYPE: $$;$
1307 PPCODE:
1308{
1309 SV *data;
1310 dREQ;
1311
1312 data = newSV (NAME_MAX);
1313 SvPOK_on (data);
1314
1315 req->type = EIO_READLINK;
1316 req->sv1 = newSVsv (path);
1317 req->ptr1 = SvPVbyte_nolen (req->sv1);
1318 req->sv2 = data;
1319 req->ptr2 = SvPVbyte_nolen (data);
1320
1321 EIO_SEND;
1322}
1323
1324void
1325aio_sendfile (SV *out_fh, SV *in_fh, SV *in_offset, UV length, SV *callback=&PL_sv_undef)
1326 PROTOTYPE: $$$$;$
1327 PPCODE:
1328{
1329 dREQ;
1330
1331 req->type = EIO_SENDFILE;
1332 req->sv1 = newSVsv (out_fh);
1333 req->int1 = PerlIO_fileno (IoIFP (sv_2io (out_fh)));
1334 req->sv2 = newSVsv (in_fh);
1335 req->int2 = PerlIO_fileno (IoIFP (sv_2io (in_fh)));
1336 req->offs = SvVAL64 (in_offset);
1337 req->size = length;
1338
1339 EIO_SEND;
1340}
1341
1342void
1343aio_readahead (SV *fh, SV *offset, IV length, SV *callback=&PL_sv_undef)
1344 PROTOTYPE: $$$;$
1345 PPCODE:
1346{
1347 dREQ;
1348
1349 req->type = EIO_READAHEAD;
1350 req->sv1 = newSVsv (fh);
1351 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1352 req->offs = SvVAL64 (offset);
1353 req->size = length;
1354
1355 EIO_SEND;
1356}
1357
1358void
1359aio_stat (SV8 *fh_or_path, SV *callback=&PL_sv_undef)
1360 ALIAS:
1361 aio_stat = EIO_STAT
1362 aio_lstat = EIO_LSTAT
1363 PPCODE:
1364{
1365 dREQ;
1366
1367 req->ptr2 = malloc (sizeof (EIO_STRUCT_STAT));
1368 if (!req->ptr2)
1369 {
1370 req_destroy (req);
1371 croak ("out of memory during aio_stat statdata allocation");
1372 }
1373
1374 req->flags |= FLAG_PTR2_FREE;
1375 req->sv1 = newSVsv (fh_or_path);
1376
1377 if (SvPOK (fh_or_path))
1378 {
1379 req->type = ix;
1380 req->ptr1 = SvPVbyte_nolen (req->sv1);
1381 }
1382 else
1383 {
1384 req->type = EIO_FSTAT;
1385 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1386 }
1387
1388 EIO_SEND;
1389}
1390
1391void
1392aio_utime (SV8 *fh_or_path, SV *atime, SV *mtime, SV *callback=&PL_sv_undef)
1393 PPCODE:
1394{
1395 dREQ;
1396
1397 req->nv1 = SvOK (atime) ? SvNV (atime) : -1.;
1398 req->nv2 = SvOK (mtime) ? SvNV (mtime) : -1.;
1399 req->sv1 = newSVsv (fh_or_path);
1400
1401 if (SvPOK (fh_or_path))
1402 {
1403 req->type = EIO_UTIME;
1404 req->ptr1 = SvPVbyte_nolen (req->sv1);
1405 }
1406 else
1407 {
1408 req->type = EIO_FUTIME;
1409 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1410 }
1411
1412 EIO_SEND;
1413}
1414
1415void
1416aio_truncate (SV8 *fh_or_path, SV *offset, SV *callback=&PL_sv_undef)
1417 PPCODE:
1418{
1419 dREQ;
1420
1421 req->sv1 = newSVsv (fh_or_path);
1422 req->offs = SvOK (offset) ? SvVAL64 (offset) : -1;
1423
1424 if (SvPOK (fh_or_path))
1425 {
1426 req->type = EIO_TRUNCATE;
1427 req->ptr1 = SvPVbyte_nolen (req->sv1);
1428 }
1429 else
1430 {
1431 req->type = EIO_FTRUNCATE;
1432 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1433 }
1434
1435 EIO_SEND;
1436}
1437
1438void
1439aio_chmod (SV8 *fh_or_path, int mode, SV *callback=&PL_sv_undef)
1440 ALIAS:
1441 aio_chmod = EIO_CHMOD
1442 aio_fchmod = EIO_FCHMOD
1443 aio_mkdir = EIO_MKDIR
1444 PPCODE:
1445{
1446 dREQ;
1447
1448 req->type = type;
1449 req->int2 = mode;
1450 req->sv1 = newSVsv (fh_or_path);
1451
1452 if (ix == EIO_FCHMOD)
1453 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1454 else
1455 req->ptr1 = SvPVbyte_nolen (req->sv1);
1456
1457 EIO_SEND;
1458}
1459
1460void
1461aio_chown (SV8 *fh_or_path, SV *uid, SV *gid, SV *callback=&PL_sv_undef)
1462 PPCODE:
1463{
1464 dREQ;
1465
1466 req->int2 = SvOK (uid) ? SvIV (uid) : -1;
