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

Comparing libeio/eio.c (file contents):
Revision 1.2 by root, Sat May 10 23:33:40 2008 UTC vs.
Revision 1.126 by root, Fri Dec 28 07:33:41 2012 UTC

1/*
2 * libeio implementation
3 *
4 * Copyright (c) 2007,2008,2009,2010,2011,2012 Marc Alexander Lehmann <libeio@schmorp.de>
5 * All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without modifica-
8 * tion, are permitted provided that the following conditions are met:
9 *
10 * 1. Redistributions of source code must retain the above copyright notice,
11 * this list of conditions and the following disclaimer.
12 *
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR IMPLIED
18 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MER-
19 * CHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO
20 * EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPE-
21 * CIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
22 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
23 * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
24 * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTH-
25 * ERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
26 * OF THE POSSIBILITY OF SUCH DAMAGE.
27 *
28 * Alternatively, the contents of this file may be used under the terms of
29 * the GNU General Public License ("GPL") version 2 or any later version,
30 * in which case the provisions of the GPL are applicable instead of
31 * the above. If you wish to allow the use of your version of this file
32 * only under the terms of the GPL and not to allow others to use your
33 * version of this file under the BSD license, indicate your decision
34 * by deleting the provisions above and replace them with the notice
35 * and other provisions required by the GPL. If you do not delete the
36 * provisions above, a recipient may use your version of this file under
37 * either the BSD or the GPL.
38 */
39
40#ifndef _WIN32
41# include "config.h"
42#endif
43
1#include "eio.h" 44#include "eio.h"
45#include "ecb.h"
46
47#ifdef EIO_STACKSIZE
48# define X_STACKSIZE EIO_STACKSIZE
49#endif
2#include "xthread.h" 50#include "xthread.h"
3 51
4#include <errno.h> 52#include <errno.h>
5
6#include "EXTERN.h"
7#include "perl.h"
8#include "XSUB.h"
9
10#include <stddef.h> 53#include <stddef.h>
11#include <stdlib.h> 54#include <stdlib.h>
55#include <string.h>
12#include <errno.h> 56#include <errno.h>
13#include <sys/types.h> 57#include <sys/types.h>
14#include <sys/stat.h> 58#include <sys/stat.h>
15#include <limits.h> 59#include <limits.h>
16#include <fcntl.h> 60#include <fcntl.h>
17#include <sched.h> 61#include <assert.h>
62
63/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
64/* intptr_t only comes from stdint.h, says idiot openbsd coder */
65#if HAVE_STDINT_H
66# include <stdint.h>
67#endif
68
69#ifndef ECANCELED
70# define ECANCELED EDOM
71#endif
72#ifndef ELOOP
73# define ELOOP EDOM
74#endif
75
76#if !defined(ENOTSOCK) && defined(WSAENOTSOCK)
77# define ENOTSOCK WSAENOTSOCK
78#endif
79
80static void eio_destroy (eio_req *req);
18 81
19#ifndef EIO_FINISH 82#ifndef EIO_FINISH
20# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0 83# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
21#endif 84#endif
22 85
… …
26 89
27#ifndef EIO_FEED 90#ifndef EIO_FEED
28# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0) 91# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)
29#endif 92#endif
30 93
94#ifndef EIO_FD_TO_WIN32_HANDLE
95# define EIO_FD_TO_WIN32_HANDLE(fd) _get_osfhandle (fd)
96#endif
97#ifndef EIO_WIN32_HANDLE_TO_FD
98# define EIO_WIN32_HANDLE_TO_FD(handle) _open_osfhandle (handle, 0)
99#endif
100
101#define EIO_ERRNO(errval,retval) ((errno = errval), retval)
102
103#define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1)
104
31#ifdef _WIN32 105#ifdef _WIN32
32 106
33 /*doh*/ 107 #undef PAGESIZE
108 #define PAGESIZE 4096 /* GetSystemInfo? */
109
110 /* TODO: look at how perl does stat (non-sloppy), unlink (ro-files), utime, link */
111
112 #ifdef EIO_STRUCT_STATI64
113 /* look at perl's non-sloppy stat */
114 #define stat(path,buf) _stati64 (path,buf)
115 #define fstat(fd,buf) _fstati64 (fd,buf)
116 #endif
117 #define lstat(path,buf) stat (path,buf)
118 #define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
119 #define mkdir(path,mode) _mkdir (path)
120 #define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
121
122 #define chmod(path,mode) _chmod (path, mode)
123 #define dup(fd) _dup (fd)
124 #define dup2(fd1,fd2) _dup2 (fd1, fd2)
125
126 #define fchmod(fd,mode) EIO_ENOSYS ()
127 #define chown(path,uid,gid) EIO_ENOSYS ()
128 #define fchown(fd,uid,gid) EIO_ENOSYS ()
129 #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
130 #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
131 #define mknod(path,mode,dev) EIO_ENOSYS ()
132 #define sync() EIO_ENOSYS ()
133 #define readlink(path,buf,s) EIO_ENOSYS ()
134 #define statvfs(path,buf) EIO_ENOSYS ()
135 #define fstatvfs(fd,buf) EIO_ENOSYS ()
136
137 #define pread(fd,buf,count,offset) eio__pread (fd, buf, count, offset)
138 #define pwrite(fd,buf,count,offset) eio__pwrite (fd, buf, count, offset)
139
140 #if __GNUC__
141 typedef long long eio_off_t; /* signed for compatibility to msvc */
142 #else
143 typedef __int64 eio_off_t; /* unsigned not supported by msvc */
144 #endif
145
146 static eio_ssize_t
147 eio__pread (int fd, void *buf, eio_ssize_t count, eio_off_t offset)
148 {
149 OVERLAPPED o = { 0 };
150 DWORD got;
151
152 o.Offset = offset;
153 o.OffsetHigh = offset >> 32;
154
155 return ReadFile ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd), buf, count, &got, &o)
156 ? got : -1;
157 }
158
159 static eio_ssize_t
160 eio__pwrite (int fd, void *buf, eio_ssize_t count, eio_off_t offset)
161 {
162 OVERLAPPED o = { 0 };
163 DWORD got;
164
165 o.Offset = offset;
166 o.OffsetHigh = offset >> 32;
167
168 return WriteFile ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd), buf, count, &got, &o)
169 ? got : -1;
170 }
171
172 /* rename() uses MoveFile, which fails to overwrite */
173 #define rename(old,neu) eio__rename (old, neu)
174
175 static int
176 eio__rename (const char *old, const char *neu)
177 {
178 if (MoveFileEx (old, neu, MOVEFILE_REPLACE_EXISTING))
179 return 0;
180
181 /* should steal _dosmaperr */
182 switch (GetLastError ())
183 {
184 case ERROR_FILE_NOT_FOUND:
185 case ERROR_PATH_NOT_FOUND:
186 case ERROR_INVALID_DRIVE:
187 case ERROR_NO_MORE_FILES:
188 case ERROR_BAD_NETPATH:
189 case ERROR_BAD_NET_NAME:
190 case ERROR_BAD_PATHNAME:
191 case ERROR_FILENAME_EXCED_RANGE:
192 errno = ENOENT;
193 break;
194
195 default:
196 errno = EACCES;
197 break;
198 }
199
200 return -1;
201 }
202
203 /* we could even stat and see if it exists */
204 static int
205 symlink (const char *old, const char *neu)
206 {
207 #if WINVER >= 0x0600
208 if (CreateSymbolicLink (neu, old, 1))
209 return 0;
210
211 if (CreateSymbolicLink (neu, old, 0))
212 return 0;
213 #endif
214
215 return EIO_ERRNO (ENOENT, -1);
216 }
217
218 /* POSIX API only */
219 #define CreateHardLink(neu,old,flags) 0
220 #define CreateSymbolicLink(neu,old,flags) 0
221
222 struct statvfs
223 {
224 int dummy;
225 };
226
227 #define DT_DIR EIO_DT_DIR
228 #define DT_REG EIO_DT_REG
229 #define D_NAME(entp) entp.cFileName
230 #define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
34 231
35#else 232#else
36 233
37# include "config.h"
38# include <sys/time.h> 234 #include <sys/time.h>
39# include <sys/select.h> 235 #include <sys/select.h>
236 #include <sys/statvfs.h>
40# include <unistd.h> 237 #include <unistd.h>
238 #include <signal.h>
239 #include <dirent.h>
240
241 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
242 #include <sys/mman.h>
243 #endif
244
245 #define D_NAME(entp) entp->d_name
246
247 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
248 #if __FreeBSD__ || __NetBSD__ || __OpenBSD__
249 #define _DIRENT_HAVE_D_TYPE /* sigh */
250 #define D_INO(de) (de)->d_fileno
251 #define D_NAMLEN(de) (de)->d_namlen
252 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
253 #define D_INO(de) (de)->d_ino
254 #endif
255
256 #ifdef _D_EXACT_NAMLEN
257 #undef D_NAMLEN
258 #define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
259 #endif
260
261 #ifdef _DIRENT_HAVE_D_TYPE
262 #define D_TYPE(de) (de)->d_type
263 #endif
264
265 #ifndef EIO_STRUCT_DIRENT
266 #define EIO_STRUCT_DIRENT struct dirent
267 #endif
268
269#endif
270
271#if HAVE_UTIMES
41# include <utime.h> 272# include <utime.h>
42# include <signal.h>
43
44# ifndef EIO_STRUCT_DIRENT
45# define EIO_STRUCT_DIRENT struct dirent
46# endif 273#endif
47 274
48#endif 275#if HAVE_SYS_SYSCALL_H
49 276# include <sys/syscall.h>
50# ifndef EIO_STRUCT_STAT
51# define EIO_STRUCT_STAT struct stat
52# endif 277#endif
278
279#if HAVE_SYS_PRCTL_H
280# include <sys/prctl.h>
281#endif
53 282
54#if HAVE_SENDFILE 283#if HAVE_SENDFILE
55# if __linux 284# if __linux
56# include <sys/sendfile.h> 285# include <sys/sendfile.h>
57# elif __freebsd 286# elif __FreeBSD__ || defined __APPLE__
58# include <sys/socket.h> 287# include <sys/socket.h>
59# include <sys/uio.h> 288# include <sys/uio.h>
60# elif __hpux 289# elif __hpux
61# include <sys/socket.h> 290# include <sys/socket.h>
62# elif __solaris /* not yet */ 291# elif __solaris
63# include <sys/sendfile.h> 292# include <sys/sendfile.h>
64# else 293# else
65# error sendfile support requested but not available 294# error sendfile support requested but not available
66# endif 295# endif
67#endif 296#endif
68 297
69/* number of seconds after which an idle threads exit */ 298#ifndef D_TYPE
70#define IDLE_TIMEOUT 10 299# define D_TYPE(de) 0
300#endif
301#ifndef D_INO
302# define D_INO(de) 0
303#endif
304#ifndef D_NAMLEN
305# define D_NAMLEN(entp) strlen (D_NAME (entp))
306#endif
71 307
72/* used for struct dirent, AIX doesn't provide it */ 308/* used for struct dirent, AIX doesn't provide it */
73#ifndef NAME_MAX 309#ifndef NAME_MAX
74# define NAME_MAX 4096 310# define NAME_MAX 4096
75#endif 311#endif
76 312
313/* used for readlink etc. */
314#ifndef PATH_MAX
315# define PATH_MAX 4096
316#endif
317
77/* buffer size for various temporary buffers */ 318/* buffer size for various temporary buffers */
78#define EIO_BUFSIZE 65536 319#define EIO_BUFSIZE 65536
79 320
80#define dBUF \ 321#define dBUF \
81 char *eio_buf; \
82 X_LOCK (wrklock); \
83 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 322 char *eio_buf = malloc (EIO_BUFSIZE); \
84 X_UNLOCK (wrklock); \ 323 errno = ENOMEM; \
85 if (!eio_buf) \ 324 if (!eio_buf) \
86 return -1; 325 return -1
326
327#define FUBd \
328 free (eio_buf)
87 329
88#define EIO_TICKS ((1000000 + 1023) >> 10) 330#define EIO_TICKS ((1000000 + 1023) >> 10)
331
332/*****************************************************************************/
333
334struct tmpbuf
335{
336 void *ptr;
337 int len;
338};
339
340static void *
341tmpbuf_get (struct tmpbuf *buf, int len)
342{
343 if (buf->len < len)
344 {
345 free (buf->ptr);
346 buf->ptr = malloc (buf->len = len);
347 }
348
349 return buf->ptr;
350}
351
352struct tmpbuf;
353
354#if _POSIX_VERSION >= 200809L
355 #define HAVE_AT 1
356 #define WD2FD(wd) ((wd) ? (wd)->fd : AT_FDCWD)
357 #ifndef O_SEARCH
358 #define O_SEARCH O_RDONLY
359 #endif
360#else
361 #define HAVE_AT 0
362 static const char *wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path);
363#endif
364
365struct eio_pwd
366{
367#if HAVE_AT
368 int fd;
369#endif
370 int len;
371 char str[1]; /* actually, a 0-terminated canonical path */
372};
373
374/*****************************************************************************/
375
376#define ETP_PRI_MIN EIO_PRI_MIN
377#define ETP_PRI_MAX EIO_PRI_MAX
378
379struct etp_worker;
380
381#define ETP_REQ eio_req
382#define ETP_DESTROY(req) eio_destroy (req)
383static int eio_finish (eio_req *req);
384#define ETP_FINISH(req) eio_finish (req)
385static void eio_execute (struct etp_worker *self, eio_req *req);
386#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
387
388/*****************************************************************************/
389
390#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
391
392/* calculate time difference in ~1/EIO_TICKS of a second */
393ecb_inline int
394tvdiff (struct timeval *tv1, struct timeval *tv2)
395{
396 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
397 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
398}
399
400static unsigned int started, idle, wanted = 4;
89 401
90static void (*want_poll_cb) (void); 402static void (*want_poll_cb) (void);
91static void (*done_poll_cb) (void); 403static void (*done_poll_cb) (void);
92 404
93static unsigned int max_poll_time = 0; 405static unsigned int max_poll_time; /* reslock */
94static unsigned int max_poll_reqs = 0; 406static unsigned int max_poll_reqs; /* reslock */
95 407
96/* calculcate time difference in ~1/EIO_TICKS of a second */ 408static unsigned int nreqs; /* reqlock */
97static int tvdiff (struct timeval *tv1, struct timeval *tv2) 409static unsigned int nready; /* reqlock */
98{ 410static unsigned int npending; /* reqlock */
99 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 411static unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
100 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 412static unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
101}
102 413
103static unsigned int started, idle, wanted; 414static xmutex_t wrklock;
415static xmutex_t reslock;
416static xmutex_t reqlock;
417static xcond_t reqwait;
418
419typedef struct etp_worker
420{
421 struct tmpbuf tmpbuf;
422
423 /* locked by wrklock */
424 struct etp_worker *prev, *next;
425
426 xthread_t tid;
427
428#ifdef ETP_WORKER_COMMON
429 ETP_WORKER_COMMON
430#endif
431} etp_worker;
432
433static etp_worker wrk_first; /* NOT etp */
434
435#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
436#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
104 437
105/* worker threads management */ 438/* worker threads management */
106static mutex_t wrklock = X_MUTEX_INIT;
107 439
108typedef struct worker { 440static void
109 /* locked by wrklock */
110 struct worker *prev, *next;
111
112 thread_t tid;
113
114 /* locked by reslock, reqlock or wrklock */
115 eio_req *req; /* currently processed request */
116 void *dbuf;
117 DIR *dirp;
118} worker;
119
120static worker wrk_first = { &wrk_first, &wrk_first, 0 };
121
122static void worker_clear (worker *wrk) 441etp_worker_clear (etp_worker *wrk)
123{ 442{
124 if (wrk->dirp) 443}
125 {
126 closedir (wrk->dirp);
127 wrk->dirp = 0;
128 }
129 444
130 if (wrk->dbuf) 445static void ecb_cold
131 { 446etp_worker_free (etp_worker *wrk)
447{
132 free (wrk->dbuf); 448 free (wrk->tmpbuf.ptr);
133 wrk->dbuf = 0;
134 }
135}
136 449
137static void worker_free (worker *wrk)
138{
139 wrk->next->prev = wrk->prev; 450 wrk->next->prev = wrk->prev;
140 wrk->prev->next = wrk->next; 451 wrk->prev->next = wrk->next;
141 452
142 free (wrk); 453 free (wrk);
143} 454}
144 455
145static volatile unsigned int nreqs, nready, npending; 456static unsigned int
146static volatile unsigned int max_idle = 4; 457etp_nreqs (void)
147
148static mutex_t reslock = X_MUTEX_INIT;
149static mutex_t reqlock = X_MUTEX_INIT;
150static cond_t reqwait = X_COND_INIT;
151
152unsigned int eio_nreqs (void)
153{ 458{
459 int retval;
460 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
461 retval = nreqs;
462 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
154 return nreqs; 463 return retval;
155} 464}
156 465
157unsigned int eio_nready (void) 466static unsigned int
467etp_nready (void)
158{ 468{
159 unsigned int retval; 469 unsigned int retval;
160 470
161 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 471 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
162 retval = nready; 472 retval = nready;
163 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 473 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
164 474
165 return retval; 475 return retval;
166} 476}
167 477
168unsigned int eio_npending (void) 478static unsigned int
479etp_npending (void)
169{ 480{
170 unsigned int retval; 481 unsigned int retval;
171 482
172 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 483 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
173 retval = npending; 484 retval = npending;
174 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 485 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
175 486
176 return retval; 487 return retval;
177} 488}
178 489
179unsigned int eio_nthreads (void) 490static unsigned int
491etp_nthreads (void)
180{ 492{
181 unsigned int retval; 493 unsigned int retval;
182 494
183 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 495 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
184 retval = started; 496 retval = started;
… …
191 * a somewhat faster data structure might be nice, but 503 * a somewhat faster data structure might be nice, but
192 * with 8 priorities this actually needs <20 insns 504 * with 8 priorities this actually needs <20 insns
193 * per shift, the most expensive operation. 505 * per shift, the most expensive operation.
194 */ 506 */
195typedef struct { 507typedef struct {
196 eio_req *qs[EIO_NUM_PRI], *qe[EIO_NUM_PRI]; /* qstart, qend */ 508 ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */
197 int size; 509 int size;
198} reqq; 510} etp_reqq;
199 511
200static reqq req_queue; 512static etp_reqq req_queue;
201static reqq res_queue; 513static etp_reqq res_queue;
202 514
203static int reqq_push (reqq *q, eio_req *req) 515static void ecb_noinline ecb_cold
516reqq_init (etp_reqq *q)
517{
518 int pri;
519
520 for (pri = 0; pri < ETP_NUM_PRI; ++pri)
521 q->qs[pri] = q->qe[pri] = 0;
522
523 q->size = 0;
524}
525
526static int ecb_noinline
527reqq_push (etp_reqq *q, ETP_REQ *req)
204{ 528{
205 int pri = req->pri; 529 int pri = req->pri;
206 req->next = 0; 530 req->next = 0;
207 531
208 if (q->qe[pri]) 532 if (q->qe[pri])
… …
214 q->qe[pri] = q->qs[pri] = req; 538 q->qe[pri] = q->qs[pri] = req;
215 539
216 return q->size++; 540 return q->size++;
217} 541}
218 542
219static eio_req *reqq_shift (reqq *q) 543static ETP_REQ * ecb_noinline
544reqq_shift (etp_reqq *q)
220{ 545{
221 int pri; 546 int pri;
222 547
223 if (!q->size) 548 if (!q->size)
224 return 0; 549 return 0;
225 550
226 --q->size; 551 --q->size;
227 552
228 for (pri = EIO_NUM_PRI; pri--; ) 553 for (pri = ETP_NUM_PRI; pri--; )
229 { 554 {
230 eio_req *req = q->qs[pri]; 555 eio_req *req = q->qs[pri];
231 556
232 if (req) 557 if (req)
233 { 558 {
… …
239 } 564 }
240 565
241 abort (); 566 abort ();
242} 567}
243 568
244static void grp_try_feed (eio_req *grp) 569static int ecb_cold
570etp_init (void (*want_poll)(void), void (*done_poll)(void))
245{ 571{
246 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 572 X_MUTEX_CREATE (wrklock);
247 { 573 X_MUTEX_CREATE (reslock);
248 int old_len = grp->size; 574 X_MUTEX_CREATE (reqlock);
575 X_COND_CREATE (reqwait);
249 576
250 EIO_FEED (grp); 577 reqq_init (&req_queue);
578 reqq_init (&res_queue);
251 579
252 /* stop if no progress has been made */ 580 wrk_first.next =
253 if (old_len == grp->size) 581 wrk_first.prev = &wrk_first;
254 {
255 grp->feed = 0;
256 break;
257 }
258 }
259}
260 582
261static int eio_finish (eio_req *req); 583 started = 0;
584 idle = 0;
585 nreqs = 0;
586 nready = 0;
587 npending = 0;
262 588
263static int grp_dec (eio_req *grp) 589 want_poll_cb = want_poll;
264{ 590 done_poll_cb = done_poll;
265 --grp->size;
266 591
267 /* call feeder, if applicable */
268 grp_try_feed (grp);
269
270 /* finish, if done */
271 if (!grp->size && grp->int1)
272 return eio_finish (grp);
273 else
274 return 0; 592 return 0;
275} 593}
276 594
277void eio_destroy (eio_req *req)
278{
279 if ((req)->flags & EIO_FLAG_PTR2_FREE)
280 free (req->ptr2);
281
282 EIO_DESTROY (req);
283}
284
285static int eio_finish (eio_req *req)
286{
287 int res = EIO_FINISH (req);
288
289 if (req->grp)
290 {
291 int res2;
292 eio_req *grp = req->grp;
293
294 /* unlink request */
295 if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
296 if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
297
298 if (grp->grp_first == req)
299 grp->grp_first = req->grp_next;
300
301 res2 = grp_dec (grp);
302
303 if (!res && res2)
304 res = res2;
305 }
306
307 eio_destroy (req);
308
309 return res;
310}
311
312void eio_grp_cancel (eio_req *grp)
313{
314 for (grp = grp->grp_first; grp; grp = grp->grp_next)
315 eio_cancel (grp);
316}
317
318void eio_cancel (eio_req *req)
319{
320 req->flags |= EIO_FLAG_CANCELLED;
321
322 eio_grp_cancel (req);
323}
324
325X_THREAD_PROC (eio_proc); 595X_THREAD_PROC (etp_proc);
326 596
327static void start_thread (void) 597static void ecb_cold
598etp_start_thread (void)
328{ 599{
329 worker *wrk = calloc (1, sizeof (worker)); 600 etp_worker *wrk = calloc (1, sizeof (etp_worker));
330 601
331 if (!wrk) 602 /*TODO*/
332 croak ("unable to allocate worker thread data"); 603 assert (("unable to allocate worker thread data", wrk));
333 604
334 X_LOCK (wrklock); 605 X_LOCK (wrklock);
335 606
336 if (thread_create (&wrk->tid, eio_proc, (void *)wrk)) 607 if (xthread_create (&wrk->tid, etp_proc, (void *)wrk))
337 { 608 {
338 wrk->prev = &wrk_first; 609 wrk->prev = &wrk_first;
339 wrk->next = wrk_first.next; 610 wrk->next = wrk_first.next;
340 wrk_first.next->prev = wrk; 611 wrk_first.next->prev = wrk;
341 wrk_first.next = wrk; 612 wrk_first.next = wrk;
… …
345 free (wrk); 616 free (wrk);
346 617
347 X_UNLOCK (wrklock); 618 X_UNLOCK (wrklock);
348} 619}
349 620
621static void
350static void maybe_start_thread (void) 622etp_maybe_start_thread (void)
351{ 623{
352 if (eio_nthreads () >= wanted) 624 if (ecb_expect_true (etp_nthreads () >= wanted))
353 return; 625 return;
354 626
355 /* todo: maybe use idle here, but might be less exact */ 627 /* todo: maybe use idle here, but might be less exact */
356 if (0 <= (int)eio_nthreads () + (int)eio_npending () - (int)eio_nreqs ()) 628 if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
357 return; 629 return;
358 630
359 start_thread (); 631 etp_start_thread ();
360} 632}
361 633
362void eio_submit (eio_req *req) 634static void ecb_cold
635etp_end_thread (void)
363{ 636{
364 ++nreqs;
365
366 X_LOCK (reqlock);
367 ++nready;
368 reqq_push (&req_queue, req);
369 X_COND_SIGNAL (reqwait);
370 X_UNLOCK (reqlock);
371
372 maybe_start_thread ();
373}
374
375static void end_thread (void)
376{
377 eio_req *req = calloc (1, sizeof (eio_req)); 637 eio_req *req = calloc (1, sizeof (eio_req)); /* will be freed by worker */
378 638
379 req->type = EIO_QUIT; 639 req->type = -1;
380 req->pri = EIO_PRI_MAX + EIO_PRI_BIAS; 640 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
381 641
382 X_LOCK (reqlock); 642 X_LOCK (reqlock);
383 reqq_push (&req_queue, req); 643 reqq_push (&req_queue, req);
384 X_COND_SIGNAL (reqwait); 644 X_COND_SIGNAL (reqwait);
385 X_UNLOCK (reqlock); 645 X_UNLOCK (reqlock);
… …
387 X_LOCK (wrklock); 647 X_LOCK (wrklock);
388 --started; 648 --started;
389 X_UNLOCK (wrklock); 649 X_UNLOCK (wrklock);
390} 650}
391 651
392void eio_set_max_poll_time (double nseconds) 652static int
393{
394 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
395 max_poll_time = nseconds;
396 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
397}
398
399void eio_set_max_poll_reqs (unsigned int maxreqs)
400{
401 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
402 max_poll_reqs = maxreqs;
403 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
404}
405
406void eio_set_max_idle (unsigned int nthreads)
407{
408 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
409 max_idle = nthreads <= 0 ? 1 : nthreads;
410 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
411}
412
413void eio_set_min_parallel (unsigned int nthreads)
414{
415 if (wanted < nthreads)
416 wanted = nthreads;
417}
418
419void eio_set_max_parallel (unsigned int nthreads)
420{
421 if (wanted > nthreads)
422 wanted = nthreads;
423
424 while (started > wanted)
425 end_thread ();
426}
427
428int eio_poll (void) 653etp_poll (void)
429{ 654{
430 int maxreqs = max_poll_reqs; 655 unsigned int maxreqs;
656 unsigned int maxtime;
431 struct timeval tv_start, tv_now; 657 struct timeval tv_start, tv_now;
432 eio_req *req;
433 658
659 X_LOCK (reslock);
660 maxreqs = max_poll_reqs;
661 maxtime = max_poll_time;
662 X_UNLOCK (reslock);
663
434 if (max_poll_time) 664 if (maxtime)
435 gettimeofday (&tv_start, 0); 665 gettimeofday (&tv_start, 0);
436 666
437 for (;;) 667 for (;;)
438 { 668 {
669 ETP_REQ *req;
670
439 maybe_start_thread (); 671 etp_maybe_start_thread ();
440 672
441 X_LOCK (reslock); 673 X_LOCK (reslock);
442 req = reqq_shift (&res_queue); 674 req = reqq_shift (&res_queue);
443 675
444 if (req) 676 if (req)
… …
452 X_UNLOCK (reslock); 684 X_UNLOCK (reslock);
453 685
454 if (!req) 686 if (!req)
455 return 0; 687 return 0;
456 688
689 X_LOCK (reqlock);
457 --nreqs; 690 --nreqs;
691 X_UNLOCK (reqlock);
458 692
459 if (req->type == EIO_GROUP && req->size) 693 if (ecb_expect_false (req->type == EIO_GROUP && req->size))
460 { 694 {
461 req->int1 = 1; /* mark request as delayed */ 695 req->int1 = 1; /* mark request as delayed */
462 continue; 696 continue;
463 } 697 }
464 else 698 else
465 { 699 {
466 int res = eio_finish (req); 700 int res = ETP_FINISH (req);
467 if (res) 701 if (ecb_expect_false (res))
468 return res; 702 return res;
469 } 703 }
470 704
471 if (maxreqs && !--maxreqs) 705 if (ecb_expect_false (maxreqs && !--maxreqs))
472 break; 706 break;
473 707
474 if (max_poll_time) 708 if (maxtime)
475 { 709 {
476 gettimeofday (&tv_now, 0); 710 gettimeofday (&tv_now, 0);
477 711
478 if (tvdiff (&tv_start, &tv_now) >= max_poll_time) 712 if (tvdiff (&tv_start, &tv_now) >= maxtime)
479 break; 713 break;
480 } 714 }
481 } 715 }
482 716
483 errno = EAGAIN; 717 errno = EAGAIN;
484 return -1; 718 return -1;
485} 719}
486 720
721static void
722etp_cancel (ETP_REQ *req)
723{
724 req->cancelled = 1;
725
726 eio_grp_cancel (req);
727}
728
729static void
730etp_submit (ETP_REQ *req)
731{
732 req->pri -= ETP_PRI_MIN;
733
734 if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
735 if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
736
737 if (ecb_expect_false (req->type == EIO_GROUP))
738 {
739 /* I hope this is worth it :/ */
740 X_LOCK (reqlock);
741 ++nreqs;
742 X_UNLOCK (reqlock);
743
744 X_LOCK (reslock);
745
746 ++npending;
747
748 if (!reqq_push (&res_queue, req) && want_poll_cb)
749 want_poll_cb ();
750
751 X_UNLOCK (reslock);
752 }
753 else
754 {
755 X_LOCK (reqlock);
756 ++nreqs;
757 ++nready;
758 reqq_push (&req_queue, req);
759 X_COND_SIGNAL (reqwait);
760 X_UNLOCK (reqlock);
761
762 etp_maybe_start_thread ();
763 }
764}
765
766static void ecb_cold
767etp_set_max_poll_time (double nseconds)
768{
769 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
770 max_poll_time = nseconds * EIO_TICKS;
771 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
772}
773
774static void ecb_cold
775etp_set_max_poll_reqs (unsigned int maxreqs)
776{
777 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
778 max_poll_reqs = maxreqs;
779 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
780}
781
782static void ecb_cold
783etp_set_max_idle (unsigned int nthreads)
784{
785 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
786 max_idle = nthreads;
787 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
788}
789
790static void ecb_cold
791etp_set_idle_timeout (unsigned int seconds)
792{
793 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
794 idle_timeout = seconds;
795 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
796}
797
798static void ecb_cold
799etp_set_min_parallel (unsigned int nthreads)
800{
801 if (wanted < nthreads)
802 wanted = nthreads;
803}
804
805static void ecb_cold
806etp_set_max_parallel (unsigned int nthreads)
807{
808 if (wanted > nthreads)
809 wanted = nthreads;
810
811 while (started > wanted)
812 etp_end_thread ();
813}
814
815/*****************************************************************************/
816
817static void
818grp_try_feed (eio_req *grp)
819{
820 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
821 {
822 grp->flags &= ~EIO_FLAG_GROUPADD;
823
824 EIO_FEED (grp);
825
826 /* stop if no progress has been made */
827 if (!(grp->flags & EIO_FLAG_GROUPADD))
828 {
829 grp->feed = 0;
830 break;
831 }
832 }
833}
834
835static int
836grp_dec (eio_req *grp)
837{
838 --grp->size;
839
840 /* call feeder, if applicable */
841 grp_try_feed (grp);
842
843 /* finish, if done */
844 if (!grp->size && grp->int1)
845 return eio_finish (grp);
846 else
847 return 0;
848}
849
850static void
851eio_destroy (eio_req *req)
852{
853 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
854 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
855
856 EIO_DESTROY (req);
857}
858
859static int
860eio_finish (eio_req *req)
861{
862 int res = EIO_FINISH (req);
863
864 if (req->grp)
865 {
866 int res2;
867 eio_req *grp = req->grp;
868
869 /* unlink request */
870 if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
871 if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
872
873 if (grp->grp_first == req)
874 grp->grp_first = req->grp_next;
875
876 res2 = grp_dec (grp);
877
878 if (!res)
879 res = res2;
880 }
881
882 eio_destroy (req);
883
884 return res;
885}
886
887void
888eio_grp_cancel (eio_req *grp)
889{
890 for (grp = grp->grp_first; grp; grp = grp->grp_next)
891 eio_cancel (grp);
892}
893
894void
895eio_cancel (eio_req *req)
896{
897 etp_cancel (req);
898}
899
900void
901eio_submit (eio_req *req)
902{
903 etp_submit (req);
904}
905
906unsigned int
907eio_nreqs (void)
908{
909 return etp_nreqs ();
910}
911
912unsigned int
913eio_nready (void)
914{
915 return etp_nready ();
916}
917
918unsigned int
919eio_npending (void)
920{
921 return etp_npending ();
922}
923
924unsigned int ecb_cold
925eio_nthreads (void)
926{
927 return etp_nthreads ();
928}
929
930void ecb_cold
931eio_set_max_poll_time (double nseconds)
932{
933 etp_set_max_poll_time (nseconds);
934}
935
936void ecb_cold
937eio_set_max_poll_reqs (unsigned int maxreqs)
938{
939 etp_set_max_poll_reqs (maxreqs);
940}
941
942void ecb_cold
943eio_set_max_idle (unsigned int nthreads)
944{
945 etp_set_max_idle (nthreads);
946}
947
948void ecb_cold
949eio_set_idle_timeout (unsigned int seconds)
950{
951 etp_set_idle_timeout (seconds);
952}
953
954void ecb_cold
955eio_set_min_parallel (unsigned int nthreads)
956{
957 etp_set_min_parallel (nthreads);
958}
959
960void ecb_cold
961eio_set_max_parallel (unsigned int nthreads)
962{
963 etp_set_max_parallel (nthreads);
964}
965
966int eio_poll (void)
967{
968 return etp_poll ();
969}
970
487/*****************************************************************************/ 971/*****************************************************************************/
488/* work around various missing functions */ 972/* work around various missing functions */
489 973
490#if !HAVE_PREADWRITE
491# define pread aio_pread
492# define pwrite aio_pwrite
493
494/*
495 * make our pread/pwrite safe against themselves, but not against
496 * normal read/write by using a mutex. slows down execution a lot,
497 * but that's your problem, not mine.
498 */
499static mutex_t preadwritelock = X_MUTEX_INIT;
500
501static ssize_t pread (int fd, void *buf, size_t count, off_t offset)
502{
503 ssize_t res;
504 off_t ooffset;
505
506 X_LOCK (preadwritelock);
507 ooffset = lseek (fd, 0, SEEK_CUR);
508 lseek (fd, offset, SEEK_SET);
509 res = read (fd, buf, count);
510 lseek (fd, ooffset, SEEK_SET);
511 X_UNLOCK (preadwritelock);
512
513 return res;
514}
515
516static ssize_t pwrite (int fd, void *buf, size_t count, off_t offset)
517{
518 ssize_t res;
519 off_t ooffset;
520
521 X_LOCK (preadwritelock);
522 ooffset = lseek (fd, 0, SEEK_CUR);
523 lseek (fd, offset, SEEK_SET);
524 res = write (fd, buf, count);
525 lseek (fd, offset, SEEK_SET);
526 X_UNLOCK (preadwritelock);
527
528 return res;
529}
530#endif
531
532#ifndef HAVE_FUTIMES 974#ifndef HAVE_UTIMES
533 975
976# undef utimes
534# define utimes(path,times) aio_utimes (path, times) 977# define utimes(path,times) eio__utimes (path, times)
535# define futimes(fd,times) aio_futimes (fd, times)
536 978
979static int
537static int aio_utimes (const char *filename, const struct timeval times[2]) 980eio__utimes (const char *filename, const struct timeval times[2])
538{ 981{
539 if (times) 982 if (times)
540 { 983 {
541 struct utimbuf buf; 984 struct utimbuf buf;
542 985
… …
547 } 990 }
548 else 991 else
549 return utime (filename, 0); 992 return utime (filename, 0);
550} 993}
551 994
995#endif
996
997#ifndef HAVE_FUTIMES
998
999# undef futimes
1000# define futimes(fd,times) eio__futimes (fd, times)
1001
1002static int
552static int aio_futimes (int fd, const struct timeval tv[2]) 1003eio__futimes (int fd, const struct timeval tv[2])
553{ 1004{
554 errno = ENOSYS; 1005 errno = ENOSYS;
555 return -1; 1006 return -1;
556} 1007}
557 1008
558#endif 1009#endif
559 1010
560#if !HAVE_FDATASYNC 1011#if !HAVE_FDATASYNC
1012# undef fdatasync
561# define fdatasync fsync 1013# define fdatasync(fd) fsync (fd)
562#endif 1014#endif
1015
1016static int
1017eio__syncfs (int fd)
1018{
1019 int res;
1020
1021#if HAVE_SYS_SYNCFS
1022 res = (int)syscall (__NR_syncfs, (int)(fd));
1023#else
1024 res = EIO_ENOSYS ();
1025#endif
1026
1027 if (res < 0 && errno == ENOSYS && fd >= 0)
1028 sync ();
1029
1030 return res;
1031}
1032
1033/* sync_file_range always needs emulation */
1034static int
1035eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
1036{
1037#if HAVE_SYNC_FILE_RANGE
1038 int res;
1039
1040 if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
1041 || EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
1042 || EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
1043 {
1044 flags = 0
1045 | (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
1046 | (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
1047 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
1048 }
1049
1050 res = sync_file_range (fd, offset, nbytes, flags);
1051
1052 if (!res || errno != ENOSYS)
1053 return res;
1054#endif
1055
1056 /* even though we could play tricks with the flags, it's better to always
1057 * call fdatasync, as that matches the expectation of its users best */
1058 return fdatasync (fd);
1059}
1060
1061static int
1062eio__fallocate (int fd, int mode, off_t offset, size_t len)
1063{
1064#if HAVE_LINUX_FALLOCATE
1065 return fallocate (fd, mode, offset, len);
1066#else
1067 return EIO_ENOSYS ();
1068#endif
1069}
563 1070
564#if !HAVE_READAHEAD 1071#if !HAVE_READAHEAD
1072# undef readahead
565# define readahead(fd,offset,count) aio_readahead (fd, offset, count, self) 1073# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
566 1074
1075static eio_ssize_t
567static ssize_t aio_readahead (int fd, off_t offset, size_t count, worker *self) 1076eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
568{ 1077{
569 size_t todo = count; 1078 size_t todo = count;
570 dBUF; 1079 dBUF;
571 1080
572 while (todo > 0) 1081 while (todo > 0)
573 { 1082 {
574 size_t len = todo < EIO_BUFSIZE ? todo : EIO_BUFSIZE; 1083 size_t len = todo < EIO_BUFSIZE ? todo : EIO_BUFSIZE;
575 1084
576 pread (fd, aio_buf, len, offset); 1085 pread (fd, eio_buf, len, offset);
577 offset += len; 1086 offset += len;
578 todo -= len; 1087 todo -= len;
579 } 1088 }
580 1089
581 errno = 0; 1090 FUBd;
1091
1092 /* linux's readahead basically only fails for EBADF or EINVAL (not mmappable) */
1093 /* but not for e.g. EIO or eof, so we also never fail */
582 return count; 1094 return 0;
583} 1095}
584 1096
585#endif
586
587#if !HAVE_READDIR_R
588# define readdir_r aio_readdir_r
589
590static mutex_t readdirlock = X_MUTEX_INIT;
591
592static int readdir_r (DIR *dirp, X_DIRENT *ent, X_DIRENT **res)
593{
594 X_DIRENT *e;
595 int errorno;
596
597 X_LOCK (readdirlock);
598
599 e = readdir (dirp);
600 errorno = errno;
601
602 if (e)
603 {
604 *res = ent;
605 strcpy (ent->d_name, e->d_name);
606 }
607 else
608 *res = 0;
609
610 X_UNLOCK (readdirlock);
611
612 errno = errorno;
613 return e ? 0 : -1;
614}
615#endif 1097#endif
616 1098
617/* sendfile always needs emulation */ 1099/* sendfile always needs emulation */
1100static eio_ssize_t
618static ssize_t sendfile_ (int ofd, int ifd, off_t offset, size_t count, worker *self) 1101eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
619{ 1102{
1103 eio_ssize_t written = 0;
620 ssize_t res; 1104 eio_ssize_t res;
621 1105
622 if (!count) 1106 if (!count)
623 return 0; 1107 return 0;
624 1108
1109 for (;;)
1110 {
1111#ifdef __APPLE__
1112# undef HAVE_SENDFILE /* broken, as everything on os x */
1113#endif
625#if HAVE_SENDFILE 1114#if HAVE_SENDFILE
626# if __linux 1115# if __linux
1116 off_t soffset = offset;
627 res = sendfile (ofd, ifd, &offset, count); 1117 res = sendfile (ofd, ifd, &soffset, count);
628 1118
629# elif __freebsd 1119# elif __FreeBSD__
630 /* 1120 /*
631 * Of course, the freebsd sendfile is a dire hack with no thoughts 1121 * Of course, the freebsd sendfile is a dire hack with no thoughts
632 * wasted on making it similar to other I/O functions. 1122 * wasted on making it similar to other I/O functions.
633 */ 1123 */
634 {
635 off_t sbytes; 1124 off_t sbytes;
636 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 1125 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
637 1126
638 if (res < 0 && sbytes) 1127 #if 0 /* according to the manpage, this is correct, but broken behaviour */
639 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 1128 /* freebsd' sendfile will return 0 on success */
1129 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
1130 /* not on e.g. EIO or EPIPE - sounds broken */
1131 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
640 res = sbytes; 1132 res = sbytes;
641 } 1133 #endif
1134
1135 /* according to source inspection, this is correct, and useful behaviour */
1136 if (sbytes)
1137 res = sbytes;
1138
1139# elif defined __APPLE__
1140 off_t sbytes = count;
1141 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
1142
1143 /* according to the manpage, sbytes is always valid */
1144 if (sbytes)
1145 res = sbytes;
642 1146
643# elif __hpux 1147# elif __hpux
644 res = sendfile (ofd, ifd, offset, count, 0, 0); 1148 res = sendfile (ofd, ifd, offset, count, 0, 0);
645 1149
646# elif __solaris 1150# elif __solaris
647 {
648 struct sendfilevec vec; 1151 struct sendfilevec vec;
649 size_t sbytes; 1152 size_t sbytes;
650 1153
651 vec.sfv_fd = ifd; 1154 vec.sfv_fd = ifd;
652 vec.sfv_flag = 0; 1155 vec.sfv_flag = 0;
653 vec.sfv_off = offset; 1156 vec.sfv_off = offset;
654 vec.sfv_len = count; 1157 vec.sfv_len = count;
655 1158
656 res = sendfilev (ofd, &vec, 1, &sbytes); 1159 res = sendfilev (ofd, &vec, 1, &sbytes);
657 1160
658 if (res < 0 && sbytes) 1161 if (res < 0 && sbytes)
659 res = sbytes; 1162 res = sbytes;
660 }
661 1163
662# endif 1164# endif
1165
1166#elif defined (_WIN32) && 0
1167 /* does not work, just for documentation of what would need to be done */
1168 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
1169 /* libeio guarantees that the file offset does not change, and windows */
1170 /* has no way to get an independent handle to the same file description */
1171 HANDLE h = TO_SOCKET (ifd);
1172 SetFilePointer (h, offset, 0, FILE_BEGIN);
1173 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1174
663#else 1175#else
664 res = -1; 1176 res = EIO_ENOSYS ();
665 errno = ENOSYS;
666#endif 1177#endif
667 1178
1179 /* we assume sendfile can copy at least 128mb in one go */
1180 if (res <= 128 * 1024 * 1024)
1181 {
1182 if (res > 0)
1183 written += res;
1184
1185 if (written)
1186 return written;
1187
1188 break;
1189 }
1190 else
1191 {
1192 /* if we requested more, then probably the kernel was lazy */
1193 written += res;
1194 offset += res;
1195 count -= res;
1196
1197 if (!count)
1198 return written;
1199 }
1200 }
1201
668 if (res < 0 1202 if (res < 0
669 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1203 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1204 /* BSDs */
1205#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1206 || errno == ENOTSUP
1207#endif
1208#ifdef EOPNOTSUPP /* windows */
1209 || errno == EOPNOTSUPP /* BSDs */
1210#endif
670#if __solaris 1211#if __solaris
671 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1212 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
672#endif 1213#endif
673 ) 1214 )
674 ) 1215 )
… …
678 1219
679 res = 0; 1220 res = 0;
680 1221
681 while (count) 1222 while (count)
682 { 1223 {
683 ssize_t cnt; 1224 eio_ssize_t cnt;
684 1225
685 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 1226 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
686 1227
687 if (cnt <= 0) 1228 if (cnt <= 0)
688 { 1229 {
… …
700 1241
701 offset += cnt; 1242 offset += cnt;
702 res += cnt; 1243 res += cnt;
703 count -= cnt; 1244 count -= cnt;
704 } 1245 }
1246
1247 FUBd;
705 } 1248 }
706 1249
707 return res; 1250 return res;
708} 1251}
709 1252
710/* read a full directory */ 1253#ifdef PAGESIZE
711static void scandir_ (eio_req *req, worker *self) 1254# define eio_pagesize() PAGESIZE
1255#else
1256static intptr_t
1257eio_pagesize (void)
712{ 1258{
713 DIR *dirp; 1259 static intptr_t page;
714 union
715 {
716 EIO_STRUCT_DIRENT d;
717 char b [offsetof (EIO_STRUCT_DIRENT, d_name) + NAME_MAX + 1];
718 } *u;
719 EIO_STRUCT_DIRENT *entp;
720 char *name, *names;
721 int memlen = 4096;
722 int memofs = 0;
723 int res = 0;
724 1260
725 X_LOCK (wrklock); 1261 if (!page)
726 self->dirp = dirp = opendir (req->ptr1); 1262 page = sysconf (_SC_PAGESIZE);
727 self->dbuf = u = malloc (sizeof (*u));
728 req->flags |= EIO_FLAG_PTR2_FREE;
729 req->ptr2 = names = malloc (memlen);
730 X_UNLOCK (wrklock);
731 1263
732 if (dirp && u && names) 1264 return page;
733 for (;;) 1265}
1266#endif
1267
1268static void
1269eio_page_align (void **addr, size_t *length)
1270{
1271 intptr_t mask = eio_pagesize () - 1;
1272
1273 /* round down addr */
1274 intptr_t adj = mask & (intptr_t)*addr;
1275
1276 *addr = (void *)((intptr_t)*addr - adj);
1277 *length += adj;
1278
1279 /* round up length */
1280 *length = (*length + mask) & ~mask;
1281}
1282
1283#if !_POSIX_MEMLOCK
1284# define eio__mlockall(a) EIO_ENOSYS ()
1285#else
1286
1287static int
1288eio__mlockall (int flags)
1289{
1290 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1291 extern int mallopt (int, int);
1292 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1293 #endif
1294
1295 if (EIO_MCL_CURRENT != MCL_CURRENT
1296 || EIO_MCL_FUTURE != MCL_FUTURE)
1297 {
1298 flags = 0
1299 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1300 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1301 }
1302
1303 return mlockall (flags);
1304}
1305#endif
1306
1307#if !_POSIX_MEMLOCK_RANGE
1308# define eio__mlock(a,b) EIO_ENOSYS ()
1309#else
1310
1311static int
1312eio__mlock (void *addr, size_t length)
1313{
1314 eio_page_align (&addr, &length);
1315
1316 return mlock (addr, length);
1317}
1318
1319#endif
1320
1321#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1322# define eio__msync(a,b,c) EIO_ENOSYS ()
1323#else
1324
1325static int
1326eio__msync (void *mem, size_t len, int flags)
1327{
1328 eio_page_align (&mem, &len);
1329
1330 if (EIO_MS_ASYNC != MS_SYNC
1331 || EIO_MS_INVALIDATE != MS_INVALIDATE
1332 || EIO_MS_SYNC != MS_SYNC)
1333 {
1334 flags = 0
1335 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1336 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1337 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1338 }
1339
1340 return msync (mem, len, flags);
1341}
1342
1343#endif
1344
1345static int
1346eio__mtouch (eio_req *req)
1347{
1348 void *mem = req->ptr2;
1349 size_t len = req->size;
1350 int flags = req->int1;
1351
1352 eio_page_align (&mem, &len);
1353
1354 {
1355 intptr_t addr = (intptr_t)mem;
1356 intptr_t end = addr + len;
1357 intptr_t page = eio_pagesize ();
1358
1359 if (addr < end)
1360 if (flags & EIO_MT_MODIFY) /* modify */
1361 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1362 else
1363 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1364 }
1365
1366 return 0;
1367}
1368
1369/*****************************************************************************/
1370/* requests implemented outside eio_execute, because they are so large */
1371
1372static void
1373eio__lseek (eio_req *req)
1374{
1375 /* this usually gets optimised away completely, or your compiler sucks, */
1376 /* or the whence constants really are not 0, 1, 2 */
1377 int whence = req->int2 == EIO_SEEK_SET ? SEEK_SET
1378 : req->int2 == EIO_SEEK_CUR ? SEEK_CUR
1379 : req->int2 == EIO_SEEK_END ? SEEK_END
1380 : req->int2;
1381
1382 req->offs = lseek (req->int1, req->offs, whence);
1383 req->result = req->offs == (off_t)-1 ? -1 : 0;
1384}
1385
1386/* result will always end up in tmpbuf, there is always space for adding a 0-byte */
1387static int
1388eio__realpath (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
1389{
1390 const char *rel = path;
1391 char *res;
1392 char *tmp1, *tmp2;
1393#if SYMLOOP_MAX > 32
1394 int symlinks = SYMLOOP_MAX;
1395#else
1396 int symlinks = 32;
1397#endif
1398
1399 errno = EINVAL;
1400 if (!rel)
1401 return -1;
1402
1403 errno = ENOENT;
1404 if (!*rel)
1405 return -1;
1406
1407 res = tmpbuf_get (tmpbuf, PATH_MAX * 3);
1408 tmp1 = res + PATH_MAX;
1409 tmp2 = tmp1 + PATH_MAX;
1410
1411#if 0 /* disabled, the musl way to do things is just too racy */
1412#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1413 /* on linux we may be able to ask the kernel */
1414 {
1415 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1416
1417 if (fd >= 0)
734 { 1418 {
1419 sprintf (tmp1, "/proc/self/fd/%d", fd);
1420 req->result = readlink (tmp1, res, PATH_MAX);
1421 /* here we should probably stat the open file and the disk file, to make sure they still match */
1422 close (fd);
1423
1424 if (req->result > 0)
1425 goto done;
1426 }
1427 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1428 return -1;
1429 }
1430#endif
1431#endif
1432
1433 if (*rel != '/')
1434 {
1435 int len;
1436
735 errno = 0; 1437 errno = ENOENT;
736 readdir_r (dirp, &u->d, &entp); 1438 if (wd == EIO_INVALID_WD)
1439 return -1;
1440
1441 if (wd == EIO_CWD)
1442 {
1443 if (!getcwd (res, PATH_MAX))
1444 return -1;
737 1445
738 if (!entp) 1446 len = strlen (res);
1447 }
1448 else
1449 memcpy (res, wd->str, len = wd->len);
1450
1451 if (res [1]) /* only use if not / */
1452 res += len;
1453 }
1454
1455 while (*rel)
1456 {
1457 eio_ssize_t len, linklen;
1458 const char *beg = rel;
1459
1460 while (*rel && *rel != '/')
1461 ++rel;
1462
1463 len = rel - beg;
1464
1465 if (!len) /* skip slashes */
1466 {
739 break; 1467 ++rel;
1468 continue;
1469 }
740 1470
741 name = entp->d_name; 1471 if (beg [0] == '.')
1472 {
1473 if (len == 1)
1474 continue; /* . - nop */
742 1475
743 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1476 if (beg [1] == '.' && len == 2)
1477 {
1478 /* .. - back up one component, if possible */
1479
1480 while (res != tmpbuf->ptr)
1481 if (*--res == '/')
1482 break;
1483
1484 continue;
1485 }
1486 }
1487
1488 errno = ENAMETOOLONG;
1489 if (res + 1 + len + 1 >= tmp1)
1490 return -1;
1491
1492 /* copy one component */
1493 *res = '/';
1494 memcpy (res + 1, beg, len);
1495
1496 /* zero-terminate, for readlink */
1497 res [len + 1] = 0;
1498
1499 /* now check if it's a symlink */
1500 linklen = readlink (tmpbuf->ptr, tmp1, PATH_MAX);
1501
1502 if (linklen < 0)
744 { 1503 {
745 int len = strlen (name) + 1; 1504 if (errno != EINVAL)
1505 return -1;
746 1506
747 res++; 1507 /* it's a normal directory. hopefully */
748
749 while (memofs + len > memlen)
750 {
751 memlen *= 2;
752 X_LOCK (wrklock);
753 req->ptr2 = names = realloc (names, memlen);
754 X_UNLOCK (wrklock);
755
756 if (!names)
757 break;
758 }
759
760 memcpy (names + memofs, name, len);
761 memofs += len; 1508 res += len + 1;
762 }
763 }
764
765 if (errno)
766 res = -1;
767
768 req->result = res;
769}
770
771/*****************************************************************************/
772
773X_THREAD_PROC (eio_proc)
774{
775 eio_req *req;
776 struct timespec ts;
777 worker *self = (worker *)thr_arg;
778
779 /* try to distribute timeouts somewhat randomly */
780 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
781
782 for (;;)
783 {
784 ts.tv_sec = time (0) + IDLE_TIMEOUT;
785
786 X_LOCK (reqlock);
787
788 for (;;)
789 {
790 self->req = req = reqq_shift (&req_queue);
791
792 if (req)
793 break;
794
795 ++idle;
796
797 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
798 {
799 if (idle > max_idle)
800 {
801 --idle;
802 X_UNLOCK (reqlock);
803 X_LOCK (wrklock);
804 --started;
805 X_UNLOCK (wrklock);
806 goto quit;
807 }
808
809 /* we are allowed to idle, so do so without any timeout */
810 X_COND_WAIT (reqwait, reqlock);
811 ts.tv_sec = time (0) + IDLE_TIMEOUT;
812 }
813
814 --idle;
815 }
816
817 --nready;
818
819 X_UNLOCK (reqlock);
820
821 errno = 0; /* strictly unnecessary */
822
823 if (!EIO_CANCELLED (req))
824 switch (req->type)
825 {
826 case EIO_READ: req->result = req->offs >= 0
827 ? pread (req->int1, req->ptr2, req->size, req->offs)
828 : read (req->int1, req->ptr2, req->size); break;
829 case EIO_WRITE: req->result = req->offs >= 0
830 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
831 : write (req->int1, req->ptr2, req->size); break;
832
833 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
834 case EIO_SENDFILE: req->result = sendfile_ (req->int1, req->int2, req->offs, req->size, self); break;
835
836 case EIO_STAT: req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
837 case EIO_LSTAT: req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
838 case EIO_FSTAT: req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
839
840 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
841 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
842 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
843 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
844 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
845 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
846
847 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
848 case EIO_CLOSE: req->result = close (req->int1); break;
849 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
850 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
851 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
852 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
853 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
854 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
855 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
856 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
857 case EIO_READLINK: req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
858
859 case EIO_SYNC: req->result = 0; sync (); break;
860 case EIO_FSYNC: req->result = fsync (req->int1); break;
861 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
862
863 case EIO_READDIR: scandir_ (req, self); break;
864
865 case EIO_BUSY:
866#ifdef _WIN32
867 Sleep (req->nv1 * 1000.);
868#else
869 {
870 struct timeval tv;
871
872 tv.tv_sec = req->nv1;
873 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;
874
875 req->result = select (0, 0, 0, 0, &tv);
876 }
877#endif
878 break;
879
880 case EIO_UTIME:
881 case EIO_FUTIME:
882 {
883 struct timeval tv[2];
884 struct timeval *times;
885
886 if (req->nv1 != -1. || req->nv2 != -1.)
887 {
888 tv[0].tv_sec = req->nv1;
889 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
890 tv[1].tv_sec = req->nv2;
891 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
892
893 times = tv;
894 }
895 else
896 times = 0;
897
898
899 req->result = req->type == EIO_FUTIME
900 ? futimes (req->int1, times)
901 : utimes (req->ptr1, times);
902 }
903
904 case EIO_GROUP:
905 case EIO_NOP:
906 break;
907
908 case EIO_QUIT:
909 goto quit;
910
911 default:
912 req->result = -1;
913 break;
914 }
915
916 req->errorno = errno;
917
918 X_LOCK (reslock);
919
920 ++npending;
921
922 if (!reqq_push (&res_queue, req) && want_poll_cb)
923 want_poll_cb ();
924
925 self->req = 0;
926 worker_clear (self);
927
928 X_UNLOCK (reslock);
929 }
930
931quit:
932 X_LOCK (wrklock);
933 worker_free (self);
934 X_UNLOCK (wrklock);
935
936 return 0;
937}
938
939/*****************************************************************************/
940
941static void eio_atfork_prepare (void)
942{
943 X_LOCK (wrklock);
944 X_LOCK (reqlock);
945 X_LOCK (reslock);
946#if !HAVE_PREADWRITE
947 X_LOCK (preadwritelock);
948#endif
949#if !HAVE_READDIR_R
950 X_LOCK (readdirlock);
951#endif
952}
953
954static void eio_atfork_parent (void)
955{
956#if !HAVE_READDIR_R
957 X_UNLOCK (readdirlock);
958#endif
959#if !HAVE_PREADWRITE
960 X_UNLOCK (preadwritelock);
961#endif
962 X_UNLOCK (reslock);
963 X_UNLOCK (reqlock);
964 X_UNLOCK (wrklock);
965}
966
967static void eio_atfork_child (void)
968{
969 eio_req *prv;
970
971 while (prv = reqq_shift (&req_queue))
972 eio_destroy (prv);
973
974 while (prv = reqq_shift (&res_queue))
975 eio_destroy (prv);
976
977 while (wrk_first.next != &wrk_first)
978 {
979 worker *wrk = wrk_first.next;
980
981 if (wrk->req)
982 eio_destroy (wrk->req);
983
984 worker_clear (wrk);
985 worker_free (wrk);
986 }
987
988 started = 0;
989 idle = 0;
990 nreqs = 0;
991 nready = 0;
992 npending = 0;
993
994 eio_atfork_parent ();
995}
996
997int eio_init (void (*want_poll)(void), void (*done_poll)(void))
998{
999 want_poll_cb = want_poll;
1000 done_poll_cb = done_poll;
1001
1002#ifdef _WIN32
1003 X_MUTEX_CHECK (wrklock);
1004 X_MUTEX_CHECK (reslock);
1005 X_MUTEX_CHECK (reqlock);
1006 X_MUTEX_CHECK (reqwait);
1007 X_MUTEX_CHECK (preadwritelock);
1008 X_MUTEX_CHECK (readdirlock);
1009
1010 X_COND_CHECK (reqwait);
1011#endif
1012
1013 X_THREAD_ATFORK (eio_atfork_prepare, eio_atfork_parent, eio_atfork_child);
1014}
1015
1016#if 0
1017
1018eio_req *eio_fsync (int fd, eio_cb cb);
1019eio_req *eio_fdatasync (int fd, eio_cb cb);
1020eio_req *eio_dupclose (int fd, eio_cb cb);
1021eio_req *eio_readahead (int fd, off_t offset, size_t length, eio_cb cb);
1022eio_req *eio_read (int fd, off_t offs, size_t length, char *data, eio_cb cb);
1023eio_req *eio_write (int fd, off_t offs, size_t length, char *data, eio_cb cb);
1024eio_req *eio_fstat (int fd, eio_cb cb); /* stat buffer=ptr2 allocates dynamically */
1025eio_req *eio_futime (int fd, double atime, double mtime, eio_cb cb);
1026eio_req *eio_ftruncate (int fd, off_t offset, eio_cb cb);
1027eio_req *eio_fchmod (int fd, mode_t mode, eio_cb cb);
1028eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, eio_cb cb);
1029eio_req *eio_dup2 (int fd, int fd2, eio_cb cb);
1030eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, eio_cb cb);
1031eio_req *eio_open (const char *path, int flags, mode_t mode, eio_cb cb);
1032eio_req *eio_readlink (const char *path, eio_cb cb); /* result=ptr2 allocated dynamically */
1033eio_req *eio_stat (const char *path, eio_cb cb); /* stat buffer=ptr2 allocates dynamically */
1034eio_req *eio_lstat (const char *path, eio_cb cb); /* stat buffer=ptr2 allocates dynamically */
1035eio_req *eio_utime (const char *path, double atime, double mtime, eio_cb cb);
1036eio_req *eio_truncate (const char *path, off_t offset, eio_cb cb);
1037eio_req *eio_chmod (const char *path, mode_t mode, eio_cb cb);
1038eio_req *eio_mkdir (const char *path, mode_t mode, eio_cb cb);
1039eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, eio_cb cb);
1040eio_req *eio_unlink (const char *path, eio_cb cb);
1041eio_req *eio_rmdir (const char *path, eio_cb cb);
1042eio_req *eio_readdir (const char *path, eio_cb cb); /* result=ptr2 allocated dynamically */
1043eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, eio_cb cb);
1044eio_req *eio_busy (double delay, eio_cb cb); /* ties a thread for this long, simulating busyness */
1045eio_req *eio_nop (eio_cb cb); /* does nothing except go through the whole process */
1046void
1047aio_open (SV8 *pathname, int flags, int mode, SV *callback=&PL_sv_undef)
1048 PROTOTYPE: $$$;$
1049 PPCODE:
1050{
1051 dREQ;
1052
1053 req->type = EIO_OPEN;
1054 req->sv1 = newSVsv (pathname);
1055 req->ptr1 = SvPVbyte_nolen (req->sv1);
1056 req->int1 = flags;
1057 req->int2 = mode;
1058
1059 EIO_SEND;
1060}
1061
1062void
1063aio_fsync (SV *fh, SV *callback=&PL_sv_undef)
1064 PROTOTYPE: $;$
1065 ALIAS:
1066 aio_fsync = EIO_FSYNC
1067 aio_fdatasync = EIO_FDATASYNC
1068 PPCODE:
1069{
1070 dREQ;
1071
1072 req->type = ix;
1073 req->sv1 = newSVsv (fh);
1074 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1075
1076 EIO_SEND (req);
1077}
1078
1079void
1080aio_close (SV *fh, SV *callback=&PL_sv_undef)
1081 PROTOTYPE: $;$
1082 PPCODE:
1083{
1084 dREQ;
1085
1086 req->type = EIO_CLOSE;
1087 req->sv1 = newSVsv (fh);
1088 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1089
1090 EIO_SEND (req);
1091}
1092
1093void
1094aio_read (SV *fh, SV *offset, SV *length, SV8 *data, IV dataoffset, SV *callback=&PL_sv_undef)
1095 ALIAS:
1096 aio_read = EIO_READ
1097 aio_write = EIO_WRITE
1098 PROTOTYPE: $$$$$;$
1099 PPCODE:
1100{
1101 STRLEN svlen;
1102 char *svptr = SvPVbyte (data, svlen);
1103 UV len = SvUV (length);
1104
1105 SvUPGRADE (data, SVt_PV);
1106 SvPOK_on (data);
1107
1108 if (dataoffset < 0)
1109 dataoffset += svlen;
1110
1111 if (dataoffset < 0 || dataoffset > svlen)
1112 croak ("dataoffset outside of data scalar");
1113
1114 if (ix == EIO_WRITE)
1115 {
1116 /* write: check length and adjust. */
1117 if (!SvOK (length) || len + dataoffset > svlen)
1118 len = svlen - dataoffset;
1119 } 1509 }
1120 else 1510 else
1121 { 1511 {
1122 /* read: grow scalar as necessary */ 1512 /* yay, it was a symlink - build new path in tmp2 */
1123 svptr = SvGROW (data, len + dataoffset + 1); 1513 int rellen = strlen (rel);
1514
1515 errno = ENAMETOOLONG;
1516 if (linklen + 1 + rellen >= PATH_MAX)
1517 return -1;
1518
1519 errno = ELOOP;
1520 if (!--symlinks)
1521 return -1;
1522
1523 if (*tmp1 == '/')
1524 res = tmpbuf->ptr; /* symlink resolves to an absolute path */
1525
1526 /* we need to be careful, as rel might point into tmp2 already */
1527 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1528 tmp2 [linklen] = '/';
1529 memcpy (tmp2, tmp1, linklen);
1530
1531 rel = tmp2;
1124 } 1532 }
1533 }
1125 1534
1535 /* special case for the lone root path */
1536 if (res == tmpbuf->ptr)
1537 *res++ = '/';
1538
1539 return res - (char *)tmpbuf->ptr;
1540}
1541
1542static signed char
1543eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1544{
1545 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1546 : a->inode < b->inode ? -1
1547 : a->inode > b->inode ? 1
1548 : 0;
1549}
1550
1551#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1552
1553#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1554#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1555
1556static void
1557eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1558{
1559 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1560 unsigned char *bit = bits;
1561
1562 assert (CHAR_BIT == 8);
1563 assert (sizeof (eio_dirent) * 8 < 256);
1564 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1565 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1566
1567 if (size <= EIO_SORT_FAST)
1568 return;
1569
1570 /* first prepare an array of bits to test in our radix sort */
1571 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1572 /* inode_bits must contain all inodes ORed together */
1573 /* which is used to skip bits that are 0 everywhere, which is very common */
1574 {
1575 eio_ino_t endianness;
1576 int i, j;
1577
1578 /* we store the byte offset of byte n into byte n of "endianness" */
1579 for (i = 0; i < sizeof (eio_ino_t); ++i)
1580 ((unsigned char *)&endianness)[i] = i;
1581
1582 *bit++ = 0;
1583
1584 for (i = 0; i < sizeof (eio_ino_t); ++i)
1585 {
1586 /* shifting off the byte offsets out of "endianness" */
1587 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1588 endianness >>= 8;
1589
1590 for (j = 0; j < 8; ++j)
1591 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1592 *bit++ = offs + j;
1593 }
1594
1595 for (j = 0; j < 8; ++j)
1596 if (score_bits & (1 << j))
1597 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1598 }
1599
1600 /* now actually do the sorting (a variant of MSD radix sort) */
1601 {
1602 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1603 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1604 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1605 int stk_idx = 0;
1606
1607 base_stk [stk_idx] = dents;
1608 end_stk [stk_idx] = dents + size;
1609 bit_stk [stk_idx] = bit - 1;
1610
1611 do
1612 {
1613 base = base_stk [stk_idx];
1614 end = end_stk [stk_idx];
1615 bit = bit_stk [stk_idx];
1616
1617 for (;;)
1618 {
1619 unsigned char O = *bit >> 3;
1620 unsigned char M = 1 << (*bit & 7);
1621
1622 eio_dirent *a = base;
1623 eio_dirent *b = end;
1624
1625 if (b - a < EIO_SORT_CUTOFF)
1626 break;
1627
1628 /* now bit-partition the array on the bit */
1629 /* this ugly asymmetric loop seems to perform much better than typical */
1630 /* partition algos found in the literature */
1631 do
1632 if (!(((unsigned char *)a)[O] & M))
1633 ++a;
1634 else if (!(((unsigned char *)--b)[O] & M))
1635 {
1636 eio_dirent tmp = *a; *a = *b; *b = tmp;
1637 ++a;
1638 }
1639 while (b > a);
1640
1641 /* next bit, or stop, if no bits left in this path */
1642 if (!*--bit)
1643 break;
1644
1645 base_stk [stk_idx] = a;
1646 end_stk [stk_idx] = end;
1647 bit_stk [stk_idx] = bit;
1648 ++stk_idx;
1649
1650 end = a;
1651 }
1652 }
1653 while (stk_idx--);
1654 }
1655}
1656
1657static void
1658eio_dent_insertion_sort (eio_dirent *dents, int size)
1659{
1660 /* first move the smallest element to the front, to act as a sentinel */
1661 {
1662 int i;
1663 eio_dirent *min = dents;
1664
1665 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1666 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1667 if (EIO_DENT_CMP (dents [i], <, *min))
1668 min = &dents [i];
1669
1670 /* swap elements 0 and j (minimum) */
1671 {
1672 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1673 }
1674 }
1675
1676 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1677 {
1678 eio_dirent *i, *j;
1679
1680 for (i = dents + 1; i < dents + size; ++i)
1681 {
1682 eio_dirent value = *i;
1683
1684 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1685 j [1] = j [0];
1686
1687 j [1] = value;
1688 }
1689 }
1690}
1691
1692static void
1693eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1694{
1695 if (size <= 1)
1696 return; /* our insertion sort relies on size > 0 */
1697
1698 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1699 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1700 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1701
1702 /* use an insertion sort at the end, or for small arrays, */
1703 /* as insertion sort is more efficient for small partitions */
1704 eio_dent_insertion_sort (dents, size);
1705}
1706
1707/* read a full directory */
1708static void
1709eio__scandir (eio_req *req, etp_worker *self)
1710{
1711 char *name, *names;
1712 int namesalloc = 4096 - sizeof (void *) * 4;
1713 int namesoffs = 0;
1714 int flags = req->int1;
1715 eio_dirent *dents = 0;
1716 int dentalloc = 128;
1717 int dentoffs = 0;
1718 eio_ino_t inode_bits = 0;
1719#ifdef _WIN32
1720 HANDLE dirp;
1721 WIN32_FIND_DATA entp;
1722#else
1723 DIR *dirp;
1724 EIO_STRUCT_DIRENT *entp;
1725#endif
1726
1727 req->result = -1;
1728
1729 if (!(flags & EIO_READDIR_DENTS))
1730 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1731
1732#ifdef _WIN32
1733 {
1734 int len = strlen ((const char *)req->ptr1);
1735 char *path = malloc (MAX_PATH);
1736 const char *fmt;
1737 const char *reqpath = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1738
1739 if (!len)
1740 fmt = "./*";
1741 else if (reqpath[len - 1] == '/' || reqpath[len - 1] == '\\')
1742 fmt = "%s*";
1743 else
1744 fmt = "%s/*";
1745
1746 _snprintf (path, MAX_PATH, fmt, reqpath);
1747 dirp = FindFirstFile (path, &entp);
1748 free (path);
1749
1750 if (dirp == INVALID_HANDLE_VALUE)
1751 {
1752 /* should steal _dosmaperr */
1753 switch (GetLastError ())
1754 {
1755 case ERROR_FILE_NOT_FOUND:
1756 req->result = 0;
1757 break;
1758
1759 case ERROR_INVALID_NAME:
1760 case ERROR_PATH_NOT_FOUND:
1761 case ERROR_NO_MORE_FILES:
1762 errno = ENOENT;
1763 break;
1764
1765 case ERROR_NOT_ENOUGH_MEMORY:
1766 errno = ENOMEM;
1767 break;
1768
1769 default:
1770 errno = EINVAL;
1771 break;
1772 }
1773
1774 return;
1775 }
1776 }
1777#else
1778 #if HAVE_AT
1779 if (req->wd)
1780 {
1781 int fd = openat (WD2FD (req->wd), req->ptr1, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1782
1126 if (len < 0) 1783 if (fd < 0)
1127 croak ("length must not be negative"); 1784 return;
1128 1785
1786 dirp = fdopendir (fd);
1787
1788 if (!dirp)
1789 close (fd);
1790 }
1791 else
1792 dirp = opendir (req->ptr1);
1793 #else
1794 dirp = opendir (wd_expand (&self->tmpbuf, req->wd, req->ptr1));
1795 #endif
1796
1797 if (!dirp)
1798 return;
1799#endif
1800
1801 if (req->flags & EIO_FLAG_PTR1_FREE)
1802 free (req->ptr1);
1803
1804 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1805 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1806 req->ptr2 = names = malloc (namesalloc);
1807
1808 if (!names || (flags && !dents))
1809 return;
1810
1811 for (;;)
1812 {
1813 int done;
1814
1815#ifdef _WIN32
1816 done = !dirp;
1817#else
1818 errno = 0;
1819 entp = readdir (dirp);
1820 done = !entp;
1821#endif
1822
1823 if (done)
1129 { 1824 {
1130 dREQ; 1825#ifndef _WIN32
1826 int old_errno = errno;
1827 closedir (dirp);
1828 errno = old_errno;
1131 1829
1132 req->type = ix; 1830 if (errno)
1133 req->sv1 = newSVsv (fh); 1831 break;
1134 req->int1 = PerlIO_fileno (ix == EIO_READ ? IoIFP (sv_2io (fh)) 1832#endif
1135 : IoOFP (sv_2io (fh)));
1136 req->offs = SvOK (offset) ? SvVAL64 (offset) : -1;
1137 req->size = len;
1138 req->sv2 = SvREFCNT_inc (data);
1139 req->ptr2 = (char *)svptr + dataoffset;
1140 req->stroffset = dataoffset;
1141 1833
1142 if (!SvREADONLY (data)) 1834 /* sort etc. */
1835 req->int1 = flags;
1836 req->result = dentoffs;
1837
1838 if (flags & EIO_READDIR_STAT_ORDER)
1839 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1840 else if (flags & EIO_READDIR_DIRS_FIRST)
1841 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1842 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1843 else
1844 {
1845 /* in this case, all is known, and we just put dirs first and sort them */
1846 eio_dirent *oth = dents + dentoffs;
1847 eio_dirent *dir = dents;
1848
1849 /* now partition dirs to the front, and non-dirs to the back */
1850 /* by walking from both sides and swapping if necessary */
1851 while (oth > dir)
1852 {
1853 if (dir->type == EIO_DT_DIR)
1854 ++dir;
1855 else if ((--oth)->type == EIO_DT_DIR)
1856 {
1857 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1858
1859 ++dir;
1860 }
1861 }
1862
1863 /* now sort the dirs only (dirs all have the same score) */
1864 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1865 }
1866
1867 break;
1868 }
1869
1870 /* now add the entry to our list(s) */
1871 name = D_NAME (entp);
1872
1873 /* skip . and .. entries */
1874 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1875 {
1876 int len = D_NAMLEN (entp) + 1;
1877
1878 while (ecb_expect_false (namesoffs + len > namesalloc))
1143 { 1879 {
1144 SvREADONLY_on (data); 1880 namesalloc *= 2;
1145 req->flags |= FLAG_SV2_RO_OFF; 1881 req->ptr2 = names = realloc (names, namesalloc);
1882
1883 if (!names)
1884 break;
1146 } 1885 }
1147 1886
1148 EIO_SEND; 1887 memcpy (names + namesoffs, name, len);
1888
1889 if (dents)
1890 {
1891 struct eio_dirent *ent;
1892
1893 if (ecb_expect_false (dentoffs == dentalloc))
1894 {
1895 dentalloc *= 2;
1896 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1897
1898 if (!dents)
1899 break;
1900 }
1901
1902 ent = dents + dentoffs;
1903
1904 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1905 ent->namelen = len - 1;
1906 ent->inode = D_INO (entp);
1907
1908 inode_bits |= ent->inode;
1909
1910 switch (D_TYPE (entp))
1911 {
1912 default:
1913 ent->type = EIO_DT_UNKNOWN;
1914 flags |= EIO_READDIR_FOUND_UNKNOWN;
1915 break;
1916
1917 #ifdef DT_FIFO
1918 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1919 #endif
1920 #ifdef DT_CHR
1921 case DT_CHR: ent->type = EIO_DT_CHR; break;
1922 #endif
1923 #ifdef DT_MPC
1924 case DT_MPC: ent->type = EIO_DT_MPC; break;
1925 #endif
1926 #ifdef DT_DIR
1927 case DT_DIR: ent->type = EIO_DT_DIR; break;
1928 #endif
1929 #ifdef DT_NAM
1930 case DT_NAM: ent->type = EIO_DT_NAM; break;
1931 #endif
1932 #ifdef DT_BLK
1933 case DT_BLK: ent->type = EIO_DT_BLK; break;
1934 #endif
1935 #ifdef DT_MPB
1936 case DT_MPB: ent->type = EIO_DT_MPB; break;
1937 #endif
1938 #ifdef DT_REG
1939 case DT_REG: ent->type = EIO_DT_REG; break;
1940 #endif
1941 #ifdef DT_NWK
1942 case DT_NWK: ent->type = EIO_DT_NWK; break;
1943 #endif
1944 #ifdef DT_CMP
1945 case DT_CMP: ent->type = EIO_DT_CMP; break;
1946 #endif
1947 #ifdef DT_LNK
1948 case DT_LNK: ent->type = EIO_DT_LNK; break;
1949 #endif
1950 #ifdef DT_SOCK
1951 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1952 #endif
1953 #ifdef DT_DOOR
1954 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1955 #endif
1956 #ifdef DT_WHT
1957 case DT_WHT: ent->type = EIO_DT_WHT; break;
1958 #endif
1959 }
1960
1961 ent->score = 7;
1962
1963 if (flags & EIO_READDIR_DIRS_FIRST)
1964 {
1965 if (ent->type == EIO_DT_UNKNOWN)
1966 {
1967 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1968 ent->score = 1;
1969 else if (!strchr (name, '.')) /* absence of dots indicate likely dirs */
1970 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1971 }
1972 else if (ent->type == EIO_DT_DIR)
1973 ent->score = 0;
1974 }
1975 }
1976
1977 namesoffs += len;
1978 ++dentoffs;
1149 } 1979 }
1980
1981 if (EIO_CANCELLED (req))
1982 {
1983 errno = ECANCELED;
1984 break;
1985 }
1986
1987#ifdef _WIN32
1988 if (!FindNextFile (dirp, &entp))
1989 {
1990 FindClose (dirp);
1991 dirp = 0;
1992 }
1993#endif
1994 }
1995}
1996
1997/*****************************************************************************/
1998/* working directory stuff */
1999/* various deficiencies in the posix 2008 api force us to */
2000/* keep the absolute path in string form at all times */
2001/* fuck yeah. */
2002
2003#if !HAVE_AT
2004
2005/* a bit like realpath, but usually faster because it doesn'T have to return */
2006/* an absolute or canonical path */
2007static const char *
2008wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
2009{
2010 if (!wd || *path == '/')
2011 return path;
2012
2013 if (path [0] == '.' && !path [1])
2014 return wd->str;
2015
2016 {
2017 int l1 = wd->len;
2018 int l2 = strlen (path);
2019
2020 char *res = tmpbuf_get (tmpbuf, l1 + l2 + 2);
2021
2022 memcpy (res, wd->str, l1);
2023 res [l1] = '/';
2024 memcpy (res + l1 + 1, path, l2 + 1);
2025
2026 return res;
2027 }
2028}
2029
2030#endif
2031
2032static eio_wd
2033eio__wd_open_sync (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
2034{
2035 int fd;
2036 eio_wd res;
2037 int len = eio__realpath (tmpbuf, wd, path);
2038
2039 if (len < 0)
2040 return EIO_INVALID_WD;
2041
2042#if HAVE_AT
2043 fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
2044
2045 if (fd < 0)
2046 return EIO_INVALID_WD;
2047#endif
2048
2049 res = malloc (sizeof (*res) + len); /* one extra 0-byte */
2050
2051#if HAVE_AT
2052 res->fd = fd;
2053#endif
2054
2055 res->len = len;
2056 memcpy (res->str, tmpbuf->ptr, len);
2057 res->str [len] = 0;
2058
2059 return res;
2060}
2061
2062eio_wd
2063eio_wd_open_sync (eio_wd wd, const char *path)
2064{
2065 struct tmpbuf tmpbuf = { 0 };
2066 wd = eio__wd_open_sync (&tmpbuf, wd, path);
2067 free (tmpbuf.ptr);
2068
2069 return wd;
1150} 2070}
1151 2071
1152void 2072void
1153aio_readlink (SV8 *path, SV *callback=&PL_sv_undef) 2073eio_wd_close_sync (eio_wd wd)
1154 PROTOTYPE: $$;$
1155 PPCODE:
1156{ 2074{
1157 SV *data; 2075 if (wd != EIO_INVALID_WD && wd != EIO_CWD)
1158 dREQ; 2076 {
1159 2077 #if HAVE_AT
1160 data = newSV (NAME_MAX); 2078 close (wd->fd);
1161 SvPOK_on (data); 2079 #endif
1162 2080 free (wd);
1163 req->type = EIO_READLINK; 2081 }
1164 req->sv1 = newSVsv (path);
1165 req->ptr1 = SvPVbyte_nolen (req->sv1);
1166 req->sv2 = data;
1167 req->ptr2 = SvPVbyte_nolen (data);
1168
1169 EIO_SEND;
1170} 2082}
2083
2084#if HAVE_AT
2085
2086/* they forgot these */
2087
2088static int
2089eio__truncateat (int dirfd, const char *path, off_t length)
2090{
2091 int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
2092 int res;
2093
2094 if (fd < 0)
2095 return fd;
2096
2097 res = ftruncate (fd, length);
2098 close (fd);
2099 return res;
2100}
2101
2102static int
2103eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
2104{
2105 int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
2106 int res;
2107
2108 if (fd < 0)
2109 return fd;
2110
2111 res = fstatvfs (fd, buf);
2112 close (fd);
2113 return res;
2114
2115}
2116
2117#endif
2118
2119/*****************************************************************************/
2120
2121#define ALLOC(len) \
2122 if (!req->ptr2) \
2123 { \
2124 X_LOCK (wrklock); \
2125 req->flags |= EIO_FLAG_PTR2_FREE; \
2126 X_UNLOCK (wrklock); \
2127 req->ptr2 = malloc (len); \
2128 if (!req->ptr2) \
2129 { \
2130 errno = ENOMEM; \
2131 req->result = -1; \
2132 break; \
2133 } \
2134 }
2135
2136static void ecb_noinline ecb_cold
2137etp_proc_init (void)
2138{
2139#if HAVE_PRCTL_SET_NAME
2140 /* provide a more sensible "thread name" */
2141 char name[16 + 1];
2142 const int namelen = sizeof (name) - 1;
2143 int len;
2144
2145 prctl (PR_GET_NAME, (unsigned long)name, 0, 0, 0);
2146 name [namelen] = 0;
2147 len = strlen (name);
2148 strcpy (name + (len <= namelen - 4 ? len : namelen - 4), "/eio");
2149 prctl (PR_SET_NAME, (unsigned long)name, 0, 0, 0);
2150#endif
2151}
2152
2153X_THREAD_PROC (etp_proc)
2154{
2155 ETP_REQ *req;
2156 struct timespec ts;
2157 etp_worker *self = (etp_worker *)thr_arg;
2158
2159 etp_proc_init ();
2160
2161 /* try to distribute timeouts somewhat evenly */
2162 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
2163
2164 for (;;)
2165 {
2166 ts.tv_sec = 0;
2167
2168 X_LOCK (reqlock);
2169
2170 for (;;)
2171 {
2172 req = reqq_shift (&req_queue);
2173
2174 if (req)
2175 break;
2176
2177 if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
2178 {
2179 X_UNLOCK (reqlock);
2180 X_LOCK (wrklock);
2181 --started;
2182 X_UNLOCK (wrklock);
2183 goto quit;
2184 }
2185
2186 ++idle;
2187
2188 if (idle <= max_idle)
2189 /* we are allowed to idle, so do so without any timeout */
2190 X_COND_WAIT (reqwait, reqlock);
2191 else
2192 {
2193 /* initialise timeout once */
2194 if (!ts.tv_sec)
2195 ts.tv_sec = time (0) + idle_timeout;
2196
2197 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
2198 ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
2199 }
2200
2201 --idle;
2202 }
2203
2204 --nready;
2205
2206 X_UNLOCK (reqlock);
2207
2208 if (req->type < 0)
2209 goto quit;
2210
2211 ETP_EXECUTE (self, req);
2212
2213 X_LOCK (reslock);
2214
2215 ++npending;
2216
2217 if (!reqq_push (&res_queue, req) && want_poll_cb)
2218 want_poll_cb ();
2219
2220 etp_worker_clear (self);
2221
2222 X_UNLOCK (reslock);
2223 }
2224
2225quit:
2226 free (req);
2227
2228 X_LOCK (wrklock);
2229 etp_worker_free (self);
2230 X_UNLOCK (wrklock);
2231
2232 return 0;
2233}
2234
2235/*****************************************************************************/
2236
2237int ecb_cold
2238eio_init (void (*want_poll)(void), void (*done_poll)(void))
2239{
2240 return etp_init (want_poll, done_poll);
2241}
2242
2243ecb_inline void
2244eio_api_destroy (eio_req *req)
2245{
2246 free (req);
2247}
2248
2249#define REQ(rtype) \
2250 eio_req *req; \
2251 \
2252 req = (eio_req *)calloc (1, sizeof *req); \
2253 if (!req) \
2254 return 0; \
2255 \
2256 req->type = rtype; \
2257 req->pri = pri; \
2258 req->finish = cb; \
2259 req->data = data; \
2260 req->destroy = eio_api_destroy;
2261
2262#define SEND eio_submit (req); return req
2263
2264#define PATH \
2265 req->flags |= EIO_FLAG_PTR1_FREE; \
2266 req->ptr1 = strdup (path); \
2267 if (!req->ptr1) \
2268 { \
2269 eio_api_destroy (req); \
2270 return 0; \
2271 }
2272
2273#define SINGLEDOT(ptr) (0[(char *)(ptr)] == '.' && !1[(char *)(ptr)])
2274
2275static void
2276eio_execute (etp_worker *self, eio_req *req)
2277{
2278#if HAVE_AT
2279 int dirfd;
2280#else
2281 const char *path;
2282#endif
2283
2284 if (ecb_expect_false (EIO_CANCELLED (req)))
2285 {
2286 req->result = -1;
2287 req->errorno = ECANCELED;
2288 return;
2289 }
2290
2291 if (ecb_expect_false (req->wd == EIO_INVALID_WD))
2292 {
2293 req->result = -1;
2294 req->errorno = ENOENT;
2295 return;
2296 }
2297
2298 if (req->type >= EIO_OPEN)
2299 {
2300 #if HAVE_AT
2301 dirfd = WD2FD (req->wd);
2302 #else
2303 path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
2304 #endif
2305 }
2306
2307 switch (req->type)
2308 {
2309 case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
2310 req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
2311 break;
2312 case EIO_WD_CLOSE: req->result = 0;
2313 eio_wd_close_sync (req->wd); break;
2314
2315 case EIO_SEEK: eio__lseek (req); break;
2316 case EIO_READ: ALLOC (req->size);
2317 req->result = req->offs >= 0
2318 ? pread (req->int1, req->ptr2, req->size, req->offs)
2319 : read (req->int1, req->ptr2, req->size); break;
2320 case EIO_WRITE: req->result = req->offs >= 0
2321 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
2322 : write (req->int1, req->ptr2, req->size); break;
2323
2324 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
2325 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
2326
2327#if HAVE_AT
2328
2329 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2330 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break;
2331 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2332 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, AT_SYMLINK_NOFOLLOW); break;
2333 case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break;
2334 case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break;
2335 case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break;
2336 case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break;
2337
2338 case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
2339 case EIO_RMDIR: /* complications arise because "." cannot be removed, so we might have to expand */
2340 req->result = req->wd && SINGLEDOT (req->ptr1)
2341 ? rmdir (req->wd->str)
2342 : unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break;
2343 case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); break;
2344 case EIO_RENAME: /* complications arise because "." cannot be renamed, so we might have to expand */
2345 req->result = req->wd && SINGLEDOT (req->ptr1)
2346 ? rename (req->wd->str, req->ptr2)
2347 : renameat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2); break;
2348 case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2, 0); break;
2349 case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
2350 case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
2351 case EIO_READLINK: ALLOC (PATH_MAX);
2352 req->result = readlinkat (dirfd, req->ptr1, req->ptr2, PATH_MAX); break;
2353 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2354 req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2355 case EIO_UTIME:
2356 case EIO_FUTIME:
2357 {
2358 struct timespec ts[2];
2359 struct timespec *times;
2360
2361 if (req->nv1 != -1. || req->nv2 != -1.)
2362 {
2363 ts[0].tv_sec = req->nv1;
2364 ts[0].tv_nsec = (req->nv1 - ts[0].tv_sec) * 1e9;
2365 ts[1].tv_sec = req->nv2;
2366 ts[1].tv_nsec = (req->nv2 - ts[1].tv_sec) * 1e9;
2367
2368 times = ts;
2369 }
2370 else
2371 times = 0;
2372
2373 req->result = req->type == EIO_FUTIME
2374 ? futimens (req->int1, times)
2375 : utimensat (dirfd, req->ptr1, times, 0);
2376 }
2377 break;
2378
2379#else
2380
2381 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2382 req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
2383 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2384 req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
2385 case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
2386 case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
2387 case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
2388 case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
2389
2390 case EIO_UNLINK: req->result = unlink (path ); break;
2391 case EIO_RMDIR: req->result = rmdir (path ); break;
2392 case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
2393 case EIO_RENAME: req->result = rename (path , req->ptr2); break;
2394 case EIO_LINK: req->result = link (path , req->ptr2); break;
2395 case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
2396 case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
2397 case EIO_READLINK: ALLOC (PATH_MAX);
2398 req->result = readlink (path, req->ptr2, PATH_MAX); break;
2399 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2400 req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
2401
2402 case EIO_UTIME:
2403 case EIO_FUTIME:
2404 {
2405 struct timeval tv[2];
2406 struct timeval *times;
2407
2408 if (req->nv1 != -1. || req->nv2 != -1.)
2409 {
2410 tv[0].tv_sec = req->nv1;
2411 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
2412 tv[1].tv_sec = req->nv2;
2413 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
2414
2415 times = tv;
2416 }
2417 else
2418 times = 0;
2419
2420 req->result = req->type == EIO_FUTIME
2421 ? futimes (req->int1, times)
2422 : utimes (req->ptr1, times);
2423 }
2424 break;
2425
2426#endif
2427
2428 case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
2429 {
2430 ALLOC (req->result);
2431 memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
2432 }
2433 break;
2434
2435 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2436 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
2437
2438 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2439 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2440
2441 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
2442 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
2443 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
2444
2445 case EIO_CLOSE: req->result = close (req->int1); break;
2446 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
2447 case EIO_SYNC: req->result = 0; sync (); break;
2448 case EIO_FSYNC: req->result = fsync (req->int1); break;
2449 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
2450 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
2451 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
2452 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
2453 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
2454 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
2455 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
2456 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
2457
2458 case EIO_READDIR: eio__scandir (req, self); break;
2459
2460 case EIO_BUSY:
2461#ifdef _WIN32
2462 Sleep (req->nv1 * 1e3);
2463#else
2464 {
2465 struct timeval tv;
2466
2467 tv.tv_sec = req->nv1;
2468 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
2469
2470 req->result = select (0, 0, 0, 0, &tv);
2471 }
2472#endif
2473 break;
2474
2475 case EIO_GROUP:
2476 abort (); /* handled in eio_request */
2477
2478 case EIO_NOP:
2479 req->result = 0;
2480 break;
2481
2482 case EIO_CUSTOM:
2483 req->feed (req);
2484 break;
2485
2486 default:
2487 req->result = EIO_ENOSYS ();
2488 break;
2489 }
2490
2491 req->errorno = errno;
2492}
2493
2494#ifndef EIO_NO_WRAPPERS
2495
2496eio_req *eio_wd_open (const char *path, int pri, eio_cb cb, void *data)
2497{
2498 REQ (EIO_WD_OPEN); PATH; SEND;
2499}
2500
2501eio_req *eio_wd_close (eio_wd wd, int pri, eio_cb cb, void *data)
2502{
2503 REQ (EIO_WD_CLOSE); req->wd = wd; SEND;
2504}
2505
2506eio_req *eio_nop (int pri, eio_cb cb, void *data)
2507{
2508 REQ (EIO_NOP); SEND;
2509}
2510
2511eio_req *eio_busy (double delay, int pri, eio_cb cb, void *data)
2512{
2513 REQ (EIO_BUSY); req->nv1 = delay; SEND;
2514}
2515
2516eio_req *eio_sync (int pri, eio_cb cb, void *data)
2517{
2518 REQ (EIO_SYNC); SEND;
2519}
2520
2521eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
2522{
2523 REQ (EIO_FSYNC); req->int1 = fd; SEND;
2524}
2525
2526eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2527{
2528 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2529}
2530
2531eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2532{
2533 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2534}
2535
2536eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
2537{
2538 REQ (EIO_SYNCFS); req->int1 = fd; SEND;
2539}
2540
2541eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2542{
2543 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2544}
2545
2546eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2547{
2548 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2549}
2550
2551eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
2552{
2553 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
2554}
2555
2556eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
2557{
2558 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2559}
2560
2561eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2562{
2563 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
2564}
2565
2566eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
2567{
2568 REQ (EIO_CLOSE); req->int1 = fd; SEND;
2569}
2570
2571eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
2572{
2573 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
2574}
2575
2576eio_req *eio_seek (int fd, off_t offset, int whence, int pri, eio_cb cb, void *data)
2577{
2578 REQ (EIO_SEEK); req->int1 = fd; req->offs = offset; req->int2 = whence; SEND;
2579}
2580
2581eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
2582{
2583 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
2584}
2585
2586eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
2587{
2588 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
2589}
2590
2591eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
2592{
2593 REQ (EIO_FSTAT); req->int1 = fd; SEND;
2594}
2595
2596eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
2597{
2598 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
2599}
2600
2601eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
2602{
2603 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
2604}
2605
2606eio_req *eio_ftruncate (int fd, off_t offset, int pri, eio_cb cb, void *data)
2607{
2608 REQ (EIO_FTRUNCATE); req->int1 = fd; req->offs = offset; SEND;
2609}
2610
2611eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
2612{
2613 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
2614}
2615
2616eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
2617{
2618 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
2619}
2620
2621eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
2622{
2623 REQ (EIO_DUP2); req->int1 = fd; req->int2 = fd2; SEND;
2624}
2625
2626eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, int pri, eio_cb cb, void *data)
2627{
2628 REQ (EIO_SENDFILE); req->int1 = out_fd; req->int2 = in_fd; req->offs = in_offset; req->size = length; SEND;
2629}
2630
2631eio_req *eio_open (const char *path, int flags, mode_t mode, int pri, eio_cb cb, void *data)
2632{
2633 REQ (EIO_OPEN); PATH; req->int1 = flags; req->int2 = (long)mode; SEND;
2634}
2635
2636eio_req *eio_utime (const char *path, double atime, double mtime, int pri, eio_cb cb, void *data)
2637{
2638 REQ (EIO_UTIME); PATH; req->nv1 = atime; req->nv2 = mtime; SEND;
2639}
2640
2641eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
2642{
2643 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
2644}
2645
2646eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
2647{
2648 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
2649}
2650
2651eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
2652{
2653 REQ (EIO_CHMOD); PATH; req->int2 = (long)mode; SEND;
2654}
2655
2656eio_req *eio_mkdir (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
2657{
2658 REQ (EIO_MKDIR); PATH; req->int2 = (long)mode; SEND;
2659}
2660
2661static eio_req *
2662eio__1path (int type, const char *path, int pri, eio_cb cb, void *data)
2663{
2664 REQ (type); PATH; SEND;
2665}
2666
2667eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
2668{
2669 return eio__1path (EIO_READLINK, path, pri, cb, data);
2670}
2671
2672eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2673{
2674 return eio__1path (EIO_REALPATH, path, pri, cb, data);
2675}
2676
2677eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
2678{
2679 return eio__1path (EIO_STAT, path, pri, cb, data);
2680}
2681
2682eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
2683{
2684 return eio__1path (EIO_LSTAT, path, pri, cb, data);
2685}
2686
2687eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
2688{
2689 return eio__1path (EIO_STATVFS, path, pri, cb, data);
2690}
2691
2692eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
2693{
2694 return eio__1path (EIO_UNLINK, path, pri, cb, data);
2695}
2696
2697eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
2698{
2699 return eio__1path (EIO_RMDIR, path, pri, cb, data);
2700}
2701
2702eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
2703{
2704 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
2705}
2706
2707eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
2708{
2709 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
2710}
2711
2712static eio_req *
2713eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2714{
2715 REQ (type); PATH;
2716
2717 req->flags |= EIO_FLAG_PTR2_FREE;
2718 req->ptr2 = strdup (new_path);
2719 if (!req->ptr2)
2720 {
2721 eio_api_destroy (req);
2722 return 0;
2723 }
2724
2725 SEND;
2726}
2727
2728eio_req *eio_link (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2729{
2730 return eio__2path (EIO_LINK, path, new_path, pri, cb, data);
2731}
2732
2733eio_req *eio_symlink (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2734{
2735 return eio__2path (EIO_SYMLINK, path, new_path, pri, cb, data);
2736}
2737
2738eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2739{
2740 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
2741}
2742
2743eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
2744{
2745 REQ (EIO_CUSTOM); req->feed = execute; SEND;
2746}
2747
2748#endif
2749
2750eio_req *eio_grp (eio_cb cb, void *data)
2751{
2752 const int pri = EIO_PRI_MAX;
2753
2754 REQ (EIO_GROUP); SEND;
2755}
2756
2757#undef REQ
2758#undef PATH
2759#undef SEND
2760
2761/*****************************************************************************/
2762/* grp functions */
1171 2763
1172void 2764void
1173aio_sendfile (SV *out_fh, SV *in_fh, SV *in_offset, UV length, SV *callback=&PL_sv_undef)
1174 PROTOTYPE: $$$$;$
1175 PPCODE:
1176{
1177 dREQ;
1178
1179 req->type = EIO_SENDFILE;
1180 req->sv1 = newSVsv (out_fh);
1181 req->int1 = PerlIO_fileno (IoIFP (sv_2io (out_fh)));
1182 req->sv2 = newSVsv (in_fh);
1183 req->int2 = PerlIO_fileno (IoIFP (sv_2io (in_fh)));
1184 req->offs = SvVAL64 (in_offset);
1185 req->size = length;
1186
1187 EIO_SEND;
1188}
1189
1190void
1191aio_readahead (SV *fh, SV *offset, IV length, SV *callback=&PL_sv_undef)
1192 PROTOTYPE: $$$;$
1193 PPCODE:
1194{
1195 dREQ;
1196
1197 req->type = EIO_READAHEAD;
1198 req->sv1 = newSVsv (fh);
1199 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1200 req->offs = SvVAL64 (offset);
1201 req->size = length;
1202
1203 EIO_SEND;
1204}
1205
1206void
1207aio_stat (SV8 *fh_or_path, SV *callback=&PL_sv_undef)
1208 ALIAS:
1209 aio_stat = EIO_STAT
1210 aio_lstat = EIO_LSTAT
1211 PPCODE:
1212{
1213 dREQ;
1214
1215 req->ptr2 = malloc (sizeof (EIO_STRUCT_STAT));
1216 if (!req->ptr2)
1217 {
1218 req_destroy (req);
1219 croak ("out of memory during aio_stat statdata allocation");
1220 }
1221
1222 req->flags |= FLAG_PTR2_FREE;
1223 req->sv1 = newSVsv (fh_or_path);
1224
1225 if (SvPOK (fh_or_path))
1226 {
1227 req->type = ix;
1228 req->ptr1 = SvPVbyte_nolen (req->sv1);
1229 }
1230 else
1231 {
1232 req->type = EIO_FSTAT;
1233 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1234 }
1235
1236 EIO_SEND;
1237}
1238
1239void
1240aio_utime (SV8 *fh_or_path, SV *atime, SV *mtime, SV *callback=&PL_sv_undef)
1241 PPCODE:
1242{
1243 dREQ;
1244
1245 req->nv1 = SvOK (atime) ? SvNV (atime) : -1.;
1246 req->nv2 = SvOK (mtime) ? SvNV (mtime) : -1.;
1247 req->sv1 = newSVsv (fh_or_path);
1248
1249 if (SvPOK (fh_or_path))
1250 {
1251 req->type = EIO_UTIME;
1252 req->ptr1 = SvPVbyte_nolen (req->sv1);
1253 }
1254 else
1255 {
1256 req->type = EIO_FUTIME;
1257 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1258 }
1259
1260 EIO_SEND;
1261}
1262
1263void
1264aio_truncate (SV8 *fh_or_path, SV *offset, SV *callback=&PL_sv_undef)
1265 PPCODE:
1266{
1267 dREQ;
1268
1269 req->sv1 = newSVsv (fh_or_path);
1270 req->offs = SvOK (offset) ? SvVAL64 (offset) : -1;
1271
1272 if (SvPOK (fh_or_path))
1273 {
1274 req->type = EIO_TRUNCATE;
1275 req->ptr1 = SvPVbyte_nolen (req->sv1);
1276 }
1277 else
1278 {
1279 req->type = EIO_FTRUNCATE;
1280 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1281 }
1282
1283 EIO_SEND;
1284}
1285
1286void
1287aio_chmod (SV8 *fh_or_path, int mode, SV *callback=&PL_sv_undef)
1288 ALIAS:
1289 aio_chmod = EIO_CHMOD
1290 aio_fchmod = EIO_FCHMOD
1291 aio_mkdir = EIO_MKDIR
1292 PPCODE:
1293{
1294 dREQ;
1295
1296 req->type = type;
1297 req->int2 = mode;
1298 req->sv1 = newSVsv (fh_or_path);
1299
1300 if (ix == EIO_FCHMOD)
1301 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1302 else
1303 req->ptr1 = SvPVbyte_nolen (req->sv1);
1304
1305 EIO_SEND;
1306}
1307
1308void
1309aio_chown (SV8 *fh_or_path, SV *uid, SV *gid, SV *callback=&PL_sv_undef)
1310 PPCODE:
1311{
1312 dREQ;
1313
1314 req->int2 = SvOK (uid) ? SvIV (uid) : -1;
1315 req->int3 = SvOK (gid) ? SvIV (gid) : -1;
1316 req->sv1 = newSVsv (fh_or_path);
1317
1318 if (SvPOK (fh_or_path))
1319 {
1320 req->type = EIO_CHOWN;
1321 req->ptr1 = SvPVbyte_nolen (req->sv1);
1322 }
1323 else
1324 {
1325 req->type = EIO_FCHOWN;
1326 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1327 }
1328
1329 EIO_SEND;
1330}
1331
1332void
1333aio_unlink (SV8 *pathname, SV *callback=&PL_sv_undef)
1334 ALIAS:
1335 aio_unlink = EIO_UNLINK
1336 aio_rmdir = EIO_RMDIR
1337 aio_readdir = EIO_READDIR
1338 PPCODE:
1339{
1340 dREQ;
1341
1342 req->type = ix;
1343 req->sv1 = newSVsv (pathname);
1344 req->ptr1 = SvPVbyte_nolen (req->sv1);
1345
1346 EIO_SEND;
1347}
1348
1349void
1350aio_link (SV8 *oldpath, SV8 *newpath, SV *callback=&PL_sv_undef)
1351 ALIAS:
1352 aio_link = EIO_LINK
1353 aio_symlink = EIO_SYMLINK
1354 aio_rename = EIO_RENAME
1355 PPCODE:
1356{
1357 dREQ;
1358
1359 req->type = ix;
1360 req->sv1 = newSVsv (oldpath);
1361 req->ptr1 = SvPVbyte_nolen (req->sv1);
1362 req->sv2 = newSVsv (newpath);
1363 req->ptr2 = SvPVbyte_nolen (req->sv2);
1364
1365 EIO_SEND;
1366}
1367
1368void
1369aio_mknod (SV8 *pathname, int mode, UV dev, SV *callback=&PL_sv_undef)
1370 PPCODE:
1371{
1372 dREQ;
1373
1374 req->type = EIO_MKNOD;
1375 req->sv1 = newSVsv (pathname);
1376 req->ptr1 = SvPVbyte_nolen (req->sv1);
1377 req->int2 = (mode_t)mode;
1378 req->offs = dev;
1379
1380 EIO_SEND;
1381}
1382
1383void
1384aio_busy (double delay, SV *callback=&PL_sv_undef)
1385 PPCODE:
1386{
1387 dREQ;
1388
1389 req->type = EIO_BUSY;
1390 req->nv1 = delay < 0. ? 0. : delay;
1391
1392 EIO_SEND;
1393}
1394
1395void
1396aio_group (SV *callback=&PL_sv_undef)
1397 PROTOTYPE: ;$
1398 PPCODE:
1399{
1400 dREQ;
1401
1402 req->type = EIO_GROUP;
1403
1404 req_send (req);
1405 XPUSHs (req_sv (req, AIO_GRP_KLASS));
1406}
1407
1408void
1409aio_nop (SV *callback=&PL_sv_undef)
1410 ALIAS:
1411 aio_nop = EIO_NOP
1412 aio_sync = EIO_SYNC
1413 PPCODE:
1414{
1415 dREQ;
1416
1417#endif
1418
1419void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2765eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1420{ 2766{
1421 grp->int2 = limit; 2767 grp->int2 = limit;
1422 grp->feed = feed; 2768 grp->feed = feed;
1423 2769
1424 grp_try_feed (grp); 2770 grp_try_feed (grp);
1425} 2771}
1426 2772
2773void
1427void eio_grp_limit (eio_req *grp, int limit) 2774eio_grp_limit (eio_req *grp, int limit)
1428{ 2775{
1429 grp->int2 = limit; 2776 grp->int2 = limit;
1430 2777
1431 grp_try_feed (grp); 2778 grp_try_feed (grp);
1432} 2779}
1433 2780
2781void
1434void eio_grp_add (eio_req *grp, eio_req *req) 2782eio_grp_add (eio_req *grp, eio_req *req)
1435{ 2783{
1436 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2784 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
2785
2786 grp->flags |= EIO_FLAG_GROUPADD;
1437 2787
1438 ++grp->size; 2788 ++grp->size;
1439 req->grp = grp; 2789 req->grp = grp;
1440 2790
1441 req->grp_prev = 0; 2791 req->grp_prev = 0;
… …
1445 grp->grp_first->grp_prev = req; 2795 grp->grp_first->grp_prev = req;
1446 2796
1447 grp->grp_first = req; 2797 grp->grp_first = req;
1448} 2798}
1449 2799
2800/*****************************************************************************/
2801/* misc garbage */
1450 2802
2803eio_ssize_t
2804eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2805{
2806 return eio__sendfile (ofd, ifd, offset, count);
2807}
2808

Diff Legend

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