1467 req->int3 = SvOK (gid) ? SvIV (gid) : -1;
1468 req->sv1 = newSVsv (fh_or_path);
1469
1470 if (SvPOK (fh_or_path))
1471 {
1472 req->type = EIO_CHOWN;
1473 req->ptr1 = SvPVbyte_nolen (req->sv1);
1474 }
1475 else
1476 {
1477 req->type = EIO_FCHOWN;
1478 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1479 }
1480
1481 EIO_SEND;
1482}
1483
1484void
1485aio_unlink (SV8 *pathname, SV *callback=&PL_sv_undef)
1486 ALIAS:
1487 aio_unlink = EIO_UNLINK
1488 aio_rmdir = EIO_RMDIR
1489 aio_readdir = EIO_READDIR
1490 PPCODE:
1491{
1492 dREQ;
1493
1494 req->type = ix;
1495 req->sv1 = newSVsv (pathname);
1496 req->ptr1 = SvPVbyte_nolen (req->sv1);
1497
1498 EIO_SEND;
1499}
1500
1501void
1502aio_link (SV8 *oldpath, SV8 *newpath, SV *callback=&PL_sv_undef)
1503 ALIAS:
1504 aio_link = EIO_LINK
1505 aio_symlink = EIO_SYMLINK
1506 aio_rename = EIO_RENAME
1507 PPCODE:
1508{
1509 dREQ;
1510
1511 req->type = ix;
1512 req->sv1 = newSVsv (oldpath);
1513 req->ptr1 = SvPVbyte_nolen (req->sv1);
1514 req->sv2 = newSVsv (newpath);
1515 req->ptr2 = SvPVbyte_nolen (req->sv2);
1516
1517 EIO_SEND;
1518}
1519
1520void
1521aio_mknod (SV8 *pathname, int mode, UV dev, SV *callback=&PL_sv_undef)
1522 PPCODE:
1523{
1524 dREQ;
1525
1526 req->type = EIO_MKNOD;
1527 req->sv1 = newSVsv (pathname);
1528 req->ptr1 = SvPVbyte_nolen (req->sv1);
1529 req->int2 = (mode_t)mode;
1530 req->offs = dev;
1531
1532 EIO_SEND;
1533}
1534
1535void
1536aio_busy (double delay, SV *callback=&PL_sv_undef)
1537 PPCODE:
1538{
1539 dREQ;
1540
1541 req->type = EIO_BUSY;
1542 req->nv1 = delay < 0. ? 0. : delay;
1543
1544 EIO_SEND;
1545}
1546
1547void
1548aio_group (SV *callback=&PL_sv_undef)
1549 PROTOTYPE: ;$
1550 PPCODE:
1551{
1552 dREQ;
1553
1554 req->type = EIO_GROUP;
1555
1556 req_send (req);
1557 XPUSHs (req_sv (req, AIO_GRP_KLASS));
1558}
1559
1560void
1561aio_nop (SV *callback=&PL_sv_undef)
1562 ALIAS:
1563 aio_nop = EIO_NOP
1564 aio_sync = EIO_SYNC
1565 PPCODE:
1566{
1567 dREQ;
1568
1569#endif
1570
1571void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2643eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1572{ 2644{
1573 grp->int2 = limit; 2645 grp->int2 = limit;
1574 grp->feed = feed; 2646 grp->feed = feed;
1575 2647
1576 grp_try_feed (grp); 2648 grp_try_feed (grp);
1577} 2649}
1578 2650
2651void
1579void eio_grp_limit (eio_req *grp, int limit) 2652eio_grp_limit (eio_req *grp, int limit)
1580{ 2653{
1581 grp->int2 = limit; 2654 grp->int2 = limit;
1582 2655
1583 grp_try_feed (grp); 2656 grp_try_feed (grp);
1584} 2657}
1585 2658
2659void
1586void eio_grp_add (eio_req *grp, eio_req *req) 2660eio_grp_add (eio_req *grp, eio_req *req)
1587{ 2661{
1588 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2662 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
2663
2664 grp->flags |= EIO_FLAG_GROUPADD;
1589 2665
1590 ++grp->size; 2666 ++grp->size;
1591 req->grp = grp; 2667 req->grp = grp;
1592 2668
1593 req->grp_prev = 0; 2669 req->grp_prev = 0;
1597 grp->grp_first->grp_prev = req; 2673 grp->grp_first->grp_prev = req;
1598 2674
1599 grp->grp_first = req; 2675 grp->grp_first = req;
1600} 2676}
1601 2677
2678/*****************************************************************************/
2679/* misc garbage */
1602 2680
2681eio_ssize_t
2682eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2683{
2684 return eio__sendfile (ofd, ifd, offset, count);
2685}
2686

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines